Proceedings ECT 2009

ISSN 1822-5934
ELECTRICAL AND CONTROL

TECHNOLOGIES
PROCEEDINGS
OF THE 4 INTERNATIONAL CONFERENCE ON
ELECTRICAL AND CONTROL TECHNOLOGIES
TH
2009
KAUNAS
ISSN 1822-5934
ELECTRICAL AND CONTROL

TECHNOLOGIES
PROCEEDINGS
OF THE 4 INTERNATIONAL CONFERENCE ON
ELECTRICAL AND CONTROL TECHNOLOGIES
TH
Sponsored by:
Lithuanian State Science and Studies Foundation
ABB UAB
KAUNAS, TECHNOLOGIJA, 2009
UDK 621.3(474.5)(06); 681.5(474.5)(06)
The 4th International Conference on Electrical and Control Technologies

ECT-2009.
Selected papers of the International Conference, May 7-8, 2009, Kaunas, Lithuania
(Kaunas University of Technology), edited: A. Navickas (general editor), A. Sauhats,
A. Virbalis, M. Aubalis, V. Galvanauskas, A. Jonaitis
The 4th International Conference on Electrical and Control Technologies ECT-2009 is

the continuation of series of annual conferences on electrical, power engineering,
electrical drives and control technologies held earlier in Kaunas Polytechnic Institute
and from 1990 in Kaunas University of Technology.
The aim of the Conference is to create the forum for presentations and discussions to
consider the newest research and application results of investigations on electrical and
power engineering, electric drives, power converters, automation and control
technologies.
All papers were reviewed by Conference Editorial Committee and Conference
Program Advisory Committee and independent peer reviewers.
For information write to:

Kaunas University of Technology
Department of Electric Power Systems
Student g. 48, LT-51367 Kaunas
Lithuania
Kaunas University of Technology, 2009
Conference organizers
Kaunas University of Technology, Lithuania
IFAC Committee of National Lithuanian Organisation
Lithuanian Electricity Association, Full Member of EURELECTRIC
Conference Chairman
Prof. J. Daunoras, Dean of Kaunas University of Technology,
Electrical and Control Engineering faculty, Lithuania
Honorary Chairman
Prof. R. J. Kays, Vice Rector of Kaunas University of Technology, Lithuania
Conference Program Advisory Committee:
Prof. R. Deksnys, Kaunas University of Technology, Lithuania
Prof. L. Fara, Polytechnic University of Bucharest, Romania
Prof. J. Gerhards, Riga Technical University, Latvia
Prof. E. Heinemann, Schmalkalden University, Germany
Dr. B. Jaekel, Siemens AG, Germany
Prof. T. Jokinen, Helsinki Technical University, Finland
Prof. V. Kaminskas, Vytautas Magnus University, Lithuania
Prof. S. Kauinis , Kaunas University of Technology, Lithuania
Prof. J. Laugis, Tallinn University of Technology, Estonia
Assoc. Prof. V. Maerauskas, Kaunas University of Technology, Lithuania
Prof. L. A. Markeviius, Kaunas University of Technology, Lithuania
Prof. V. Mikinis, Lithuanian Energy Institute, Lithuania
Prof. A. Morkvnas, Kaunas University of Technology, Lithuania
Prof. A. Narglas, Kaunas University of Technology, Lithuania
Prof. A. Nemura, Lithuanian Energy Institute, Lithuania
V. Pakeviius, President of Lithuanian Electricity Association, Lithuania
Prof. L. Ribickis, Riga Technical Univesity, Latvia
Assoc. Prof. J. Sa da Costa, Technical University of Lisbon, Portugal
Prof. A. Sauhats, Riga Technical University, Latvia
Prof. R. Simutis, Kaunas University of Technology, Lithuania
Prof. A. W. Sowa, Bialystok Technical University, Poland
Prof. R. Strzelecki, Zielona Gora University, Poland
Prof. M. Valdma, Tallinn University of Technology, Estonia
Conference Editorial Committee:
General Editor: Prof. A. Navickas, Kaunas University of Technology, Lithuania
Prof. A. Sauhats, Riga Technical University, Latvia
Prof. A. Virbalis, Kaunas University of Technology, Lithuania
Assoc. Prof. V. Aubalis, Kaunas University of Technology, Lithuania
Assoc. Prof. V. Galvanauskas, Kaunas University of Technology, Lithuania
Assoc. Prof. A. Jonaitis, Kaunas University of Technology, Lithuania
Organizing committee:
Assoc. Prof. M. Aubalis, Kaunas University of Technology, Lithuania
Assoc. Prof. V. Galvanauskas, Kaunas University of Technology, Lithuania
Assoc. Prof. A. Jonaitis, Kaunas University of Technology, Lithuania
CONTENTS
A new approach based on support vector machine to analyse and control the REM-sleep-stage
Maria Krautwald, Rudolf Baumgart-Schmitt, Daniel Trommer, Christian Walther
11
University of Applied Sciences Schmalkalden, Faculty of Electrical Engineering, Germany
Adaption of fuzzy rules by different multi criteria optimization procedures to control stages of anaesthesia
Christian Walther, Rudolf Baumgart-Schmitt, Maria Krautwald, Daniel Trommer
15
Investigation of the influence of fuzzy set structure to the color segmentation

Toma Nakvosait, Vidas Raudonis, Rimvydas Simutis, Audrius Pukeviius
19
Department of Control Technologies, Kaunas University of Technology, Lithuania,

Department of Process Control, Kaunas University of Technology, Lithuania
Development and investigation of an efficient algorithm for alternative cursor control

Audrius Pukeviius, Vidas Raudonis, Rimvydas Simutis, Toma Nakvosait
23
Department of Control Technologies, Kaunas University of Technology, Lithuania

Optimal control of robots by dynamic programming

Rudolf Baumgart-Schmitt, Stephan Liebetrau, Christian Walther, Maria Krautwald, Daniel Trommer
27
Map drawing of unknown environment using autonomous mobile robot

arnas Germanas, Gintautas Narvydas
31
Department of Mathematics Research in System, Kaunas University of Technology, Lithuania

Influence of dimensions of a mobile robot on navigation using vector marks

Virginijus Baranauskas, Stanislovas Bartkeviius, Kastytis arkauskas
36
40
Optimization of hexapod robot locomotion

Tomas Luneckas, Dainius Udris
Vilnius Gediminas Technical University, Lithuania
Faster autonomous mobile robot learning using decision correction

Gintautas Narvydas, Petras Raanskas
44
Voice controlled telephony services

Rytis Maskelinas, Algimantas Rudionis, Kastytis Ratkeviius, Vytautas Rudionis
48
Kaunas University of Technology; Vilnius University, Kaunas Faculty, Lithuania
54
Data acquisition in OPC-based industrial it systems

Roman Kwiecie, Elbieta Szychta, Leszek Szychta
Technical University of Radom, Faculty of Transport and Electrical Engineering, Poland
Synthesis of program modules for typical boiler-house applications

Gediminas Danileviius, Darius Ezerskis
58
Distributed substation units control system in district heating network

Julius Kalueviius, Vytautas Deksnys, Virgilijus Pamaktis
62
Kaunas University of Technology; AB Axis Industries
Estimation of typical safety circuits for medium power industrial boilers

Algirdas ebatorius, Darius Ezerskis
66
ABB UAB, Lithuania; Kaunas University of Technology, Lithuania
Positioning of stepper motors in graphical data acquisition system

Raimundas Liutkeviius, Andrius Davidsonas
Department of Systems Analysis, Vytautas Magnus University, Lithuania
70
Visibility of text in multimedia presentation

Stanislovas Masiokas, Migl Kriuglait-Jarainien, Valdas Paknas
75
Kaunas University of Technology, Department of Electrical Engineering, Lithuania
Automatic PLC-source-code generation of C4.5 decision trees

Elmar Heinemann, Jolanta Repyt, Andr Wenzel
79
University of Applied Sciences Schmalkalden, Germany; Kaunas University of Technology, Lithuania
A new wireless solution to transmit and evaluate biomedical signals under low signal noise ratio to control
the state of relaxation
Daniel Trommer, Markus Greulich, Rudolf Baumgart-Schmitt, Maria Krautwald
Christian Walther, Andreas Wenzel
83

Fraunhofer-Institute for Information and Data Processing, Applications Center System Technology, Germany
Correction of measuring track processing errors in vibration measurements of coal wagon

Mirosaw Luft, Radosaw Cio, Daniel Pietruszczak
87
Kazimierz Puaski Technical University of Radom, Faculty of Transport and Electrical Engineering, Poland
Regression modelling using adaptive basis function construction

Gints Jekabsons
93
Institute of Applied Computer Systems, Riga Technical University, Latvia
Categorizing sequences of laryngeal data for decision support

Adas Gelinis, Evaldas Vaiciukynas, Edgaras Kelertas, Marija Bacauskiene,
Antanas Verikas, Virgilijus Uloza, Aurelija Vegiene
99
Kaunas University of Technology, Lithuania; Halmstad University, Sweden; Kaunas University of Medicine, Lithuania
103
System parameters identification in the state space

Leonas Balaeviius, Kstas Rimkus
Department of Control Technology, Kaunas University of Technology, Lithuania
107
Model of current space vector control

Robertas Janickas, Algirdas Smilgeviius
Parameter estimation of induction motor at locked rotor

Andrius Petrovas, Roma Rinkeviien, Saulius Lisauskas
111
115
Modelling of ammonia production process

Giedrius Oberauskas, Vytautas Galvanauskas
Process Control Department, Kaunas University of Technology, Lithuania
Mathematical model for adaptive control of dissolved oxygen concentration in biological wastewater
treatment process
Kstutis Jonelis, Kstutis Brazauskas, Donatas Leviauskas
121
Adaptive control system for dissolved oxygen concentration in Kaunas wastewater treatment plant
Jolanta Repyt, Ieva Cebatorien, Andr Wenzel
125
Kaunas University of Technology (KTU), Lithuania

University of Applied Sciences Schmalkalden (fhS), Germany
Analysis of efficiency of a pump set operating in an irrigation system

Elbieta Szychta, Leszek Szychta, Radosaw Figura
131
Different simulation models in exemplary employment for pump aggregate pipeline systems
Anna Koziorowska, Jacek Bartman
University of Rzeszw, Institute of Technology, Poland
135
Application of simulation tools for a fertilizer granulation circuit

Gediminas Valiulis, Rimvydas Simutis
139

Electrical Engineering Department, iauliai University, Lithuania
Innovative moisture control method for concrete mixing systems

Gintautas Lengvinas, Vytautas Deksnys
143
Department of Electronics and Measurement Systems, Kaunas University of Technology, Lithuania
The independent fuzzy control with correction of two dimensional process

Gediminas Liauius, Vytautas Kaminskas
146
Investigation of adaptive pH control system

Vytautas Galvanauskas
152
Building of securities valuation IT system using multi-agents approach

Andrius Jurgutis, Rimvydas Simutis
156

AB Bankas Finasta, Lithuania
Modelling of ATM cash turnover

Vaidas Kazlauskas, Vygandas Vaitkus
162
Scientific service company GTV, Kaunas University of Technology, Lithuania
Lithuanian energy after 2009

Arvydas Galinis
166
Lithuanian Energy Institute, Lithuania
Analysis of Latvian power system development in the Kurzeme region using multicriterion optimization
Anatoly Mahnitko, Alexander Gavrilov
170
Riga Technical University, Power Engineering Institute, Latvia
Balance management in the Baltic countries and possibilities for harmonization

Aura Parait, Mindaugas Krakauskas, Giedrius Radvila, Vidmantas Baltakis
176
Lithuanian Energy Institute; Lietuvos Energija AB;

Sprendim erdv, Lithuania
Integration of tracking standard into electricity markets of Baltic states

Inga Konstantinaviit, Dalius Tarvydas
182
Tendencies of nuclear fuel utilization modes

Gediminas Adlys, Alvydas Jotautis, Ramnas Naujokaitis Pranas Kanapeckas
186
Nuclear energy mix in the Baltic see region

Gediminas Adlys, Olegs Linkevis, Saulius Raila, Rta Adlyt, Helmut Alt
192
Kaunas University of Technology, Lithuania; Latvenergo, Latvia;

FH Aachen Fachbereich Elektrotechnik, Germany
Perspectives of nuclear cogeneration technologies

Gediminas Petrauskas, Gediminas Adlys, Diana Adlien, Kstutis Gediminskas, Algimantas Valineviius
196
Kaunas University of Technology, Lithuania;

Ignalina nuclear power plant, Lithuania
Allocating production cost of CHP plant to heat and power

Viktoria Neimane, Antans Sauhats, Guntars Vempers, Inga Tereskina
Vattenfall Research & Development, Sweden;
Riga Technical University, Latvia;
Riga Technical University, Latvia
JSC Siltumelektroprojekt, Latvia
200
Cross-border capacity auctions

Rimantas Deksnys, Robertas Staniulis, Giedrius Radvila
205
Kaunas University of Technology, Lithuania,

AB Lietuvos energija, Lithuania
The neural network as the inductive engines states analyser

Boguslaw Twarog, Zbigniew Gomolka
211
University of Rzeszow, Institute of Technology, Poland
Investigation of the induction motor start up process using star-delta start

Marina Konuhova, Karlis Ketners, Elena Ketnere, Svetlana Klujevska
215
Rational geometry of a magnetic circuit of axial inductor generator

Svetlana Orlova, Vladislav Pugachov, Nikolaj Levin, Edmunds Kamolins
219
Institute of Physical Energetics, Latvia;

The edge effects influence on the braking force of linear induction motor
Bronius Karalinas, Edvardas Matkeviius
223
227
Modeling of two-mass system with elasticity

Sigitas Juraitis, Roma Rinkeviien, Jonas Kriauinas
The means of electromagnetic flow meter magnetic circuit testing

Valdas Jonas Paknas, Juozapas Arvydas Virbalis
231
The attenuation of the electric reactor magnetic field strength

Jevgenij Morozionkov, Juozapas Arvydas Virbalis
235
Digitally selected speeds for a direct current (DC) motors applications

Sambwa Adoko, Onyekachi Adoko
239
Asa-Lambda Technology Ltd., Nigeria;

Benson Idahosa University, Nigeria
Intensification of heat exchange in refrigeration technologies by use of corona field

Stasys ebrauskas, Irmantas Beniulis
243
Kaunas University of Technology; Lithuania,

Lithuanian University of Agriculture, Lithuania
247
Using of corona fields in building material industries

Stasys ebrauskas, Riardas Bariauskas
A comparative study between small traditional induction motors and small conducting sleeve induction
motors using 3D finite element transient analysis
Seyed Mohsen Hosseini, Javad S. Moghani, Akram Hosseini
251
Amirkabir University of Technology, Iran
Graphical method of estimating optimum values for the main dimensions of a D.C. Machine
C. Easwarlal, V. Palanisamy, M.Y. Sanavullah
256
Sona College of Technology, India,

Principal Info Institute of Engineering, India
K.S.R. College of Technology, India
The research of causes and consequences of parameters changes of grounding electrical installations
Marina Kizhlo, Arvids Kanbergs
Riga Technical University, Institute of Power Engineering, Latvia
261
A simple transient-based wide-area selectivity technique for earth-faults in isolated and compensated
neutral medium voltage networks
Mohamed F. Abdel-Fattah, Matti Lehtonen
266
Helsinki University of Technology (TKK), Espoo, Finland
Sensors for on-line partial discharge detection in covered-conductor overhead distribution networks
G. Murtaza Hashmi, Matti Lehtonen, Mikael Nordman, Muzamir Esa
272
Power Systems and High Voltage Engineering, Helsinki University of Technology (TKK), Espoo, Finland
Ensto Utility Networks, Porvoo, Finland
278
Failures and diagnostics of power transformers

Sandra Vitolina, Janis Dirba
Influence of the distribution of the axial, radial and circumferential component of the magnetic induction
vector within the space between switches on rupturing capacity of vacuum switches
Bogdan Kwiatkowski, Robert Pkala
282
Rzeszow University , Technology and IT Institute; Rzeszow University, Technology and IT Institute, Poland
The research of the influence of the boundary conditions of the atmospheric overvoltages on the insulation
of high voltage equipment
Saulius Gudius, Linas Andronis Markeviius, Alfonsas Morkvnas
286
Insulation limit resistance performance assessment model

Alfonsas Morkvnas, Renata Stanionien, Povilas Valatka
290
The models of sources of electromagnetic interferences over UHV line for transient processes investigation
Linas A. Markeviius, Vytautas ioinys
294
Application of actual power deficiency value on load-shedding

Vladimir Chuvychin, Nikolay Gurov, Santa Kiene
298
Technical advance in the automation of distributed network control

Virginijus Radziukynas, Arturas Klementaviius, Sigitas Kadia, Algimantas Leonaviius, Neringa Rutkauskait,
Antanas Nemura, Feliksas Bielinskis, Viktorija Nezvanova
302
308
Distributed generators: definition identification

Dainius Slunys
New accurate fault location method for multi-terminal power lines

Przemyslaw Balcerek, Marek Fulczyk, Jan Izykowski, Eugeniusz Rosolowski, Murari Mohan Saha
312
ABB Corporate Research Center, Poland;

Wroclaw University of Technology, Poland;
ABB Power Technology Systems, Sweden
318
Transient analysis of variable speed wind turbine

Djilali Kairous, Bachir Belmadani, Mostapha Benghanem
Faculty of Science and Sciences Engineering, UHB, Chlef, Algeria, Dpt. of Electrotechnics, UST, Oran, Algeria
323
Selected high-frequency power supply systems

Elbieta Szychta, Kamil Kiraga
Technical University of Radom, Poland
Identification of parameters of synchronous generators dynamic models

Mindaugas Aubalis, Vaclovas Aubalis, Audrius Jonaitis, Ramunas Ponelis
Kaunas University of Technology, Lithuania;
Lietuvos energija AB, Lithuania
327
Potential of non-traditional energy sources for cold recovery in absorption refrigerating machines
Michail G. Berengarten, Semyon I. Vainshtein, Aleksandr S. Pushnov, Vytautas Adomaviius
333
Moscow State University of Ecological Engineering, Russian Federation

The Centre of Renewable Energy Technologies at KTU, Lithuania
Possible perspectives of cooling tower application for electric energy production

Aleksandr S. Pushnov, Michail G. Berengarten, Semyon I. Vainshtein, Vytautas Adomaviius
337

Application of renewable energy sources for feeding of water recycling systems technological equipment
Michail G. Berengarten, Semyon I. Vainshtein, Aleksandr S. Pushnov, Vytautas Adomaviius
343

The Centre of Renewable Energy Technologies at Kaunas University of Technology, Lithuania
Wind power plant with redox flow battery inverter system converter energy conversion processes
research
Povilas Balinas, Povilas Norkeviius
347
Kaunas University of Technology

Hybrid water heating system based on solar collectors and heat exchanger fed from the centralised
heat supply grid
Vytautas Adomaviius, Gytis Petrauskas
352
The Centre for Renewable Energy at KTU, Lithuania

Optimisation of hybrid water heating system based on solar collectors and heat exchanger in order to maximize
356
solar heat energys share
Vytautas Adomaviius, Gytis Petrauskas
Control of energy fluxes in the grid-tied domestic power system based on wind energy
eslovas Ramonas, Vytautas Adomaviius, Vytautas Kepalas
362

Research of the power conversion processes in the system of power supply from a number of wind
turbines over the one grid-tied inverter
eslovas Ramonas, Vytautas Adomaviius, Vytautas Kepalas
368

Comparison of small scale wind turbines properties

Vytautas Adomaviius, Thomas Watkowski, Edmundas ilinskas, Alfonsas Adomaviius
374
The Centre for Renewable Energy Technologies at Kaunas University of Technology, Lithuania
Lithuanian-German joint venture UAB Tometa, Germany
Joint Stock Company UAB Saules energija, Lithuania
Kaunas Business Machines Company, Lithuania
Peculiarities of wind power plants work in the small power system

Tomas Deveikis, Enrikas Nevardauskas, Renata Stanionien
10
380
The 4th International Conference on

Electrical and Control Technologies
May 7-8, 2009, Kaunas, Lithuania
A NEW APPROACH BASED ON SUPPORT VECTOR MACHINE TO ANALYSE

AND CONTROL THE REM-SLEEP-STAGE
Maria KRAUTWALD, Rudolf BAUMGART-SCHMITT, Daniel TROMMER, Christian WALTHER
Keywords: nonlinear classification, EEG, REM-sleepstage, support vector machine.
The conventional method of analyzing the sleep-stage is

to collect the EEG with a 2-channel ElectroEncephalograph at least and classify this EEG by an
expert manually. To support the scientists and to
disburden this procedure a mobile device with a onechannel EEG-receiver shall be developed for automatic
online classification.
A Support-Vector-Machine (SVM) approach shall be
used. This is a mathematic method to differ between
two or more classes. The SVM is especially suitable for
solving two-class-problems.
To get a mobile device it is necessary to implement the
SVM on a microcontroller. Therefore this paper should
give a perspective for realizing this problem.
Furthermore by knowing REM or Non-REM-stage it is
possible to control externally the stage by
administration of pharmaceutical or distortions of the
REM-sleep.
1. Introduction
2. Methods
The purpose of this paper is to show a solution for

analyzing the sleep-stages concerning the REM or NonREM-sleep-stages.
Besides the REM-sleep there are four other sleepstages. From Stage I, the drowsy sleep, to Stage IV, the
deep sleep, the depths of the sleep increase and in order
that the relaxation of the body increase too. Coevally
the threshold for awakening becomes more and more
high.
The REM-sleep, the dream sleep, has a special part
there. It gets this name because this stage is
characterized by rapid eye movements especially in the
early morning hours.
Currently there are only hypotheses concerning the
meaning of REM-sleep which is the aim of many
experimental studies. Searchers try to find this meaning
by analyzing the REM-sleep and coevally the memory
of the subject or even deprive REM-Sleep and observe
the consequences [1]. But there is still further need for
research.
The Support Vector Machine (SVM) is a mathematical

method to solve classification problems. For training a
SVM a set of training data are required [2]. In our cases
that means the training data consists of a set of EEGdata and the associated classification which was done
by an expert. With this training data it is possible to
define a decision function. To valuate the SVM it has to
be tested with testing data. The SVM calculates with the
EEG-data and the decision function the class. A
comparison of the artificial classification to the manual
classification of the expert is done afterwards. At the
end it is possible to estimate the performance of the
SVM and the decision function which is a result of the
training data. To demonstrate the functionality an easy
type of classification is shown in the following.
The easiest type is the separation of two classes in a two
dimensional decision space. The advantage of this
example is that it can easily visualize. So it is possible
to imagine the mathematical relations.
Abstract: The sleep stage of rapid-eye movement

(REM) should be recognized in real-time to evaluate the
current sleep quality. A robust recognition on the basis
of only one frontal electro-encephalogram (EEG)
channel is necessary to give the sleep stage researcher
the opportunity to control the following sleep-stages of
the patient.
A new approach based on a Support Vector Machine
was developed. We classify the REM-stage with
different data sets. An adapted kernel with RBF (radial
basis function) topology was applied. On the base of
these we try to implement the algorithm on a
microcontroller-platform. By this way a mobile wireless
device can diagnose the actual sleep stage and the
further sleep-profile can control by medical treatments.
11
Figure 1 shows a two dimensional room with a trainingdataset {xi,yi} illustrate by a set of white circles (class
1) and a set of black circles (class 2).
The distance d is defined as:
w
b
d H , x i yi
, xi
,
w
w
where ||w|| is the norm of the vector w which is defined

as:
d+
d-
H+
w, w .
(6)
For this reason:

H
d 1 w, x b
H-
class 1 (y = +1)
support vector
d 1 w, x b
Fig. 1. Two classes in a two dimensional decision space, with

class 1 as white circles, class 2 as black circles, H as the
hyperplane, M as the margin, H+ as positive shifted
hyperplane, H- as negative shifted hyperplane, d as the
distance from the hyperplane to a shifted hyperplane and the
grey circle as the support vectors
M d d
(1)
H 2 x | w, x b 1.
(4)
(9)
w, x i b 1 if yi 1
(10)
w, x i b 1 if yi 1 .
(11)
But usually the datasets arent completely separable like

it is shown in Figure 2. For this case parameter C, the
penalty factor, and , the slack variable, are established.
C controls the trade off between errors and margin
maximization. If C is too small there are too many
training errors and the SVM is under fitted, if C is too
large the SVM is over fitted, the hyperplane is to
complex, there are many support vectors and the margin
becomes small.
The slack variable is the distance of error vectors to
their correct position. The sum of i for i = 1,2,,n
should has its minimal value.
Now there are two restrictions: the margin and the
number of errors have to be minimized. Therefore the
goal of minimizing is modified to:
Of cause there exist infinite possibilities for such a

hyperplane. So it is necessary to define the restriction:
the margin M has to be maximized. The maximum
margin depends on the vectors next to the hyperplane.
These vectors are called support vectors (marked in
Figure 1). Thats the reason why this method is called
support vector machine.
Furthermore the data were scaled so that the support
vectors are in a distance of 1 to the hyperplane. That
means that yi are only +1 or -1. So the straight lines H1
and H2 are defined as:
(3)
w
2
.
, x x
w
w
and
(2)
H 1 x | w, x b 1
w
b
1 , (8)
, x
w
w
w
That means to maximize the margin requires

minimizing ||w|| which is equivalent to minimize ||w||2.
So the constrains are:
where w is the normal vector, which is orthogonal to the

hyberplane, b the displacement and <w,x> describe the
dot product.
It result the decision function:
f ( x new ) sign w, x new b .
(7)
the margin M result from:
The index i = 1,2,,n and describe a consecutive

numbering where n is the quantity of data pairs {x,y}.
For identifying a decision function a straight line (the so
called hyperplane) has to be placed between both sets.
The hyperplane is defined as:
H x | w, x b 0,
b
w
1
, x
w
w
w
and
class 2 (y = -1)
and
(5)
n
1 2
w Ci ,
2
i 1
12
(12)
with the constraints:
Ld ( ) i
w, x i b 1 i if yi 1
(13)
w, x i b 1 i if yi 1 .
(14)
i 1
1 n
i j yi y j x i , x j ,
2 i , j 1
(19)
with the constrains:
and
i 0
and
y
i 1
0.
(20, 21)
Then i can be calculated such that Ld is maximized.

The decision function can be written as:
5
f ( x new ) sign i yi x i , x new b .

i1
Thats the solution for a linear and separable problem.

But most of the problems in the nature are nonlinear. To
avoid the nonlinear problem there is the possibility to
go in a higher dimension where the problem becomes
linear. Figure 3 and 4 visualize this circumstance.
class 1 (y = +1)
(22)
support vector
class 2 (y = -1)
Fig. 2. Linear non-separable case of data classification, with

class 1 as white circles, class 2 as black circles, grey circles as
support vectors, i with i = 2, 5 as the distance of error vectors
to their correct position
An approach for solving this optimization problem is

the Lagrange-function:
L ( x, ) f ( x ) h ( x ) ,
(15)
where w and b have to minimize and have to

maximize ( min max L( w, b, ) ). For this purpose the
db
(17)
K ( x i , x j ) exp x i x j
i 1
and
dL p
dw
0 w i yi x i .
(23)
For the results a Raial Basis Function(RBF)-Kernel was

used.
0 i yi 0
Fig. 4. Simple classification

in a three dimension decision
space
f ( x new ) sign i yi K x i , x new b .

i 1
first derivation after w and after b is set equal to zero.

One obtains:
dL p
class 2 (y = -1)
In the two dimensional space the dot product are used.

In a higher dimension the dot product have to be
converted. That can be very complex or not possible. In
this situation the so called kernel trick afford an
answer. A kernel is a function which behave in a higher
dimension like the dot product in the two dimensional
space. The decision function has the form:
1 2 n
w i yi w, x i b 1 , (16)
2
i 1
w ,b
class 1 (y = +1)
class 2 (y = -1)
Fig. 3. Complex classification in a two dimensions

decision space
with optimization function f(x), constrain in 0-form h(x)

and Lagrange multiplier .
For separable case is the primal form of L:
L p ( w, b, )
class 1 (y = +1)
with the real valued parameter
(18)
(24)
0.
i 1
In the following the cycle of training and testing a SVM

is described for our sleep data set based on the approach
of [3]. At first 53 features are calculated with a rawEEG data and the associated classification. Then a file
with the relative time of an epoch, the 53 features and a
eight class classification is created. In the next step the
Inserted in Lp it follows the dual form:
13
eight-class-file is converting to a two-class-file with the

classes REM or Non-REM. Now the epochs were sorted
according to the manually classified classes. This
procedure is executed with the training- and the testing
data.
Of cause the Non-REM class is more dominant then the
REM class. Because of that the same of samples are
taken from each class in a randomized mode. Thats
avoiding one class proffered. Next step is the scaling of
the training data. After that the SVM try to optimize the
parameter C and by testing C and in a specific grid.
For testing the SVM the cross validation is applied.
That means all training datasets except one are used for
training and the remaining one for testing. This
procedure (one has to leave out) will be repeated for all
possible combinations of data.
If C and are defined the training of the SVM begins.
Subsequent to this also the testing data are scaled and
the SVM predict the classes of the test data and
compare the result with existing classification.
4. Conclusion
The results of the SVM have to observe critically
because of a possible preference of one class.
The SVM is suitable for a two class classification. Even
for classification of biological signals it can be used but
it doesnt achieve excellent percent values.
The suitability of the SVM concerning two-class
classification of biological signals was also shown
before with anaesthetic classification between awake not awake and stage A4 - not A4. The results were over
80% by using different datasets [4].
In multiclass-classification a SVM show similar results
compared to Fuzzy-Logic and neuronal networks [5].
Every learnt SVM only need one prediction algorithm
and appending upon Training-data different Input data.
That is the reason why a SVM could be very efficient
on a microcontroller platform.
5. References
1. Rasch B., Pommer J., Diekelmann S. & Born J.
Pharmacological REM sleep suppression
paradoxically improves rather than impairs skill
memory Nature Neuroscience advance online
publication Published online: 5 October 2008
doi:10.1038/nn.2206
2. Suykens J.A.K., Van Gestel T., De Brabanter J., De
Moor B. & Vandewalle J. Least squares support
vector machines World Scientific Publishing Co.
Pte. Ltd: 2002 ISBN 9812381511
3. Hsu C.-W., Chang C.-C. and Lin C.-J. A Practical
Guid to Support Vector Classification. Dec 20 2007.
www.csie.ntu.edu.tw/~cjlin
4. Walther C., Baumgart-Schmitt R. and Backhaus K.
Support Vector Machines and Optimized Neural
Networks Adaptive Tools for Monitoring and
Controlling the Depth of Anaesthesia. Proceedings
of International Conference Electrical and Control
Technologies 2008, Kaunas, 2008. p. 11-14.
5. Baumgart-Schmitt R., Walther C. and Backhaus K.
Design and Assessment of Fuzzy Rules by Multi
Criteria Optimization to Classifiy Anaesthetic
Stages. 4^th European Congress of the International
Federation for Medical and Biological Engineering,
Antwerp, Belgium, 2008.
3. Results
Because of the individualism of every human it is
difficult to classify EEG-signals on the basis of training
data from another patient.
In consideration of this the result of 83.07% correct
classification is satisfying.
The results have to observe critically. Often one class
was preferred. And even if the SVM can classify to
90% it is disuse because it only classified Non-REM
which was present to 90%. There are two possibilities
to avoid this situation. First is to test the SVM with
Non-REM and REM in a data record ratio 1:1. Another
is to adjust prediction of the SVM that both classes are
predicted percent similar.
The results were analyzed by calculating how much
percent of each class was classified correctly. Like it
shown in Table 1 class Non-REM were classified by
83.84 % and REM by 79.26 %. These values are nearly
similar. Then the SVM is suitable for online
classification.
Table 1. Results of classification by the Support Vector
Machine
testing-data
number of epochs
number of epochs in %
right classify epochs
right classify epochs in %
Non-REM
10793
83.23
9049
83.84
REM
2175
16.77
1724
79.26
For online classification at first all necessary data for

classification have to save on microcontroller. That
includes mainly the model file which is between 30kB
and 100kB. At second the EEG-data have to sample and
scale. Next 53 features need to calculate. Then the
prediction-program can make a decision. For real time
classification this procedure must take four milliseconds
at most.
14

ADAPTION OF FUZZY RULES BY DIFFERENT MULTI CRITERIA

OPTIMIZATION PROCEDURES TO CONTROL STAGES OF ANAESTHESIA
Christian WALTHER, Rudolf BAUMGART-SCHMITT, Maria KRAUTWALD, Daniel TROMMER
different stages of anaesthesia and one wake stage.

During the training procedure of the fuzzy sets
evolutionary operators and multi criteria optimization
algorithms were used. We optimized the fuzzy models
by three different multi objective evolutionary
algorithms (MOEA). The final result is represented by a
set of solutions in the multi criteria space called Pareto
set. The first implemented algorithm was the Nondominated Sorting Genetic Algorithm II (NSGAII)
developed by Deb et al. [7]. To test the hypothesis that
the results are depended on the used approach further
MOEAs were employed:
The Strength Pareto
Evolutionary Algorithm II (SPEAII) and the IndicatorBased Evolutionary Algorithm (IBEA) were included.
The last two evolutionary algorithms were developed by
Zitzler et al. [8, 9].
Abstract: Fuzzy rules were designed to classify four

different stages of anaesthesia and the wake stage
additionally. Feature sets of one frontal EEG channel
measured during operations in hospitals were used to
create and adapt fuzzy models. The topology and the
parameters of the models were optimized by
evolutionary algorithms based on three different multi
criteria optimization procedures: NSGA II, SPEA II and
IBEA. Typical fuzzy models were selected from the
amount of trained fuzzy rules contained in the resulting
Pareto set. The performance of the fuzzy models to
classify and control the depth of anaesthesia was tested
by independent data sets.
Keywords: Fuzzy modelling and control, multi criteria
optimization, evolutionary algorithms, anaesthesia,
frontal EEG.
2. Materials and Methods

1. Introduction
The frontal one-channel electroencephalogram from
different patients was recorded during the operation
procedure in cooperation with our hospital partners in
Schmalkalden and Zella-Mehlis (Germany). The
corresponding autonomous parameters like heart rate,
blood pressure or breathing information from the
patients were also saved. Based on this vital information
the anaesthesists were able to rate the whole data sets
and to divide the EEG into different epochs of stages of
anaesthesia. They distinguish between the stages "W wake", "A1 - light sedation", "A2 - chirurgical
tolerance", "A3 - deep anaesthesia" and "A4 - BurstSuppression" with A4 as the deepest stage of
anaesthesia. Every stage describes an EEG epoch with
the length of 10 seconds within the whole EEG
recording of a patient. We divided the rated EEG
recordings of 20 patients into two different data sets for
training and validation. Basically we used the rated
EEG data sets from 10 different patients for training
purposes (Table 1).
In [3] we have already presented an opportunity to
classify the frontal one-channel EEG by using fuzzy
models. Furthermore we used artificial neural networks
The analysis of the depth of anaesthesia is a challenge

for the anaesthesiologists. Because of the difficulties to
predict the exact patients reaction to medical drugs a
lot of experiences are necessary up to now. To get
information about the actual patient state they use vital
and breathing parameters of the patient. The whole
procedure of anaesthesia is very complex. Both the
consciousness and pain are gradually suppressed by
different kinds of medicinal drugs. The muscles are
relaxed too by this way to permit an optimal chirurgical
access to the operation site. The information from vital
parameters, breathing activity and dosages of the
administered medicinal drugs are used by the
anaesthesiologist to control the depth of anaesthesia of
the patient. Another way to get information about the
consciousness stages of the patient is to analyze the
electroencephalogram (EEG). But in this case the
anaesthesiologist has to be an expert in analyzing the
EEG in real-time. To support the anaesthesiologist we
developed algorithms to calculate the depth of
anaesthesia based on the frontal one-channel EEG
[1,2,3]. Different fuzzy rules are used to classify four
15
Table 1. General information about the data sets recorded

from different patients used in the training set
and support vector machines to solve this nonlinear

classification problem [1,2]. To increase the
performance of the selected fuzzy model we introduced
more than one fuzzy rule for one class. The number of
fuzzy rules which can be used for classification is
theoretically unrestricted. Up to now one rule per class
was used [3]. To improve the performance of the
classifier on the one hand and to limit the computing
time on the other hand the number of fuzzy rules were
restricted to three per class in this case.
To optimize the structure and the parameters of the
fuzzy model by the evolutionary procedure we tested
three different MOEAs: the NSGAII, SPEAII and
IBEA. The implemented algorithms are described in
[4,6,7,8] and [9]. All three MOEAs have in common an
internal elitist population of individuals which is
updated throughout the optimization process. In our
case one individual is represented by a certain fuzzy
model, which will be optimized. A set of individuals
form a population.
The multi criteria optimization procedures use four
different quality values to decide if a fuzzy model is
able to classify the five anaesthetic stages correctly and
if it is working effectively. The first quality value q1
evaluate the number of epochs of EEG in the training
set which were correctly classified by the fuzzy model.
The second goal function q2 is defined by a mean value
of correctly classified epochs for each class of
anaesthesia separately. The third criterion q3 evaluates
the worst rated class by the fuzzy model. The last
quality criterion q4 is a value of effectiveness. It
represents the counted activated fuzzy sets within the
fuzzy model as a value in percentage. If this value is
high, the fuzzy model is using a lower number of fuzzy
sets for classification.
At the beginning of the optimization process a
population is randomly initialized (fig. 1). In a first step,
after the calculation of the criteria for every individual
of the population, the population is sorted. For every
individual a value of density is calculated within the
MOEA. This value represents the position of the
individual as an element inside the Pareto set and
describes the relationship to the other individuals of the
population. In a last step the best individuals are
selected out of the population to create an updated elite
pool, which is saved for the next iteration. Apart from
this elite population a further child population is
generated by the genetic operators recombination and
mutation. This child population forms the new
population of individuals for the next iteration.
The MOEAs differs in internal sorting procedures and
calculation of the density values. These algorithms were
implemented to find and select the best fuzzy models
for the genetic operators to generate the next generation
of new fuzzy model populations. When a new fuzzy
model is generated by the recombination process it
could be randomly influenced by a mutation operator
(fig. 1). The two genetic operators are responsible for
the inclusion of single fuzzy sets, which belongs to the
resulting fuzzy model.
Data Set
Gender
Age
Epochs
0066
0007
0006
0010
0011
0009
0072
0012
0021
0024
F
F
F
F
F
F
F
F
F
F
52
62
63
45
66
36
75
37
42
41
1180
1438
449
1286
947
355
1828
614
1068
995
Fig. 1. General approach to generate and optimize fuzzy

models, which are able to classify anaesthetic stages. Multiobjective evolutionary algorithms are used for optimization.
A population consists of different individuals. One individual
represents one adapted fuzzy model
Table 2. General information about the data sets recorded
from different patients used in the validation set. The capital F
means female and M means male
16
Data Set
Gender
Age
Number of
Epochs
0121
0118
0117
0116
0125
0127
0126
0134
0135
0128
F
F
M
M
F
F
F
F
M
M
73
77
47
65
84
24
45
45
67
55
1745
385
546
371
399
440
283
299
467
1389
Table 3. Used selection weight modes applied to the criteria

of every single individual to extract a subset of individuals
from the Pareto-optimal front. The criteria are multiplied by
the presented weight factors. The sum of these weighted
single criteria forms the applied selection criterion
The generally used structure of membership function

(x,a,b) for each fuzzy set is:
(x,a,b) = e[-0.5 ((x-a)/b)] ,
(1)
with x as normalized input value and a and b as real

parameters. The two parameters a and b are changed by
the mutation operator in a defined range.
To end up the whole optimization procedure a stopping
criterion is needed. For that purpose a subset of EEG
epochs from the training set was extruded to check the
performance of the actual population of fuzzy models.
To calculate the stopping criteria every fuzzy model of
the elitist population has to classify this EEG subset.
Furthermore the median on all resulting classifications
is used to represent the performance of all fuzzy models
together. If the quality of the Pareto set significantly
decreases in the actual population, the optimization
procedure is stopped and the elitist population of one
generation before is saved as the final result.
To validate the performance of the trained fuzzy models
a validation set of 10 other patient EEG recordings was
used (table 2). These data sets were not used to train the
fuzzy models.
Criteria
q1
q2
q3
q4
Table
Weight
mode Aw
Weight
mode Bw
0.4
0.3
0.3
0.0
4,6,7,8
0.2
0.4
0.4
0.0
Table 4. Validation results of the fuzzy models, which were

optimized using the non-dominated sorting genetic algorithm
II and selected from the Pareto-front using weight mode Aw.
Each row represents the classification estimation given by an
anaesthesiologist for all epochs of this class. Each column
presents the degrees of concordance of the automatic
classification done by the selected fuzzy models with the
manually rated epochs
NSGAII
W (E)
A1
(E)
A2
(E)
A3
(E)
A4
(E)
3. Results
To get the final results, which are represented in the
tables 4, 5, 6 and 7 a sub set of fuzzy models was
extracted from the Pareto-optimal front for every
MOEA-Run. At the end of a multi-criteria optimization
process a suitable number of solutions have to be
selected from the Pareto set. In our case a general
selection criterion was defined by the sum of the
weighted criteria q1 to q4 for every individual of the
Pareto-optimal front. If this selection criterion has a
high value the evaluated individual implies a high
performance within the search space. The used weight
modes Aw and Bw with different weight vectors are
presented in table 3.
After the selection of the individuals, these different
optimized fuzzy models were used to classify the
validation data (table 2). To explain the problem of
finding the best solutions using the two different weight
modes Aw and Bw the validation results of the respective
individuals are presented in table 4 and 5. Both sets of
individuals were trained in the optimization process
using the NSGAII algorithm. The fuzzy models used to
calculate the solutions presented in table 5 show a
slightly better performance in comparison to the
solutions of table 4. The performance is measured by
the degrees of concordance between the automatic
classification in each column and the given ratings of
the anaesthesiologists in each row. Beside the diagonal
fields there are also some grey highlighted rectangles in
the tables 4, 5, 6 and 7. This means that confusions
indicated by these values inside the rectangles are
negligible. The fuzzy models which have performed the
results of table 5 were only found by using the weight
mode Bw.
W
19.8
3.2
A1
47.8
43.6
A2
9.9
16.5
A3
2.7
21.4
A4
19.8
15.3
0.1
4.9
32.7
60.4
1.7
0.0
0.3
14.3
82.4
3.0
0.0
0.8
5.7
31.3
62.1

optimized using the Non-dominated Sorting Genetic
Algorithm II and selected from the Pareto-front using weight
mode Bw
NSGAII
W (E)
A1
(E)
A2
(E)
A3
(E)
A4
(E)
W
21.6
5.2
A1
44.1
42.3
A2
8.1
17.4
A3
1.8
13.3
A4
24.3
21.8
0.1
5.4
35.4
57.7
1.5
0.0
0.3
16.7
78.5
1.5
0.0
1.4
4.3
27.4
66.9

optimized using the Strength Pareto Evolutionary Algorithm
II and selected from the Pareto-front using weight mode Aw
SPEAII
W (E)
A1
(E)
A2
(E)
A3
(E)
A4
(E)
17
W
27.9
5.24
A1
43.2
44.8
A2
4.5
18.2
A3
7.2
16.1
A4
17.1
15.7
0.6
4.9
27.6
63.3
3.5
0.1
0.9
8.4
86.6
4.0
0.3
2.0
7.6
30.8
59.4

optimized using the Indicator-Based Evolutionary Algorithm
and selected from the Pareto-front using weight mode Aw
IBEA
W (E)
A1
(E)
A2
(E)
A3
(E)
A4
(E)
5. Conclusions
W
63.1
10.5
A1
10.8
41.5
A2
0.0
17.3
A3
2.7
5.7
A4
23.4
25.0
0.0
4.4
43.1
46.5
6.0
0.0
1.8
20.6
70.9
6.8
The performance of the fuzzy models measured by the

validation set is mainly dependent on the selection
criterion. All members of the Pareto set represent a
knowledge base, which was optimized during the multiobjective evolutionary optimization process.
This knowledge base established during training is used
to generalize. By means of the weight vectors the
generalizing can be controlled in a final step.
0.1
6.0
11.9
17.2
64.8
6. References
1. Walther C., Baumgart-Schmitt R. and Backhaus K.
Support Vector Machines and Optimized Neural
Networks Adaptive Tools for Monitoring and
Controlling the Depth of Anaesthesia. Proceedings
of International Conference Electrical and Control
2. Baumgart-Schmitt R., Walther C., Backhaus K.,
Reichenbach R., Sturm K.-P. and Jger U. Robust
Nonlinear Adaptive Network Classification of
Anaesthesia. 2008 IAPR Workshop on Cognitive
Information Processing, Santorini, Greece, 2008.
3. Baumgart-Schmitt R., Walther C. and Backhaus K.
Design and Assessment of Fuzzy Rules by Multi
Criteria Optimization to Classifiy Anaesthetic
Stages. 4th European Congress of the International
Federation for Medical and Biological Engineering,
Antwerp, Belgium, 2008.
4. Hoffmann
T.
Erprobung
von
Methoden
mehrkriterieller Optimierung zur nichtlinearen
Klassifikation am Beispiel von simulierten
Evolutionsstrategien. Diploma thesis. University of
Applied Sciences Schmalkalden, Germany, 2009.
5. Kim M.-S., Kim C.-H. and Lee J.-J.
Structure/Parameter Optimization of Fuzzy Models
by Evolutionary Algorithm. Proceedings of the
IASTED International Conference on Intelligent
Systems and Control, Salzburg, Austria, 2003.
6. Laabidi K., Bouani F. and Ksouri M. Multi-criteria
optimization in nonlinear predictive control.
Mathematics and Computers in Simulation, Volume
76 , Issue 5-6, 2008. p. 363-374.
7. Deb K., Pratap A., Agarwal S. and Meyarivan T. A
Fast and Elitist Multiobjective Genetic Algorithm:
NSGA-II. IEEE Transactions on Evolutionary
Computation, Volume 6, Issue 2, 2002. p. 182-197.
8. Zitzler E., Laumanns, M. and Thiele L. SPEA2:
Improving the Strength Pareto Evolutionary
Algorithm. TIK-Report 103, Swiss Federal Institute
of Technology Zurich, Switzerland, 2001.
9. Zitzler E. and Knzli S. Indicator-Based Selection in
Multiobjective Search. Proceedings of the 8th
International Conference on Parallel Problem
Solving from Nature (PPSN VIII), Birmingham,
UK, 2004. p. 832-842.
The performances of the generated fuzzy models, which

were optimized using the evolutionary algorithms
SPEAII and IBEA are presented in the tables 6 and 7. In
table 8 the degrees of concordance are additionally
presented for a two class problem, based on the
distinction between the deepest narcotic stage A4 to all
other stages.
optimized using the Non-dominated Sorting Genetic
Algorithm II and selected from the Pareto-front using weight
mode Aw. The fuzzy models were trained to solve a two-class
problem to distinguish between the set of the stages W, A1,
A2, A3 and the stage A4
NSGAII
W,A1,A2,A3 (E)
A4 (E)
W,A1,A2,A3
89.4
31.6
A4
10.6
68.4
4. Discussion
Despite using different multi criteria procedures the
results differ only slightly in classification of the
validation data.
Differences in the degrees of concordance obviously
only depend on the chosen individuals and the used
selection criteria, which were defined by a weight
mode. Mainly the three multi-objective evolutionary
algorithms differ in sorting the individuals and
calculating the density value for each fuzzy model. This
causes a different speed of the optimization process
itself for each algorithm. In every case a population of
optimized individuals is generated as a result.
Measurements of the optimization speed of such
algorithms (NSGAII, SPEAII and IBEA) were not
analyzed within this work. To estimate the efficiency of
the multi criteria optimization procedures defined test
functions were employed by coworker Hoffmann [4].
The tables 4, 5, 6 and 7 contain higher degrees of
confusions of the stage W and A1 to the stage A4.
These values are a result of the higher dimensionality of
the fuzzy models in the training procedure in
comparison to [3]. It is possible to distinguish correctly
between Wake, A1 and A4, which was shown in [1] and
[3]. This is underlined by the results shown in Table 8.
18

INVESTIGATION OF THE INFLUENCE OF FUZZY SET STRUCTURE TO THE

COLOR SEGMENTATION
Toma NAKVOSAIT*, Vidas RAUDONIS**, Rimvydas SIMUTIS**, Audrius PUKEVIIUS*
*Department of Control Technologies, Kaunas University of Technology, Lithuania,
**
proposed. FCH considers the color similarity

information by spreading each pixels total membership
value to all the histogram bins. Other approaches
measures fuzzy homogeneity in path of image pixels
[7], where resemblance between color features is
calculated on the basis of distances in the HSI color
space.
In this work a potential of using fuzzy sets for color
segmentation in the image is investigated. We use an
Artificial Neuro-Fuzzy Inference System (ANFIS),
which requires to be trained. Pixel values of an image in
RGB color space was used as an input for the developed
algorithm and black - white image is obtained as an
output. ANFIS allows choosing different structures of
fuzzy sets (quantity and shape of membership
functions). Our task in this work was to find the most
efficient fuzzy set structure, which can be used in realtime eye position estimation.
This paper is organized in 4 sections. Basic idea of
ANFIS algorithm is presented in the second section,
investigation algorithm is presented in the third, more
detailed description of the experiments and their results
are presented in the fourth.
Abstract: An Adaptive Neuro-Fuzzy Inference System

(ANFIS) with a hybrid learning algorithm is used for
color segmentation in this work. An algorithm for
testing different fuzzy set structures was composed and
it was used to extract the iris of a human eye in a color
image. An efficiency of each tested fuzzy set structure
was evaluated according to its applicability to an eye
position tracking system. A desirable precision of
extraction was reached with the simplest ANFIS
structure possible. Image processing rate depends on the
structure of the fuzzy set.
Keywords: Adaptive Neuro-Fuzzy Inference System,
color segmentation, fuzzy set.
1. Introduction
Over the last decade there has been growing interest
within the computer vision and pattern recognition
communities in the problem of analyzing the human eye
behaviour in the video. A real-time eye tracking
requires a method and algorithm, which allows effective
and fast determination of a human eye position in to two
dimensional data space. The position of the eye pupil
can be tracked using various techniques based on the
measurements of the potential differences which appear
between electrodes mounted around the user eye or
using video cameras. One of the relatively cheapest
ways to gather information for this purpose is to use a
web camera, so this leaves us with analysis of color
images of an eye.
Color segmentation of an image for feature extraction is
researched and applied widely. Many segmentation
methods and algorithms are developed and tested, such
as brightness invariant segmentation using B-spline
curve fitting [1], context-dependent segmentation [2],
segmentation using luminance histograms [3], spatial
color histograms [4]. However most of the proposed
methods are suitable only for particular application.
Several approaches of color segmentation based on the
fuzzy logic are proposed in [5, 6, 8] instead of
traditional CCHs, Fuzzy Color Histogram (FCH) is
2. Theoretical background
General fuzzy inference is used in systems whose rule
structures are predetermined by the user and
membership functions are fixed and arbitrarily chosen.
ANFIS allows generating fuzzy inference system
automatically, without having predetermined structure
based on characteristics of variables in the system.
Membership function parameters are set during learning
process of the inference system. Adaptive network
based fuzzy inference systems (ANFIS) is a FIS
implemented in the framework of an adaptive fuzzy
neural network. Neuro-adaptive learning techniques
provide a method for the fuzzy modelling procedure to
learn information about a data set, in order to compute
the membership function parameters that best allow the
associated fuzzy inference system to track the given
input/output data. ANFIS constructs a fuzzy inference
19
system, (FIS) whose membership function parameters

are adjusted using either a backpropagation algorithm
alone, or in combination with a least squares type of
method (used in this work). The computation of
parameters associated with membership functions is
facilitated by a gradient vector, which provides a
measure of how well the fuzzy inference system is
modelling the input/output data for a given set of
parameters. Once the gradient vector is obtained, any of
several optimization routines could be applied in order
to reduce some error measure.
ANFIS only supports Sugeno-type systems that follow
requirements:
Be first or zeroth order Sugeno-type systems.
Have a single output, obtained using weighted
average defuzzification.
Have no rule sharing. Different rules cannot share the
same output membership function.
Have unity weight for each rule. A typical rule in a
Sugeno fuzzy model has the form
If the system with two inputs and one output is used,
where first input is equal to x and second is equal to y,
then output can be written as:
the iris usually differs from its surroundings. Images in

RGB color space are used as ANFIS learning data. Data
sample for learning all FIS structures is selected
manually from one of the images.
After ANFIS trains certain FIS structures, segmentation
of experimental data is performed. Output of ANFIS is
an intensity image, where pixel values closest to 1
belong to the region of the iris. Areas that are not likely
to belong to the identified iris are filtered.
Start
RGB
images
of an eye
ANFIS learning
Segment
images?
No
Yes
Image
segmentation
(1)
z = ax + by + c
Selection of the
requisite segment
For a zero-order Sugeno model, the output level z is a

constant (a=b =0).
The output level zi of each rule is weighted by the firing
strength wi of the rule. For example, for an AND rule
with Input 1 = x and Input 2 = y, the firing strength is
wi = AndMethod (F1(x), F2(y)),
where F1,2 are the membership functions for Inputs 1
and 2. An AND operation of 2 triangular fuzzy sets A
and B is shown in Fig. 1. The result of intersection is a
fuzzy set A and B.
1
ANFIS
learning
data
selection
Noise reduction
Data for
analysis
Finish
Fig. 2. Algorithm of experiments
4. Results of experiments
0
1
Experimental data consists of RGB images, gathered

with ordinary web-camera. Size of the gathered and
later processed frame is 320x240 pixels.
18 different fuzzy set structures trained by ANFIS
algorithm were tested. Structures differ by number of
membership functions for each input, their shape and
number of training iterations (epochs). All inference
systems are trained using the same learning data sample
from random RGB image. Learning data consists of two
manually selected regions: one from the iris region
(1144 pixels), another from its surroundings (1666
pixels). Notifiable that effectiveness of learning FIS and
segmentation depends on color composition of the
selected regions. In example, the sample that represents
the iris should have all its colors proportionally to the
entire iris.
Experiments with each FIS were performed using 200
images of an eye that is looking straight to the camera,
up, down, right and left (40 frames each).
A and B
Fig. 1. The AND operation of two membership functions
The final output of the system is the weighted average

of all rule outputs, computed as
N
w z
i i
FinalOutput
i 1
N
(2)
i 1
3. Segmentation algorithm
An algorithm for image segmentation using different
trained fuzzy set structures is presented in Fig. 2. It
detects iris of a human eye. The color composition of
20
Fig. 3. The data base of the images with human eye positioned
in five different positions
As an input of trained FIS are red, green and blue

values of image pixels. Fuzzy inference system gives
segmented intensity image of an eye (Fig. 4, upper
right). The brightest region is selected according to
intensity threshold equal to 0.7 (Fig. 4, lower left, colors
in Fig. 4 upper right, lower left and right are inverted).
Then small areas that dont belong to the iris are cleared
(Fig. 4, lower right). After iris of an eye is extracted
from the image, the center coordinates of it are
calculated.
Fig. 6. Calculated centers of the iris when FIS has 3

triangular-shaped MFs for each input, trained in 100 epochs
Fig. 7. FIS structure when each input has 3 membership

functions
Fig. 4. Processing of an image when FIS has 8 fuzzy rules,

trapezoidal-shaped MFs, trained in 20 epochs
Training time increases significantly when ANFIS

trains 3 MFs for each input instead of 2 (Table 1).
Table 1. Training and processing time
Structur
e No
Learnin
g epochs
Number
of MFs
MF shape
Training
time, s
Processin
g speed,
fps
1
2
3
20
Triangular
Trapezoidal
Gaussian
curve
Triangular
Trapezoidal
Gaussian
curve
Triangular
Trapezoidal
Gaussian
curve
Triangular
Trapezoidal
Gaussian
curve
Triangular
Trapezoidal
Gaussian
curve
Triangular
Trapezoidal
Gaussian
curve
1.49
1.44
8.14
8.27
1.53
10.78
10.83
7.28
5.09
5.26
10.96
3.33
3.51
4.65
8.17
8.26
3.72
26.71
26.96
7.23
5.10
5.32
27.66
6.79
6.71
4.74
8.23
8.35
7.49
54.05
54.14
7.30
5.02
5.19
55.50
4.60
4
5
6
7
8
9
Fig. 5. Calculated centers of the iris when FIS has 2 Gaussian

curve-shaped MFs for each input, trained in 20 epochs
50
10
11
12
The results of segmentation with all FIS structures are

similar. An algorithm finds the iris in all of the
processed images. Centers of the extracted regions are
grouped into 5 apparent clusters (Fig. 5, 6), which
represent 5 different eye positions. In these figures
center points that represent eye position up are lower
than centers of eye looking down, because y coordinates
of image pixels starts from 1 at the top of the image
(reverse y axis).
13
14
15
16
17
18
21
100
Therefore structures with 4 or more MFs per input

werent investigated. Calculated images processing
speed also depends on shape of input membership
functions. Using trapezoidal MFs caused a bit faster
processing than using triangular shaped. Gaussian
curve-shaped MFs processing is much slower, than the
rest two shapes investigated. All investigated fuzzy set
structures are listed in Table 1. They differ by number
of training iterations (learning epochs), number of
membership functions per input, and shape of
membership functions. Variations of center coordinates
are shown in Fig. 8 and 9. Numbering of each structure
(1-18) is according to structure numbering in Table 1.
Greatest variations calculated while processing images
with eye directed straight and using Gaussian curve
shaped membership functions.
showed that segmentation quality doesnt depend on

FIS structure and number of training iterations.
Desirable results were reached with the simplest FIS
structure (with 2 triangular shaped membership
functions). Training time increases significantly when
number of membership functions is increased by 1.
Images processing speed is greater with triangular and
trapezoidal membership functions up to 8.35 frames
per second. The results of this paper can be used for the
development of efficient associative communication
systems, human face tracking and systems which
analyzes medical images. The future work will
incorporate the development of the position classifier
and applications of the external device control.
6. Acknowledgment
The authors appreciate the financial support of the
Lithuanian state science and studies foundation.
7. References
1. Chi-Ho Kim, Bum-Jae You, Hagbae Kim, Color
Segmentation Robust to Brightness Variations by
Using
B-spline
Curve
Modeling,
Intelligent Robots and Systems, 2006 IEEE/RSJ
International Conference, 2006, p. 4874 4879.
2. Palma-Amestoy, Rodrigo A., Guerrero, Pablo A.,
Vallejos P.A., Ruiz-del-Solar J., Context dependent color segmentation for Aibo robots,
Robotics Symposium, 2006. LARS '06. IEEE 3rd
Latin American, 2006, p. 128 136.
3. Albiol A.; Torres L.; Delp E.J., An unsupervised
color image segmentation algorithm for face
detection applications, Image Processing, 2001.
Proceedings. 2001 International Conference,
Volume 2, 2001, p. 681 - 684 vol.2.
4. Aibing Rao; Srihari, R.K.; Zhongfei Zhang, Spatial
color histograms for content-based image retrieval,
Tools
with
Artificial
Intelligence,
1999.
Proceedings. 11th IEEE International Conference,
1999, p.183 186.
5. Ju Han, Kai-Kuang Ma, Fuzzy color histogram and
its use in color image retrieval, Image Processing,
IEEE Transactions on Volume 11, Issue 8, 2002, p.
944 952.
6. Yongzhong Wang, Quan Pan, Chunhui Zhao,
Yongmei Cheng, Fuzzy Color Histogram Based
Kernel Tracking under Varying Illumination, Fuzzy
Systems and Knowledge Discovery, 2007. FSKD
2007. Fourth International Conference on Volume 4,
2007, p. 235 239.
7. Chamorro-Martinez J., Sanchez D., PradosSuarez B., Galan-Perales E. Fuzzy Homogeneity
Measures
for
Path-Based
Color
Image
Segmentation, Fuzzy Systems, 2005. FUZZ '05. The
14th IEEE International Conference, 2005, p. 218
223.
8. El-Feghi I., Aboasha H., Sid-Ahmed M.A.,
Ahmadi M, Content-Based Image Retrieval based
on efficient fuzzy color signature, Systems, Man and
Cybernetics, 2007. ISIC. IEEE International
Conference, 2007, p. 1118 1124.
Fig. 8. Variation of x coordinates of the iris center
Fig. 9. Variation of y coordinates of the iris center
5. Conclusions and discussions

An algorithm for FIS structure evaluation is presented.
FIS structures are trained automatically by Adaptive
Neuro-Fuzzy Inference System (ANFIS), and it requires
an expert to select training data. Results of experiments
22

DEVELOPMENT AND INVESTIGATION OF AN EFFICIENT ALGORITHM FOR

ALTERNATIVE CURSOR CONTROL
Audrius PUKEVIIUS*, Vidas RAUDONIS**, Rimvydas SIMUTIS**, Toma NAKVOSAIT*
*
Department of Control Technologies, Kaunas University of Technology, Lithuania
**
in visual information processing area. They are

successfully used in road signs recognition [3], [4],
human face recognition [5], [6]. It is possible to
combine neural networks with other methods [7] an in
result to get an effective hybrid system for skin lesion
detection.
In Neuro-Gaussian algorithm statistical Gaussian
method is used to form a training material for neural
network. With Gaussian mixture model color
segmentation is applied to static background.
Mahanalobis distance is calculated between the pixels
of the new picture with users hand and backgrounds
regions. The pixels of the hand region do not combine
with any earlier segmented regions and are classified as
object pixels, which is above background. Pixels of
hand region and some of background pixels are used to
form training data for neural network. This algorithm
helps to avoid an expert intervention in system
calibration process. In further process pixels are
classified with trained neural network, because it
performs faster and more accurate then Gaussian
method.
The paper is organized in four main sections. In the
second section is overviewed theoretical background of
Neuro-Gaussian algorithm. Third section gives more
information about hand tracking system, an algorithm of
hand position detection and cursor control.
Experimental results are represented in section 4.
Abstract: The development and investigation of

alternative computer mouse control system is presented
in this work. The system uses 2D visual information,
which is acquired from ordinary web-cam, and controls
cursor position in the monitor by tracking the
movements of the hand. Hand tracking is achieved by
using an algorithm, witch combines two methods:
statistical Gaussian method and artificial neural
networks. This algorithm helps to avoid an expert
intervention in system calibration process. Before
starting to control cursor, the user have to put hand in
front of web-cam and wait a few moments while system
calibrates it self.
Keywords: object tracking, neural networks, visionbased human-computer interface.
1. Introduction
Nowadays cursor control in computer monitor is
realized in various ways: computer mouse, touch-pad,
touch-screen. An alternative algorithm is represented in
this work, witch uses 2D visual information from
ordinary web-camera. With this algorithm it is possible
to control cursor position by moving hand, placed in
front of camera. In this case user has an opportunity to
control cursor position without any physical contact
with cursor control devices.
Cursor is realized with Neuro-Gaussian algorithm,
witch combines statistical Gaussian mixture model [1]
with neural networks. There is known other statistical
methods for background subtraction by using color
segmentation, such as mean, median background
models [2]. It is known that for making such models it
is necessary to use more than one frame with static
background picture. It is important to choose right
values for pixel classification constants in order to gain
accurate background extraction. Gaussian mixture
model can be made from one frame with static
background. Statistical Gaussian method is combined
with neural networks, which nowadays are widely used
2. Theoretical background
Background subtraction is widely used to detect moving
object from static cameras. It is usually regard as one of
the most important step in applications such as traffic
monitoring, human motion capture, recognition and
video surveillance, etc. In order to get a good
performance of the whole system, the background
subtraction method should not be so time and space
consuming, and the accuracy is also required. Gaussian
mixture model is a robust background subtraction
method and is widely used ever since it is proposed.
23
Gaussian mixture model has parameters, which

describes backgrounds color features. Static background
picture is clustered in to chosen number of familiar
color regions. Every region is parameterized like a
Gaussian distribution and described with its pixels mean
value i and covariance matrix Ki in three color spaces
(c = (r, g, b), v = 3). Gaussian distribution is represented
in (1) expression.
pi (c)
1
(2)v/ 2 detKi
(1)
1
exp( (c i )T Ki1(c i ))
2
Mahanalobis distance between new pictures pixel and ith region is calculated as follows in (2) expression:
M (c, i) (c i ) T K i1 (c i )
Fig. 2. Cursor control systems web-cam position: webcams capturing region horizontal to xy plane; 1 vertically
down directed web-cam; 2 users hand. Hand movement
directions are horizontal to plane
(2)
Pixels of the new picture are classified by using

Gaussian mixture model and classification is realized
with (3) expression:
1 if M ( x, y, t )
M( x, y, t )
0 otherwise
It is taken in account that the users hand movements

are parallel to plane. Visual information and
calculations are processed with PC. Cursor can be
controlled when camera captures the hand of the user.
The Neuro-Gaussian algorithm begins from the
evaluation stage of static background. After calculations
represented in 2nd section, Gaussian background
mixture model is formed. After model is formed, picture
with hand in front of camera is captured and regarding
to Gaussians background mixture model, pixels are
classified in two classes: first class background pixels,
second foreground object (users hand region) pixels.
Neural networks training data is formed from hand
region pixels, extracted with statistical Gaussian
method, and background pixels. This helps to avoid an
expert intervention in neural network training process.
Other subsequent picture pixels are classified with
trained neural network. After extracting hand region,
top extreme point is detected as it shown in figure 3.
Usually this point is the top of the hands middle finger.
Cursors positioning in computers monitor is realized
regarding to this point position in picture.
(3)
Mahanalobis distance is calculated to all regions. If this

distance is bigger than constant , then its considered
that pixel belongs to foreground object. After first
classification stage, classified pixels are used to form
the training data for neural network.
Neural network consists of three inputs for RGB color
spaces, one hidden layer with three neurons and one
output with sigmoid activation function. Neural network
is trained with Levenberg Marquardt algorithm.
Neural networks structure is represented in 1 figure.
Input
layer
Hidden
layer
Output
layer
Output
B
Fig. 1. Neural networks structure
As a result binary picture is generated, where

background is marked with 0 and users hand region
with 1.
Fig. 3. Extracted hands region handling. 1 top extreme

hand region point that controls cursor position; a, b ellipse,
that has the same second-moments as the region, axes
3. Hand tracking algorithm
By measuring eccentricity of hands region, left mouse

button press is initialized. The eccentricity is the ratio of
the distance between the foci of the ellipse and its major
axis length. This value changes when fingers are spread
or folded together. Now it is realized that left mouse
Alternative cursor control system visual information

gathers from web-cam, fixed in front of hand and
directed vertically down as it shown in 2 figure.
24
button is initialized when fingers are folded. This

Neuro-Gaussian algorithm is represented in fig. 4.
Fig. 5. Hand recognition results with statistical Gaussian

method: a) static background picture; b) background
clustered in 5 similar color regions; c) picture with hand in
front of web-cam; d) extracted white hand region
With recognized hand region pixels and part of

background pixels neural network was trained. Further
pixel classification was made with neural network.
During experiment hand was moved in circles in front
of camera and recognized hand coordinates was logged.
On the same frames expert selected real hand position
coordinates they where compared with recognition
results (Fig. 6).
X coord. Recognised and real hand cordinates when picture size: 120x160
150
1
2
100
50
20
40
60
80
100
120
140
160
180
200
Y coord. Recognised and real hand cordinates when picture size: 120x160
100
80
60
Fig. 4. Neuro-Gaussian algorithm
40
20
First four steps in Neuro-Gaussian algorithm are system

calibration process and its done automatically without
expert intervention. Other steps represent normal
functioning of alternative cursor control system.
2
0
20
40
60
80
100
120
140
160
180
200
Fig. 6. Recognized and real hand coordinates : 1

recognized, 2 real coordinates
From comparison results in Fig. 6 it is visible that

recognized and real coordinate curves are similar. Error
function of x and y coordinates on Fig. 7 also shows
that tracking algorithm is quite accurate: the offset
between recognized and real coordinates is not bigger
than five pixels.
4. Experiments and results

Experimental data consists of 200 frames with different
hand positions. Experiments where made with three
different frame sizes (160x120, 320x240, 640x480) in
purpose to evaluate dependence on frame size and one
iteration time, cursor positioning accuracy. First of all
hand tracking system was calibrated: Gaussian
background mixture model is calculated and neural
network is trained. On Fig. 5 are shown hand
recognition results during system calibration process.
X coord. Error function when picture size: 120x160

5
-5
20
40
60
80
100
120
140
160
180
200
180
200
Y coord. Error function when picture size: 120x160

2
0
-2
-4
-6
20
40
60
80
100
120
140
160
Fig. 7. Error functions between recognized and real hand

coordinates
25
5. Conclusions and discussions
As it was mentioned above, experiments were mane

with three different size frames. Results displayed in
Fig. 8 shows that one iteration time increases with
bigger frame resolution.
A Neuro-Gaussian algorithm was presented. Statistical

Gaussian method enables to avoid an expert
intervention in system calibration process. System is
calibrated automatically. Neural network usage enables
to reach good performance in hand position tracking
task. Experimental results showed that Neuro-Gaussian
algorithm can perform an accurate and fast hand
position tracking and cursor positioning on monitor.
Alternative cursor control system works faster with
smaller resolution frames, but cursor positioning on
computer monitor is more accurate with higher
resolution frames.
One iteration time in s. with different picturte size

0.7
0.6
0.5
0.4
0.3
6. Acknowledgment
0.2
The authors appreciate the financial support of the

Lithuanian state science and studies foundation.
0.1
0
160x120
320x240
640x480
7. References
1. Zhen Tang, Zhenjiang Miao. Fast Background
Subtraction and Shadow Elimination Using
Improved Gaussian Mixture Model. Haptic, Audio
and Visual Environments and Games. IEEE
International Workshop on 12-14 Oct. 2007. p. 38
41.
2. Shi Peijun, Jones E.G., Zhu Qiuming. Median
model for background subtraction in intelligent
transportation system. Proceedings of the SPIE,
Volume 5298, 2004. p. 168-176.
3. Alberto Broggi, Pietro Cerri, Paolo Medici, Pier
Paolo Porta. Real Time Road Signs Recognition.
Proceedings of the 2007 IEEE Intelligent Vehicles
Symposium Istanbul, Turkey, June 13-15, 2007. p.
981 986.
4. Shuangdong Zhu, Lanlan Liu. Traffic Sign
Recognition based on Color Standardization.
Proceedings of the 2006 IEEE International
Conference on Information Acquisition August 20 23, 2006. p. 951 955.
5. Yang Kun, Zhu Hong, Pan Ying-jie. Human Face
Detection Based on SOFM Neural Network.
Proceedings of the 2006 IEEE International
Conference on Information Acquisition August 20 23, 2006. p. 1253 1257.
6. Paliy I., Kurylyak Y., Kapura V., Sachenko A,
Lamovsky D, Sadykhov R. Combined Approach to
Face Detection for Video-Surveillance. IEEE
International Workshop on Intelligent Data
Acquisition and Advanced Computing Systems:
Technology and Applications 6-8 September, 2007.
p. 594 598.
7. Andy Chiem, Adel Al-Jumaily, Rami N. Khushaba.
A Novel Hybrid System for Skin Lesion Detection.
1-4244-1502-0/07, 2007. p 567 572.
Fig. 8. One iteration time in seconds with different picture

size
By increasing frame size 16 times, number of iterations

per second decreased approximately from 200 to 2.
Whereas cursor movement steps on monitor with
1440x900 resolution decreased, but error dispersion
increased (Table 1). That invokes cursor position
instability on the monitor.
Table 1. Experimental results
160x120
320x240
640x480
t, s
0.005
0.184
0.665
2(x)
2(y)
1.241
6.797
59.911
0.667
2.393
16.185
X step
9
5
2
Y step
8
4
2
Cursor control system was tested with other

applications. User used this system to draw in windows
application Paint. Results are represented in figure 9.
Fig. 9. Drawing results with represented cursor control system
26

OPTIMAL CONTROL OF ROBOTS BY DYNAMIC PROGRAMMING

Rudolf BAUMGART-SCHMITT, Stephan LIEBETRAU, Christian WALTHER,
Maria KRAUTWALD, Daniel TROMMER
algorithms is introduced by [6] which support the

collaboration of a set of robots.
In our contribution a swarm of robots navigate in a
complete unknown environment and try to find an
optimal path from the start to the final area. The robots
are able to recognize hindrances by distance sensors and
use compass modules to be able to correct turnings and
courses. The relevant data are transmitted to the central
data processing unit in a wireless mode by Bluetooth.
The unit collects and processes the information. A two
dimensional model with a pre selected spatial resolution
of the three dimensional unknown world are updated.
The model is adapted to the actual dimensions of the
robots by rules of cellular automata. Than the unit
calculates the optimal path for each robot by Discrete
Dynamic Programming (DDP) and transmits the next
commands to the robots wireless. Each valid transition
of the robots from one position to the next is connected
with a fixed effort. The accumulated effort of the
complete route should be minimized. That means the
reinforcement learning strategy is used. The processing
of the modules
Abstract: A swarm of robots are adapted to an unknown

environment by reinforcement learning. Each of the
robots is able to scan the environment and to measure
the distance to the next hindrance. The information
collected by the robots are transferred in a wireless
mode by the serial Bluetooth protocol to the central
unit. This unit performs the backward recursion of
Dynamic Programming to get the optimal route from
one place to another for each of the robots. The general
goal consists of finding the route with the minimal
number of steps from the initial to the final point of a
two dimensional area. The performance of the approach
can be demonstrated by both the simulation and the real
robots.
Keywords: control of robots, cellular automata, dynamic
programming.
1. Introduction
The autonomous navigation of mobile robots is the
main topic in a growing number of papers. The different
approaches in soft computing like neural networks,
genetic algorithms and fuzzy rules are successfully
applied to control mobile robots [1, 2, 3]. The quality of
the control strategies or the fitness of the robots in the
context of genetic algorithms can be evaluated in
different ways and serve as feedback for the optimal
adaptation. Several optimization procedures have been
implemented to find the global or local optima of the
optimization problem with restrictions. That task should
be performed in real time. By means of Dynamic
Programming invented by Bellman the global optimum
can be found but at the expense of strongly increased
computation time if the dimension of the state space is
large. In [4] a hierarchical Dynamic Programming
approach is used for path planning to reduce the effort
and make the approach more efficient. Low-level and
high-level maps were formalized as Markov Decision
Processes. The idea of cooperation between mobile
robots supported by Petri-Net communication models is
presented in [5]. A new reinforcement learning
the scanning of the environment by the sensors

of the robots,
the updating of the model by the central
computing unit,
the calculation of the optimal path by the
central computing unit,
the transmission of the further command to the
robots and
the executions of the commands by the robots,
has to be repeated until the goals are reached by all

participants of the running.
2. Sensors and Microcontroller Hierarchy of the
Robots
Six robots RP6 developed by the company Arexx were
used as basic experimental platforms. The RP6 is an
autonomous mobile robot system. The navigation is
27
originally supported by several sensors: bumper

sensors, light and infrared sensors. The RP6 is
controlled by the microcontroller ATMEGA32L with 8Bit RISC architecture. To improve the abilities of RP6
the expansion module with a second ATMEGA32
microcontroller and further accessories were mounted
on the top of the robot [7]. The distance sensor sharp
GP2D12, the compass module CMPS03 and the
Bluetooth module BTM222 were also included. Fig. 1
shows the side view of the modified robot. The local
processing units can be seen at three layers which are
connected by the IIC Bus. The distance sensor GP2D12
uses infrared to detect barriers up to 80 cm away. There
is a nonlinear output voltage of the sensor in relation to
the distance. After analogue digital conversion the
voltages were processed by means of a look up table.
The sensor was positioned on a servomotor and it can
scan the distances to the next hindrances in an angle of
180 degrees with a resolution of 2 degrees. In this way
the measuring tolerances of the distances because of
robot moving can be minimized.
broader in the model world to shift the optimal path

calculated by the optimization procedure. The third
problem which has to be solved by cellular automat
before the optimization procedure starts is the
correction of false edge detection by the distance
sensor.
4. Discrete Dynamic Programming
To find the shortest distance from start to destination
the backward recursion of discrete dynamic
programming was used. That means the stages and
states had to be defined at first. The destination should
be the stage N with one state only. Than the places with
immediate transitions to the destination are the two
dimensional states xN-1 of stage N-1 and so on.
Recognized hindrances are restrictions which interrupt
the possible path. The scalar decisions dn at stage n are
limited to 4. The robots can go right, left, upward and
downward. The scalar returns rn at stage n are
accumulated. The transitions xn tn ( xn-1 , d n-1 ) for n =
1,2, N are defined by the two dimensional table.
According to the principle of optimality developed by
Bellman [8, 9] the backward recursion of the Discrete
Dynamic Programming can be formulated by the
following recursion equation:
f n -1 ( xn-1 ) min {rn -1 ( xn-1 , d n-1 ) f n ( xn )}

dn
(1)
-1
with n = N, N-1, , 2, f n -1 ( xn-1 ) as the optimal return

from the stages N, N-1,n-1 and f n ( xn ) =0 . To find
the optimal path to the final state the set of optimal
decisions should be applied to the transitions starting
with the values of the initial state.
5. Reinforcement Learning
Fig. 1. Robot RP6 produced by Arexx and modified by

Liebetrau (2008). The autonomous driving RP6 is controlled
by the microcontroller ATMEGA32L at 8 MHz clock with an
external 32KB SPI EEPROM. The expansion module with a
second microcontroller (ATMEGA32 at 16 MHz), an
additional distance sensor sharp GP2D12, compass module
CMPS03 and the Bluetooth module BTM222 are included
Reinforcement learning can be classified as a nonsuper- wised method of machine learning. In our case
the agent (robot) tries to minimize a long term negative
reward (return) in goal directed moving through an
unknown environment. The on-line learning takes place
during the complete actions. The state specific decisions
are the results of (1).
3. Cellular Automata
Cellular automata were successfully applied in image
processing [10]. The optimization procedure bases on a
model which is created by sensor data and modified by
cellular automata. We solved three different image
processing problems by a cellular automat. The first
problem is connected with recognized gaps between the
hindrances which are too small for passing by the
selected robots. The rules of the defined automat are
responsible to close the gates before the optimization
procedure starts. The next problem arise if the selected
resolution of the two dimensional model of the world is
too small compared to space which is necessary for one
robot. The optimal path calculated by DDP can lead
beside the hindrances too narrow for passing by. The
cellular automat is used to make the walls virtually
6. Simulation
To test the reinforcement learning and the
implementation in C++ the world for path finding was
simulated by software. The simulated robots were
placed on one square only according the pre selected
spatial resolution. Therefore there was no need to apply
the rules of the cellular automat. By means of a drawing
program the model can be created on the selected
spatial resolution without any limitation. Up to 6 robots
can be started. Because of the high speed of this
simulation the different models of the world can be
tested and evaluated very efficiently. Fig. 2 presents the
output of the program in simulation mode. The path of
28
the simulated robot is drawn by gray coloured circles.

The hindrances are in black colour.
measured by the compass module. The compass had to

be adjusted before using. After the execution of the next
step the actual position of the robot had to be corrected
too if necessary. Because of the danger to end up in
undefined places, the original bump and infrared
sensors were switch off.
8. Summary and further research
The system works reliable. The wireless communication
is efficient. The learning and controlling procedure is
able to find the optimal path. Up to now the distance
sensor takes to much time for scanning. Therefore it
should be exchanged. During the experiences with the
robot swarm we recognized the necessity for a stronger
parallelizing the actions of the robots. The parallel
scanning of the robots can lead to inferences of the
reflected light. Furthermore the detection of the exact
location by a camera system described in [12] should be
included to improve the accuracy. The Discrete
Dynamic Programming approach will be extended to
handle multiple criteria and time varying constraints in
the next working steps [10].
9. References
1. Narvydas G., Simutis R. and Raudonis V.
Autonomous Mobile Robot Control Using Fuzzy
Logic and Genetic Algorithm. Intelligent Data
Acquisition and Advanced Computing Systems:
Technology and Application, IDAACS 2007 4.th
IEEE Workshop, 2007, p. 460-464.
2. Narvydas G., Simutis R. and Raudonis V.
Autonomous Mobile Robot Control Using If-Then
Rules Fuzzy and Genetic Algorithm. ISSN 1392124X Information Technology and Control, 2008
Vol. 37, No 3.
3. Janglova D. Neural Networks in Mobile Robot
Motion. Int. J. of Advanced Robotic Systems, Vol. 1
Nu. 1 (2004).
4. Bakker B., Zivkovic Z. and Krse B. Hierarchical
Dynamic Programming for Robot Path Planning.
IEEE / RSJ International Conference on Intelligent
Robots and Systems 2005 (IROS 2005).
5. Stambolov
G.
Development
of
Petri-Net
Communication Model for Cooperative Mobile
Robot System. The 3 rd International Conference on
Electrical and Control Technologies, Kaunas,
Lithuania, 2008.
6. Kartoun U., Stern H., Edan Y., Feid C., Handler J.,
Smith M. and Gillam M. Collaborative Q()
Reinforcement Learning Algorithm A Promising
Robot Learning Framework. In: Robotics and
Appplications (RA 2005), Gerhardt, L. A. (Editor),
Cambridge MA USA.
7. Liebetrau St. Entwurf von Algorithmen zur
optimalen Navigation von Robotern unter Nutzung
von Reinforcement Learning Strategien und ihre
Erprobung mittels C/C++ Softwaremodulen.
Diploma thesis. Faculty of Electrical Engineering
University of Applied Sciences Schmalkalden,
Germany, 2008.
Fig. 2. Simulated path finding. The path is indicated by gray

coloured circles
Fig. 3. Training and testing area of the robots. The movement

directions are shown by arrows
7. Experiences with Mobile Robots

The firmware of the robots was written in C and was
processed by bus connected microcontroller. The robots
training and testing area can be seen in Fig. 3. It shows
three robots which move to the destination. In contrast
to the simulation processing the rules of the cellular
automat had to be included. They were used to balance
the deficits of the sensors in recognizing edges and to
adapt the dimensions of the hindrances to the spatial
resolution. The optimization procedure worked very
successfully and efficiently. The command for the next
step was calculated and transmitted after correcting the
actual position of the robot by means of direction
29
8. Bellman R. Dynamic Programming. Princeton

University Press, 1957.
9. Robinett III R., Wilson D., Eisler G. and Hurtato J.
Applied Dynamic Programming for Optimization of
Dynamic Systems, Society for Industrial and
Applied Mathematics (SIAM), Philadelphia, 2005.
10. Popovici A. and Popovici D. Cellular Automata in
Image
Processing,
Fifteenth
International
Symposium on Mathematical Theory of Cellular

Automata, 2002.
11. Santos dos R. Steffen V. and Saramago S. Robot
path planning in a constrained workspace by using
optimal control techniques, Multibody Syst Dyn
(2008) 19, p. 159-177.
12. Altenburg J. and Altenburg U. Mobile Roboter.
Hanser, 2002.
30

MAP DRAWING OF UNKNOWN ENVIRONMENT

USING AUTONOMOUS MOBILE ROBOT
arnas GERMANAS*, Gintautas NARVYDAS**
*Department of Mathematics Research in System, Kaunas University of Technology, Lithuania
**Department of Process Control, Kaunas University of Technology, Lithuania
of the terrain is basalt or clay. It uses the absorption and

reflection patterns of the terrain. Later this robot could be
sent to Mars to make maps of this planet.
The area of map drawing could be divided to two parts
topological maps and metric approaches. Metric maps
deal with geometric properties, and topological maps
describe with the connectivity of different objects [10, 5].
In our work the robot E-puck is programmed to explore
unknown environment and to make a clear map of that
environment without any human influence. The main task
is to use data of the sensors of the robot that we could see
obstacles. We have initial robot position and direction.
Also robots position every moment is calculated.
Together the positions of all sensors are known. The robot
sends to computer the values of proximity sensors, which
show the distance between the current sensor and the point
of the obstacle. With this distance and the sensor
coordinate we can determine the coordinate of the point of
the obstacle. With the speed of information sending speed
we can get the values of the coordinates of obstacles.
When the program determines that the robot is very close
to the obstacle it sends instruction it to go in the other
direction.
Abstract: This paper focuses on map drawing using

autonomous mobile robot. A small robot E-puck with
eight infrared proximity sensors and two wheels is
programmed to detect obstacles and walls, and in this way
to draw a map of unknown environment. Program in
computer analyses this information and draws points
which are calculated as obstacle points.
Position of the robot at any time is calculated using wheel
odometry. The values of the sensors are calculated to the
distance measured in metric system. Every time robot tries
to detect the obstacles and when it approaches too close to
the obstacle, it changes the direction of the movement.
Keywords: Map drawing, odometry, E-puck, autonomous
mobile robot.
1. Introduction
Mobile robots are widely used in industry, military,
research and entertainment [9, 8, 1]. Robots can help
people to make works that are simple for machine and
perhaps complex for human. Unfortunately robots can not
do intellectual works, such as creative work, solving
abstract situations. When there are very big amount of
variants of decisions, it is very hard to determine the
decisions in all possible situations.
Important area of robots application is drawing maps of
unknown environments. There could be environments that
are too dangerous or impregnable for humans. So we must
use mobile programmed robots [6, 2]. One must put some
instructions to the robot, which could let the robot to make
some exercises, for example, searching for mines, making
the map, detecting conditions of the environment. The
example is the Mars project [7], where mobile robots are
sent to the Mars and there they must search for life, take
the pictures and do not get into dangerous situations.
There is a project which teaches the robot Zoe [3] to draw
a map of the desert. It must understand if the composition
2. Robot E-puck
E-puck [4] is small and elegant robot. It is comfortable to
use in research because it is small, light and fast. Its view
from the top is showed in Figure 1.
The diameter of it is 70mm. The structure of this robot can
be defined by three parts the body, motors and printed
circuit. The body ensures other parts, including the
battery. The battery is connected with printed circuit by
two contacts. The battery is based on LiION technology,
has 5Wh capacity and is sufficient for about 2-3 hours of
intensive use.
The motor uses steppers with gear reduction. There are 20
steps per revolution, the gear reduction is 50:1. The Robot
31
has two wheels, which have diameter of 41mm and the

distance between them of 53mm. The maximum speed of
wheel spinning is 1000 steps/s. 1000 steps correspond to
12.8 cm. The wheels can spin to both directions. When
spinning the wheels to special directions robot can go by
any angle.
The robot has information about the count of steps driven
by both wheels. These values are used to calculate other
parameters, for example, the coordinates of the robot.
Sensor values
3500
3000
2500
2000
1500
1000
500
0
0.5
1.5
2.5
3.5
4.5
5.5
Distance
Fig. 2. Analysis of the 2nd sensor
The first task is to understand what type of function

relates the distance and the sensors value. After several
guesses the most appropriate form is
y a xb ,
(1)
where y represents the distance, and x represents the value

of the IR proximity sensor.
The least square method is used to find out the parameters
a and b. Though sensors are the same type, but their
returned values are different for the same distance.
Therefore for every sensor we get different parameters.
These different functions are used in the later research.
The next task is to create the algorithm with which the
coordinates of the robot could be known at every moment
during the wandering of the robot. In this model the
coordinates (named global coordinates) contain three
variables: x coordinate, y coordinate and the angle of the
robot in Cartesian coordinate system. The problem is that
using information sent by robot we calculate different
parameters (local coordinates): x and y coordinates, and
the angle of the robot. The value of the angle sent by robot
is measured in different system than the angle we need in
Cartesian coordinate system. To be precise the angle sent
by robot could also measured in Cartesian system, but all
these angles are transformed changing the coordinate
quarters. The figure 3 shows this situation more
apparently.
Fig. 1. The view from the top of the robot e-puck
There are eight infrared (IR) proximity sensors. They are

placed around the robot. There are 4 sensors in the front of
the robot, two sensors on the sides, and two sensors in the
back side.
Also the robot has dsPIC family processor. The robot
connects the host computer via Bluetooth interface.
3. Research
First of all we need to examine the sensors. The function
which relates the value of the sensor and the distance
between that sensor and the obstacle must be found. So
the robot is put in different distances from the obstacle
(the wall) to the examining sensor. With a different
distance the sensor gives a different value. The distances
varied from the initial 0.5 cm to the final 6 cm
increasing the distance by 0.5 cm. At every distance the
sensor was requested to give 1000 values. Then the mean,
standard deviation and median were calculated. Then
there were eight pairs of quantity the distance and the
mean of the sensors value. The diagram of the example
experiment is shown in figure 2. The vertical axis
represents values of the sensor and the horizontal axis
represents the distance.
The values of the sensors are not linear depending on the
distance. So, we need get the analytical function which
relates values of the sensors and the distance. Then we
always could have the distance value and work in metrical
space.
Fig. 3. Diagrams which show change of local and global angle

systems
32
The formulas which relate the angle of the global

coordinates and the angle of the local coordinates appear
so:
90 local
270 local
global
, local [90;180]
.
, local (180;90]
d right dleft
cos( local ),
x x0
2
y y d right dleft sin( ),

local
0
2
d d
local _ new local left right ,
d
(2)
global and local vary in range of (-180; 180].

Formula (2) is represented in figure 4:
where D d baseline
200
global angle
100
50
0
-50
-100
-150
-100
100
200
local angle
Fig. 4. Relationship of global and local coordinate systems
Analogously:
, global [90;180]
90 global
.
local
270
global
global ( 180;90]
The x and y coordinates also must be recalculated:

x global y local
.
y global x local
(3)
(9)
Here ( Sx k , Sy k ) is the coordinate of the kth sensor,

k is the angle of the kth sensors in global system.
This angle is calculated in this way:
k k 0 90 local ,
(10)
where k 0 is the angle of the robot in global system.

So now we have the coordinates of the sensors and the
distance from the obstacles to the sensors. We must make
the same thing which we made when we were calculating
the coordinates of the sensors.
(5)
When the robot turns to left i.e. dleft < drigh,,

d right dleft
local)
x x0 D sin(local) D sin(
d
d
d
y y D cos( ) D cos( right left )
0
local
local
d
; dbaseline is distance
(Sxk , Syk ) (xglobal, yglobal) 3.5 (cos(k ),sin(k )) .
(4)
The robot does not send his coordinates in Cartesian form

but it sends the amount of the steps of every wheel. There
are used formulas which calculate the coordinate of the
robot center point and the angle of the robot using data
sent from the robot.
When the robot turns to right i.e. dleft > dright,
d right dleft
local)
x x0 D sin(local) D sin(
d
d right dleft
y y D cos( ) D cos(
local)
0
local
d right d left
(8)
between the wheels; x0 and y0 are old coordinates; local is

old angle; x, y are new coordinates; local_new is new angle;
dright and dleft are number of steps of right and left wheels.
But formulas work only in local coordinate system. The
coordinates are renewed in every program cycle, so there
are used two odometry values from the next moments. So
only the change of the coordinates is calculated. Therefore
this change must be added to the old values. Because old
values are calculated in global system and the formulas
work in local system, old information must be converted
to local system before it is used when adding the update to
it. After adding updated information must be converted
back to global system.
The next task is to calculate the coordinates of the
committed obstacle. To do that we must use two before
calculated values distance from the obstacle to the
sensor and the coordinates of the robot. The coordinates of
the sensors are calculated easily. Every angle of the sensor
coordinate vector in global system is known (The initial
angle of the robot in our experiment is 90). The distance
between the sensor and the center of the robot is 3.5 cm.
So, as follows:
150
-200
-200
d right d left
(7)
(Oxk , Oyk ) ( Sxk , Syk ) y (cos( k ), sin( k )) , (11)
(6)
where (Ox k , Oy k ) is the coordinate of the obstacle

committed by the kth sensor; y is the distance between the
kth sensor and the obstacle.
When the robot moves straight i.e. dleft = dright,
33
So, now we can calculate dynamically the coordinates of

the obstacles during the wandering of the robot.
The robot is asked to avoid the obstacles. When the values
of the distances between the robot and the obstacle are too
small, the robot must change direction.
drawn with points. So, more research must be done and

other methods must be applied.
We tried to control robot in distance not looking directly
to the unknown environment but just looking to the
information received from the robot sensors. Such
situation could be performed when human can not enter
some dangerous environment but wants get the map of
that environment. So he sends the robot to the
environment and controls it by control device. In this
work as control device was used the joystick. With this
device a person who controls robot can see dynamically
drawn map. Regarding to the information in the map he
controls the robot. Then if the robot appears in difficult
situation the person can help it. Results of the controlled
robot are shown in figure 6.
4. Results
The result of the experiment must be a clear map of the

given unknown environment. At first the map is made of
the points which represent points of the obstacles. The
example result is given in figure 5. The initial coordinate
of the center of the robot is (0, 0). Primarily the robot
starts to move when it is parallel to the Y-axis. Outer
points represent detected obstacles, inner points represent
the trajectory of the robot. Black line represents the real
obstacles of the environment. Measure unit is centimeter.
Fig. 6. The map got using joystick
We get better results but there also is some noise and the
slide of the wheels. Perhaps robot must turn around slower
because it has problems then.
Fig. 5. Map of unknown environment drawn by points
As seen in the picture there is some noise while detecting

the points. This noise is caused by errors of the sensors
values and by the slide of wheels. At the point (3, 0)
wheels of the robot have slipped. After that we can see
that the right part of the map is turned down.
Approximately the error of points position is not bigger
than 2.5 cm. Despite this we can get enough clear map.
The next task is to convert this data to the more discrete
form. To achieve this, the point grouping algorithm was
used. The essence of it is to divide the area of the map to
the small squares (for example, the side of the square
could have length of 0.5 cm). The number of the points
which fall to the square is counted. The marginal number
of points is defined. If the number of points is sufficient
the square is painted.
Unfortunately the result is not passable. If the marginal
number of the points is relatively large, then we get too
less walls and the map is not correct. Then we could take
smaller marginal number. But in this project we get
passable results only when we take marginal number
equal to 1. Then we get the same result like in the map
5. Conclusions and future works
The robot can send enough information to draw fairly

correct map. Approximately the error of points position is
not bigger than 2.5 cm.
The process of points of the map descreeting is not
passable. Either robot must drive slower or different
method must be used.
The way how the robot must behave in the environment
was very simple. Robot must only avoid obstacles. In this
way it theoretically must explore more and more territory.
But practically sometimes the robot gets into difficult
situations and struggles. Hence, the more intellectual and
intended behaviour must be programmed for robot. In
future works robot could be programmed to follow the
wall. Also the problems caused by unexpected situations,
for example, wheel slide, fast wall reaching, must be
solved.
Interesting and practical solve of robot wandering problem
is use of control device. It can be used when human can
34
5. Hafner V.V. Evaluating cognitive maps for mobile

robot navigation behaviour. Proceedings of the
AISB'03 Symposium on "Scientific Methods for the
Analysis
of
Agent-Environment
Interaction",
Aberystwyth, UK, 2003.
6. Hhnel D., Schulz D., Burgard W. Map Building with
Mobile Robots in Populated Environments. Intelligent
Robots and System, 2002. IEEE/RSJ International
Conference on Intelligent Robots and Systems,
Lausanne, Switzerland, 2002, vol.1, p. 496-501.
7. Hunstberger T.L., Baumgartner E.T., Aghazarian H.,
Cheng Y., Schenker P.S., Leger P.C, Iagnemma K.D.,
Dubowski S. Sensor-Fused Autonomous Guidance of
a Mobile Robot and Applications to Mars Sample
Return Operations. Proceedings of the SPIE
Symposium on Sensor Fusion and Decentralized
Control, 1999, Boston, USA, vol. 3839, p. 2-8.
8. Myers G.K., Julia L.E., Cheyer A.J., Gerbi S.
Multimodal User Interface. Communications of the
ACM, New York, USA, 2004, p. 57-59.
9. Setalaphruk Y., Uneno T., Kono Y., Kidode M.
Topological Map Generation from Simplified Map for
Mobile Robot Navigation. Proceedings of the Annual
Conference of JSAI, 2002, p. 109-110.
10. Thrun S. Robotic Mapping: A Survey. Exploring
artificial intelligence in the new millennium,
San Francisco, USA, 2002, p. 1-35.
not enter the territory but wants get the map of that
territory
6. Acknowledgements
We would like to express our acknowledgement to

Lithuanian State Science and Studies foundation for the
received support.
7. References
1. Autonomous Robot Vision.

http://www.cis.udel.edu/~cer/arv, retrieved February
25, 2009.
2. Batavia P.H., Nourbakhsh I. Path Planning for the Cye
Personal Robot. Intelligent Robots and System, 2002.
IEEE/RSJ International Conference on Intelligent
Robots and Systems, Takamatsu, Japan, 2000, p. 1520 vol.1.
3. Calderon F., Thompson D.R., Wettergreen D.
Autonomous Rover Reflectance Spectroscopy with
Dozens of Targets. International Symposium on
Artificial Intelligence, Robotics and Automation in
Space (iSAIRAS), Los Angeles, USA, February, 2008.
4. E-puck official webpage. www.e-puck.org, retrieved
February 9, 2009.
35

INFLUENCE OF DIMENSIONS OF A MOBILE ROBOT ON NAVIGATION

USING VECTOR MARKS
Virginijus BARANAUSKAS, Stanislovas BARTKEVIIUS, Kastytis ARKAUSKAS
Abstract: Vector marks allow us to find shortest path

between initial position of the robot and target point.
This path is shortest, when the robot is viewed as a
material point. According to the dimensions of the
robot, not only the length of the path changes, but there
situations, when the robot could not pass existing
openings may arise. The purpose of the article is to
create robot with real dimensions navigation algorithm,
keeping simplicity of the usage of vector marks. Also,
the aim is to find shortest path. Algorithm and
modelling results of navigation are represented.
2. Assessment of robot dimensions in model of

environment
Model (map) of the environment, in which robot moves,
is known to robot or to the attendant supervisor
program. This model comprises from vector marks
(fig. 1) and inventory of obstacles. It is easy to describe
a set of marks as a tree. Then all possible paths are
visible.
Keywords: mobile robots, path planning, navigation.

1. Introduction
The problem of robot navigation in known environment
with obstacles is the purpose of the article. The simplest
case is when there are no obstacles between robot and
target point. This case can be solved using scanners. If
the robot and the target point are not on the same
straight visible line, the problem of target point search
evokes. [1, 2] suggest using vector marks, i. e. virtual
vector marks to target, allowing robot to orientate in the
environment and to choose direction of movement to
target point. A set of vector marks is attribute of the
target and describes all possible paths to the target
point. Dimensions of the mobile robot are not fixed, for
example, the dimensions of the transport robot depend
of the clearance of the consignment. Therefore not all
paths, described using vector marks fits for particular
robot with known dimensions, because there could be
narrow places, through which robot could not pass. The
simplest solution for this problem is described [3], is to
enlarge artificially all obstacles (in all directions) by the
half of a robot width. This solution closes all narrow
openings and all passing paths are remaining. But, if the
dimensions of the robot are not known a priori, early
search of the path becomes impossible. Search of the
path, when robot moves, slows this process, because
this search is complicated.
Fig. 1. Model of the environment. A target point
Each vector mark is specified with weight coefficient,

i.e. distance from tail point of the mark to the target.
Robot orients according to the directly visible marks
and chooses the easiest. So, it moves at the shortest
path.
Robot, which has finite dimensions, could have various
forms, which, depending on the direction of the
movement, could change. For example, when robot
stretches out or folds up the manipulator. It is rational,
especially when you are modelling, to put robot into
cocoon, geometrical figure, limitary dimensions of
which involves not only clearance of robot, but also
necessary openings between robot and obstacles, not
elaborating dimensions of the robot. Such figure could
be ellipse or circle. In this work, robots cocoon is circle.
Conflict emerges, when cocoon intersects obstacle.
36
cases, when there is the same obstacle in both sides

fig. 3. So, the number of obstacles with which robot
cocoon crosses, could not be criterion for determining
narrow opening.
3. Solution of the conflicts

It is necessary to predict robot reaction to conflict.
Robot, reacting to possible conflict, must change the
direction of the movement, i.e. retire from possible path.
Fig. 3. Robot in narrow opening
It is rational to find too narrow opening using decision

procedures of conflict with obstacle. If cocoon of the
robot crosses obstacle K in several places (see fig. 3), it
is necessary to solve conflict in one of the intersection
places. Lets say, that we solve conflict, associated with
chord AB. If then, i.e. we move cocoon of the robot
along the vector D, from position P1 to position P2,
conflict with each of the obstacle persists the opening
is too narrow. Algorithm of determination of too narrow
opening uses algorithm of solution of conflicts and its
stages are:
1. Solve the conflict, if it emerges;
2. Verify if new conflict does not emerges;
3. If so opening is too narrow.
If narrow opening is determined, we must form new
reaction of the robot it must find other path.
Fig. 2. Direction of the robot displacement after conflict with

corner of the obstacle
If there is collision between robot cocoon and obstacle

in the predictable robot position fig. 2 thick line, it is
rational to move robot in such way, that its movement
would be minimal. As we can see from fig. 2, it would
be then, when the robot will be moved along the
direction of vector K, i.e. athwart cocoon and obstacle
intersection chord D-C. Movement must be such, that
the distance between straight line MN passing the
circle centre H, parallel to chord D-C and nearest
obstacle point G would be not less than the radius of the
circle. Point G can be described as a farthest point from
chord D-C of obstacle fragment, situated in the indoor
of the circle. Algorithm of conflict solution must
perform such stages:
1. Verify, that robot cocoon does not cross any
obstacle;
2. Determine cocoon and obstacle cross chord
(D-C);
3. Determine direction of the movement (vector
K), perpendicular to cross chord;
4. Find farthest, from chord point, which is
situated in the indoor of the circle, obstacle
point (G);
5. Determine distance between founded point (G)
and part of the circle, overlapping with
obstacle, moving perpendicularly to crossing
chord (section G-L).
6. Move cocoon in the direction of vector K by
distance GL.
5. Closing of the opening, trim and supplement of

the marks tree
It is possible to start new search of the path from the
start position of the robot and to use usual algorithm of
the path search. It is necessary to close determined too
narrow openings in order to not find them again. In
order to do it, the array of the pseudo obstacles are
formed, these supposed obstacles are segments, closing
narrow openings. The length and place of each segment
are determined too narrow opening. Too narrow
opening, showed in fig. 3, is closed using segment of
conjunctive nearest points A and E of chords end
points, which were found after initial and repeated
conflict. The coordinates of these points determines
place of the pseudo obstacle. Pseudo obstacles, during
the search of the path, are treated as real obstacles.
Robot searches the path using the tree of vector marks,
which could lead robot to closed opening. So it is
necessary to modify tree of the marks. It is necessary to
eliminate the branch (branches), which is cut by new
pseudo obstacle. Also, all other branches, outgoing
from the branch, that we cut, becomes pointless. In
4. Determination of too narrow opening

There could be openings, through which robot could not
pass, in the robot path. It is important to find such
openings. It is evident, that in to narrow opening robot
cocoon must cross obstacles in both sides. There are
37
other words, it is necessary to cut all fragment of

marks branch, which was following from the branch
that we cut. Such recast of vector marks tree is
shoved in fig. 4.
Fig. 5. Path of the first size robot in maze. Robot chooses

shortest path, depending on vector marks and its weight
coefficients
Fig. 4. Modification of vector mark tree, when to narrow

opening is found; a tree before alteration, b tree after
alteration
If opening, through which goes tree branches E and D,

is to narrow, then, if we close it with pseudo obstacle
AB, not only the branches E and D are eliminated, but
also branch F, emerged from point E is eliminated too.
If we want to find all possible paths, when these
branches are eliminated, it is necessary to add branch C
into the tree. Recast of the tree do not requires finding it
newly, it is enough to augment pollard. Stages of the
tree modification:
1. Find branches of the tree, which intersects
pseudo obstacle;
2. Eliminate it and its childs from tree;
3. Add new tree branches, according to [1]
formed conditions.
7.
Fig. 6. Path of the second size robot in maze. Robot selects

shortest path, according to vector marks, but finds contraction
in the maze, which is to small. Robot, according to described
methodology, searches new shortest path
Robot in the maze
Generation of vector marks and the movement of a

robot, according to its dimensions, are viewed in
mazes, using different size of corridors and possible
paths between initial point and the target point. Robot
navigation is executed according to vector marks, which
has weight coefficients, which enables robot to choose
shortest path between initial position and the target
point. Three examples of different size robot are
selected and the functionality of created system is
viewed.
Fig. 7. Path of the second size robot in maze. Robot finds

another shortest path and reaches exit of the maze. The
obstacle of the shortest path, which does not let robot to pass
over it, is viewed as circle, with diameter equal to robots
diameter, near the obstacle
38
Fig. 8. Path of the third size robot in the maze. Robot finds
first too narrow opening and searches for new shortest path
Fig. 11. Path of the third size robot in the maze. Search of the
shortest path was successful. Circles in maze shows places,
where robot does not pass obstacle
7. Conclusions
Suggested method allows using of vector marks to find
nearly shortest path for a robot with finite dimensions as
well as in the case of a robot represented by a material
point. That makes the finding the path quite simple and
quick.
Too narrow openings can be detected and closed by
artificial obstacles to avoid in future. The tree of vector
marks is modified cutting not useful and adding new
necessary branches, when artificial obstacles added.
The algorithm finds the shortest possible way with
respect to too narrow openings in quite complicated
environments (mazes), as shown in examples.
Fig. 9. Path of the third size robot in maze. Robot finds
another (too narrow opening) obstacle and searches for new
shortest path
8. References
1. Baranauskas V., arkauskas K., Bartkeviius S.
Target Retrieval in Known Enviroment. Electronics
and Electrical Engineering, 2008, No. 8(88). p.9194.
2. Baranauskas V., Bartkeviius S., arkauskas K.
Modified Algorithm of Vector Marks Formation.
Proceedings of the Conference on Electrical and
Control Technologies, 2008, Kaunas, 2008. p. 1518.
3. Pimenta L.C.A., Fonseca A.R., Pereira G.A.S.,
Mesquita R.C., Silva E.J., Caminhas W.M., Campos
M.F.M. Robot Navigation based on Electrostatic
Field Computation. IEEE Transactions on
magnetics, Vol. 42, No.4, April 2006. p. 1459-1462.
Fig. 10. Path of the third size robot in maze. Robot finds new
possible shortest path, but another to narrow corridor is found
and robot must found new shortest path, according to vector
marks
39

OPTIMIZATION OF HEXAPOD ROBOT LOCOMOTION

Tomas LUNECKAS, Dainius UDRIS
Abstract: Hexapod robot moving through irregular

terrain must adapt its locomotion in order to sustain
stable and fast movement. Chosen gaits and control
methods have to vouch an optimal leg positioning and
coordination. An inverse kinematics method is
introduced for positioning legs to a needed point. Means
of leg contact with ground indications using tactile
sensors are discussed and their influence to robot
control is discussed. Simple obstacle detection method
using IR sensors is presented. Aim is to adapt all these
methods into a real-time control system.
Keywords: Hexapod robot, inverse kinematics, DenavitHartenberg method, IR sensors, tactile sensors, obstacle
detection.
1. Introduction
Fig. 1. Hexapod robot model
2. Robot control methods
For a movement of a hexapod robot (Fig. 1) on irregular

terrain one of described gaits with different features can
be chosen [1]:
Tripod gait the most popular hexapod robot gait. In
tripod gait case three legs are in air at given moment
while other three are on the ground, positioning the
body. This form of locomotion is good for navigating
through relatively smooth surface giving maximum
speed for the robot. The drawback of tripod gait is a
difficulty to control each leg simultaneously.
Wave gait only one leg is in the air at a time giving
robot five points of contact with the surface. This gives
robot the stable posture, making this gait ideal for
moving over unfamiliar and particularly irregular
terrain. It is easier to control each leg, but robot speed
suffers greatly.
Ripple gait this gait is very similar to wave gait as two
separate waves are formed on both sides of the robot.
This makes the robot body balancing easier.
Biped gait it is a mediate variant of tripod and wave
gaits as simultaneously two legs are repositioned. This
gives robot mediate stability and speed than tripod or
wave gaits.
Considering these gait descriptions it is important that
robot would change its gait in order to maintain set
speed or stability on different surfaces. The decision
must be taken according to task given to a robot.
Robot control refers to the way in which the sensing and

action of a robot are coordinated. There are infinitely
many possible robot programs, but they all fall along a
well-defined spectrum of control. Along this spectrum,
there are four basic practical approaches being used
today [2]:
Deliberative Control: Think hard, then act.
Reactive Control: Dont think, (re)act.
Hybrid Control: Think and act independently, in
parallel.
Behaviour-Based Control: Think the way you act.
No single approach is the best for all robots; each has its
strengths and weaknesses. Control requires some
unavoidable trade-offs, as follows [2]:
Thinking is slow.
Reaction must be fast.
Thinking allows looking ahead (planning) to avoid
bad actions.
Thinking too long can be dangerous (e. g., falling off
a cliff, being run over).
To think, the robot needs (a lot of) accurate
information.
The world keeps changing during thinking, so the
long time for it can cause inaccurate solutions.
40
As a result of these trade-offs, some robots are made

dont think at all, while others mostly think and act very
little. It all depends on the robots task and the
environment. If the task and environment require the
robot to move and react very quickly, there is usually no
time for thinking, such as in automated fast-moving
cars, or in soccer playing robots. On the other hand, if
the environment does not change much and the robot
has enough time, it can plan far ahead to find the best
action, such as playing chess, monitoring a warehouse at
night, or assembling a complicated object [2].
Control method of the robot depends on time it has for
planning or acting. If robots task is not required to be
carried out as quickly as possible, it is better to give
some time for the coordination and planning ahead.
Energy saving is one of the important criteria of

autonomous robot. Energy saving could be achieved by
optimizing angles and speeds at which motors have to
turn ensuring their low loads. Angle optimization is
done by giving the minimal angles needed to position
legs into a needed point. For example, when robot is
moving over smooth terrain its not needed to lift legs
very high as there is no need to step over obstacles. In
other words, it is energetically inefficient for robot to
lift legs higher than needed. Horizontal foot motion is
always dependent from speed.
4. Indication of foot contact with a surface
Robots body will swing when walking over irregular
terrain if leg lifting and lowering angles are static. To
avoid this, robot has to stop lowering the leg when it
touches ground. In order to do this each foot must be
fitted with tactile sensor. When leg touches ground,
sensor indicates it to the controller. Controller stops the
corresponding motor at a present angle, and memorizes
it. In order to avoid too deep cavities, leg lowering angle
limit must be set. This is the maximum angle that motor
can turn down.
3. Gait and control method connection

The gait and control methods of a robot strongly
depends on environment it woks in and time available
for decision making and planning. Thus gait and control
method are closely connected. This suggests such gate
and control combinations:
1. Tripod gait reactive control. This combination is
suitable for fast moving over familiar terrain when
planning and careful acting is not a priority.
2. Wave gait deliberative control. This combination
ensures that robot is acting deliberately, slowly
moving over unfamiliar surface and planning its
decisions far ahead.
Hybrid and behaviour-based control methods are
separate cases because planning and locomotion are
performed simultaneously. So it is not so important on
what terrain robot is walking.
During robots movement some locomotion parameters
must be respected. These parameters can be freely
chosen by robot designers, can be given or obvious from
working area.
The main purpose of a walking robot is navigation over
irregular terrain, so there is a great importance of its
body stability. Stability can be described as retention of
bodys given position and orientation independently
from terrains irregularity. In other words, robot legs
have to adapt to terrain. This can be achieved by using
sensors for detecting nearing obstacle (distance and
height). According to this information angles for each
legs motor are calculated at which they have to turn in
order to position foot to a needed point.
In most cases the robot has to move quickly and
independent from a surface travelling on. The speed
usually is set by the robots task. Moving through the
well known and smooth terrain a tripod gait (fastest
gait) and reactive control should be chosen. In this way
robot just reacts to external obstructions not thinking
much about leg positioning or planning. While moving
through irregular terrain two variants are possible. The
first is just as for moving through a smooth terrain.
However robot most likely will lose balance, run into an
obstacle and etc. The second variant is to change robots
gait to biped, wave or ripple and control method to
deliberative or hybrid. In this case robot will devote
more time for leg positioning and body balancing.
Fig. 2. Algorithm for checking is the leg on the ground or has

it reached an angle limit
In the block 1 (Fig. 2) a check is performed if leg is on

the ground (DI is a controllers digital input port,
connected to the legs tactile sensors; n code of a
checked leg). If DI value is equal to a code indicating
that one or more legs are on the ground, program stops
the motors at present angles. If not all needed legs are
on the ground (Fig. 2, block 2), the program compares
present legs angles to the limits (angle present motors
angle; dgr set permissible angle limit). When motors
present angle equals to the set angle, corresponding
motors are stopped. This way robot will be able to avoid
too deep cavities.
5. Obstacle detection using IR sensors
Robot can adapt to terrains irregularity, but it is needed
to obtain information about surroundings. Most
important information is the distance to and the height
of the incoming obstacle. That is needed to avoid a
collision with obstacle and make a decision: step over,
lift the body higher or turn around the obstacle. The
simplest way to solve this problem is to put two IR
41
cos 1
sin
1
T0,1
0
sensors on each of the front foot, pointed to the front,

and one IR sensor at the bottom of robots body also
pointed to the front. All sensors must be configured to
detect obstacles at a given distance. Preferable distance
is a bit longer than the steps length. In this way only
obstacles interfering with recent step will be detected.
While nothing is detected within defined area robot
continues movement with no changes to its locomotion.
If one of the foot sensors detect an obstacle robot must
stop legs horizontal motion and lift this leg until there
is no signal from IR sensor, meaning that foot is higher
than obstacle. Then robot can resume legs horizontal
motion. If IR sensors signal does not disappear while
robot is lifting leg, that means that obstacle is too high
to step over and robot must look for way around it.
When bodys IR sensor detects obstacle, it means that
there is an obstacle in front of the robot, and it must lift
the body higher to avoid collision with the obstacle.
0
0
sin 1
cos 1
cos 2
sin
2
T1,2
0
sin 2
cos 2
0
0
cos 3
sin
3
T2,3
0
sin 3
cos 3
0
0
0
0
,
l1
1
l2 cos 2
l2 sin 2
,
l3 cos 3
l3 sin 3
.
(1)
(2)
(3)
Each transformation matrix describes how one

kinematic pairs coordinate system translates to
previous. In order to find transformation matrix to
translate foots coordinates to bases coordinate system,
must be
transformation matrices T0,1 ,T1,2 ,T2,3
6. Inverse kinematics for one leg using

Denavit-Hartenberg method
The hexapod leg consists of four links and three
kinematic pairs (Fig. 3). Each one has given a
coordinate system.
multiplied:
T T0,1 T1,2 T2,3 .
(4)
This multiplication of matrices was done using package

MatLab and in a Fig. 4 presented task.
syms l1 l2 l3
syms t1 t2 t3
T1=[cos(t1) 0 -sin(t1) 0; sin(t1) 0 cos(t1) 0; 0 -1 0 -l1; 0 0
0 1]
T1 =
[ cos(t1),
0, -sin(t1),
[ sin(t1),
0, cos(t1),
[
0,
-1,
0,
[
0,
0,
0,
0]
0]
-l1]
1]
T2=[cos(t2) -sin(t2) 0 l2*cos(t2); sin(t2) cos(t2) 0

l2*sin(t2); 0 0 1 0; 0 0 0 1]
Fig. 3. Kinematic layout of robots leg
T2 =
[ cos(t2), -sin(t2),
[ sin(t2), cos(t2),
[
0,
0,
[
0,
0,
All Denavit-Hartenberg parameters for every kinematic

pair in one leg are written out into Table 1.
T3=[cos(t3) -sin(t3) 0 l3*cos(t3); sin(t3) cos(t3) 0

l3*sin(t3); 0 0 1 0; 0 0 0 1]
T3 =
[ cos(t3), -sin(t3), 0,
[ sin(t3), cos(t3), 0,
[
0,
0,
1,
[
0,
0,
0,
Table 1. Denavit-Hartenberg parameters for one leg

Kinematic
pair
0, 1
Kinematic
pair type
Rotational
Link
No.
1
1, 2
Rotational
2, 3
Rotational
Parameters
d
a
0 90
l1
l2
l3
0,
0,
1,
0,
l2*cos(t2)]
l2*sin(t2)]
0]
1]
l3*cos(t3)]
l3*sin(t3)]
0]
1]
T=T1*T2*T3
Fig. 4. MatLab task for matrix multiplication
Result of these calculations

transformation matrix T (5).
Transformation matrices for each kinematic pair are

presented in (13):
42
is
homogenous
C1C2 C3 C1S2S3
S C C S S S
1
2
3
T 1 2 3
S2 C3 C2S3
C1C2S3 C1S2 C3
S1C2S3 S1S2 C3
S1
C1
S2S3 C2 C3
C1C2 l3C3 C1S2 l3S3 C1l2 C2

S1C2 l3C3 S1S2 l3S3 S1l2 C2
,
S2 l3C3 C2 l3S3 l2S2 l1
(5)
Sometimes such calculation task can be hard to solve or

even unsolvable. Other solution is to use iterations,
when foot coordinates are recalculated at different
motor angles and compared to the known coordinates. If
result is not equal, angles are increased or decreased
further. Coordinates are recalculated again and
compared to known coordinates. This is repeated until
set and present coordinates become equal, meaning that
foot is at needed point. This method can be unsuitable if
fast leg repositioning is needed. So calculation method
must depend on motion requirements, and searching
optimization methods like simplex search [3] can be
used to reduce time and increase reliability.
Each trajectory must contain several points with
coordinates xn , yn ,zn representing different parts of
trajectory. Each leg is positioned to these points in
sequence, forming a full step.
where: S = sin, C = cos.

Three first members of last column represent foot
coordinates in bases coordinate system:
x cos 1 cos 2l3 cos 3 cos 1 sin 2l3 sin 3 cos 1l2 cos 2 ;
y sin 1 cos 2l3 cos 3 sin 1 sin 2l3 sin 3 sin 1l2 cos 2 ;
z sin 2l3 cos 3 cos 2l3 sin 3 l2 sin 2 l1 .
7. Step forming using derived homogenous
transformation matrix
As it is necessary to position foot to a needed point,

each motors angle must be known. These angles are
1 ,2 ,3 and must be worked out from equation
system. Irrespective of chosen calculation method, a
step for one leg can be represented as a three
dimensional trajectory (Fig. 5).
8. Conclusions
Hexapod robot movement over smooth terrain can be

organized using tripod gait and reactive control method.
That way controlled it only reacts to external
obstructions. While navigating over irregular terrain
requires more precise leg positioning, and wave gait and
deliberative control method becomes very suitable for
this task.
The simplest way to control robots vertical position
(feet are on the ground or in the air) is by using tactile
sensors on each foot. For robot to be able to step over or
avoid obstacles IR sensors can be used. To detect
incoming obstacle and evaluate height the IR sensors on
each of the front foot and under the robot base can be
used.
An inverse kinematics calculation for each leg and robot
as a whole is needed so that robot could calculate step
trajectories and body position. Using this information
more precise decision can be inducted, so the next task
for this investigation is to describe transformation
matrix for a whole robot.
For unsolvable by calculation tasks a direct search
algorithm can be implemented into the main program.
9. References
1. Analysis of Multi-Legged Animal + Robot Gaits.

http://www.oricomtech.com/projects/leg-time.htm,
last accessed April 10, 2008.
2. Mataric M.J. The Basics of Robot Control.
http://www-robotics.usc.edu/~maja/robotcontrol.html, last accessed February 15, 2009.
3. Dambrauskas A. Statistical Theory of Simplex
Search. (In Lithuanian), Vilnius, 2007. 260 p.
Fig. 5. Possible one step trajectories
43

FASTER AUTONOMOUS MOBILE ROBOT LEARNING

USING DECISION CORRECTION
Gintautas NARVYDAS, Petras RAANSKAS
While it is perfectly acceptable for a robot to have a

human controller who would make all of the decisions
for it, usually the circumstances require the robot to
have at least some degree of autonomity. This
automatically raises a whole batch of control problems,
how to make robot respond adequately to whatever
situation it might find itself in. Many different
algorithms have been proposed throughout the years for
various applications.
A subset of the proposed algorithms deal with a
powerful mathematical construct named an artificial
neural network. The most prominent feature of artificial
neural networks is their ability to learn. Given a set of
data and a corresponding set of results to be obtained
from these data, an artificial neural network can be
adapted using one of the many known algorithms to
approximate those given results and detect general
relationships between various parameters, which could
then be used to make appropriate decisions even for the
sets of data not previously encountered. This makes an
artificial neural network perfect for robot navigation
purposes, since a mobile robot is usually embedded into
a dynamic environment that is constantly evolving and
thus accounting for each and every possible situation
would be simply unfeasible or even impossible.
In this paper we will attempt to define an algorithm
which would teach a mobile robot with an artificial
neural network to solve a classic problem of following a
wall on the right using decision correction technology.
Abstract: In this paper an approach to teaching artificial

neural network by decision adaptation is described. Our
task is to define an algorithm for an autonomous mobile
robot e-puck to learn to follow a wall utilizing a
mechanism, which would alter the already made
decisions post factum toward a better solution by
evaluating their outcome. The algorithm is then tested
for performance and briefly compared with other known
approaches for advantages and disadvantages.
Keywords: autonomous mobile robot, artificial neural
network, control system, decision correction, machine
learning.
1. Introduction
Robotics is a branch of science that is currently gaining
importance and acclaim very fast. New uses for robots
are discovered every day, including but not limited to
industrial robots for car assembling, transportation of
heavy machinery, welding and other tasks [1], military
robots for reconnaissance, minesweeping and combat
missions [2], exploration robots, which can be sent into
various hazardous places of scientific interest, for
example, volcanoes [3], deep ocean [4] or even outer
space [5] and other planets beside Earth [6]. On other
frontiers, toy robots help children learn various basic
concepts of robotics and allow them to explore robots
from an early age.
There are many advantages to use robots in various
fields. First and foremost, robots can withstand a lot
bigger tensions and loads than human beings, they are
able to survive extremely acidic and caustic
environment for longer periods of time and are overall
more robust. Secondly, robots do not exhibit as fast
degradation of their work quality as humans over time,
meaning that they could work without stopping for
days, weeks or even months. Also, maintaining a single
robot is frequently a lot cheaper than its equivalent
group of human workers. Finally, robots do not get
bored, depressed or otherwise psychologically inhibited.
2. Algorithm overview
The main idea of the algorithm is that the decisions are
immediately evaluated for their success and a special
subsystem tries to adjust them so that they would
perform better in the next generation.
The robot obtains proximity sensor information and
passes it to an artificial neural network, saving a copy of
sensor values in an array. After this, both the sensor
information and the decision of the network are
forwarded to a primary IF-THEN test. The exact
content of this test is unimportant for our purposes,
44
however, minimally it should be able to detect and

appropriately override network decision in these
situations:
when the robot is about to make contact with a
wall;
when the neural network attempts to move the
robot backwards;
robot exceeding its maximum speed;
robot away from the wall.
After the final decision is established, it is forwarded to
engines of the robot. A little timeout is given, after
which new proximity information is obtained. Here we
enter a second batch of IF-THEN tests, however, this
time they are not meant to override decisions but to
apply corrections to them. In our experiment, the
following correction rules were applied:
if a wall is approaching in front of the robot,
encourage turning left;
if a wall on the right is getting closer,
encourage turning left slightly (Fig. 1);
if a wall on the right is getting further,
encourage turning right slightly;
if everything went well, increase the speed.
frame 70mm in diameter, which holds two engines and

a battery, as well as circuitry board.
Miniature stepper motors are used, which have 20 steps
per revolutions, which are further reduced with 1:50
gear reduction. Gears and the wheels are mounted
directly on the axis. The wheels are made of lightweight
plastic and have 41mm in diameter. The maximum
speed of a single engine is 1000 steps per second, which
is equivalent to a single revolution of a wheel, or
129mm.
The e-puck robot determines its position in space
using eight infrared proximity sensors (Fig. 2),
deployed around the perimeter of the body four in
front, two to the sides and two in the rear. Each of the
sensors has about 40mm effective range, which further
depends on the noise level of the surrounding area. The
returned values fall into a range from 0 to 3000, higher
numbers meaning closer object. The relationship is nonlinear, but irrelevant to our further discussion.
Fig. 2. The sensor layout of an e-puck mobile robot
e-puck is also equipped with three microphones, a

640x480 pixel camera, accelerometer and ground
sensors which were not used in this particular
experiment.
Fig. 1. An example of a made decision (solid arrow) and an

adjusted decision (dashed arrow)
Encouraging certain behaviour is achieved simply by

adding a certain amount of speed to one of the wheels,
unless its speed is already approaching maximum, when
the speed of an opposite wheel is reduced instead for
turning, or nothing is changed at all for accelerating.
The old proximity data and the adjusted decision are
saved into a database. After a certain predetermined
amount of iterations, where the data are acquired and
decisions are made and adapted, the workflow is
temporarily halted, the old artificial network destroyed
and a new created instead. This new network is taught
all the corrected decisions that were obtained, and then
the workflow resumes.
4. Experiment setup
For our experiment the e-puck robot was connected to
MATLAB computing environment, which was then
used to implement the algorithm and obtain various
debugging information in an accessible graphical
representation. The connection was achieved via
Bluetooth technology, with all data sent to and read
from an emulated COM serial port.
The robot was put into a box approximately 60
centimetres long and 40 centimetres wide, with several
cardboard obstacles inside (Fig. 3). Both the inside
walls of the box and obstacles were covered with sheets
of white paper to enhance their reflectivity to infrared
waves. Obstacles were not fixed to the box in any way
to make changing the configuration of the test maze in
the middle of an experiment possible.
Fluorescent lights were turned off in the surrounding
area to reduce the noise picked by infrared sensors.
3. e-puck mobile robot

e-puck mobile robot is a low cost mobile robot,
constructed specifically for various research
applications with simplicity and robustness is mind. The
whole structure is built around a single-part plastic
45
keep the value of the third proximity sensor

stable, which translates into keeping a stable
distance to the right side wall;
keep the sum of wheel speeds as stable and as
high as possible, which means that robot would
navigate the maze as fast as it can;
keep the difference of wheel speeds as stable
as possible, which means that robot would
avoid
changing
direction
of
travel
unnecessarily.
5. Results
For our experiments we restricted ourselves with 20

generations of the neural network and limited the speed
of the robot to 500 steps (6.3 centimetres) per second.
Both IF-THEN routines were absolute for the first
experiment, meaning that they didn't account for the
size of an error, just for its type. The first generation
was a zero network, thus the robot was guided around
the maze by the first IF-THEN routine. Since we elected
to use a discrete control scheme, the path of the robot
was noticeably rough. Also, the IF-THEN routine was
designed to keep the travelling speed constant at 200
steps (2.5 centimetres) per second, which is rather slow.
After the first generation finished, a new artificial
neural network was taught everything the previous
network collected. The smoothing of the trajectory was
evident immediately, and the maximum speed of the
robot rose up to 300 steps (3.8 centimetres) per second,
with the average being around 230 steps (2.9
centimetres per second).
On the fifth generation the average speed was already in
the 400 steps (5 centimetres) range, however, some
problems became evident. Since we used constant
corrections for our data, situations started to arise where
the neural network started taking evidently wrong
decisions which would have led into sudden spikes of
speed, minor collisions with obstacles or turning more
than necessary to negotiate a corner due to
overcompensation, which led to IF-THEN routine
becoming more active again (Fig. 5).
Fig. 3. An example setup of the testing box with an

approximation of the best route marked
On the software side of things it was decided to use a

feed-forward back propagation artificial networks with
a single hidden layer, made of four tangential sigmoid
neurons, and an output layer, consisting of two
tangential sigmoid neurons (Fig. 4). To reduce the
dependence on the exact type of robot used and increase
the speed of learning, both the input (proximity sensor)
and output (speed) data were normalized to the range
0.0-1.0.
S
1
S
2
S
3
Lef
t
S
4
S
5
S
6
S
7
S
8
Righ
t
Fig. 4 The structure of the used neural network

Good
A single artificial neural network was used for 500

work cycles, collecting proximity data and adjusted
decisions. After the life cycle was over, the network
would be destroyed, another one of the same structure
would be created and it would be taught according to
the batch of data its predecessor had collected. After
this switch, the new network would proceed with
collecting new, improved data for its successor.
During the experiment the robot was observed visually
as well as some of its parameters, namely the value of
the third proximity sensor, the sum and the difference of
wheel speeds, were monitored through MATLAB
environment. A well-behaved robot would do the
following:
Bad
1st
Generation
20th
Fig. 5. The evaluation of the performance of the neural

network compared to its generation
46
From the tenth generation onward the IF-THEN

override percentage rose dramatically to the range of
40-60% because the data correction algorithm began to
under-compensate the errors at an increasing rate,
which led to robot going too fast for its maximum data
transfer rate and sometimes even losing the wall on the
right, and then driving very slow because of the
override, resulting in very twitchy control and bad
overall performance.
For our second experiment the corrections were made
proportional to the speed the robot was travelling at the
moment. This reduced the under-compensation problem
significantly, allowing robot to negotiate right-hand
corners a lot more effectively, however, the problems of
spiking speed, colliding into the walls and turning too
much in the left-hand corners still persisted somewhat,
making IF-THEN override percentage rise again after
several generations of staying below 10%.
With this algorithm the burden of the problem is

transferred from actual network manipulation into
finding those well formed control blocks. Here lies the
main advantage and disadvantage of this method at the
same time. On one hand, for sufficiently simple tasks
describing a rule set for adapting behaviour could be a
very easy and intuitive task, on another hand, when the
difficulty rises, finding the appropriate rules may prove
to be bordering impossible, and very complex rule sets
may be very time consuming to parse through, which
could inhibit the responsiveness of a system.
Compared to evolutionary algorithms, this method
showed a potential for a lot faster convergence,
measurable in minutes instead of hours or days. On the
other hand, expert teaching algorithms are of
comparable convergence speed; however, the input
from human expert is constantly needed, which is a
waste of human resources.
6. Future work
8. Acknowledgements
The algorithm has shown a great potential to teach a

robot to navigate a maze in a matter of minutes as
opposed to hours. However, the simplest variants of the
correction rule set showed convergence only for several
generations and then started to diverge away. One
obvious way to take care of the problem would be to
reformulate the rule sets to take into account not only
the speed of a robot, but the size of an error as well.
Also, this dependence of a correction from the size of
an error might not be linear. A further work in that
direction could bring some more satisfactory results.
We would like to express our gratitude to Lithuanian

State Science and Studies foundation for the received
support and to Vidas Raudonis and Arnas Lipnickas
for their help and suggestions along the way.
9. References
1.
2.
7. Conclusions
3.
With well formed primary and secondary IF-THEN
blocks, this algorithm is capable of converging to a
good solution very quickly. In our tests it was revealed
that the robot was capable of navigating the maze
sufficiently fast with a minor percentage (below 10%)
of overridden network decisions in very few iterations
of the algorithm without any further input from a human
expert. However, after a while the quality of the
artificial network started degrading considerably, with
increasing twitchiness of control by neural network,
which led to IF-THEN overrides and thus decreased
average speed. A further work on avoiding such
degradation or finding a best neural network and
stopping the teaching process is required.
4.
5.
6.
47
Zavlangas P.G., Tzafestas S.G. Industrial

Robot Navigation and Obstacle Avoidance
Employing Fuzzy Logic, 2004.
McLain T., Beard R., Kelsey J. Experimental
Demonstration of Multiple Robot Cooperative
Target Intercept, 2002.
Caltabiano D., Muscato G. A Robotic System
for Volcano Exploration, 2005.
Whitcomb L., Yoerger D., .Singh H,
Howland J. Advances in Underwater Robot
Vehicles for Deep Ocean Exploration:
Navigation, Control, and Survey Operations,
1999.
Brown G.M., Bernard D.E., Rasmussen R.D.
Attitude and Articulation Control for the
Cassini
Spacecraft:
Afault
Tolerance
Overview, 1995.
Maki J.N., Litwin T., Schwochert M.,
Herkenhoff K. Operation and Performance of
the Mars Exploration Rover Imaging System
on the Martian Surface, 2005.

VOICE CONTROLLED TELEPHONY SERVICES

Rytis MASKELINAS*, Algimantas RUDIONIS*, Kastytis RATKEVIIUS*, Vytautas RUDIONIS**
*Kaunas University of Technology; **Vilnius University, Kaunas Faculty, Lithuania
main source of spoken language communication, faces a

challenge of development of own speech recognition
tools. Two major ways for solution are as follows [2]:
- to develop own speech recognition engine from
scratch;
- to adapt the foreign language based engine for
the recognition of the native language.
The first approach has potentially higher capabilities to
exploit peculiarities of selected language and hence to
achieve higher recognition accuracy. But the drawback
of such approach is the same why the providers of major
speech technologies avoid the implementation of such
languages in their products high costs in the general
sense of this word.
The second approach [3-5] has the potential to achieve
some practically acceptable results faster than
developing entirely new speech recognition engine [6].
Another advantage of this approach is potential to
achieve faster compatibility with the existing
technological platforms. Such advantage is often
important for business customers, since they need to
follow various technical specifications in order to
guarantee consistent functioning of the enterprise. This
approach also requires careful investigation of the ways
of adapting and optimizing adaptation algorithms [7].
This paper deals with the attempts to evaluate
possibilities to adapt some foreign language oriented
speech engines for the development of Lithuanian
spoken language based telephony services. Chapters 2
and 3 briefly introduces with the modern call control
technologies used in our applications. Chapter 4
presents the results of experiments carried out to
optimize performance of some Lithuanian speech
recognition using telephony services.
Abstract: Paper deals with voice controlled telephony

services. Main speech server and TAPI 3.0 telephony
control activities are reviewed. Microsoft speech server
based voice control prototype telephony application
interactive voice controlled phonebook is described.
Experiments on Lithuanian names and surnames
recognition are offered.
Keywords: telephony
commands recognition.
control,
Lithuanian
voice
1. Introduction
Speech based interfaces for the control of various home
appliances and as a basic modality for the control of
telephony based services has enormous potential and
advantages. In recent years we saw significant number
of new such type services and prototypes presented and
developed in various countries and in various fields.
There were two important factors that influenced
development and caused growing number of
applications: appearance of well established and easily
available telephony application program interface
(TAPI) and tools for development of such type
interfaces and even more importantly appearance of
speech application programming interfaces (SAPI) with
appropriate speech engines. Such speech recognition
engines or engines integrated into Microsoft Speech
Server (OCS-SS) enables to achieve highly accurate
speaker independent recognition of speech commands
necessary for the control of various speech based
telephony services.
Speech recognition engines have one very important
factor: they are not language independent and depend on
the language spoken by the users of the system [1].
Microsoft and other major speech technologies
providers develop and market speech engines for
popular languages having many native speakers
(English, Spanish, French, etc.). Many other less widely
used languages remains out of the scope of interest for
the major speech recognition solution providers.
In such situation businesses and state institutions, in
countries were such less popular languages are used as a
2. Call control technologies for Windows

This chapter introduces with the telephony application
programming interfaces (TAPI) that were used in our
experiments and for the development of Lithuanian
speech programming interfaces.
Telephony Application Programming Interface (TAPI)
3.0 is an evolutionary API providing convergence of
48
both traditional PSTN telephony and VoIP telephony.

TAPI 3.0 supports standards-based H.323 conferencing
and VoIP multicast conferencing.
Similar to TAPI 3.0, Speech Server supports VoIP for
telephone calls over the Internet, using Session Initiation
Protocol (SIP).
There are four major call control components to TAPI
3.0:
TAPI 3.0 COM API - this API is implemented
as a suite of COM objects. Moving TAPI to the
COM model allows component upgrades of
TAPI features. It also allows developers to
write TAPI-enabled applications in any
language.
TAPI Server - this process abstracts the TSPI
(TAPI Service Provider Interface) from TAPI
3.0 and TAPI 2.1, allowing TAPI 2.1
Telephony Service Providers to be used with
TAPI 3.0, maintaining the internal state of
TAPI.
Telephony Service Providers - (TSPs) are
responsible for resolving the protocolindependent call model of TAPI into protocolspecific call-control mechanisms. TAPI 3.0
provides backward compatibility with TAPI
2.1 TSPs.
Media Stream Providers - TAPI 3.0 provides a
uniform way to access the media streams in a
call, supporting the DirectShowTM API as the
primary media-stream handler.
In the OCS-SS case the following activities allow the
developer to control call connectivity:
AnswerCall - Answers a call from a telephony
device using the Computer Supported
Telecommunications Applications (CSTA)
AnswerCall service.
DisconnectCall - Disconnects a call using the
Computer Supported Telecommunications
Applications (CSTA) ClearConnection service.
When the DisconnectCall control completes its
operation successfully, the Speech Platform
terminates the dialog.
MakeCall - Initiates a telephone call using the
Computer Supported Telecommunications
Applications (CSTA) MakeCall service. The
MakeCall control initiates an outbound call to
the specific number on the telephony device
when activated by RunSpeech.
BlindTransfer used to initiate a blind call
transfer.
DetectAnsweringMachine allows detecting
answering machines.
DeclineCall refuses the incoming call.
ISDN lines. Then it must enumerate the supported

address types of each Address object (for example, a
phone number, IP address), choose an Address object,
based on queries for support for appropriate media
(audio, video, etc.) and address types, use the CreateCall
method of the Address object to create a Call object
associated with a particular address. Then it must select
the appropriate Terminals on the Call object and finally
call the Connect method of the Call object to place the
call.
In order to answer a call the client similarly must create
and initialize a TAPI object, to use the TAPI object to
enumerate all available Address objects on a computer,
to enumerate the supported address types of each
Address object, to choose an Address object, based on
queries for support of appropriate media and address
types, to register an interest in the specific media types
with the appropriate Address object, to register a call
event handler with the Address object. TAPI notifies the
application of a new call through ITCallNotification and
creation of a Call object, to select the appropriate
Terminals on the Call object and finally to process
incoming call using the Answer method of the Call
object to answer the call.
In the case if Speech Server registers client application
with an SIP server then SIP server is giving certain data,
such as its location, IP address, and the protocol it
supports. When a user wants to place a call to another
registered user, an INVITE message is sent to the
registered user, based on the data that client gave the
SIP server when it registered. Once the client accepts
the INVITE, the two clients can communicate. The
reason for the INVITE is that a client can be registered
at multiple locations, such as a mobile device and a
computer. The first location which accepted the INVITE
will receive subsequent messages. If a network does not
currently support VoIP protocols, the traditional
telephony setup may be used: a telephony card and the
associated Telephone Interface Manager (TIM)
software.
Speech
Server
provides
backward
compatibility via the Telephony Interface Service (TIS),
which serves as an interface between TIM and the
telephony application proxy (TAP). The TAP acts very
much like a SIP redirect server, as it interprets requests
to and from the TIS and other SIP peers, and routes
them to the appropriate server.
Speech server allows direct combining of call controls
with proprietary voice input and output processing
activities. For TAPI 3.0 speech controls (i.e.
system.speech namespace) from .NET framework 3.0
must be used.
The usage scenario of the interactive voice controlled
phonebook (Fig. 1) application was developed to imitate
phone book directory service by most popular
information access ways telephony calls using regular
GSM phone.
After the user calls the application on speech server, the
telephony answerCall activity is started and the dialog
workflow is initiated. At this stage the telephone user
needs to ask a person to give by voice some information
(to tell the last and/or first name of the person he is
3. Call control in speech processing system

In order to place a call in TAPI 3.0 the client has to
create and initialize a TAPI object, then use the TAPI
object to enumerate all available Address objects on a
computer (for example, network cards, modems, and
49
looking for) or to enter the last name by letters using

phone keypad (letters abc=2, def=3 and so on, i.e.
maskeliunas=6275354627)).
4. Experiments
The prototype of Lithuanian speech recognition system
for telephony services aims to recognize a set of
Lithuanian names and family names. The examples of
such services could be the phone directory of small or
medium enterprise or information service aimed to the
companys personnel. Since the variety of names and
particularly family names is very big weve limited the
number to 100 names and surnames.
4.1. Speech corpora
For the Lithuanian first and family names speech
database 40 different speakers uttered each name and
surname once. The total number of combinations first
name + family name was equal to 100 in this case (4000
utterances total). Two speech databases were applied in
the experiments: the initial corpora and the corrected
corpora. Corrected database was freed from various
inadequacies and mistakes that were present in the initial
database. Between inadequacies and mistakes were the
pronunciation errors. Most of such pronunciation errors
were situations when speaker used other phoneme than
the phoneme present in the original family name, but
still getting grammatically correct and often really
existing name (in example, speaker said Pojarskaite
instead of Bojarskaite or Laimis instead of Laimius).
Other errors were related with such problems as a
stammer near family name or some technical spoilage
such as a truncated part of the word (most often at the
end of the name the error of automatic segmentation
system).
4.2. Lithuanian
experiments
Fig. 1. Interactive voice controlled phonebook visual

studio workflow window
proper
names
recognition
The experiments were carried out using the 100

Lithuanian names data. In these experiments we used the
transcription formation methodology proposed in [3].
Microsoft SAPI 5.3 based English speech recognition
engines were used as a basis for the adaptation.
The experiments were performed for the male and
female speakers separately and also for all speakers
together.
Before going to the deeper analysis of experimental
evidence let us introduce several definitions used
further. The following abbreviations are used in tables
of results:
C Correctly recognized voice commands;
U Unintelligible voice commands;
S Substitution;
O Omission;
I Initial speech corpora version;
V Verified speech corpora version.
Table 1 show the results obtained in the experiments.
Last row in the table presents results obtained using
corrected speech corpora.
Based on the results the male speakers recognition
accuracy was slightly higher than females speakers.
Both voice and DTMF recognition grammars are

controlled using speech QuestionAnswer activity,
utilizing W3C standard based grammars. Lithuanian
language recognition is possible using cross lingual
language models [8]. If the user said only the last name,
and there were found more than one alternatives, n-best
search algorithm is fired, system offers the best possible
choices and asks to choose the correct one at any time
(i.e. user may say this is the one Im looking for, this
one, stop, etc. or press # key). Once the person
searched has been determined, workflow initiates code
activity ieskau_adresato and searches for the details of
that person (phone number in this case) in the SQL
database. Once the required data is fetched, the direct
transfer can be initiated (in this case the Speech Server
issues a command to a telephony gateway to dial the
associated number and transfer the call (using a
telephony blindTransfer activity). After a successful
transfer, telephony disconnectCall activity starts and
dialog is stopped and switched to the ready state for
another caller.
50
One of the most interesting observations was that the

recognition accuracy for the initial and corrected
databases was almost equal which show, that a quite
robust recognition system for the small pronunciation
errors may be developed (rather probable in
applications, where we need to recognize people
names). The biggest number of errors was the
unintelligible voice commands. The omissions were the
least often observed type of errors. Substitution errors
are the worst type of errors, because the system accepts
wrongly recognized word as the correct one.
Table 3. Examples of errors that speaker RYTMIK made in

the initial corpora
speaker RYTMIK
Should be
Adamonyt Asta
Aliauskien Jadvyga
Anereviit Giedr
Andziulyt Loreta
Averjanovien Violeta
Bartukait Aura
Bilat Eugeniu
Binkeviius Ramnas
Bojarskait Egl
Brzga Juozas
Buinskas Artras
Bulachien Danguol
Table 1. Experiment results using verified and initial speech

corpora
Corpora
Version
I
I
I
V
Speakers
20 males
20 females
all 40
speakers
all 40
speakers
C, %
89.2
91.2
90.1
90.8
Recognition results
U, % S, %
6.8
3.8
4.5
3.9
5.7
3.8
5.4
O, %
0.2
0.4
0.4
3.6
Butkus Valdas
Daugla Andrius
Domarkien Inga
Diaugys Audrius
Giedraitis Laimius
Gudas Audrius
Ignataviius Gintaras
Ivanceviit Erika
0.2
Detailed error analysis (Table 2, 3) showed, that speaker

RYTMIK made as much as 19 errors (he mixed
(pronounced wrongly) at least one phoneme in name or
surname while reading names), but all words were
recognized correctly.
4.3. Analysis of errors

In order to improve the recognizer performance and to
find the ways to optimize the adaptation procedure, the
detailed error analysis was done.
There were 300 substitution and insertion errors (126
substitutions and 174 insertions) for 40 speakers (4000
phrases) and it was natural to expect that not all names
will produce the equal number of errors. Table 4 shows
the 5 names that produced the largest number of errors
in the experiments.
The 5 most confusing names produced almost 40% of
all substitution errors and slightly more than 35% of all
insertion errors. So the concentration of errors is big
and more attention to the names that resulted in larger
amounts of errors is necessary.
Table 2. The error list that speaker RYTMIK made in the

initial corpora
speaker RYTMIK
Initial corpora
Verified corpora
U Aleksandraviit Ramut
Biaigo Sandra
Drungilien Gintar
Recognized
as
Gedutis
Eugenijus
Butkus
Valdas
Dutkeviit Ieva
Should be
Giedra
Nerijus
Gudas
Audrius
Ivanceviit
Erika
Corpora error
Adomonyt Asta
Aliaustien Jadvyga
Reading in syllables
Andiulyt Loreta
Averjovien Violeta
Bartu.kait Aura
Bilat Audeniu
Binnkeviius Ramnas
Pojarskait Egl
Background noise, pant
Buinskas Arnas
Buloctien
Danguol,
background noise
background noise
Daudla Andrius
Domartien Inga
Diaugys Andrius
Giedraitis Laimis
background noise
First letter cut
...aneviit Erika
Aleksandraviit Ramut
Anereviit Giedr
Andziulyt Loreta
Bartukait Aura
Biaigo Sandra
Bojarskait Egl
Bulaxien Danguol
Domarkien Inga
Drungilien Gintar
Recognized Should be
as
Gedutis
Giedra
Eugenijus
Nerijus
Butkus
Gudas
Valdas
Audrius
Table 4. Five names that result in the largest number of the

recognition errors in the 100 Lithuanian names recognition
experiment
Substitution errors (S)
Name
Rec.
errors
Gudas
17
Audrius
Baublys Algis
12
Biaigo Sandra
6
This wasnt a clear result, as the names list has a lot of

phonetically correct names.
In the original corpora the recognition accuracy for
speaker RYTMIK was 94 % (C = 94), ASR
unrecognized 3 phrases (U = 3), and substituted 3
phrases (S = 3).
There were 20 various mistakes in the RYTMIK
corpora. Using the verified corpora the 89 % (C = 89)
recognition accuracy was achieved (much lower than for
unverified corpora), ASR system unrecognized 9
phrases (U = 9), substituted 2 phrases (S = 2).
Balcius
Ernestas
Gailiunas
Rytis
Unintelligible errors (U)

Name
Rec.
errors
Biaigo Sandra
16
16
11
Dolgij Andrej
Grigonis
Audrius
Baublys Algis
Giedra Nerijus
10
Detailed view to the most confusing names in these

experiments showed that most of those names dont
have particularly difficult phonetic structure (the family
name Biaigo may be treated as the more difficult case).
51
The bigger number of errors obtained by the name

Gailiunas Rytis may be explained by the presence of the
name Gailiunas Vytautas in the same list. But most of
those errors cant be explained straightforwardly.
For example, the name Gudas often was confused with
the name Butkus. Such situation shows, that theres still
plenty of room for the optimization and further
investigation and the better recognition accuracy is
obtainable.
For the name Beliukeviciute 27 transcription variations

were recognized for the first speaker and 21
transcription variations were recognized for the second
speaker. It is worth noting that in both cases the first
speaker tended to activate more different transcriptions
than the second.
Even more important observation is that the most often
recognized (as the most similar) transcriptions arent the
same for the different speakers. For example, for the
first speaker the most often used transcriptions for the
word Beliukeviciute were transcription variations 111th,
99th, 15th and 3d. At the same time for second speaker
the most often used transcriptions were the transcription
variations 505th, 121st, 504th and 507th. Similar situation
has been observed for the name Varanauskas.
These results allowed drawing conclusion that the usage
of multiple transcriptions in the adaptation of the foreign
language speech recognition engine to the Lithuanian
language is a reasonable step and it is worth further
investigation and implementation in the practical
applications.
4.4. Experiments using multiple transcriptions for

the same word
In order to improve recognition accuracy we tried to use
multiple transcriptions instead of single transcription for
the same Lithuanian word. The idea here was that for
different speakers, different transcriptions will be better
suited and will allow achievement of higher overall
recognition accuracy. This assumption is based on the
belief that there is no single mapping from a phonemic
units space in one language to the phonemic units space
in another language, thatll be equally efficient for all
speakers.
In this experiment the two family names from the list of
100 Lithuanian names Beliukeviciute and Varanauskas
were selected. These family names are longer and
have more phonemes in their phonemic structure and
allow generating more phonetic transcriptions that will
be sensible and may produce valuable result during the
recognition. In the case of family name Beliukeviciute
1152 transcriptions were generated for this experiment
and for the family name Varanauskas 188 transcriptions
were obtained. Then the two speakers pronounced each
of the family names 100 times and the recognition
system was coded to select which of the transcription is
the most likely for that speaker and that name. Table 6
shows the four most popular transcriptions for each of
the speakers investigated and could serve as a good
basis for a further development of the Lithuanian foreign
language based speech recognizer adaptation system.
One of the observations that could be seen from the data
in the Table 5 is that a big number of transcription
variations were recognized as the most similar ones for
each of the speaker.
4.5. Experiments using noisy and clipped speech

The following experiments were carried out using clean
and noisy or clipped speech. The aim was to evaluate
the robustness of recognizer for the speech signal
distortions. 10 speakers participated in this experiment
and several SNR levels were selected and clipping
coefficients were used (using the algorithms below).
Signal to Noise Ratio (SNR) has been measured using
formula (1):
SNR 20 lg
smax
,
nmax
(1)
Clipping level has been obtained using formula (2):

sc smax ;
(2)
where:
smax max(abs ( si )) ,
i1, N
(3)
sic itself has been determined using
Clipped signal
equations (4):
s,
sic c
sc ,
if
si sc
if
si sc
Table 5. The four most frequently recognized transcription for the two speaker uttering 2 Lithuanian names
Name
Beliukeviit
speaker
ALGRUD
transcr.
occur.
111
23
99
20
15
13
3
10
Varanauskas
speaker
RYTMAS
transcr.
occur.
505
18
121
12
504
11
507
11
ALGRUD
transcr.
occur.
67
24
130
11
4
8
70
8
52
RYTMAS
transcr.
occur.
19
10
166
8
144
8
6
8
(4)
Table 6 shows the results obtained in those experiments.
The usage of multiple transcriptions in the adaptation of

the foreign language speech recognition engine to the
Lithuanian language is a reasonable step and it is worth
further investigation and implementation in the practical
applications.
The robustness of the speech recognizer used the
robustness was verified using the distorted speech
corpora. The performance of the recognizer began to
deteriorate significantly when the SNR level dropped
below 30 dB and was in principle unacceptable at the
SNR levels below 20 dB. Using the high clipping
coefficient ( = 0.1) resulted the significant drop in
recognition accuracy.
Table 6. Recognition accuracy (10 speakers) of 100

Lithuanian names using different quality of speech corpora
Corpora
unmodified
SNR 40 dB
SNR 30 dB
SNR 20 dB
Clipping 0,3
Clipping 0,1
C, %
96.2
95.8
91.1
43.0
92.5
81.0
Recognition results
U, %
S, %
O, %
1.3
2.3
0.2
1.8
2.3
0.1
1.9
6.9
0.1
12.0
30.4
14.6
5
2.5
0.0
13.2
4.8
1.0
The performance of the recognizer began to deteriorate

significantly when the SNR level dropped below 30 dB
and was in principle unacceptable at the SNR levels
below 20 dB. So the performance cant be treated as
robust one looking from the SNR variations point of
view. Using the clipping coefficient 0.3 the
recognizer
performance
dropped
relatively
insignificantly while the clipping coefficient 0.1
resulted in the much bigger loss in accuracy.
6. References
1. Fung, P., Schultz, T. Multilingual spoken language
processing. In: IEEE Signal Processing Magazine,
Vol. 25, Issue 3, pp. 89-97 (2008).
2. Zgank A., et al. The COST278 MASPER initiative
crosslingual speech recognition with large telephone
databases. In: Proc. of 4th International Conference
on Language Resources and Evaluation LREC04,
pp. 2107 2110, Lisbon (2004).
3. Kasparaitis P. Lithuanian Speech Recognition Using
the English Recognizer. In: INFORMATICA 2008,
Vol. 19, No 4, pp. 505516, Institute of Mathematics
and Informatics, Vilnius (2008).
4. Beyerlein, P., et.al. Towards Language Independent
Acoustic Modeling, In: Proc. of IEEE ICASSP00,
Vol. 2, pp. II1029-II1032, Istanbul (2000).
5. Lindberg B., et al. Noise Robust Multilingual
Reference recognizer Based on Speech Dat. In: Proc.
Of ICSLP 2000, vol. 3, pp. 370373, Beijing (2000).
6. Villaseor-Pineda L., et. al. Toward Acoustic
Models for Languages with Limited Linguistic
Resources, In: CICLing-2005, Lecture Notes in
Computer Science, Vol. 3406/2005, pp. 433-436,
Springer-Verlag (2005).
7. Schultz T. and Kirchhoff, K. (Eds.), Multilingual
Speech Processing, Academic Press, 2006.
8. Maskeliunas R. Modeling Aspects of Multimodal
Lithuanian Human - Machine Interface. A. Esposito
et al. (Eds.): Multimodal Signals, LNAI 5398, pp.
7582, Springer-Verlag Berlin, Heidelberg (2009).
5. Conclusions
Speech recognition engines are not language
independent and depend on the language spoken by the
users of the system. Two major ways for Lithuanian
speech recognition solution are: a) developing own
speech recognition engine from scratch; b) adapting the
foreign language based engine for the recognition of the
native language.
The application of the English based engine for the 100
Lithuanian names recognition allowed achieving
recognition accuracy of more than 90 %. These results
show that the implementation of longer commands and
transcription generation method proposed by [3] were
confirmed.
The 5 most confusing names produced almost 40 % of
all substitution errors and slightly more than 35 % of all
insertion errors. Most of those names dont have
particularly difficult phonetic structure (the family name
Biaigo may be treated as the more difficult case). The
bigger number of errors obtained by the name Gailiunas
Rytis may be explained by the presence of the name
Gailiunas Vytautas in the same list. But most of those
errors cant be explained straightforwardly. For
example, the name Gudas often was confused with the
name Butkus. Such situation shows, that theres still
plenty of room for the optimization and further
investigation and the better recognition accuracy is
obtainable.
The most often recognized (as the most similar)
transcriptions arent the same for the different speakers.
A big number of transcriptions were recognized as the
most similar ones for each of the speaker (27
transcription variations of the name Beliukeviciute 27
for the first speaker and 21 for the second speaker).
Each speaker tended to activate more different
transcriptions than the second.
53

DATA ACQUISITION IN OPC-BASED INDUSTRIAL IT SYSTEMS

Roman KWIECIE, Elbieta SZYCHTA, Leszek SZYCHTA
fundamental purpose of this system is to control

equipment to implement production objectives and to
collect information which will serve to utilize an
appropriately managed IT system in the decision-making
process.
Abstract: This article illustrates utility of data

transmission as part of the manufacturing process when
OPC technology is used. It presents results of research
of loading of Ethernet for the purpose of verification of
manner of transmission of process data in this
technology. The results also verify that current standard
of OPC technology is a good solution.
Client
Keywords: OPC technology, OPC Foundation, OPC

specification, COM, DCOM, OPC UA, SOAP, XML
Client
Client
1. Introduction
Client
Client
Ethernet
SERVER
Data acquisition in industrial IT systems is the key link

in management of a manufacturing company [1, 7]. Data
obtained from the lowest strata of the engineering
process lead to appropriate decision making in higher
strata, which should enhance effectiveness of
manufacturing and improve product quality. However,
development of the information technologies increases
information flows across the entire IT system of an
enterprise.
MES
Systems
SCADA/HMI
Systems
PLC/IPC
PLC
Fielbus
ERP
Systems
Fielbus
Industrial
devices
Fig. 1. Model of IT system infrastructure representing the

engineering process
Industrial automatics IT systems can be divided into two

subsystems [1], creating a model that consists of two
pyramids (Fig. 1):
manufacturing subsystem (bottom pyramid), where
the supervisory unit (server) reads processing data
(from industrial automatics equipment [4]) and
gathers them in an industrial database. The data
become a component of the information describing
functionality of an enterprise.
management subsystem (top pyramid, oriented
towards the bottom one), where the supervisory
units of the highest production management strata
play the main role. Using the data in the industrial
database and their engineering and economic
analysis, decisions are made to adequately
represent functioning of the engineering process
control systems.
Servers, linking two subsystems in the information
sharing process, are key elements of the IT system. They
supply process data to clients in accordance with a
clearly defined communication Ethernet protocol. OPC
2. Model of IT system infrastructure which

represents the engineering process
A growing trend can be observed in industrial
automatics towards building real-time systems [8].
These become increasingly important in the process of
information sharing. Production process management
systems are implemented with a view to information
sharing systems, for instance (Fig. 1):
ERP (Enterprise Resource Planning),
MES (Manufacturing Execution Systems),
SCADA
(Supervisory
Control
and
Data
Acquisition).
At lower echelons of the engineering process, higherlevel control and data acquisition SCADA, industrial
CNC
(Computer
Numerical
Control),
PLC
(Programmable Logic Controller), IPC (Industrial PC),
systems, sensors, actuators, and other industrial
automatic equipment are employed. This equipment and
its software build a real-time IT system. The
54
(OLE for Process Control) technology is the current

information sharing standard in the client-server scheme
of industrial automatics equipment [6].
specification, only current process data can be viewed or

changed.
Due to complexity of processes executed by OPC
server, process data are logically divided into groups
(Fig. 2). The process data in these groups are
characterized by various scan times, reading modes, etc.
Depending on OPC DA version, two data readout modes
are possible:
synchronous readout always occurs at identical
time intervals,
asynchronous readout occurs when certain data
change levels can be defined for the readout to
occur when they have been exceeded.
Access to data by means of OPC DA can be gained in
three ways:
using COM/DCOM [2],
using XML (eXtensible Markup Language) and
SOAP (Simple Object Access Protocol),
via .NET Remoting technology, offering broader
capability than DCOM (supporting different
formats and communication protocols, easy
communication by the Internet).
OPC DA comes in a number of versions, with 3.0. being
the latest (each version provides another range of
interfaces, though reverse compatibility should be
maintained).
OPC specifications define communication in the clientserver scheme as vertical communication. On the basis
of defined OPC DA specification, horizontal
communication can be executed between the OPC
servers (Fig. 3). This is implemented by an intermediary
application called DX (Data eXchange).
3. Technique of OPC information transmission

Work on a standard of communication among industrial
equipment led OPC Task Force to develop, in 1996, the
first OPC specification, which defined access to process
data. Since then, activities have been coordinated to
maintain and publish new OPC specifications. In effect,
the following specifications are currently available [3]:
OPC Data Access (OPC DA) provides real-time
access to current process data.
OPC Historical Data Access (OPC HDA)
provides for reviewing and analysis of collected
historical data, e.g. to assess system efficiency or
anticipate errors. A client gains access to archive
data (readings of a device, etc.) by filing enquiries
with the OPC HDA server.
OPC Alarms & Events (OPC A&E) informs of
events and alarms in the system. An alarm is
defined as an abnormal condition of an object
which requires special attention. An event may be
related to a condition, e.g. a transition of a given
value to an alarm condition, or not related to
a condition, like configuration changes or system
errors.
OPC Security provides security of access to data
offered by OPC servers. It enables correct
verification of a client wishing to gain access and of
correct transmission (whether data have not been
altered).
OPC Batch required in batch management.
OPC and XML integrating OPC and XML
(eXtensible Markup Language) to build Internet
applications.
OPC Unified Architecture enables communication
between different operating systems via SOAP
(Simple Object Access Protocol a protocol for
calling remote access to objects using XML).
OPC DA specification provides access to a single
process datum with reading or recording capability, each
having a value, timestamp, type, and quality (Fig. 2).
DA
DA
DX
SERVER
SERVER
DA
DA
CLIENT
Fig. 3. Communication in the server-server scheme of OPC

technology
4. Testing of network's load-carrying capacity

OPC
Client
OPC DA
OPC
SERVER
Communication
protocol
Computer
network
The system shown in Fig. 4 is employed to test loadcarrying capacity of a computer network using OPC.
PC1 and PC2 computers are connected to Access Point
AP TP-LINK, TL-WR543G. Access Point acts as a link
in the process of information sharing among computers
in a LAN following TCP/IP protocol.
OPC server: "Matrikon Simulation Server" process
data simulator - is installed on PC1. Readout of one type
of process datum 4-bite integer is assumed, as shown
in Fig. 5. Software of OPC client, created by Delphi
compiler of Borland, educational version, is installed on
PC2. OPC "Sentrol" components are used to create the
application, accelerating its construction at the stage of
design and sending orders to OPC server in the
OPC Groups (Collection)

OPC Group
OPC Items (Collection)
OPC Browser
Value
OPC Item
Timestamp
Quality
Fig. 2. Architecture of access to process variables via OPC

DA
A device or OPC server can generate the timestamp (if

the device does not have this capability). Using this
55
executive mode. The author's own application provides

for automated creation of any quantities of process data.
Characteristic curves in IEEE 8002.3

5000
4500
AP
4000
[kbps]
3500
PC 1
PC 2
3000
2500
2000
1500
1000
Fig. 4. Diagram for testing load-carrying capacity of a

computer network using OPC
AP Access Point AP TP-LINK, TL-WR543G; PC1
Notebook HP compaq nx9020 (1,30GHz, 736MB RAM, card
Realtek RTL8139 Family PCI Fast Ethernet NIC 100Mbps);
PC2 Notebook Toshiba 1410 (1,80GHz, 256MB RAM, card
Intel(R) PRO/100 VE Network Connection 100Mbps)
500
0
0
2000
4000
6000
8000
10000
12000
Item s/s
Dow nload DL
Upload UL
OPC DL+UL
Fig. 7. Characteristic curves of IEEE 8002.3 load-carrying

capacity using OPC
5. Analysis of LAN load-carrying capacity when data

are transmitted by means of database server
protocol
Cyclical loading of process data can be affected by
means of a communication protocol, which is currently
utilised between a client and database server. Data are
tabulated in this protocol. In respect of a process datum,
three fields (Fig. 2) can be distinguished, similar to
those found in OPC:
Value, depending on the data type it is 4 bytes for
Int4,
Timestamp 8 bytes,
Quality 1 byte.
Each table must have a unique name of maximum 16
bites. In consideration of the 8-bytes protocol header,
the size of 4-bytes data can be formulated:
Fig. 5. Selection of process datum type of 4-bite integer Int4
N 24 13 n ,
(1)
where:
N the size of 4-bytes communication protocol data,
n quantity of process data,
24 protocol header including the table name = 8 +
16 = 24 bytes,
13 value + quality + timestamp = 4 + 1 + 8 = 13
bytes.
Fig. 6. LAN load-carrying capacity when +500, +500, +500,

+500, and +1000 of process data are initiated, respectively
Theoretical loading of Ethernet can be determined

according to the formula. Given this value as dependent
on the quantity of process data, the network's loadcarrying capacity is tested when data are loaded as
tabularic datasets. The characteristic curve in
comparison to OPC technology is shown in Fig. 8.
CIM (Crystal Internet Meter) is used to measure loadcarrying capacity of the network. This software enables
to measure quantities of UL (Upload) and DL
(Download). The measurement is possible in KB/s
(kilobyte per second) and kbps (kilobit per second). A
process of LAN load-carrying capacity is shown in
Fig. 6.
The measurements are carried out for several different
loads of process data read out. The measurement results
are presented in Fig. 7.
56
In addition, according to the objective of this paper, it

has been established that the method of process data
transmission using OPC technology is not the optimum
solution. Primarily, it overloads a computer network
approximately four times over and above the actual
quantity of loaded data (when the protocol of tabularic
data transmission is utilised). This is due to the fact that,
for purposes of client OPC server communication,
DCOM servers are employed, which additionally
include data needed only for DCOM technology in their
communication protocol structure. Globally, it increases
electricity consumption, which is in direct proportion to
the continually increasing quantity of data loaded in
Ethernet [5].
Com parative characteristic curves in IEEE 8002.3

5000
4500
4000
[kbps]
3500
3000
2500
2000
1500
1000
500
0
0
2000
4000
6000
8000
10000
12000
Item s/s
OPC DL+UL
Measured DataSet
7. References
Counted DataSet
1.
Fig. 8. Comparative characteristic curves of LAN loadcarrying capacity for OPC and information transmission by
means of DataSet technology
2.
3.
6. Conclusion
OPC standard is based on Microsoft-developed COM

(Component Object Model) and DCOM (Distributed
Component Object Model) technologies. These
technologies are a kind of communication platforms
which define the way OPC applications communicate
with one another and share process data. Unfortunately,
when OPC applications are configured in network
operating
conditions,
certain
drawbacks
of
COM/DCOM based communication become apparent,
namely:
difficulties with configuring the connection,
problems with establishing the connection,
problems with maintaining the client server
connection.
In extreme cases, sharing of process data may be
interrupted and the manufacturing process may be
stopped.
4.
5.
6.
7.
8.
57
Chrupek R. Data Acquisition in Industrial Systems.

(In Polish), Napdy i sterowanie, issue 4, April
2008.
DCOM: http://www.microsoft.com/com/
Iwanitz F.,
Lange J.
OPC:
Fundamentals,
Implementation and Application. Huthig Verlag,
Heidelberg, Germany, 2002.
Luft M., Cio R. Increase of Accuracy of
Measurement Signals Reading From Analog
Measuring Transducers. Silesian University of
Technology Scientific Publications 2005, Transport
vol. 59 (1691), Gliwice, 2005.
Moore G.
Energy-Efficient
Ethernet,
IEEE
Spectrum, Int 9, May 2008.
OPC: http://www.opcfoundation.org
Skura K. Issues of IT System Integration in
Automation of Manufacturing Processes Based on
OPC Technology. (In Polish), Napdy i sterowanie,
issue 10, October 2007.
Uasiewicz J. Real Time systems QNX6 Neutrino.
(In Polish), BTC, Warsaw, 2007, ISBN 978-8360233-27-6.

SYNTHESIS OF PROGRAM MODULES FOR TYPICAL BOILER-HOUSE

APPLICATIONS
Gediminas DANILEVIIUS, Darius EZERSKIS
Abstract: A Paper suggests automated control program

synthesis for typical boiler-house of small to medium
power. Due to typical boiler-house architecture even
part of control algorithms for the equipment or overall
program structure can be ready made. This paper
presents the way to build those control algorithms and
automated generation of the control program for
programmable logic controller. Process signal coding,
special PLC templates and meta-program for the boilerhouse control program is also presented.
2. Finding of typical control algorithms

2.1. Decomposition of a process
Boiler-house is a complex of equipment, where are the
boiler for heat generating, and auxiliary technological
equipment for the heat supply [1]. Small and medium
power water heating boiler-houses can be defined as a
technological unit.
Keywords: Boiler-house, meta-program, PLC

programming.
Inlet water
A significant part of control problems are similar for

small to medium boiler-house applications. For
instance, control solutions for boiler control and
protection, heating network pressure ant temperature
control, supplement water supply into network are
general for any boiler-house. Therefore those typical
control algorithms can be well documented and like
framework used for program building in such
applications.
Similarly a structure of control program, especially
using programming systems conform to programming
standard IEC 61131-3 [4], do not depend on a specific
boiler-house. Moreover due to local regulations for the
boiler-houses they have similar architecture of the
technological equipment. This makes it possible to build
a program structure, which can be used as core
structure.
In the following chapter typical algorithms are
discussed and the implementation of a network pump
algorithm is presented. Those algorithms will be used as
basic algorithms for the automated generation of the
PLC program for the boiler-house control.
Essential steps for automated program synthesis as
process signal coding in order to order them for right
algorithms to meta-program, which builds PLC program
code, are presented in following chapters.
TP
NP2
MV1
Outlet water
1. Introduction
NP1
PIC
TT
TIC
B1
B2
Supplement water
Gas
Air
NS1
NS2
Fig. 1. Typical boiler-house technological scheme
To better understand the system, it is convenient to

expand a technological unit into components technological branches [2]. Technological unit is
decomposed into technological branches according to
technological or energy transfer operations.
There are several typical technological branches in a
boiler-house from Fig.1:
Fuel tract that consists of gas, air supply equipment.
Heat generation tract that consists of boilers B1 and
B2.
Heated agent tract that consists of water network
equipment e.g. network pumps NP1 and NP2,
temperature control valve MV1.
Network supplementation tract, that consists of
supplementation pumps NS1 and NS2.
Every technological branch consists of a group of
devices which are typical in control terms. Its best to
58
group technological devices according to their design

purpose, for example:
Valves: positional, discontinuous and continuous
control.
Motors (pumps, fans); discontinuous and continuous
control.
Control loops.
Typical control algorithms as function blocks for those
technological devices can be designed. Water heating
boiler [1] builds normally an isolated control system
with well specified algorithm, interfaces for status
supervision and mode control.
the water pressure is higher as the max specified

pressure, as presented in Fig. 2.
Pressure
max
min
Q1
Q2
2.2. Specific requirements for a boiler-house control
T2 T2
T1 T1
Lithuanian requirements for boiler-house equipment

installation [3] regulates the automation requirements of
the boiler-house, which can be regarded as typical
control task for all small and medium power water
heating boiler-houses.
Power control task - ensuring the maintenance of the
basic operational boiler-house parameters, for example,
a supply water temperature to the district heating
network.
Device reservation task - ensuring the installed
performance of the boiler-house, for example, additional
network pump activation in case of faults.
Security task - the boiler-house technological scheme
and control equipment that guarantees secure and
automated processes operation.
User interface task - boiler operating parameters for
monitoring information system must show the values of
parameters and alarms deviations.
According to these control tasks, following control
algorithms are common for boiler-houses:
Boiler cascade control
Network pump control
Water pressure in a network control
Water temperature to network
In a following chapter one of these algorithms and its
implementation is presented in details.
Time
Fig. 2. The network pumps pressure control diagram
In this control diagram Q1 and Q2 are representing the

pump contactors status, T1 and T2 are the pumps
start-up and braking times. Start and stop times are
required to record the measured parameter changes
during this time.
Frequency
max
min
Q1
Q2
T2
T1
Time
Fig. 3. The network pumps frequency control diagram
If one of the pumps is controlled by frequency converter

and the pumping capacity can be controlled by analogue
signals, an algorithm should implement a closed loop
pressure control algorithm. Then the feedback for the
starting and stopping of other pumps is the first pump
rotational speed as presented in Fig. 3. Q1 and Q2 are
the pump contactors, T1 and T2 are the pumps startup and braking times that should be considered in the
algorithm.
2.3. Network pump control algorithm example

Two as presented in Fig. 1 or more network pumps are
normally installed in a boiler-house to ensure water
circulation reliability. Therefore set-point for network
pump operation is water pressure to boilers. In a general
case capacity of one pump is not enough for normal
boiler-house operation. Thus, the typical algorithm
should be designed for the control of at least two pumps
with easy formal extension capabilities for more pumps.
Pressure control algorithm depends on the pump control
equipment, signals and their type. If the pumps are
operated by circuit breakers, control system action to
the system can be only discrete signals to switch-on one
of the pumps according to the water pressure in the
heating network. System feedback is measured water
pressure to the network and status of each pump. In this
case the control algorithm should switch-on the pump if
water pressure less than min and switch-off the pump if
PV
M1
Reservation
Digital
Analogue
SP
Controller
M2
M3
Faults
Fig. 4. Pressure control algorithm block diagram
59
Q1
Q2
Q3
SP
Both algorithms are similar and can be combined into a

single control block as presented in Fig. 4. The control
block designed in object-oriented design manner [4].
Algorithm consists of a set of mutually cooperated
object each of them represents clearly declared and
isolated control task like closed loop controller, digital
signal evaluation, analogue signal evaluation, basic
motor control, reservation algorithms. These objects
have all possible connections and operations between
objects.
On the left side of the block diagram are control loop
inputs: measured pressure (P), the pump with a
frequency converter frequency (PV), set point parameter
(SP) and the external fault signals. The control block
combines the two previously described cases in which
the network pumps are controlled by the pump pressure,
or under the speed of rotation. Depending on what are
the input and output signal types, control block uses the
Digital or the Analogue and the Controller
control objects. The object named Reservation
controls pumps M1, M2 and M3 and ensure to
the election in case of fault. On the right side of the
block are control loop outputs: pump switching
contactors and frequency converters control signal - the
rotation speed or frequency (SP).
description of the equipment. This means that the design

of equipment must be described according to certain
rules. The issue: the real object signals must be
unambiguous encoded because of the meta-program
could respectively attribute them to the objects groups.
Such a decision should be known to the power industry.
The energy industry is often subject to KKS codification
system [5]. Based on this system all of the low average
power and water heating boiler equipment may be
specified as presented in Table 1.
Table 1. KKS codification system
0
G F0
A~N N
Main idea of program synthesis algorithms is to identify

signals at detailed design phase to which technological
branch they belong and order them to corresponding
function block inputs and outputs.
Work project
KKS codification
I/O list
MS Excel file (*.csv)
ASP
VBScrip
CoDeSys file (*.exp)
CoDeSys sub-program
F2
A
2
3
F3 Fn A1A2 An A3 B1B2 Bn
A NN AA NNN A AA NN
It is worthwhile to note that the level 0 and the level

1 first column F0 indicates the technological unit of
code, which during the event is not important. Level 1
second column F1F2F3 refers to the pipeline, but
during the event is important only in the last subscheme of F3, but the general case, the pipeline must
be used to describe the entire column, F1F2F3 code.
This code is quite accurately refers to the device
function, such as the pipeline network pumps boilers
always labelled as NDC code.
Level 2 is by far the most important in the present
proceeding, as describing the device type. For example,
using the VBScript text in the identification of all the
tools are selected, all records mentioning the column
F1F2F3 code, and then from the list drawn up by the
election of the engine or valve describing the level of
2 code.
Design software EPLAN has the possibility to export
of the work project KKS signal list, which contains the
information elements are separated by semicolons.
Useful information is deemed to be that which will
require automation of the sub-creation. Without this
information the next automated creation are impossible.
It is noted that each signal functional text also has a
strict form, where must be the signal KKS code and
signal function as a commentary.
3. Meta-program of the control algorithm
IIS
1
F1
A
FB
Typical FB
Fig. 5. Automated control algorithm functional scheme
Automated program synthesis process is presented in

Fig. 5. From work project at basic design stage, the
information will be exported into comma separated
value format to analysis program that generates
instances of typical function blocks and assigns
corresponding O/I channels to them. As the result a
program code for industrial PLC programming package
CoDeSys is generated.
Input-output data from the detailed design project are
uploaded in to the meta-program using project I/O list
export. In order to identify I/O signal exported data
must have the certain restrictive conditions to provide
complete information on the work project Fig. 5. The
main restrictive condition is a formal technical
Fig. 6. Data input form
Continuing the development of an automated subprogram exported KKS signal list must be examined.
For this purpose a prototype program using visual Basic
60
programming language is created. User interface and

file reading functions are implemented by using Internet
Information Service (IIS) platform based scripting
forms. Example is presented in the index.html file
form Fig. 6. In this form the user can specify the file and
then by clicking Continue button to carry out its
analysis.
Uploaded data analysis is performed in another file,
entitled submit.asp. There is all of the analysis scripts
called as a meta-program Fig. 7 in this file.
Meta-program analyses data file and split them into the
array. One column array consists of rows of data file
lines. Data array is filtered by removing the elements
that will not be used for further analysis.
The result as presented in Fig. 8 last line is formed in

the export file (which has extension *.exp). Otherwise if
export file can not be created because of bad input file
format the meta-program returns explanations of error.
File.csv
A=Filter(File, KKS)
A(n)
F
n > -1
FB(i)=C
Fig. 8. Data output form
Exported file has file format for CoDeSys programming

package and can be imported into this package. After
this import a program and basic functions for boilerhouse control are generated.
FB(i)=A(n)
6. Conclusions
File.exp
Analysis of boiler-house control algorithms offers

possibility to make them typical and implement as
function blocks. This analysis leads to typical program
for the boiler-house control too.
In order to assign input and output channels to
corresponding technological functions, widely used in
energetic KKS coding is used. The meta-program
analyses KKS code of the signal and orders them to
corresponding technological branch and unit. Finally the
instances for program modules and function blocks are
created and import file for industrial PLC programming
package is generated.
Fig. 7. Meta-program algorithm block scheme
Array is analyzed for KKS codes the global inputs and

outputs list. The latter list should be indicated in the
sub-logic of the variables used as unique names [5].
Each variable corresponding to the physical address,
KKS code, inputs and outputs addresses and
commentary.
Global input-output list is designed using the following
rules:
type (DI, DO, AI, AO);
number (01, 02, ..., 10, 11, ...);
name (KKS functional text);
physical address (for example: AT% IX62.0);
type of variable (BOOL, INT, WORD, ...);
comments (functional text).
After the signals are described with unique names and
addresses, sub-programmes can be designed. Subprogrammes consist of blocks of typical control
algorithms, such as a network pumps control block
which is described in the user library. This block was
made on the basis of a representative pressure control
equipment control algorithm.
In this case, in process of automation of control subprogrammes are possible to conclude the block inputs
and outputs to options, and assign them to the block
input and output.
7. References
1. Gimbutis G. Thermo-technology. (In Lithuanian),
Vilnius, 1993. 333 p.
2. Ezerskis D. Development and Application of
Programming Methodology for Automation Control
Systems, Based on Design Patterns. Dissertation,
Kaunas, 2006. 128 p.
3. LRS. Requirements for boiler-house equipment
installation. (In Lithuanian), Vilnius, 2006.
http://www3.lrs.lt/.
4. IEC. IEC 61131-3 Programmable controllers Part
3: Programming languages, Second edition 2003.
http://www.plcopen.org/.
5. VGB. Identification System for Power Stations
(KKS). http://www.vgb.org/.
61

DISTRIBUTED SUBSTATION UNITS CONTROL SYSTEM IN DISTRICT

HEATING NETWORK
Julius KALUEVIIUS*, Vytautas DEKSNYS*, Virgilijus PAMAKTIS**
*Kaunas University of Technology; **AB Axis Industries
3 empty lines using 10-point font with single spacing
3 empty lines using 10-point font with s3 empty lines using 10-point font with single spacing
data communication capabilities. Growing requirements
for safety and security, form new tasks for maintenance
and support personnel.
Growing energy prices and rising customers
requirements for quality of service, forces heat energy
and hot water suppliers to look for new ways of
lowering production costs while increasing quality of
service and reliability.
There has been inherited district heating network from
former Soviet times. All these energy supply networks
have their pros and cons that require constant
investments into renovation. Part of these plants are
already modernized and insulated, equipped with
automated control units responsible for heating
substation control and maintenance. Heating substation
control units are mainly equipped with powerful
controllers that realize building heating models and
places heating control in much higher comfort level.
1 empty line using 10-point font with single spacing
Abstract: Drastically rising energy production costs,

continuously increasing demand for environmentfriendly and reliable technologies require rational and
economic usage of energy recourses. This creates
objectives to develop an advanced metering and energy
control system.
Rising energy consumption could be reduced by
optimizing the energy supply chain. This can be done in
many ways: using real time control of building
environment (heating, air conditioning and ventilation),
increasing measurement accuracy, distributing heating
consumption network and optimizing substation units
control algorithms.
There is a list of automated building environment and
substation unit control devices that optimizes plant
energy consumption and ensures maximum comfort
level. While these devices are saving energy used for
single house heating they are misbalancing energy
transportation system. This effect can be easily seen
during peak period failures.
Misbalance of district heating network leads to tangible
energy losses and increased damage warning level. To
solve this problem, it is needed to control peak period
failures that could be avoided or their impact minimized
by using efficient and intelligent distributed control
system. It covers all substation units on the district
heating network.
Such distributed system equipped with intelligent load
management algorithms is able to minimize peak load
influence to the network.
Keywords: district heating, distributed control,
substation unit control.
1. Introduction
District heating and heat transportation network
characteristic to former Eastern block countries. The
Bigger majority of these plants were built before several
decades and at the moment are being renovated or
planned to be renovated. It is expected that all morally
and technically worn out network parts will be changed
into new ones meeting requirements formed by new
Fig. 1. Automated district heating substation module

The majority of automated substation controllers are
regulating building heating depending on air
temperature. All the devices in the district heating
network react simultaneously to any weather change.
62
process it (feeding heat carrier temperature) has to be

recalculated using following dependency [2]:
dt ( )
T3' = 3
+ t3 ( ) = K (1 ) t e ( )
d
,
(5)
dt ( )
+ t3 ( ) = K t e ( )
T3'' = 3
d
where windows and infiltration areas ratio with
outer thermal insulation area.
Coefficient K is calculated taking heat carrier
temperature change and air temperature change ratio:
This is the reason why peak load balancing and weather

anomalies became much more important than several
years ago.
Substation units automation has solved lots of
problems but there has emerged new ones that have to
be coped with in much more sophisticated way.
2. Heating regulation model of the building
Heating mode or regime of the building depends on
various climatic factors (such as air temperature, wind
speed and direction, solar radiation and similar), also
heat sources (heating system, people, home appliances
and other devices radiating heat). All these influences
(except heating systems) change randomly and have
different effect for diversity of living quarters [2].
According to the previous statement, heating system
should adjust feeding line energy separately for every
room, taking into account factors influencing heating
mode and ensuring stable temperature in controlled
area. In the majority of the houses it is not possible
while heating infrastructure is not suited for such
precise control. In these plants there are used heating
substations that control heat stream for the whole
building. The problem of room temperature stabilization
is solved by centralizing heating supply for the whole
building.
Using qualitative regulation model feeding line
temperature is controlled according to air temperature.
Heat carrier temperature fed to heating system
dependency upon air temperature expressed in the
following equation [2]:
(1)
t3 = tin + t A + t B ,
t +t
t t
t A = 3n 2n tin in e
2
tin ten
0.75
K=
t3
,
te
(6)

where t3 heat carrier temperature change in feeding
line, te air temperature change.
Coefficients T3 and T3 describes inertia of thermal
insulation. It is calculated using equation [2]:
C
,
(7)
T3 =
2 e + i + i R
where e return coefficient of outer thermal insulation,
R thermal resistance of outer insulation, C outer
thermal insulation heat receptivity coefficient.
Described mathematical model for heating regulation of
the building is suited for automated substation unit
control. It evaluates changing air temperature and
describes feeding as well as return line heat carrier
temperature recalculation and control modes.
Using this model there will be saved energy recourses in
the scope of one building but they can be wasted
viewing to the whole district heating network.
(2)
t t
2n tin te ,
(3)
t B = 3n

t
t
2
in en

where t3 heat carrier temperature, t3n heat carrier
temperature under standard conditions, t2n return heat
carrier temperature under standard conditions, tin room
temperature, te air temperature, ten air temperature
under normal conditions [2].
Returned heat carrier temperature is calculated in
analogues way as t3 (heat carrier in feeding line) (1).
Heat quantity which is given by the heating device can
be calculated [2]:
t +t
(4)
Q = c g (t3 t2 ) = sp Fsp 3 2 ti ,
2

where c heat carrier specific heat, g heat quantity,
sp - heat supply coefficient of heating device; Fsp
heating device area.
Using this model automated heating regulator has to
control temperature according to equation (1). When
there is a change of air temperature during regulation
Fig. 2. Heat carrier temperatures in feeding and return lines

dependencies on RS3 device
63
60
increases comfort. Looking from customers point of

view it is a very beneficiary function but for energy
providers it brings a complicated energy distribution
network management problem.
Peak loads is another huge problem that automated
district heating regulators has to cope with. Normally
during critical situations standard regulators can even
worsen the situation. If heat carrier is not capable to
deliver enough energy, usual heating regulator will try
to take as much energy as it needs. This process can
become exponentiation if there hasnt been taken any
precautions. In such cases, only prioritized task control
can offer a quick solution. Heat water preparation and
heating control are two separate processes that are
controlled depending on heat carrier status and circuit
priority.
Water cycle in distributed heating network is about 24
hours. So heating energy preparation must be started at
least 24 hours before expected air temperature changes.
Because of this district heating network is very sensitive
to weather forecasts accuracy and weather anomalies
such as sudden temperature changes (more than 10C in
one or two hours) that are unpredictable. Automated
heating regulation might even worsen such situation
while it does not have feedback from the distributed
heating network and its current capabilities. During
sudden air temperatures all heating all controllers in the
same area start or stop energy usage affecting water
pressure in their pipe branch.
There are existing ways how soften this situation by
implementing self adaptable air temperature filters,
monitoring pressure in feeding and return lines, setting
up heating and hot water preparation priorities.
The most efficient way to solve or at least minimize
effect of this problem is to create a distributed network
of all district heating substation units. This way every
control device will have a possibility to receive remote
information about network capabilities.
4. Distributed control network
New communication technologies and their provided
possibilities can easily add remote access functionalities
to almost every control or monitoring device. In todays
world it is very hard to imagine automation unit which
is responsible for precise and complex processes control
that would not have remote access.
Automated substation unit controllers equipped with
Ethernet or GSM/GPRS modems become more and
more attractive to the market due to their simple remote
access and configuration possibilities. When there is an
electronic automation device in every district heating
substation module there has evolved an idea to
centralize such devices management. Connecting all
automated district control units into one common
information network opens new ways for district
heating network maintenance.
12
10
50
8
T, C
40
6
4
30
2
20
Tfeed, C
Tret, C
Tair, C
-2
10
-4
:3
0
:2
0
:0
5
:5
5
:4
5
:4
0
03
01
23
21
18
:2
5
:1
0
:0
0
:5
0
:3
5
16
14
12
10
08
05
03
01
:4
0
-6
:3
0
time
Fig. 3. Weather anomaly when air temperature increases 10 C

in less than two hours

3. Automated substation unit control
Automated substation unit controllers (Fig. 4) are the
most important part of building heating system. The
whole comfort of the house (heating, air conditioning
and ventilation, hot water preparation) depends on this
relatively small but full of intelligence electronic
device. Automated substation unit saves not only
maintenance costs but also energy recourses. Both sides,
heat energy provider and energy user, are interested to
install automated substation units.
These devices have control algorithms that can ensure
comfortable living conditions for rational usage of
energy recourses. So today, it is hard to imagine a new
building or renovation that would not have had
automated control of its heating system.
Almost every automated substation control unit that is
on the market fulfils heating regulation model described
above or similar.
In most cases heat carrier temperature function of air
temperature is given in the graphic form (Fig. 2).
Depending on thermal insulation of the building,
regulation curve steepness can be readjusted.
Fig. 4. Automated substation unit controller

Regulation of all these devices depends on air
temperature. If it drops, heat carrier temperature is
increased and if it grows up heat carrier temperature is
decreased. This is a feature which lowers energy
consumption that is needed for building heating and
64

5. Conclusions
Rising energy prices as well as maintenance and support
costs made perfect situation for district heating network
as well substation units renovation. Benefits could have
been seen by customers (higher comfort level, lower
heating and hot water preparation costs), maintenance
and support companies (easier remote maintenance
capabilities, lower requirements for support personnel),
and heat producing companies (more rational usage of
supplied energy, capabilities to adapt and more exactly
fulfill mathematical network modeling). Nevertheless
peak load balancing and temperature fluctuations
became more important problems. Also due to
automated regulation there has emerged new risk of
hydraulic network misbalancing. Right here should
come new ideas of connecting whole district heating
network into distributed substation units network.
Where will be possibilities to control groups of
substations paying attention to global network situation.
6. References
1. Pamaktis V., Ragauskas A., Kalueviius J., Petkus
V., Bagdonas A., Borodias P. Advanced energy
metering and control system, ISSN 1392 1215,
Elektronika ir elektrotechnika. N 7 (87), 2008.
2. Morkvnas R. ilumos tiekimo pastat ildymo
reguliavimo matematinis modelis. Delta T, 2008
gegu, p. 27-31.
Fig. 5. Remote substation unit control interface

Monitoring actual power capacity at the most vulnerable
district heating network points, it is possible to detect
dangerous situations that can led to degradation of
working parameters (decrease of heat carrier
temperature, pressure loses, and similar). Getting
warning events about parameters degradation there will
be sent an adjusted set of control parameters to the
group of district heating substation units. This way
every device will have a feedback about global network
situation.
Automated regulation units might misbalance heat
distribution network during peak loads or temperature
anomalies. The impact to the pipes network is
minimized or abolished by adjusting control parameters
for a large scale of substation modules installed on the
same pipe network branch.
Using distributed network monitoring and remote
control functions, heat energy providers acquire new
and exceptional tools for network maintenance. These
tools can be used as a prevention of accidents and
simplified global distributed network remote control.
Fig. 6. Monitorized distributed substation units network
65

ESTIMATION OF TYPICAL SAFETY CIRCUITS FOR MEDIUM POWER

INDUSTRIAL BOILERS
Algirdas EBATORIUS*, Darius EZERSKIS**
*ABB UAB, Lithuania; ** Kaunas University of Technology, Lithuania
Abstract: A safety level evaluation of medium power

industrial boiler safety circuits according to international
standard EN61508 is presented in this paper. According
to safety level estimation requirements for the
components and overall circuit is made. Also there is
made an analysis of technological boilers safety system
and found out typical decisions. After that the
calculation of average probability of failure for typical
safety systems is made to estimate the safe integrity
level. After calculation the possible safety problems and
safety improvement suggestions are made.
2. Determining the SIL of boiler protection circuits

Safety integrity level of middle power steam boiler
protection system is estimating using reliability analysis
based on risk evaluation graph presented in next chapter.
W3
a
W2
-
W1
-
A1
E1
A2
E2
A1
A2
E1
E2
D1
D2
Keywords: Functional safety, safety integrity level,

control system analysis and design, fail safe
programming, IEC 61508.
D3
1. Introduction
D4
Medium power industrial steam boiler produce from 10

t/h to 50 t/h of steam. Typical life time of an industrial
medium power steam boiler is more than 25 years.
International safety standards are obtaining importance
during the last 10 years [2]. Boilers can be classified as
potentially dangerous equipment therefore mach of
boiler protection systems to actual safety standards is
very important.
Historically three types of protection system solutions
are used for boiler protection: relay based, relay based
with programmable logic controller (PLC) guided start
up system and fail safe PLC based. In the chapter 3 a
average probability of failure on demand, PFDavg for the
components and overall circuits are presented.
According to calculations fulfilment of international
standard IEC 61508 [5] is evaluated. The most critical
components of protections circuits and safety
improvements of existing protection circuits are
presented.
In order to evaluate estimated safety integrity level of
typical boiler protection systems first required level for
industrial boilers should be determined. This is
presented in following chapter.
Fig. 1. Risk graph of boiler protection system
2.1. Risk graph

Risk evaluation graph according to IEC 61508-5
appendix D [6] of boiler protection system is presented
in Fig. 1. Typically a boiler is driven by one man
therefore is assumed that damage due to boiler
malfunction is restricted to one person (D2).
Furthermore it is assumed that boiler is permanently
observed by authorized personal in the vicinity of boiler
and thus in the danger zone (E2). Danger avoidance is
depending on the boiler construction. If explosion flaps
are installed then (A1) and (A2) when not. Finally the
probability that and explosion can occur by protection
system malfunction is very high (W3) due to explosive
gas supply into burner furnace.
Required reliability analysis shows that required safety
integrity level for boiler without explosion flaps
protections system must meet SIL3 and for the boiler
with explosion flaps protection system must meet SIL2.
66
2.
3.
4.
5.
6.
3. Determining SIL of a safety circuit

Typically standard automation equipment has no
reliability data from supplier. In this case an average
probability of failure on demand (PFDavg) for 1oo1
systems can be calculated according to (1) [2, 4].
Water level in boiler drum LIC1LH over limits

Too low traction in furnace PIC4H
Air pressure PIC2 over limits
Gas pressure PIC3 over limits
No flame at any burner
M
PFDavg T / 2 ,
GPS
PIC4
(1)
Steam
PIS5
where - failure rate, T system usage time.

Some automation products manufacturers gives mean
time between failure parameter (MTBF) instead of PFD
and its proportional to failure rate as shown in (2) [2,
4].
LIC1
water
1/ MTBF ,
(2)
Natural
gas
After calculation of PFDavg for every component

continue to system PFDavg. System PFDavg depends from
circuit connection. If components are connected in
series PFDavg is sum of each components PFD as shown
in following formula [2, 3]
(3)
where PFDsys system probability of failure on

demand, PFDi, components probability of failure on
demand.
Safety integration level evaluation according to PFDavg
is presented in following Table 2 [2, 4].
Table 2. SIL equivalents to PFDavg
SIL
SIL1
SIL2
SIL3
SIL4
V1
PIC3
PS7
Boilers start-up and shut-down typical algorithm is a

strictly sequence of operations [3]:
1. No fire in furnace, then next step
2. Furnace ventilation, after set time next step
allowed
3. Fuel system leakage test
4. Fuel pressure to main valve check
5. Air pressure to burner check
6. Traction in furnace check
7. Igniters start-up
8. Igniters flame check
9. Natural gas shut-off valve opening
10. Burner start up
11. Burner flame check
If there is any fault condition at any step boiler is
shutting down by closing gas shut-off valve V1. Control
problem in that algorithm is to start boiler, because
flame control unit doesnt let to open fuel shut-off valve
if there is no burner flame. Main problem is how safely
to disable flame control condition during boiler start-up.
When such function is in control algorithm there is
additional risk possibility that it will start when burner is
working and on such case flame control will be lost.
That can initiate explosion in the furnace. Three typical
safety system solutions are discussed in following
chapters.
failure rate, h-1

2,1810-4
4,9610-5
1,9110-6
4,2310-4
4,56410-5
3,21110-6
PFD sys PFDi
Burner
No.2
Fig. 2. Medium power steam boiler diagram
Table 1. Example data from reliability databases [1, 7]

Device
Differential pressure transmitter
Pressure relay
Relay
Flame scanner
Natural gas shut-off valve
Boiler PLC
Burner
No.1
air
where MTBF mean time between failure; [2, 3].

For older equipment that has no reliability data from
manufacturer typical failure rates from special reliability
databases in chemistry and offshore can be taken as
shown in Table 1 [1, 7].
No.
1
2
3
4
5
6
PS6
PIC2
4.1. Relay based protection system
PFDavg
10-2<PFDavg<10-1
10-3<PFDavg<10-2
10-4<PFDavg<10-3
10-5<PFDavg<10-4
This type medium power steam boiler safety system is

controlled by standard relays and flame control unit
disabling is initiated by operator command. In Fig. 3
there is shown scheme of such system.
4. Evaluation of safety circuits

Typical medium power steam boiler pipe and
instrumentation diagram is shown in Fig. 2 [3]. Typical
sop conditions for steam boiler are:
1. Operator command
Fig. 3. Relay based protection system
67
K 2.1
K 52.1
(4)
K 46.1
E 63,08
E 62,15
E 62,14
E 62,13
SR 3
322
320
318
1.1
323
K 51.1
314
313
312
316
T T p n1 n 24 365 10 87600 h
E 62,12
E 62,11
E 62,10
E 62,09
E 62,08
ZP1
07 KT 97
8 DI
Control algorithm in such type protection system is

realized by simple standard components: relays
(pressure, level, temperature and interface). Such type of
systems is without diagnostic function. According to
Table 2 and (1) system probability of failure on demand
is calculated for 10 years [2].
DG 1.1
DG 2.1
325
K2
Fig. 4. Relay based with PLC protection system
Where T life time, Tp day hours , n1 day amount in

year, n year amount;
Control algorithm in such type protection system is

realized by simple standard relays and programmable
logic controlled, which also realize diagnostic functions.
From schematic view system has only series
connections, so for calculations will be used in chapter 3
given formulas.
Protection system life time for the system is the same
like in chapter 4.1 analyzed system.
T=87600 h, where T system life time;
Probability of failure on demand for every system
components is the same as (5), (6), (7), (8), so
calculation will be made only for programmable logic
controller using (1).
PFD p 4.96 10 5 h 1 87600h / 2 2.1725 , (5)
where PFDp pressure relay probability of failure on

demand, p pressure relay failure rate;
PFDLD 4.23 104 h 1 87600h / 2 18.5274 , (6)
where PFDLD flame scanner probability of failure on

demand, LD flame scanner failure rate;
PFDR 1.91 10 6 h 1 87600h / 2 0.083658 , (7)
PFDPLC 4.695 10 6 h 1 87600 h / 2 0.20564 , (10)
where PFDR relay probability of failure on demand,

R relay failure rate;
where PFDPLC programmable logic controller

probability of failure on demand, PLV programmable
logic controller failure rate.
Overall probability of failure on demand for relay based
with PLC protection system is calculated using (3).
PFDATK 4.564 10 5 h 1 87600h / 2 1.99903 , (8)
where PFDATK fuel shut-off valve probability failure

of demand, ATK fuel shut-off valve failure rate
Overall probability of failure on demand is calculated
using (3).
PFDRPLC PFDR PFDR PFDP PFDP

PFDP PFDP PFDLD PFDATK
(11)
PFDLD PFDPLC 48.11679,
PFDSISR PFDR PFDR PFDP PFDP PFDP (9)
where PFDRPLC - medium power steam boiler relay

based with PLC protection system probability of failure
on demand;
The result doesnt meet any reliability requirements.
This type of system is even worse that system in chapter
4.1, because standard PLC rises up probability of failure
on demand. This is also due to standard system
equipment and it doesnt satisfy even SIL2 requirements
for boiler with explosion flaps.
PFDP PFDLD PFDATK PFDLD 47.91115,
where PFDSISR medium power steam boiler relay

based protection system probability of failure on
demand;
The result doesnt satisfy reliability requirements for any
SIL as presented in Table 3. This is mainly due to using
non fail safe sensors and actuators in a system. Relay
protection circuit itself could satisfy SIL2 and could by
used for boiler protection with integrated explosion
flaps.
4.3. Fail safe PLC based

Control algorithm in such type system is fully realized
by fail safe PLC. According to chapter 2.1 it is clear that
all devices in system must meet SIL3 requirements. But
there is left third probability to make fail safe system
standard. There is human mistake probability in such
systems, because control algorithm for PLC is written by
programmer. To lower this probability it is better to use
standard manufacturers programming methods, for
example standard data and function blocks. Such type of
programming solutions meets all required safety
requirements.
4.2. Relay based with PLC

Such type medium power steam boiler safety system is
controlled by standard relays, but it has differences from
relay based protection systems, because flame scanner
signal in boiler start-up cycle is simulated by
programmable logic controller and there are diagnostic
functions. In Fig. 4 is shown scheme of such system.
68
demand ; FSPLC fail safe PLC failure rate;

Overall probability of failure on demand is calculated
using (3).
PFDSISSIL PFDPR PFDPR PFDP

PFDP PFDP PFDDP PFDLD
, (17)
PFDPR PFDATK PFDLD 5.18103 ,

where PFDSISSIL - medium power steam boiler fail safe
PLC based protection system probability of failure on
demand;
The result fully satisfies the requirements for the boiler
with explosion flaps but it is still not enough for boiler
without explosion flaps. This is mainly due to using only
fail safe devices and fail safe software solutions, but
there are not used reservation methods. That is why
system meets only SIL2 safety requirements.
Fig. 5. Fail safe PLC based protection system

There is shown system structure in Fig. 5, but system
works like relays connected in series.
Data for system probability off failure on demand is
taken from fail safe devices database. [8]. All fail safe
devices failure rate is listed in Table 3.
Table 3. Fail safe devices probability of failure on
demand [8]
No.
1
2
3
4
5
Device
Differential pressure relay
Pressure relay
Flame scanner
Natural gas shut-off valve
Boiler fail safe PLC
5. Conclusions
The risk analysis of middle power steam boiler shows
that protections system of such boiler should fulfil SIL3
according to EN 61508. Analysis of a typical existing
protection system shows demand for improving existing
protection systems.
Relay based systems should be used only for boilers
those require SIL2 protection equipment. In this case
only certified SIL sensors and actuators could be used.
Using standard PLC in a start-up sequence of a boiler
for flame sensor bypass improves overall system PFD
and makes system unsafe.
Only fail safe PLCs should be used for protection
systems and boiler start-up system.
failure rate, h-1
8.710-9
6.110-9
1.2710-8
4.2810-8
7.6910-9
According to Table 3 and (1) system probability of

failure on demand is calculated for 10 years [2]. System
life time is the same as (4), T=87600 h
PFD p 6.1109 h 1 87600h / 2 0.00026718 , (12)
where PFDp pressure relay probability of failure on

demand, p pressure relay failure rate;
6. References
1. American Institute of Chemicals Engineers
Guidelines for process equipment reliability data
with data tables, New-York, 1989. 312 p.
2. Boercsoek J. Functional Safety. Basic Principles of
Safety-related Systems Heidelberg, 2007. 538 p.
3. Ezerskis D. Development and Application of
Programming Methodology for Automation Control
Systems, Based on Design Patterns. Dissertation,
Kaunas, 2006. 128 p.
4. Exida Safety engineering Exida, 2005. 275 p.
5. IEC
61508-1.
Functional
safety
of
electrical/electronic/programmable electronic safetyrelated systems. General requirements. Geneva,
1998. 115 p.
6. IEC
61508-5.
Functional
safety
of
electrical/electronic/programmable electronic safetyrelated systems. Examples of methods for the
determination of safety integrity levels. Geneva,
1998. 57 p.
7. OREDA Ofshore reliability data book, 4th edition,
Trodheim, 2002. 835 p.6.
8. http://www.exida.com/applications/sael/index2.asp fail safe equipment database
PFDLD 1.27 10 8 h 1 87600 h / 2 0.00055626 , (13)
where PFDLD flame scanner probability of failure on

demand, LD flame scanner failure rate;
PFDDP 8.7 109 h 1 87600h / 2 0.00038106 , (14)
where PFDDP differential pressure relay probability of

failure on demand, R differential pressure relay
failure rate;
PFDATK 4.28 108 h1 87600h / 2 0.00186 , (15)
where PFDATK fule shut-off valve probability of

failure on demand, ATK fuel shut-off valve failure
rate;
PFDFSPLC 7.69 109 h1 87600h / 2
0.0003368,
(16)
where PFDFSPLC fail safe PLC probability of failure on
69

POSITIONING OF STEPPER MOTORS IN

GRAPHICAL DATA ACQUISITION SYSTEM
Raimundas LIUTKEVIIUS, Andrius DAVIDSONAS
Another important factor in such graphical applications

is a huge amount of the process data and complex
process tasks. Stand-alone PLC system is not suitable
for whole object scanning and reconstruction process,
which consist of: control of the 3D scanner, acquisition
of the data, pre-processing it, forming a 3D model from
scanned data and displaying it. The acquisition of data,
processing of this data and graphical representation of
3D model is transferred to non real time PC-based
application to deal with a problem. Non-real time means
that it is impossible to acquire data from 3D scanner at
equally spaced intervals and to calculate the position of
a scanned object based on motor rotation time or other
characteristics. There are many stepper motors
positioning and position tracking techniques, but they all
differ in quality, speed, etc. and not all of them are
appropriate for 3D scanning and other similar tasks. In
this paper stepper motor positioning and position
tracking solutions, which help to produce sufficient
quality for 3D object reconstruction is presented and
analyzed.
Abstract: This paper analyses stepper motors position

tracing problems, found in 3D scanner application.
Attention is paid to the position track methods of the
stepper motors in a PC-based application and the use of
position data along with the data from optical sensor for
the construction of 3D model of the scanned objects.
The methods were implemented into the real time 3D
data scanning and visualization application and were
analyzed in several aspects: the execution time and the
positioning precision.
Keywords: PLC application, stepper motor control,
stepper motor positioning, 3D data modelling, 3D
scanning.
1. Introduction
PLC controlled stepper motors are one of the most
versatile forms of positioning systems. They are
frequently used in different precision positioning
equipment such as 3D scanners, image/flatbed scanners,
computer printers, plotters, floppy disk drives, CNC
machines, fluid control systems, packaging machinery
and many other devices and systems. To compare with
competitive closed loop servo motors, they are simpler,
cheaper and more rugged. Both motor types can be
easily digitally controlled using different PLC systems.
Despite this fact, accurate position tracking of stepper
motors may become a challenging task. Thats why it is
main object of this paper.
In this paper the use of stepper motors for the 3D
scanner is described. The tasks of stepper motors are: 1)
to rotate an object; 2) and to move vertically optical
sensor that reads distance to the scanned object. Position
tracking of each stepper motor is very important to
accurately acquire objects surface coordinates and to
create a sufficient 3D model. Even small position
tracking errors can cause considerable deflections in a
3D model and the result can become unrecognizable.
So, stepper motors positioning and position tracking
techniques must be taken into account when
programming 3D scanning or similar applications.
2. Graphical data acquisition system

3D scanning systems, analyzed in this paper, consists of
a scanner, a PLC and a PC. For the control of stepper
motors and the acquisition of data, Beckhoff PLC
controller CX1020 is used (Fig. 1).
Fig. 1. PLC controlled graphical data acquisition system
CX1020 is an embedded PC with Intel Celeron M 1.0

GHz processor, 1GB RAM, hard drive, Windows XP
Embedded OS and ability to attach selected I/O
70
modules/converters. Three modules/converters for the

entire control of 3D scanner are used: 2xKL2541 (step
motor controllers) and KL3052 (analogue data input
module). All modules are integrated into one 3D
scanning system and are programmed using TwinCAT
software.
The 3D scanner consists of two stepper motors, the
optical distance sensor and additional proximity sensors.
During the scanning process the scanner rotates an
object about its centre and constantly measures distance
to its surface. Another stepper motor is used to move the
optical sensor up or down (Fig. 2).
X R cos( ),
Y R sin( ),
Z H.
(1)
(2)
(3)
Converted data can be saved in internal data storage for

later visualization using other 3D reconstruction
methods such us point based rendering [3], power crust
or cocone methods described in [3, 4, 5].
It is obvious that quality of the 3D model of a scanned
object depends on three main factors: 1) the precision of
the stepper motors position tracking; 2) the quality of
the distance measurements; and 3) the quality of the
algorithms used for the construction of the 3D model. It
was experimentally shown that only small inaccuracies
of 3D scanning system can be digitally corrected by 3D
modelling algorithms and different filters [2]. So, big
attention should be paid to the accuracy of the
calculation of the scanned objects points coordinates.
Careless control of 3D scanning system usually leads to
digitally un-correctable errors and incorrect 3D model.
In the next sections the aspects, that cause such errors
and how to avoid them by selecting right way to control
3D scanning system are analyzed in more details.
4. Control of stepper motors
The rotation angle and the height of the scanned object
are evaluated depending on stepper motors position. So,
the calculation of the exact rotation angle and optical
sensor lift up position is mandatory in order to get the
correct coordinates of the scanned object. Inaccurate
calculations result in the deformation of 3D object
model - the twists and bends of the model are inevitable
(Fig. 7-8). The stepper motor is an open loop system.
This means that stepper motors rotation is controlled
without any feedback mechanism. This is the main
reason of the positioning problem, because angular
position of the motor shift cannot be estimated. It
doesnt mean stepper motors are inaccurate. On the
contrary, stepper motors can be positioned as precise, as
closed loop based systems, however they cant return the
feedback about the current position reached [6].
Depending on the stepper motor type the rotation angle
of its shaft can be calculated in several different ways: a)
evaluating the rotation time, b) rotating a motor a
specified number of micro steps and c) if a motor is
equipped with the encoder reading its data. These
methods are analyzed in the next paragraphs.
Fig. 2. 3D scanner. Dashed lines show movement directions,

straight lines scanner components
Both stepper motors are controlled with separate

KL2541 modules stepper motor controllers. Each
module has two types of internal memory: temporary
(RAM) and persistent (EEPROM) and can be
programmed using internal registers or manually
controlled through TwinCat System Manager
application [1].
Object surface coordinates are measured with optical
position sensor, which measures distance to particular
point of scanned object. KL3052 module is used to
digitize analogue data and transfer it to PLC.
3. Acquisition of 3D coordinates
In order to construct 3D model of a scanned object, the
precise acquisition of surface points coordinates are
required. This can be achieved by tracking the positions
of both - object rotating and sensor lift up stepper
motors and acquiring the data from optical sensor. The
motor, that rotates an object, produces angle , the lift
motor height H and the optical position sensor radius
R. As a result object surface points in cylindrical
coordinate system are acquired. At this point, they can
be transferred and visualized with the real time
visualization system as described in [2], or easily
converted to Cartesian coordinates using formulas:
4.1. Positioning, based on the evaluation of time

Rotation speed of the stepper motor is expressed in
terms of steps and micro-steps per second. Every step
can be translated into rotation angle s (degrees) or liftup height Hs (mm). Stepper motor is a precise system
and can rotate the exact number of steps during the same
period of time. So, the particular position of the motors
shaft can be easily calculated from its operation time
using the following formulas:
71
V s t ,
H V H s t.
revolutions and resetting the timer at the beginning of

each turn. Both methods apparently prolong the overall
scanning process, but are capable to only slightly reduce
displacement errors. Using the first method it is possible
to decrease the time skews and using the second one to
decrease the displacement error. Exact position of the
turn should be determined in order to completely avoid
the growth of the errors. The position, indicating the
beginning of a new turn can be determined by using an
additional proximity switch and a special mark on the
shaft of the motor.
(4)
(5)
Where V is the rotation/lift speed (steps/ms); s/Hs one

step angle/distance; and t time difference. In this
case motors control algorithm of 3D scanner can be:
1. Make the shaft of the motor to rotate the specified
number of micro seconds, necessary to rotate the
object 360 around its centre. Every t2 seconds:
Read objects surface coordinate R from
optical distance sensor;
Calculate rotation angle using (4);
Write objects surface point coordinate
pi(R, ,H) to memory or disk;
2. Stop rotating the motor;
3. Move the optical sensor up for the specified number
of micro seconds;
4. Calculate new optical sensor position value H using
(5).
5. Repeat 1-5 steps while entire object is scanned.
The main problem with this method is how to resolve
the exact motor rotation time. In order to use this
method a real time PLC system should be used. A PC
based control systems are built on top of other non-real
time operating systems, cannot guaranty equal cycle
execution periods (Fig. 3) and causes wrong evaluation
of objects surface coordinates.
Fig. 4. Motor position error growth depending on stepper

motor speed (v) and revolutions (rev.). s standard deviation
4.2. Positioning based on the motor control

The rotation angle of the stepper motor can also be
tracked by programming a PLC to rotate the stepper
motor for the exact number of steps. Most PLC systems
support this directly if digital encoder is present or
through referencing (homing) mode [1]. In both cases
rotation angle can be determined using this algorithm:
1. Rotate the motor shaft a specified number of steps
x;
Read surface coordinate R from optical
distance sensor;
Calculate rotation angle: = + x s;
Write object surface point coordinate pi(R,
,H) to memory or disk;
2. Repeat 1 step until the full revolution (360
degrees) is completed;
3. Move the optical sensor the specified number of
steps y.
4. Calculate optical sensors new position value
H = H + y Hs.
5. Repeat 1- 4 steps until the entire object is scanned.
Fig. 3. Execution of commands in real and non-real time

control systems. In real time systems all commands is
executed in e.g. exact 0.2s intervals, but non-real time system
cannot guaranty that
These small time skews, displayed in Fig. 3, appears

because of the fact that in non-real time systems
execution time is not controlled by PLC application, but
by the operating systems tasks supervisor. The task
supervisor executes the commands in the order they
arrive and depending on their priority, but not at the
exact time. As a result, coordinate measurement errors
grow depending on the commands execution frequency
(motor speed) and the overall system load. In addition to
that, coordinate displacement errors continuously grow
(Fig. 4).
These problems can be reduced in two ways: 1) by
decreasing the velocity of stepper motors, or 2) by
decreasing the time intervals between different
This approach is one of the slowest and suffers from the

limited precision. Even if the digital encoder is present,
a motor can rotate for at least one step and most microstepping advantages are lost. In the used system one step
is equivalent to 1.8. This means that if radius of the
72
object increases, the quality of the reconstructed object

decreases and this fact cannot be avoided (Fig. 5).
Another disadvantage of this method for the calculation
of the motor shaft rotation angle is that the motors
positioning commands usually are much slower than the
direct commands, because communication with PLC
modules and data exchange at register level is required.
This leads to additional slow down of scanning process.
Speed up of the scanning process can be achieved if the
motor, that controls the optical sensors position, can be
rotated x>1, y>1 steps and timer controlled angular
positioning is used. In this case the desired trade-off
between speed, precision and error rate can be achieved.
Fig. 6. Operation of distance sensor. T light transmitter, R

receiver [2]
Operation of optical position sensor is based on the

properties of light. Optical sensor have transmitter T,
which project red laser beam on to the object. Object
surface reflects part of the light back to the receiver R
with different radius, depending on its distance from
sensor. Position of object surface point using this angle
is calculated.
6. The experiments
Fig. 5. The dependency of the precision of the coordinate

measurements on the rotation angle and the objects size (R
radius of an object)
The methods for the evaluation of the motors shaft

rotation angle were experimentally analyzed. The
difference between methods is graphically demonstrated
in the Fig. 7-9. The cube was taken as the test object and
its models were constructed using different coordinates
calculation methods. Fig. 7 shows the layers of the
models. First two models, Fig. 7 a) and b) are
constructed using timer based shaft angle calculation
and the third one using encoders data. Experiment
proves that the increase of the scan speed decreases the
accuracy of the model, Fig. 7 b. Because timers are reset
every 360, positioning lines are coincident. Otherwise
twists of object surface would be noticeable as shown in
Fig. 8. The decrease of the scanning speed can solve the
problem, but still the model is not accurate enough - the
model of the object does not maintain the exact
proportions (Fig. 7 a). Best results are achieved when
encoders data are used (Fig. 7 c).
4.3. Positioning based on the motor encoder

In case stepper motors are equipped with internal
encoders, sufficient positioning accuracy can be
achieved by using method similar to time controlled
positioning. In this case the time tracking algorithm
should be substituted with tracking of the encoder
increments. An internal encoder cannot show the exact
position of the stepper motors shaft position, because it
along with the stepper motor remains an open loop
system. Nevertheless, encoders can provide the
information about the change in the shaft position angle.
The quality of acquired object surface points then will
depend on 1) the number of increments, supported by
the internal encoder, 2) the ratio of the motor speed and
data acquisition frequency. Because of an encoder,
position tracking errors are avoided even at higher shaft
rotation speeds (Fig. 7).
5. Operation of optical position sensor
Last thing to be mentioned is the acquisition of the
radius R (the distance to the scanned surface). The
optical position sensor measures distance from the
imaginary origin to the particular point of object surface
Rabs only. Actual radius Rt can be calculated using the
characteristics of the optical sensor (6):
Rt
Rabs Rt _ max
Rabs _ max
l ,
Fig. 7. The layers of the cubes 3D models, that are created

using different objects position calculation methods and scan
speeds:
a) 5 steps/s, time based evaluation,
b) 20 steps/s, time based evaluation, and
c) 50 steps/s, encoder based evaluation
(6)
where Rt_max is the measurement range, Rabs_max max

resolution of the PLC channel, l shift between real
and imaginary origin [2].
73
3. Pajarola R. Stream-Processing Points. Proceedings

of 16th International Conference IEEE Visualization
2005 (VIS 2005), 2005. p. 31.
4. Matukas V., Paulinas M., Uinskas A., Adakeviius
R., Mekauskas R., Valinius D. A Survey of Point
Cloud Reconstruction Methods. Proceedings of
International Conference Electrical and Control
5. Dey T.K. Curve and Surface Reconstruction.
Algorithms with Mathematical Analysis. Cambridge
University Press, New York, 2007. p. 59 112.
6. Jones D.W. Control of Stepping Motors. A Tutorial,
The University of Iowa Department of Computer
Science, 1995. Internet source:
http://www.cs.uiowa.edu/~jones/step.
Fig. 8. Evaluation of the angular position of the motor shaft

using time based aproach, errors are not reset after complete
turn
Fig. 9. a) Real object, b) model of the real object, acquired

using time based approach, rotation speed 5 steps/s.
c) model of the real object, acquired using encoder based
approach, rotation speed 50 steps/s
7. Conclusions
In the paper, stepper motor positioning and position
tracking problems, typical in the programming of 3D
scanning systems are presented and the solutions to the
problems are suggested.
It was experimentally shown that, motors, used in the
3D scanner should have encoders in order to increase
scanning speed and quality. It was also demonstrated,
that even if it is possible to precisely control the rotation
of the stepper motor, it is impossible to precisely
calculate its shaft rotation position evaluating the time in
PC-based systems without the encoders.
In case stepper motors are not equipped with encoders,
the modelling errors can be reduced by installing
external proximity switches that indicate some positions
of the motor shaft. An intermediate solution can also be
used where the motor control based positioning is
applied. This method is slow; nevertheless rate of the
positioning errors can be reduced to the desirable level
at the expense of the scanning speed. It should be
noticed that not all PLC systems supports this method
also called referencing or homing.
Time controlled positioning should be avoided in nonreal time systems but can be used in PLC systems were
equal cycle execution periods are ensured.
8. References
1. TwinCat Information System. Documentation for
KL2531 and KL2541, One Channel Stepper Motor
Terminals, Version 1.5, 2006.
2. Davidsonas A., Liutkeviius R. A PLC Application
for Data Scanning and Visualization. Proceedings of
74

VISIBILITY OF TEXT IN MULTIMEDIA PRESENTATION

Stanislovas MASIOKAS, Migl KRIUGLAIT-JARAINIEN, Valdas PAKNAS
Abstract: Digital projection on screen is the main way

of information visualization in science, teaching and
business. The main drawback of presentations is
deficient text visibility, which depends on text
parameters and its contrast with background. Visibility
determines identifying speed. With reference to
published results of experimentation accomplished with
a view to establish relation between luminance, contrast
and visual angle, in this paper identifying speed reliance
on abovementioned factors confidence probability being
different was analysed. Results bespeak, that identifying
speed depends on angular dimension most of all. In
accordance with results practical recomendations were
formulated.
Fig. 1. Visual angle in slide and its projection on screen

Generally, visibility depends on physiological
characteristics of the eye and physical factors. There are
four top ones: physical size of an object, its contrast
with background, observation time and luminance [1].
Increasing all these factors increases visibility. In case
of text visibility in slide observation time is not
restricted. Luminance of an task entirely is not so
important if theres a high contrast with background.
While creating a presentation we are free to choose high
contrast task and backround combinations. Therein lies
that there are two main determinants deciding text
visibility in a slide visual angle and contrast. In this
case we can manage both of them.
Keywords: visualization, presentation, visibility, visual

angle, contrast, identifying speed, text.
1. Introduction
Digital projection on screen is the main way of
information visualization today. Science, teaching and
business areas can not manage without it. Though lots
of presentations are made, only negligible part of them
accomplish their purpose are well seen for the
observer. Poor text visibility is the main problem of
mostly all presentations. Text on the screen must be
clearly seen and easily red. That depends on basic
parameters of text (font ant its characteristics) ant its
contrast with background. Visibility determines
identifying (and reading) speed.
High visibility not only accelerates identifying speed,
but also tires vision less. The aim is to ascertain a
relation between identifying speed and deciding factors:
visual angle, background luminance and contrast
between task and background. By analysing graphical
data we can ascertain which factors influence increment
of identifying speed most of all. Furthermore, an
optimal value of visual angle which would give a proper
identifying speed and information output will be set.
Also we will determine practicable boundaries of
background luminance and contrast values.
2. Main factors, determining visibility in a slide

Visual angle is relevant size, which characterizes
visual size of an object and depends on physical
dimension of an object, the angle of inclination of the
target from normal to the line of sight and the distance
from the viewer [2]. It is angle which object subtends at
the eye (Fig. 1). Visual angle helps to determine how
well an object can be seen.
= 2 arctan
d cos
,
2l
(1)
where d dimension of an object; angle of

inclination of the target from normal to the line of sight;
l distance from the viewer to an object. = 0 in case
of projection on screen.
75
bigger targets in slide, visibility is more sensitive

function of contrast than it is of size [1].
3. Literature analysis
There are no many researches related to visibility of text
in slide. Visibility in chromatic regard is better explored
interaction between task and background colours and
influence of luminance contrast over visibility was
analysed [3]. It is none the less important to explore
influence of text parameters over visibility.
Quite comprehensive study of four visibility
determining factors was taken by H. Richard Bleckwell.
Light discs of different luminance and visual angle
values in the center of darker background were used.
Observers task was to identify the time interval which
was needed to notice the disk. However, results of this
experiment are not most suitable for visibility of text
research as it was performed in too wide visual angle
range (164) with not practiceable values for a slide (1
4) [1].
Especially significant in this area are results of
researches made by A. S. Shaikevic (A. C. )
[4]. A research was made with a view to explore, how
identifying speed depends on visual angle (), contrast
betwen task and background (C) and background
luminance (Lb).
Observers with normal vision were watching light
Landolts rings in dark background. The task was to
identify position of a cutout. In Shaikevics research
Visual angle varied from 0.7 to 10. Values of contrast
between task and background and background
luminance were also transcriptive data.
In accordance with collected data, diagrams were
ploted, which show identifying speed reliance on visual
angle and contrast for different background luminance
values and confidence probability 0.7 [4]. Therefore we
can evaluate identifying speed, i. e. efficiency of a
visual task.
It can be seen that increment of identifying speed is not
linear function of contrast between task and
background. Its especially noticeable in range of
sizeable visual angle. Magnification of background
luminance positively influences identifying speed,
especially in low luminance values.
Using these diagrams necessary value of contrast or
background luminance may be determined so that
required identifying speed would be obtained under
certain conditions.
In this paper redrawn, modified and press-fitted
diagrams are given (Fig. 3, 4, and 5). Narrowed
research area will be further analysed, which is relevant
for the visibility of text case.
Fig. 2. Diagram of negative (a, c) and positive (b, d)
luminance contrast
A target can only be seen if it differs in luminance from
background. On this understanding luminance contrast
which defines visibility may be calculated. There are
several ways of defining luminance contrast, though
here is traditional one:
C = (Lt Lb) / Lb ,
(2)
where Lt task luminance; Lb background luminance.

Agreeably to luminance values of task and background
there may be positive or negative contrast. Dark task in
lighter background gives negative contrast (C < 0),
which ranges from 0 to minus 1 (Fig. 2a). Light task on
dark background is frequent in presentations and gives
positive contrast (C > 0), which ranges from 0 to
infinity (Fig. 2b). Such wide variation of contrast range
aggravates comparison of image visibility. Its essiantial
to say that in both cases lighting of the room remains
the same, so greater part of adaptational field maintains
the same luminance value. For more convenient
comparison of different contrasts classic formula may
be modified by changing Lb value to higher luminance
value in denominator.
C = (Lt Lb) / Lmax ,
(3)
where Lmax highest luminance value in slide which

may be both luminance of target or background.
This interpretation with regard to negative contrast
(positive image) Lb is replaced with Lmax (Fig. 2c). In
case of positive contrast (negative image) Lb = Lmax
(Fig. 2d).
After changes made positive contrast (black text on
white paper) ranges from minus 1 to 0, negative
contrast (white text on dark background) ranges from 0
to 1. Either way, adaptational luminance La, which
depends on lighting of the room, remains the same.
Such changes make comparison ef experiment results
easier.
Contrast and visual size are closely interdependant. For
objects smaller than 1.5, a small change in visual size
requires a big change in contrast, so that the same
visibility level would be rereached. As we talk about
4. Selecting practiceable values of visual angle and

background luminance
Research area must be narrowed so that only practicable
conditions would be analysed. A new diagram for
ascertaining identifying speed reliance over contrast,
background luminance and visual angle needs to be
plotted (Fig. 6). Contrast range from 0.5 to 0.8 is under
76
consideration. Values under 0.5 are not practicable,

because it is range of poor visibility. Realistically
contrast 0.8 can be achieve today. Fig. 6 shows that
there is linear dependence between logarithm of
visibility and contrast. We talk only about contrast of
slide here, which corresponds to absolutely dark room.
It can be apparently seen that identifying speed grows
most rapidly when visual angle does not exceed 2.
Growth above 2 is not so speedy. When background
luminance Lb varies from 42 to 205 cd/m2 increment of
identifying speed is not sizeable. Contrast and visual
angle influence increment of identifying speed most of
all, especially up to 2.
Fig. 5. Identifying speed reliance on contrast and visual

angle, when Lb = 205 cd/m2

angle, when Lb = 6.2 cd/m2
Fig. 6. Influence of contrast and background luminance

over identifying speed

angle, when Lb = 42 cd/m
To perform a visual task we need to determine visual

angle of the smallest detail that needs to be seen. If the
thinest text line subscripts and superscripts is
attributed to 2, then according to publishing
77
requirements main text should be 1 time bigger, i. e.

3. When optimal visual angle is stated, necessary line
thickness on display may be calculated:
=
l w
3.44 W
universities), where is a risk to ruin vision of lots of

young people.
(4)
where l distance from viewer to screen; w width of

slide in computer display; W width of slide in
projection screen.
If there is enough space in a slide and it is aesthetic,
large font, corresponding to more than 3, may be used.
However, capacity of a slide approximately
commensurates with square height of a letter, therefore
large fonts restrict information output of a slide.
For further analysis Lb = 6.2 cd/m2 value is eliminated
as it corresponds to 3040 lx (when reflectance of walls
is about 0.500.65) and that is too dark room. Screening
presentations in such dark rooms is unacceptable today.
Value Lb = 205 cd/m2 corresponds to 10001300 lx. Till
these levels of illuminance become standard in major
part of scholastic institutions and other close purpose
rooms, it is believable hardware of displaying
presentations may change essentially.
Value Lb = 42 cd/m2 corresponds to 200260 lx and
may be held as todays norm (when reflectance of walls
is about 0.500.65).
Though major part of illuminance norms for communal
and esspecially educational rooms in most of EU
countries is 300-500 lx, however these results are
closest to real situation.
Fig. 7. Influence of confidence probability over identifying

speed, when Lb = 42 cd/m2, = 2
6. Conclusion
1. Optimal value of visual angle in slide is 2 for
subscripts and superscripts and 3 for main text. If
its aesthetic and there is free space in slide larger
fonts can be used.
2. In case of good lighting (300500 lx) contrast
between text and background must be no less than
0.5, i. e. screen illuminance created by projector
must be no less than twice bigger than illuminance
created by surroundings.
3. If visual angle is greater than 3 and contrast is
higher than 0.5, identifying speed with confidence
probability 0.99 exceeds practicable avarage reading
rate of english text.
5. Influence of confidence probability and proper

identyfying speed
So far all analysed data coresponded to confidence
probability 0.7. Later derived data of the same research
was offered for other confidence probability values [5].
New diagram for ascertainig confidence probability
influence over identifying speed was plotted (Fig. 7). It
emerged that the highest identifying speed is for lowest
confidence probability value 0.5. The lowest speed is
for confidence probability 0.99. The identifying speed
between these two cases differs more than 7 times.
After sifting results of this experiment quite reliable and
easy read footing of text presentation may be defined.
In proper visual situation reading rate of english text is
about 15-20 print symbols per second and faintly
depends on visibility level (VL ranges from 2 to 10) [2].
Considering that we may state that in case of visual
angle 2, contrast C 0.5 and confidence
probability 0.99, identifying speed is higher than regular
(28 s1) identifying speed.
In case of higher contrast or lower confidence
probability, identifying speed (i. e. reserve of
identifying speed) signally exceeds practiceable value.
Under these circumstances we can create high-grade
slides wherein displayed text is clearly seen and does
not tire vision even in a long run. It is especially
talking-point to educational process (at schools,
7. References
1. Murdoch J. B. Illumination Engineering from
Edisons Lamp to the Laser. New York, 1985.
p. 542.
2. The IESNA lighting handbook. New York, 2000.
p. 800.
3. Masiokas S., Kriuglait M., Otas K. Colour
luminance contrast of digital projection image.
Electronics and Electrical Engineering, no. 7(79).
Kaunas, 2007. p. 19-22.
4. Shaikevic
A. S.
About classifying visual
perfomances. Lighting engineering, no 5. Moscow,
1958. p. 13-20 (in Russian).
5. Shaikevic A. S. Nomogram for assessing identifying
speed and confidence probability. Lighting
engineering, no 1. Moscow, 1964. p. 19-22
(in Russian).
78

AUTOMATIC PLC-SOURCE-CODE GENERATION

OF C4.5 DECISION TREES
Elmar HEINEMANN*, Jolanta REPYT **, Andr WENZEL*
* University of Applied Sciences Schmalkalden, Germany; ** Kaunas University of Technology, Lithuania
source code generation is useful if it is often necessary

to develop new decision trees for various applications, if
application tests make the chance of a decision tree
necessary, to avoid mistakes by the implementation of
complex decision trees, to be independent from a
personal programmer style and to have an identical
interface of the decision tree unit in every application.
To achieve an automatic source-code generation out of
feature data is an integrated design draft needed.
This paper describes the entire process of C4.5 tree
generation, pruning and conversion into IEC 61131-3
[1] conform PLC source-code.
Abstract: Decision trees are widely-applied for

classification problems. One of the most common
methods to create a decision tree is Quinlans C4.5.
Beside the decision tree development is the
implementation of a decision tree on a PLC very
extensive especially if a tree is changing during the
development and test phase. In this case a manual tree
implementation is not economically justifiable.
Therefore a method and tool are shown which
automatically generates a C4.5 decision tree and the
associated PLC source-code out of feature data.
Keywords: decision tree, PLC, automatic source code
generation.
2. Decision Trees
Goal of the classification is the assignment of a target
variable attribute to an unknown object. The
classification is done by a classifier which learns the
classification based on classified training data. To build
a classifier can be several methods like neuronal
networks [9], support vector machines [4], fuzzy
systems [8,6] or trees [2,3,5,7] used. The classifier in
this work based on trees.
Trees parting a feature space into a subset of rectangles.
An attribute of a target variable, a label, is assigned to
every rectangle. There exist two different types of tree
based models; regression trees and classification trees.
Regression trees work with continuous target variables
and try to minimize the distance between the label and
the target variable values in the feature space subset.
Classification trees work with discrete target variables.
They try to achieve that a subset only contains attribute
for the target variable which are identical with the
subset label. [5]
This work uses classification trees and as learning
method Quinlans C4.5 algorithm.
The C4.5 algorithm is an improved version of the ID34
algorithm.[7] The learning algorithm searches for a
hierarchical split structure to divide the data set T into
subsets Ti with the goal that a subset at the lowest split
layer only contains objects with the same class Ck.
1. Introduction
Decision trees are widely used for classification
problems. Their advantages compared with other
classification techniques are:
Decision trees can be effectively applied to any data
structure, in particular to discrete, continuous or
mixed data. [3]
Decision trees are very robust against outliers. [3]
Prediction rules are easy to interpret. [3]
Decision trees perform a stepwise variable selection
and complexity reduction. [3]
A decision tree is generated out of evaluated feature
data. The most common methods to create a decision
tree are C4.5 [7], CART1 [3] and CHAID2 [2]. The
generated tree is often very complex and overfits the
classification problem. Hence to improve the
classification ability is a tree pruning necessary. Pruned
trees are the base for a real word classification
application. Therefore is a decision tree implementation
on a target environment required, for example on a
microcontroller or a PLC3.
A manual conversion of the develop decision tree into a
target source code is often not expedient. Automatic
1
2
3
Classification and Regression Trees

Chi-square Automatic Interaction Detectors
Programmable Logical Controller
79
Iterative Dichotomiser 3
Every split point in the hierarchical structure depends on

the attributes of one feature variable and is called a
node. Final sub sets which contain only data with the
same label are called leaf. For the selection of the
splitting variable at a certain hierarchical layer uses
C4.5 the normalized information gain. The normalized
information gain of a variable is calculated by the
following schema.
In the first step is the information content, the entropy,
of the variable for the classes in the dataset with (1)
calculated. This is the information content before the
split.
test data set. But every pruning indicates a loss of

information. To measure the information loss by the
pruning is it feasible to associate confidence
information with the assigned leaf or node. As
confidence measure can be the misclassification or error
probability of the pruned tree used.
As test and comparison environment for decision tree
classifier is the in [6] and Fig. 1 displayed system used.
(1)
info information content
pm probability distribution of the classes
The second step, displayed in (2), calculates the

weighted information content of the probability
distribution for each possible subsets generated by the
splitting attributes of the variable. This can be
interpreted as the information content after the split.
(2)
infox information content after split
Ti
data subset
T
data set
Fig. 1. Testing system
The information gain, shown in (3), is the difference

between the information contends before and after the
split.
The test environment contains a process with two

operating points described as PT3-systems and a PID
controller for each operating point. The system switches
between the operating points when the system output YS
is below zero. The function of the classifier is to detect
the change of the operating point and select the correct
controller. Reference [6] uses the system for the
analysis of bayes and fuzzy classifier.
The learning base contains the feature vector and the
according operating point for a rectangle set value W.
The feature vector ok consists of the last three values for
Y at the time k; ok = (yk yk-1 yk-2). The quality of the
learned classification is tested with rectangle and sinus
function for the set value W with and without
disturbance on the output Y. Fig. 2 shows the generated
tree and Tab. 1 a comparison of the results with [6].
(3)
gain information gain of the split
To achieve a good comparability a normalization of the

information gain is necessary. Thereto is the
information content of the variable class distribution
calculated with (4) and the information gain divided
with it (5).
(4)
split info information content of class distribution
(5)
gain ratio normalized information gain
The C4.5 chooses the variable with the maximum gain

ratio as split variable on the hierarchical layer. The
procedure continues for each out coming subset until a
subset contains only one class.
The accrued decision tree has a complex structure with
long path through the tree to a leaf. The tree achieves
very good classification quality for the learning data but
mostly a very worse classification quality for unseen
data because the tree is overfitted to the learning data.
To improve the classification quality is a pruning of the
accrued decision tree necessary. At the pruning process
are whole sub trees, nodes, replaced by a leaf or another
sub node. There exist several pruning methods. It is
possible to define a maximum depth of the tree, a
frequency threshold or to minimize the error rate of a
Fig. 2. Decision tree
80
Table 1. Simulation results
continuous features which differ for every decision tree.

The output of a decision tree consists of the resulting
class and error probability. To achieve a small and
always equal interface are structures used. With the use
of structures is the POE head for every decision tree
implementation identical except the different output
handling by the implementation as function or function
block. Fig. 4 shows the structures for the example tree
and Fig. 5 the head of the example tree implemented as
function POE. On the input side is an object structure
defined that comprise the itemized features. In a loop
along the whole decision tree is the structure filled with
all occurring features. For categorical features is the
data type string and for continuous feature the data
type real used. As POE output is a solution
structure developed which contains the classification
result in the data type string and the probability error
in a real data type.
Relative Error (nwrong / n) [%]

Set value
function
Rectangle
Za = 0
Sinus
Za = 0
Sinus
Za = N(0,0.04)5
Fuzzy
classifier
Bayes classifier
direct
crestline
regression regression
Tree
2.0
2.0
2.0
0.0
2.0
2.0
2.0
2.0
4.5
38.5
3.5
4.0
The generated tree reduces the feature space from three

to one dimension and classifies the non-disturbed set
value function with no error because the tree detects the
correct operating threshold. The error for the disturbed
sinus set value function is smaller as by the fuzzy
classifier and bayes classifier with direct regression.
Only the bayes classifier with crestline regression is
better as the tree classifier.
If the learning data consists of the disturbed sinus set
value data then a more complex tree outcomes that has
to be pruned. The resulting tree is shown in Fig. 3 and is
the base for the automatic source code generation.
Fig. 4. Interface structures
Fig. 5. Head of a decision tree implementation as function
The second step is the decision tree implementation.

The realization of the decision tree is carried out as
nested ifelseifend_if statements. Every hierarchical
layer of the decision tree is a nesting level. For the
nesting generation is the decision tree first of all pass
form the top node through to the first leaf on the left
hand side. In every hierarchical layer is an
if statement opened with the comparison of the
assigned variable on the node and the branch value to
the next layer. The access to the value of the variable is
done by the variable name and their association in the
object structure. On the last layer, the leaf, are the
associated class and error probability stored on the
solution structure. As second of all is a retrace from
the leaf too the node one level above done. In this node
is an elseif statement for the next branch generated
and the source generation for all nodes and leaf among
this branch done. As third of all is after the last branch
of a node the closing of a nesting level with an
end_if statement done and a retrace too the node
above performed. The result of this recursive algorithm
for the example is shown in Fig. 6.
At the end of the automatic source code generation are
the global variables feature as object and
classification as solution created. In the feature
variable stores the end user the actual feature values and
receives in the classification variable the decision tree
result. Therefore is a program POE created that calls the
Fig. 3. Complex decision tree
3. Automatic Source-Code Generation

As decision tree implementation environment are PLCs
selected because they are widely used for process
control. The PLC programming languages and program
organization units (POE) are defined in the IEC 61311-3
norm. [1][10]
The whole PLC control program is parted by POEs into
several encapsulated duties. A decision tree is one of
these duties and accordingly it is the goal to generate a
corresponding POE. The decision tree POE is based on
the POE-types function or function block. As
programming language is structured text selected.
To achieve automatic POE generation is the definition
of a standardized POE layout necessary. Many parts of
the layout are defined by the IEC 61131-3. The
definition of the POE interface and the implementation
of the decision tree itself are left to define.
The first step is the definition of the POE interface. The
input of the decision tree contains categorical and/or
5
Gaussian distribution = 0; = 0.04
81
decision tree POE and stores the output on the

classification variable. The end user binds the
program POE in a suitable position in the control
program. The global variables and the program POE are
displayed in Fig. 7.
7. References
1. Norm. IEC 61131-3:2003. Speicherprogrammierbare
Steuerungen Teil 3: Programmiersprachen
2. Bankhofer U. Datenanalyse und Statistik: Eine
Einfhrung fr konomen im Bachelor. Gabler.
Wiesbaden, 2008.
3. Breiman L. Classification and regression trees.
Chapman & Hall. New York, 1993.
4. Cristianini N. An introduction to support vector
machines and other kernel-based learning methods.
Cambridge Univ. Press. Cambridge, 2002.
5. Hastie T. The elements of statistical learning: Data
mining, inference, and prediction. Springer. New
York, 2001.
6. Koch M. Klassifikatorkonzepte zur Steuerung
dynamischer Prozesse. Ilmenau, 1995.
7. Quinlan J. C4.5 programs for machine learning.
Kaufmann. San Mateo, 1993.
8. Russo M., Jain C. Fuzzy learning and applications.
CRC Press. Boca Raton, 2001.
9. Seiffert U. Self-organizing neural networks: Recent
advances
and
applications.
Physica-Verl.
Heidelberg, 2002.
10. Seitz M. Speicherprogrammierbare Steuerungen:
Von den Grundlagen der Prozessautomatisierung bis
zur vertikalen Integration. Fachbuchverl. Leipzig im
Hanser-Verl. Mnchen, 2003.
Fig. 6. Implemented decision tree as function
Fig. 7. Global variables and program POE
5. Decision Tree Tool

For the realization is a software tool developed that
supports the user by the decision tree building and
implementation process.
The decision tree tool contains the following
components:
1.
import of the feature data,
2.
creating training and validation data set,
3.
decision tree generation,
4.
decision tree pruning,
5.
PLC decision tree source code generation and
6.
PLC source code export for CoDeSys6 V2.3
The tool is programmed in Microsoft Visual Studio.net
with C# and the generated PLC source code is tested
with CoDeSys V2.3.9.3. and executed on
a Wago 750-841 controller.
6. Conclusions
In times of decreasing innovation cycles becomes the
area of automatic source code generation more
imported. This challenge is also valid for automation
engineering. Hence is essential to develop methods for
automatic source code generation in the field of PLC
applications. The paper shows a small introduction into
this area on the example of decision trees and can be
seen as motivation for further research activities.
6 A common PLC programming tool developed by 3S
82

A NEW WIRELESS SOLUTION TO TRANSMIT AND EVALUATE BIOMEDICAL

SIGNALS UNDER LOW SIGNAL NOISE RATIO TO CONTROL THE STATE OF
RELAXATION
Daniel TROMMER*, Markus GREULICH*, Rudolf BAUMGART-SCHMITT*, Maria KRAUTWALD*
Christian WALTHER*, Andreas WENZEL**
*University of Applied Sciences Schmalkalden, Faculty of Electrical Engineering, Germany
**Fraunhofer-Institute for Information and Data Processing, Applications Center System Technology, Germany
Abstract: A handheld data recording system originally

developed in our laboratory to extract EEG features was
extended in two ways: to transmit the data wireless and
to evaluate ECG.
By means of these features it is possible to control the
state of relaxation and the general status of
cardiovascular system e.g. before the operation.
Especially in medical environments it is necessary to
take the low signal noise ratio into consideration.
According to this the serial protocol of Bluetooth and
the Sigma- Delta- Analog Digital converter were
applied.
The prototype of the developed system is able to preprocess the data in real time mode and to stores the raw
data and the extracted features for classification,
forecasting and controlling purposes.
The EEG was originally measured by electrodes wired

with the processing system. A low- pass filter and a
sigma delta analog-digital converter were used to get a
digital EEG signal for processing.
There are 10 light emitting diodes (LED) with red,
yellow and green color which are showing the states of
relaxation.
The idea of the device is to measure and display the
states of relaxation of a subject. The subjects control
their own state of relaxation with techniques like
breath- control, defocusing a.e., they get the feedback of
the success of their techniques by looking at the LEDs.
They should continuously improve their skills in
obtaining the state of relaxation. Depending on the level
of their skills it could be possible to relax later without
the support by this device.
To measure the EEG we used wired bipolar one channel
frontal site EEG electrodes as shown in Fig.1 up to
now.
Keywords: Bluetooth, ECG, EEG, microcontroller,

relaxation, SD, wireless transmission.
1. Introduction
In some environments, such as medical and industrial, it
is necessary to integrate components for data
acquisition close to the subjects to be controlled so there
is no place for the processing unit.
Different measurement and control tasks are requiring
real time processing, storing data, to be robust against
low signal noise ratio and small mobile devices.
The solution presented in this paper was developed and
configured to fulfill these requirements.
The purpose of the device called Quilax presented in
this paper is to control the states of relaxation of a
subject. The device was originally developed to control
the states of relaxation by measuring the electrical
activity
produced
by
the
brain,
the
Electroencephalography (EEG).
Fig. 1. Draft of a wired bipolar one channel frontal site

EEG electrode
83
We suppose that the state of relaxation is related with

the relative power of the alpha- band.
The alpha-band of the EEG Signal is placed between 8
and 13 Hz. The relative power of the alpha- band
relPalpha is calculated by
relPalpha
Palpha
Ptotal
100% ,
2. Methods
2.1. Integration of the SD Card device
For integrating a SD card slot into a microcontroller
driven device it is necessary to implement an algorithm
to manage the file system. After this the SD card can be
easily red by a personal computer (PC) as well as in
other handheld devices with SD card support.
For implementing the file allocation table (FAT)
management on the microcontroller device we use the
Portable FAT Library for MCU Applications [3].
There are two different ways to communicate with a SD
card. On the one hand in the fast 4- bit communication
mode and on the other hand in the simple serial
peripheral interfaces (SPI) mode. We decided to use the
SPI mode. This modus turned out to be as fast enough
to transmit the data during one sampling period. This
mode is efficient enough to implement it on a
microcontroller with small program memory.
The device with integrated SD- card data storing is
shown in Fig.2.
(1)
with Ptotal as the accumulated power of the EEG

between 0 64 Hz, Palpha as the accumulated power of
the EEG between 8 and 13Hz.
We measured the EEG with a sampling rate of 250
samples per second.
The NyquistShannon sampling theorem
f sample 2 f max ,
(2)
with fsample as the sampling rate restricts fmax as the

maximum frequency of the evaluable EEG band.
That means it is possible to analyse signals up to 125
Hz. To avoid the aliasing effect it was necessary to
integrate a low pass (anti-aliasing) filter with a -3dB
frequency of 110 Hz. The 16 bit AD7715 from Analog
Devices, a sigma delta analog digital converter (ADC), was applied to have the useful feature according
to over sampling of the analog signal. The signal noise
ratio was dramatically improved in the interested signal
range.
To process the spectrum of the EEG signal we are using
a simplified fast real time sliding DFT algorithm (RFT)
[1, 2]. In contrast to the original versions of discrete
Fourier transformation (DFT) algorithms the processing
time of our recursive version of the DFT do not depend
on the number of elements of the input vector.
Furthermore the algorithm used a Look- Up Table
(LUT) to store the values of the periodic trigonometric
functions which have been processed at the
initialization of the algorithm. To save further
processing time it is possible to process only
frequencies we are interested in. Therefore the complete
processing time could be minimized.
The extracting of the EEG feature and the classification
is performed by the microcontroller MSP430F1611.
We proposed to extend the described device in several
ways. At first we integrate a Secure Digital (SD) Card
slot to store raw and extracted data for a further
processing. Secondly we implemented a wireless one
channel EEG electrode. It is now possible to collect data
of the subject in a wireless mode. Third we
implemented an electro cardiogram processing (ECG)
in the handheld device.
Fig. 2. Picture of the Quilax in revision with integrated

SD- card device
The microcontroller MSP430F1611 has two integrated
SPI ports. The original device uses one SPI port for the
communication with the - ADC. Therefore the
second SPI port can be used for communication with
the SD Card.
84
Our intention was that the microcontroller stores the

raw data values in real time. That means the raw data
value should be stored in one sample period. In fact of
the use of a FAT management by the microcontroller
the transmission of one data value only is not efficient,
because the FAT has to be updated too. The updating
time of the FAT is not linearly dependent on the number
of bytes stored before. Therefore we try to store the raw
data values in blocks by using of two memory buffers.
The memory space for both buffers is allocated with
512 bytes. With 512 bytes for each buffer 128 ADC
values with 16 Bit and 128 extracted features with 16
Bit can be stored in each buffer.
If there is a new analog value completely converted an
interrupt occurs. With the new raw value the features
will be extracted and displayed. After this there is time
to perform the data storing.
Over the period of time the first buffer will be filled
with the data the other buffer is ready for storing.
Because of this buffer management the data can be
stored without delaying the acquisition and extraction
process.
The safe transmission refers to the protection of

personal rights. That means no one should get the data
without the agreement of the owner and the electrode
should be able to communicate with almost every
handheld device or PC. Therefore we decided that
Bluetooth is the right transmission method for this
application. Up to now we developed and tested the first
prototype successfully.
The wireless one channel electrode adapter consists of
four main parts:
- the power supply,
- the amplifier and data acquisition,
- the Bluetooth module and
- the msp430F2013 microcontroller.
As a power supply we use a 500mAh lithium- ionicaccumulator. With an overall power consumption of
40mA (in transmission mode) and the 500mAh
accumulator the electrode has an operation time of more
than 10 hours approximately. With a deep discharge
protection, by using a mirror comparator, the use of
regular accumulators from mobile devices with higher
capacities is possible. Operation times with more than
40 hours are supposable.
The amplifier and data acquisition part as shown in
Fig.4 is necessary for being robust against low signal
noise ratio. Only with an efficient amplifier the
extraction and classification oft the EEG signal can be
performed.
2.2. Implementation of a wireless electrode

The electrode band as shown in Fig.3 uses a cable to
realize the connection to the microcontroller driven
device of Fig. 2. Our aim was to make the connection
more comfortable to the relaxing subject by a wireless
data transmission.
Fig. 4 The block diagram of the amplifier and data

acquisition consists of the VIN+ input of the EEG
electrode (1), the VIN- input of the EEG electrode (2),
the common input of the EEG electrode (3), the
instrumental amplifier (4) and the impedance converter
(5).
The instrumental amplifier increases the values of the
inputs VIN+ VIN- with a factor of 5 referred to a voltage
based on the common input of the electrode. After this
Fig. 3 Picture of a wired one channel EEG electrode

We consider to use a method that has a safe
transmission and which could be easily implemented.
85
an active anti-aliasing filter and a - ADC follows in

Fig. 4. The sampling rate of the - ADC is 512
samples per second.
For the management and control between the Bluetooth
module and the - ADC we use a MSP430F2013
microcontroller. The firmware of the microcontroller
only has the task to fetch the data from the - ADC
and send it to the Bluetooth module on its internal serial
port.
The btm-112 from RAYSON was selected as Bluetooth
module. This module is able to send data by the serial
protocol of Bluetooth. The module has a self
management for the frequency hopping and the
connection process.
methods for multi feature processing of the state of

relaxation.
3. Conclusion
We extended the Quilax device in several ways.
By implementing a wireless one channel EEG electrode
we afford the possibility to become in relaxation in a
more comfortable way for the subject. Furthermore we
offer a small mobile alternative to measure and control
units of the market. The device we developed has the
potential to fulfill several measurement tasks with
signals up to 250Hz.
The implementation of the ECG processing methods
allows us to assist our EEG analysis to predict the state
of relaxation. Only with the multi feature processing
with EEG and both ECG methods a measurement and
control of the state of relaxation is accurate enough.
By the integration of the SD- card device we are able to
collect data for post- processing, analysis and to archive
training results of the relaxation techniques.
2.3. Implementing of ECG processing

The implemented processing for EEG is one method to
measure the state of relaxation. Another method is to
process ECG features. There are several ECG features
which are representing the state of relaxation of a
subject.
One feature which reflects the state of relaxation is the
power density at 0.1Hz [4]. We implemented the
extraction of the spectrum, in the same way like the
power- spectrum extraction of the EEG signal, with the
RFT. Therefore we calculate the relative power relP0.1Hz
of the 0.1Hz band with the equation
relP0.1Hz
P0.1Hz
100% ,
Ptotal
4. References
1. Baumann, C. RFT: a simplified fast real time sliding
DFT algorithm, Convict Episcopal de Luxembourghttp://www.restena.lu/convict/Jeunes/ultimate_stuff
,RFT.pdf, 2005.
2. Baumgart-Schmitt, R. systemanalytische Anstze
in der kognitiven Psychophysiologie unter
besonderer Bercksichtigung stationrer visueller
Strukturbildungsprozesse, Humboldt Universitt
Berlin, 1986.
3. Sham I., Hue W., Rizun P. Portable FAT Library for
MCU
Applicationshttp://www.circuitcellar.com/
magazine #176, March 2005.
4. Peng, C.-K. Heart rate dynamics during three forms
of meditation, International Journal of Cardiology,
February 2003.
5. Oida E., Moritani T., Yamori Y. Tone-entropy
analysis on cardiac recovery after dynamic exercise
Proceeding of 4th eMBEC 2008
6. Kamiya N., Yokoyama K., Niwa S. Evaluation of
Parasympathetic Nerve Activity Level Monitoring
System, IEIC Technical Report, VOL.102, NO.727,
2003.
(3)
with Ptotal as the accumulated power of the ECG

between 0 64 Hz, P0.1Hz as the power of the ECG at
0.1Hz. The state of relaxation is also shown by the
LEDs.
Furthermore we implemented the Percentage Index (PI)
of heart period duration and its probability [3].
The Percentage Index PI(n) of heart period variation is
calculated by
PI (n)
[ H (n) H (n 1)] 100

,
H ( n)
(4)
for
1 n N 1,
(5)
with n as the serial number of beats, H(n) as the time

series of the heartbeats, N is the total number of
heartbeats in the range.
The PI becomes positive when the heart rate accelerates
and it becomes negative when the heart rate inhibits.
Oida et.al. supposes that the probability of the PI
represents the autonomous nerve activity which reflects
the state of relaxation [5, 6].
We implemented and tested these two methods of
measuring the state of relaxation on our device. To
analyse in future if it is possible to connect these
86

CORRECTION OF MEASURING TRACK PROCESSING ERRORS

IN VIBRATION MEASUREMENTS OF COAL WAGON
Mirosaw LUFT, Radosaw CIO, Daniel PIETRUSZCZAK
Kazimierz Puaski Technical University of Radom, Faculty of Transport and Electrical Engineering, Poland
Abstract: The paper outlines investigation outcomes of

decreasing the dynamic errors resulting from non-ideal
transformation of the vibration measuring track of a coal
wagons, Faals type, vibrations. Obtained measuring
errors with correction were compared to errors without
correction. Requirements of the UIC 518 norm are
given.
SIMULINK, a MATLAB-based interactive software
package turned out to be an irreplaceable tool here. It
offers a possibility of an analysis and synthesis of
continuous and discrete dynamic systems.
of these errors during the on-line dynamic

measurements. Ref.: [6, 9, 13].
During vibration measurements of coal wagons which
aim at defining the vehicle safety criteria, problems with
accuracy and speed of measurements occur, especially
in on-line measurements. They are a result of properties
of the transducer input signal reproduction which are
non-ideal, and the requirements imposed by norms.
The UIC-518 norm specifies the measuring operating
frequency range as being between 0.4 Hz and 10 Hz as
well as the band-pass filter used to this end and located
in the measuring chain after the analogue-digital
transducer. The tolerance of the filter is 0.5 dB in the
operating frequency range and 1 dB outside it. The
filter transfer function is written as:
Keywords: railway car vibration, UIC 518 norm,

software
correction,
error
of
measurement,
measurement transducers.
1. Introduction
Before being referred for use as well as during their
operation, means of transport undergo different
examinations that are to confirm their safety and
parameters determining their usefulness for specific
applications in transport. One of such types of
examinations is vibration measurements during a
vehicle movement. Depending on the type of means of
transport different norms are binding which specify the
way of their manufacture. In Europe rail vehicles must
meet requirements included in the document of the
International Union of Railways UIC 518. Ref.:
[3, 5, 13].
The focus of this paper is how the input value is
influenced by measurement errors resulting from the use
of non-ideal devices transforming this input value. They
concentrate on dynamic vibration measurements of a
coal wagon and examine errors caused by the
transformation properties of measuring transducers and
the errors resulting from the filtering methodology
(enforced by the necessity to apply the UIC-518 norm).
The paper shows the software method of correction of
transformation characteristics which diminishes values
A( s )
4 2 f 22 s 2
(1)
2 2 f 1
2 f 2
s
s 4 2 f 12 s 2
s 4 2 f 22
Q1
Q1
where f1 = 0.4 Hz, f2 = 10 Hz, Q1 = 0.71.

For measuring the signal frequency from 0.4 Hz and
10 Hz, the signal delay after filtration compared to the
signal before filtration can be between 625 ms to
25.5 ms, respectively. Measurement errors result from
non-ideal reproduction of the measurement signal by
measuring transducers, amplifiers and other elements of
the measuring track. The measurement of the relative
error depends mostly on the type and accuracy of the
applied measuring transducers. For example, for
measuring transducers revealing sensitivity of 10
mV/ms-2 (used in a coal wagon vibration research), in
dynamic measurements the relative value stands at
about 30 %. A decrease in the measurement errors and
the measurement signal delay can be achieved with the
help of a method of software correction.
87
2. Vibration examinations of railway wagons

according to UIC 518 norm
Amplitude
Filter work
Vibration examinations of rail vehicles are a type of

examinations which are to confirm safety of vehicles
and vibration parameters determining their usefulness
for specific transport applications. The UIC-518 norm
specifies:
conditions of test performance,
measured magnitudes referring to dynamic
behaviour of a vehicle,
conditions allowing for automatic and
statistical measurements.
Rail vehicles are passed as fit for use on the basis of the
test-determined parameters:
the movement of a vehicle on the track of
specific geometry and with a specified
velocity,
wheel-track contact geometry,
vehicle condition,
measurement of a magnitude referring to
dynamic performance of a vehicle,
fulfilling
requirements
of
automatic
measurements and statistical data processing,
assessment of measured magnitudes,
determining border values of measured
magnitudes.
Apart from the above said, the norm defines:
the way of conducting the full measurement
procedure (type of a tested zone of track
geometry, vehicle load),
the way of conducting partial tests,
measuring methods:
correct method (of measuring
interaction of forces affecting the
wheel set and/or measurements of the
wheel set acceleration, vehicle frame
and body).
Simplified method (measurements of
side forces affecting the wheel set
and/or wagon frame and body).
The results obtained can be presented in the form of
characteristics of:
forces of wheel-rail contact pressure in a
crosswise and longitudinal direction on the two
smallest external axes of a wagon or carriage,
depending on whether the wagon/carriage is
equipped with separate axes,
horizontal and vertical acceleration at the
wagon body ends above the carriage or above
the axis inside the carriage box. Measurements
must be carried out on the carriage floor
surface or, when the carriage has no floor on
the top part of the frame at the point indicated
in the test report. Ref.: [5, 13].
Figure 1 shows the response of the band pass-filter
to the sinusoidal input signal at the frequency of 5 Hz.
The signal after filtration is delayed by 175 ms in
relation to the input signal.
Input
signal
Time [s]
Fig. 1. Response of the band-pass filter
Amplitude
Figure 2 shows filter response to the same signal

with addition of the Gauus noise.
Time [s]
Fig. 2. Response of the band-pass filter to the disturbed

sinusoidal signal
Amplification [dB]
Figure 3 depicts theoretical characteristics of filter

amplification and tolerance assumed in the UIC-518
norm.
0
-3
> - 12 dB/octave
>12 dB/
/octave
Tolerance [dB]
0.4
+ 1 dB
0.25
10
+ 0.5 dB
0.4
[Hz]
+ 1 dB
10
100
[Hz]
Fig. 3. Characteristics of correction and tolerance of the filter

specified in the UIC-518 norm.
The UIC-518 norm does not specify what type of

sensors is to be used and what measuring class they
should represent. Freedom in applied equipment causes
that correctness and accuracy of the signal depend on
the designer of the measuring track. Due to signal
processing properties depending on the applied
equipment and their great number it is necessary to
draw particular attention to errors made by them (often
reproduced by the following elements of the measuring
track). Ref.: [13].
88
acceleration determined by means of the model sensor is

a A1 / a A2 1 . Magnitude characteristics for the adopted
border value of amplification 2dB Wr is within the
frequency range above 117 Hz. Ref.: [1, 9]
Transfer function (2) has been determined
experimentally by identification method ARX on the
basis of measurements made on electrodynamics shaker
with high-class standard accelerometer. Ref.: [2, 7]
The transfer function (2) has been determined
experimentally with the use of the ARX identification
method on the basis of measurements taken on an
electro-dynamic shaker with a high-class accelerometer.
Filter transfer function describes dynamic properties of
the band-pass filter (1). The software correction block
represents the correction algorithm. Figure 6 shows the
block diagram of this algorithm. Equations of the
algorithm of correction look as follows:
In conformity with the UIC-518 norm requirements,

vibration diagnostics tests are carried out in two stages:
1. Taking and registering data from measuring
sensors
2. Filtration and developing results obtained from
measurements in laboratory conditions
according to the methodology included in the
UIC-518 norm.
The main reasons for errors in the measurement process
are:
Reproduction of measured quantities by the
measuring system,
Performance properties of the band-pass filter
In order to reduce the filter response delay to the
input signal, software correction was applied. Figure 4
shows a block diagram of the correction system.
Measurements were taken in the MATLAB
environment.
x ( k )
u1( k 1)
1
y ( k ) u1( k )
b2
11u1( k ) 12 u 2( k ) 1 x ( k )
(3)
u 2( k 1) 21u1( k ) 22 u 2( k ) 2 x ( k )
Fig. 4. Block diagram of system correction, awe transducer
input acceleration, awz standard acceleration, ap transducer
output acceleration, af acceleration after filtration, ak
acceleration after correction
y(k-1)
+-
u^1(k-1)
delay
Transducer transfer fcn (represents transfer function

of real accelerometer) is written down as follows:
21
0.03215s 2 1319.6 s 1.338 106

s 2 4.678 10 4 s 2.309 107
11
u^1(k)
(2)
12
u^2(k-1)
Figure 5 presents the magnitude and phase

characteristics of the system whose transmittance is (2).
High amplification visible in the magnitude
characteristics results from the application of sensors of
different sensitivities for examining voltage signals.
22
+
+
+
+++
delay
G ( s)
^
x(k-1)
1/b2
u^2(k)
Fig. 6. Block diagram of correction algorithm: x ( k 1) - the

estimate of input quantity at moment k-1 calculated on the
basis of measurement result y(k-1) and estimation of variable 1
from the previous step; 11...22, 1...2 factors of discrete
model
Wr
Figure 7 presents characteristics of acceleration after

filtration, standard acceleration and acceleration after
use of the correction algorithm. Negative effects of the
filter operation in the form of amplitude of acceleration
change and delays of response were reduced by the
correction algorithm.
The values of 20-25 ms representing a delay of signal
characteristics after filtration in relation to the standard
signal were reduced after correction to 0.1-1 ms.
The results given are for the transducer operation time
of over 0.1 s from the start of the input signal. Table 1
shows average values of maximum acceleration
amplitudes (a series of 2000 measurement samples).
Fig. 5. Frequency characteristics of the measuring track with

accelerometer
The referential level of amplification is determined

as a ratio of a measuring transducers sensitivity to a
model one in the logarithmic scale Wr = -29.87 dB. For
the above value, the ratio of acceleration determined on
the basis of the measuring sensor voltage signal to the
89
3. Correction of processing errors in vibration

investigations of the coal wagon, Faals type
af
Acceleration [m/s2]
awz
ak
Fig. 9. Coal wagon, Faals type
Examination of a coal wagon, Faals type was conducted

on the basis of a measuring system applied in CNTK
Warsaw for determining parameters of railway wagon
vibrations. On the basis of the data obtained in the
measurements, a decision is made about accepting the
vehicle for operation, which is conformable with the
guidelines included in the UIC-518 norm binding in
Europe. These parameters are determined on the basis
of sensor readings. Sensors are located in carefully
specified places of the vehicle moving on the track of
specific geometry and with the set velocity. The basic
elements on which sensors are located include: wagon
frame, wagon body and axle-boxes. The values of
crosswise and vertical acceleration and forces are read
out from them.
The following magnitudes have been selected for the
examination of the impact of reproducing transducer
properties on the measurement result and possibilities of
eliminating errors resulting from this:
1. Crosswise acceleration on the frame of the first
car moving along the straight-line.
2. Crosswise acceleration on the frame of the first
car moving along an arch
3. Vertical acceleration of an axle-box in the first
wagon-set on the right in its movement along
the straight line
4. Vertical acceleration of an axle-box in the first
wagon-set on the right moving along an arch
5. Vertical acceleration of the car-frame above
the first pin of the turn
6. Crosswise acceleration of the car-frame over
the first pin of the turn
The magnitudes describing and at the same time
affecting the behaviour of the wagon on the track are its
velocity and geometry of the track. Examinations used
measurement results obtained at the wagon velocity of
120 km/h on the straight line track and along arches of
the turn radius being 600 m and 90 km/h along the arch
of turn having the radius from 400 m to 600 m.
Effects of the method of software correction of
measuring system transformation errors have been
checked in vibration research of a coal wagon, Faals
type. It is a six-axle self-discharging coal wagon of
special construction and is allowed to move with the
maximum speed of 100 km/h. Figure 10 shows the
research measuring system. The Data from wagon Faals
type block contains measurement results of velocity,
acceleration and coal wagon deflection.
Time [s]
Fig. 7. Characteristics of standard acceleration, acceleration

after filtration and acceleration after correction
Table 1. Acceleration amplitudes of the measuring-track
simulation with correction
Acceleration [m/s2]
Awz
Ap
Af
Ak
3.15
16.66
5.89
3.31
Estimator response to the signal without noise which is

a component of sinus run with frequency: regarding to
a) 5 Hz, 8 Hz, 50 Hz, 200 Hz and 500 Hz and regarding
to b) 1 Hz, 5Hz, 50 Hz, 200 Hz and 500 Hz are shown
in figure 8.
The filter (1) is responsible for elimination of
components that are outside the frequency range 0.4 Hz
10 Hz. In both cases correction algorithm operation
improves the transducer replay characteristic, bringing it
closer to the standard characteristic.
a)
af
awz
Acceleration [m/s2]
ak
Time [s]
b)
af
ak
awz
Time [s]
Fig. 8. Correction of the complex sinusoidal signal
90
Acceleration [m/s2] Relative error [%] Acceleration [m/s2] Gyroscope [deg/s]
Velocity [m/s]
The Transducer model and filtration block represents

laboratory determined dynamic equation of real
accelerometer and dynamic equation of band pass-filter
described by the UIC-518 norm. The algorithm of
correction block includes equations of transformation
error correction.
Fig. 10. Block diagram of the researched measuring system
Standard accelerometer
Measuring accelerometer
Before correction
*1
*2
After correction
Standard accelerometer
Measuring accelerometer
Relative error [%]
Figure 11 presents the measurement results for velocity,

deflection and lateral acceleration at wagon body ends.
A standard accelerometer whose sensitivity was
approximately 30 times higher than that normally used
in railway car vibration practice was used for reading
the standard acceleration value.
The characteristic of measured acceleration after
correction is nearly the same as standard acceleration
characteristics. Relative error values *1 and *2 were
reduced to the values of 217 % and 51 %, respectively.
The mean square error is a measure that indicates high
values of measurement errors and is used in the practice
of dynamic measurements. Characteristics of mean
square errors for a measurement with and without
correction are depicted in Figure 11. A comparison of
the two characteristics reveals a significant decrease in
the error in the case of a measurement with correction
(by about 10,000 times). It is possible to conclude that
after correction a capability of the measuring system to
reproduce the input quantity was improved and
sensitivity of the system which accounts for gross errors
was decreased.
The characteristics of the dynamic absolute error with
and without correction are illustrated in Fig. 12. The
difference between the measurements is grade 103 in
favour of the correction. The absolute error in the case
of dynamic measurements without correction is between
-7.143 and 6.674 m/s. The values of the absolute error
stand at 0.0026 m/s2 and 0.0035 m/s2 for measurements
with correction. Ref.: [5, 6, 10, 13].
Before correction
After correction
*1
*2
Time [s]
Fig. 11. Crosswise vibration of coal wagon body ends
(ym(t)- yrz(t))
Before correction
After correction
Time [s]
Fig. 12. Mean square error measurement
91
Before correction
Absolute error [m/s2]
3.
4.
After correction
5.
6.
Time [s]
Fig. 13. Absolute error measurement
The Table below shows the values of measured

quantities in vibration research of coal wagon Faals
type.
7.
Table 2. The values of measuring quantities

Before
correction
After
correction
Measuring time range t [s]
15
Velocity range [m/s]
128.89 129.28
Railway car deflection

range [deg/s]
0.17 0.06
Relative error range w [%]
1102
263
14.5
217.7
Median of relative error

range M [%]
325.2
0.01
Standard acceleration range

a w [m/s2]
Value of absolute error
range [m/s2]
8.
9.
2.054 2.196
7.143
6.674
10.
0.0026
0.0035
4. Conclusion
11.
Absolute errors in vibration measurements of a coal

wagon Faals type as well as the median of relative error
were considerably decreased when the algorithm of
correction was applied.
Owing to the said algorithm the problems with accuracy
of the measuring signal reproduction are significantly
reduced. As a result the requirements of the UIC 518
norm are fulfilled.
12.
13.
5. References
1.
2.
Analog devices technical references books:

Practical design techniques for sensor signal
conditioning, Analog Devices, Inc., USA 1999
Andrae J., Hons C., Sawla A..: High precision
force transducer with new method for online
92
compensation of parasitic effects, Measurement

Vol.33, Elsevier, Ltd., 2003
Beards C. F.: Engineering vibration analysis with
application to control systems, Edward Arnold,
London 1995
Cakar O., Sanliturk K.Y.:
Elimination
of
transducer mass loading effects from frequency
response functions, Mechanical Systems and
Signal Processing Vol. 19, Elsevier, Ltd. 2005
Cio R.: Correction of measuring transducers
dynamic characteristics in vibration of railway car,
Doctoral
dissertation,
Biblioteka
Gwna
Politechniki Radomskiej, Radom 2007, (in Polish)
Cio R.: ARX method in the context of
identification of dynamic correction parameters of
vibration measuring system, Prace naukowe PRad,
Elektryka 1(9), Radom 2005, (in Polish)
Cio R., Luft M.: Valuation of software method of
increase of accuracy measurement data on example
of accelerometer, Advances in Transport Systems
Telematics, Monograph, Faculty of Transport,
Silesian University of Technology, Katowice,
2006.
Cio R., Luft M.: Correction of transducers
dynamic characteristics in vibration research of
means of transport part 1 simulations and
laboratory research, 10th International Conference
Computer Systems Aided Science, Industry and
Transport, Transcomp 2006, vol.1, Zakopane,
2006.
Jakubiec J.: Real-time numerical reconstruction of
instantaneous values of dynamic input quantities of
nonlinear transducers, Politechnika lska,
Zeszyty Naukowe nr 964, Gliwice 1988, (in
Polish)
Luft M., Cio R.: Increase of accuracy of
measurement signals reading from analog
measuring
transducers,
Zeszyty
Naukowe
Politechniki lskiej 2005, Transport z. 59, nr kol.
1691, Gliwice 2005.
Luft M., Cio R., Pietruszczak D.: Increasing the
accuracy of non-linear measuring transducers
processing, Advances in Transport Systems
Telematics, p. 189-196, (ISBN 978-83-206-17153), WK, Warszawa 2008.
Szychta E.: Zero voltage switched multiresonant
converters analysis and design, Monograph
University of ilina, 2008
Testing and approval of railway vehicles from the
point of view of their dynamic behaviour Safety
Track fatigue Ride quality, 3rd edition, Leaflet
UIC 518, International Union of Railways,
October 2005.

REGRESSION MODELLING USING

ADAPTIVE BASIS FUNCTION CONSTRUCTION
Gints JEKABSONS
Institute of Applied Computer Systems, Riga Technical University, Latvia
corresponding sample. Using the finite number n of

training samples (x i , y i ) , (i 1,2,..., n) one wants to
Abstract: The task of estimating good predictive

models from available finite (training) data acquired
through, e.g., experiments or simulations is common in
virtually all fields of science and engineering. The most
commonly used predictive models in the case of
continuous dependent variable are polynomial
regression models acceptable predictive performance of
which are usually attempted to achieve by adjusting a
maximal degree and using subset selection techniques
requiring non-trivial exponential runtime. The paper
outlines a different approach not requiring manual
degree adjustments and running in a polynomial time
and demonstrates its efficiency in a number of
applications in the field electrical and control
technologies.
build a model F that allows predicting the output values

for yet unseen input values as closely as possible.
In practical applications of regression most commonly
low-degree polynomial models are used. Generally, a
linear regression model may be defined as a linear
expansion of basis functions
k
F ( x) ai f i ( x) ,
where a (a1 , a 2 ,..., a k ) are models parameters, k is

the number of basis functions included in the model
(equal to the number of models parameters), and fi(x)
( i 1,2,..., k ) are basis functions of the input x. As such
models are linear in the parameters, the estimation of
their parameters is typically done using the Ordinary
Least-Squares (OLS) method [6,22] minimizing the
squared-error:
Keywords: Regression modelling, approximation,

Adaptive Basis Function Construction, polynomial
regression, sparse polynomials.
1. Introduction
The task of estimating good predictive models from
available data is common in virtually all fields of
science and engineering. The goal of the task is to
estimate unknown (input, output) dependency (or
model) from training data (consisting of a finite number
of samples acquired through, e.g., experiments or
simulations) with good prediction (generalization)
capabilities [2,6].
One of the specific model estimation tasks is regression
estimation of an unknown real-valued function in the
relationship
y G (x) ,
(2)
i 1
2
a arg min y i F (x i ) .
a
(3)
i 1
The basis function representation enables moving

beyond pure linearity, by defining nonlinear
transformations of x (e.g., products of input variables in
polynomials) while still working with linear models
(and employing OLS).
For example, for d = 1 a polynomial model of fixed
degree p can be expressed as follows
(1)
F ( x) ai x i .
where is independent and identically distributed

random noise with zero mean, x x1 , x 2 ,..., x d is d-
(4)
i 0
A low-degree polynomial has low number of unknown

parameters and tends to smooth out noise in the data.
However, it may exhibit too little flexibility for
modelling highly nonlinear behaviours (causing
underfitting). Higher-degree polynomials can be used,
dimensional input, and y is scalar output. The estimation

is made based on a finite number of samples (training
data) provided in form of matrix x of input values for
each sample and vector y of output values for each
93
but they may contain too many parameters and therefore

either overfit the data or prohibit parameter estimation
because the number of parameters exceeds the number
of training data samples.
The most popular approach to controlling models
complexity is subset selection [6,22,23]. The goal of
subset selection is from a fixed full predetermined
dictionary of basis functions to find a subset that
corresponds to a model (a sparse polynomial) of the
best predictive performance.
Before performing the actual subset selection the user
must first predefine the dictionary basis functions of
which will be used for model generation. This is usually
done by setting the maximal degree of a full polynomial
and taking the set of its basis functions.
In polynomial regression, increase in the full models
degree leads to exponential growth of the number of
basis functions in the dictionary [2,6] leading to doubleexponential growth of the number all possible subsets
of basis functions. Efficient heuristic subset selection
methods such as Sequential Forward Selection, SFS
(also known as Forward Selection or Stepwise Forward
Selection [6,22]) and many others (see, e.g., [23])
considerably reduce the time. However, it is still
exponential in the degree and the number of input
variables [10].
The problem is that the approach of subset selection
assumes that the chosen fixed finite dictionary of the
predefined basis functions contains a subset that is
sufficient to describe the target relation sufficiently
well. However, generally the required maximal degree
is not known beforehand and needs to be guessed (or
found by additional search over the whole subset
selection process) since it will differ from one
regression task to another. In many cases (especially
when the studied data dependencies are complex and
not well studied) this means either a non-trivial and
long trial and error process or acceptance of a possibly
inadequate model.
In [11,12] we proposed an alternative approach to that
of typical subset selection Adaptive Basis Function
Construction (ABFC). The goal of this procedure is to
overcome some of the limitations associated with the
subset selection approach outlined above. ABFC is
developed for sparse polynomial regression model
building without restrictions on models degree and
enables model building in polynomial time instead of
exponential. The required basis functions are
automatically adaptively constructed specifically for the
data set at hand without using a restricted fixed finite
user-defined dictionary. The dictionary of ABFC is
infinite and polynomials of arbitrary complexity can be
generated.
In this paper we will shortly review the ABFC approach
(Section 2), briefly characterize two of its special cases
(Section 3), and demonstrate the efficiency of the
ABFC methods in a number of applications in the field
electrical and control technologies (Section 4).
2. Adaptive basis function construction

The basis functions in a polynomial regression model
generally can be defined as a product of original input
variables each with an individual exponent:
d
r
f i ( x) x jij ,
(5)
j 1
where r is a k d matrix of nonnegative integer

exponents such that rij is the exponent of the j-th
variable in the i-th basis function. Note that when for a
particular i-th basis function j : rij 0 , the basis
function is the intercept term.
Given a number of input variables d, matrix r with a
specified number of rows k and with specified values of
each of its elements completely defines the structure of
a polynomial model with all its basis functions.
Moreover, as neither the upper bounds of r elements
values nor the upper bound of k are defined, it is
possible to generate polynomials of arbitrary
complexity, i.e., of arbitrary number of basis functions
each with arbitrary exponent for each input variable.
In order to efficiently build a sufficiently good
regression model for a particular dataset an efficient
search mechanism is required enabling searching in an
infinite space of polynomial models. The search
mechanism of ABFC is organized as follows (see
[11,12] for details).
The search is started from the simplest model the
model with one basis function which corresponds to the
intercept term. New models are generated using socalled model refinement operators which enable adding,
copying, modifying, and deleting the rows of r, i.e.,
adding, copying, modifying, and deleting the basis
functions of the model (not only adding and deleting,
like in the subset selection methods). Next an efficient
search strategy and a model evaluation measure are
required. In the special case of ABFC termed Floating
ABFC (F-ABFC) the search strategy of Sequential
Floating Forward Selection [20,21] is adapted and the
corrected Akaikes Information Criterion (AICC) [8] is
employed. The termination condition of the search
process is met when the algorithm has generated a
model which cannot be further refined using any of the
refinement operators.
Additionally, in order to lower the general model
building issues of selection bias and selection instability
[1,9,16,19,23], a technique of model averaging (also
called ensembling or combining) is carried out. This
leads to the second special case of ABFC termed
Ensemble of F-ABFC (EF-ABFC) [9].
The ABFC methods attempt to model arbitrary
dependencies in data with little or no knowledge of the
system under study. The user is normally not required to
tune any hyperparameters. However, if there is
sufficient additional domain knowledge outside the
specific data at hand it may be appropriate to place
some constraints on the final model. If the knowledge is
94
fairly accurate, such constraints can improve the

accuracy while saving computational resources.
For example the constraints might be one or more of the
following: 1) limiting the maximal degree of all the
basis functions (similar to the subset selection), i.e.,
d
i : 0
rij p ; 2) limiting the maximal value of
weight function [3,14]; 6) Multivariate Adaptive

Regression Splines (MARS) with and without crossvalidated degrees of freedom [4]; 7) multiquadric
Radial Basis Function (RBF) interpolation [5]; 8)
Kriging interpolation [17]. Note that for FP the maximal
degree p is varied in such interval that the number of
basis functions m is always lower than or equal to the
size of the training set n. This is because with higher
values of p the number of full models parameters
exceeds the number of training data samples.
All the methods are implemented in the VariReg
software tool version 0.9.13 freely available at
http://www.cs.rtu.lv/jekabsons/. For all the methods all
the parameters had default values except were stated
otherwise.
The modelling methods are compared in terms
predictive performance of the built models estimated
using resampling techniques [15]. The specific
resampling technique used in an application depends on
the amount of the available data. The errors were
measured using the Relative Root Mean Squared Error
(RRMSE) criterion
j 1
the exponent for each particular input variable in all the

basis functions, i.e., i : 0 rij p j , where p j is
maximal exponent of the j-th variable; 3) restricting
contributions of specific input variables that are not
likely to interact with others so that those variables can
enter the model in basis functions only solely with
exponents of all other variables fixed to zero.
These constraints, as well as far more sophisticated
ones, can be easily incorporated in the ABFC. However,
note that in all the experiments described in this paper
we did not use any kind of constraints.
Finally it must be noted that in special cases when the
data is of low dimensionality (e.g., d 4 ) and/or the
existing structure in the data is very complex (so a very
complex model is required) the search algorithm may
get stuck in a local minimum too early in the search
returning a too simple and underfitted model [10,11].
As a remedy for this problem some of the refinement
operators are applied recursively [10,11]. In this way
the probability of getting stuck in too early local minima
is considerably decreased (though not removed).
A more detailed description of F-ABFC, and EF-ABFC,
as well as detailed empirical comparisons to subset
selection methods and other popular regression
modelling methods is given in our previous studies
[9,10,12]. The biggest disadvantage of the EF-ABFC
compared to F-ABFC is that it requires larger
computational resources. However, the fact that before
the model combining the models are built completely
separately allows for an easy parallelization of the
process.
In [13,14] we applied the ABFC methods in
metamodelling for optimum design of laser welded
sandwich structures and glass fibre composite
structures. However, the methods of ABFC are really a
general-purpose regression modelling tools.
The next section will demonstrate the efficiency of the
ABFC methods in a number of applications in the field
electrical and control technologies.
RMSE
y i F (x i ) , (6)
100% i
2
SD
y i y
2
RRMSE 100%
where RMSE is Root Mean Squared Error calculated in

the test data, SD is standard deviation either in the test
data or in all the data (depending on the specific
application), and y is the mean of the y values. The
closer the value of RRMSE is to zero the more accurate
is the model.
Additionally, note that for the problems in which the
dataset had only two input variables (the first and the
last) the refinement operators of F-ABFC and EFABFC were applied recursively four times in each
iteration.
3.1. Ball thrower simulation
The problem considered here (P1) is prediction of the
accuracy of a robotic arm throwing a ball to the target.
The aim is to predict the distance from the ball to the
target as closely as possible when the ball is hitting the
ground while taking different other parameters into
account. Accurate prediction of the distance enables
optimization of arms control parameters for high
throwing accuracy.
Fig. 1 depicts some of the ball thrower simulation
parameters. The simulations are performed using free
software available at http://www.cs.rtu.lv/jekabsons/.
The following parameters are tuneable: 1) initial
velocity of the ball in the moment of its release (x1); 2)
the angle of the arm when the ball is released (x2); 3) the
distance from the arms base to the target (x3); 4) the
length of the robot's arm (x4); 5) air resistance (x5); 6)
the weight of the thrown ball (x6); 7) gravity (x7); 8) the
distance from the arms base to the wall (x8).
3. Applications
This section describes a number of regression modelling
applications where the ABFC methods were used as
well as demonstrates results of empirical comparisons
of the ABFC methods to other well known and widely
used regression modelling methods.
The methods compared are the following: 1) full
polynomials (FP) of different degree; 2) sparse
polynomials of different degree built using one of the
most popular subset selection methods SFS utilizing
AICC as the evaluation criterion; 3) F-ABFC; 4) EFABFC; 5) Locally Weighted Polynomials (LWP) of
different degree utilizing cross-validated Gaussian
95
performance estimation of the regression modelling

methods the 64 samples were randomly selected from
the available data while the remaining 8128 samples
were used as a test set. The process was repeated 10
times and the results were averaged. The SD was
calculated using test data sets.
3.3. Servomechanism rise time
rise time of a servomechanism (the time required for the
system to respond to a step change in a position set
point) in terms of two gain settings and two choices of
mechanical linkages. The dataset as well as a more
detailed description of the problem are available at the
UCI repository (http://archive.ics.uci.edu/ml/).
The properties of the dataset are the following: d 4
and n 167 . For predictive performance estimation of
the regression modelling methods 10-fold CrossValidation was used. The SD was calculated using test
data sets.
Fig. 1. Ball thrower parameters (see text for details)
The problem pursued here is the prediction of the

distance from the ball to the target when the ball is
hitting the ground while varying the ball release angle
(from 45 to 135 degrees) and the distance from the
arms base to the target (from 120 to 400). All the other
variables are held fixed x1 = 85, x4 = 100, x5 = 0.08,
x6 = 1, x7 = 9.81, x8 = 490 (here the ball is never hitting
the wall).
and n 100 (a regular 10 10 grid). For predictive
methods 10-fold Cross-Validation [15] was used. The
SD was calculated using test data sets.
Fig. 2 shows the surface of the model built by EFABFC predicting the distance. When the distance from
the arm to the target (x3) is known, balls distance to the
target (i.e., throwing accuracy) can be minimized by
changing ball release angle (x2).
3.4. Robot vision

shoulder-joint angle of a six-joint robot arm learning to
grasp an object (a small brick) presented visually on a
table surface. The dataset is taken from a study
described in [7]. Each data sample in the dataset
combines the visual information of the brick with the
grasping arm posture of the robot. To collect a data
sample, the robot put the object on the table, recorded
the corresponding arm posture, removed the arm, and
took a picture of the brick. The visual information
consists of a 4 4 -pixels grid providing a coarsegrained view of the table surface and a histogram
showing the edge distribution over four orientations
within the camera image.
and n 100 (from available 3368). For predictive
methods the 100 samples were randomly selected from
the available data while the remaining 3268 samples
were used as a test set. The process was repeated 10
times and the results were averaged. The SD was
calculated using test data sets.
3.5. Electromagnetic
Reluctance Drives
Fig. 2. Surface of the model built by EF-ABFC
processes
in
Switched
The problem considered here (P5) is prediction of flux

linkages between the poles of the rotor and the stator in
a Switched Reluctance Drive [11,18]. The problem has
two input variables, phase current and rotor position,
and an output variable, flux linkage. The data is
generated using finite element simulations. A more
detailed description of the problem is given in [18].
and n 24 . As the number of samples is very small,
for predictive performance estimation of the regression
modelling methods Leave-One-Out Cross-Validation
3.2. Robot arm forward kinematics

distance of the end-effector of an 8 link all-revolute
robot arm from a target. The aim is to predict the
distance, given the angular positions of the joints. The
datasets from the simulations as well as a more detailed
description of the problem are available at the Delve
repository (http://www.cs.toronto.edu/~delve/).
and n 64 (from available 8192). For predictive
96
[15] was used. The SD was calculated using all the

available data.
whole modelling process in order to evaluate the

methods reliably.
The ABFC methods on the other hand do not require
the user to choose the degree. Instead they
automatically adapt to the data at hand.
Additionally, in the performed experiments the ABFC
methods also compare rather well to the state-of-the-art
non-polynomial methods MARS, RBF, and Kriging.
3.6. Analysis of the results

Table 1 presents the results of the performed regression
modelling experiments. In the first four problems (P1,
P2, P3, and P4), where a larger amount of data is
available, EF-ABFC gives better or similar results than
those of F-ABFC. In the fifth problem (P5) however
F-ABFC outperforms EF-ABFC. This is because if
there is too little data, the gains achieved via an
ensemble may not compensate for the decrease in
accuracy of individual models, each of which now sees
an even smaller training set.
In P1, P2, and P5 the subset selection method SFS
either outperforms full polynomials only a little or
behaves worse. In all the three problems LWP
succeeded to outperform SFS. However, with P5 the
best results are reached using full polynomials while
apparently both SFS and LWP are just too complex
modelling methods for such a little amount of data. In
P3 and P4 the SFS outperforms FP as well as LWP.
4. Conclusions
In this paper we shortly reviewed the ABFC approach,
briefly characterized two of its special cases F-ABFC
and EF-ABFC, and demonstrated the efficiency of the
methods in a number of applications in the field
electrical and control technologies while comparing
them to other widely used methods.
The ABFC methods, in contrast to the methods of
subset selection, do not require the user to predefine the
maximal degree or the dictionary of the basis functions.
Instead they automatically adapt to the particular data at
hand.
Overall, in the performed empirical experiments it is
concluded that in the considered problems the adaptive
model building methods of ABFC can outperform the
other polynomial regression modelling methods and
compare rather well to the other state-of-the-art
methods MARS, RBF, and Kriging. The ABFC
methods can be viewed as competitive tools for
regression modelling.
Directions of future research include employment of
ABFC or derivations of it in different real-world
applications, further theoretical studies concerning the
efficiency of ABFC, as well as adaptation of ABFC for
other types of tasks, e.g., classification.
F-ABFC and EF-ABFC together with many other
regression modelling methods are implemented in
VariReg software tool freely available for noncommercial research and educational purposes at
http://www.cs.rtu.lv/jekabsons/.
Table 1. Average predictive performance of the modelling

methods in terms of RRMSE
Method
FP, p = 1
FP, p = 2
FP, p = 3
FP, p = 4
FP, p = 5
SFS, p = 1
SFS, p = 2
SFS, p = 3
SFS, p = 4
SFS, p = 5
LWP, p = 1
LWP, p = 2
LWP, p = 3
LWP, p = 4
LWP, p = 5
F-ABFC
EF-ABFC
MARS
MARS, CV
RBF
Kriging
P1
91.88
57.79
31.48
27.65
18.06
91.88
56.83
35.65
29.66
27.12
28.70
20.35
23.01
16.17
16.38
17.60
12.77
14.69
15.60
11.74
14.45
P2
P3
29.94 79.29
35.37 52.30
- 44.44
- 61.85
- 382.65
29.94 79.29
28.58 54.31
49.67 49.36
53.91 41.06
72.73 39.59
25.75 48.84
35.68 45.90
- 48.81
- 127.84
- 305.94
41.15 41.66
26.25 42.04
35.54 38.63
34.88 38.24
27.19 39.67
28.02 34.02
P4
12.33
13.04
12.02
16.04
12.32
15.98
9.36
17.77
15.13
16.60
9.38
P5
36.19
29.83
13.20
12.76
7.80
36.19
28.79
13.21
9.49
10.09
12.89
15.51
9.43
10.05
12.41
8.00
14.58
36.31
28.72
12.60
4.66
5. References
1. Breiman L. Heuristics of instability and stabilization
in model selection. Annals of Statistics, vol. 24,
1996. p. 2350-2383.
2. Cherkassky V., Mulier F.M. Learning from Data:
Concepts, Theory, and Methods, 2nd ed. WileyIEEE Press, 2007. 538 p.
3. Cleveland W.S., Devlin S.J. Locally weighted
regression: An approach to regression analysis by
local fitting, Journal of the American Statistical
Association, vol. 83, 1988. p. 596-610.
4. Friedman J.H. Multivariate Adaptive Regression
Splines (with discussion), The Annals of Statistics,
vol. 19, no. 1, 1991. p. 141.
5. Gutmann H.-M. A Radial Basis Function Method
for Global Optimization, Journal of Global
Optimization, vol. 19, 2001. p. 201-227.
6. Hastie T., Tibshirani R., Friedman J. The Elements
of Statistical Learning. Springer, 2003. 552 p.
In all the problems the ABFC methods either

outperformed FP, SFS, and LWP or at least gave
comparable results. Additionally it must be noted that in
practice with FP, SFS, and LWP very commonly the
modelling task is done in a one-shot manner choosing
the degree p only once, which makes it difficult to
obtain good results (for instance in the experiments
performed here the best value for p was almost each
time a different one). An alternative is to check results
for all the values of p in some interval but this takes a
much larger amount of computational resources and
demands for an additional cross-validation loop over the
97
7. Hoffmann H., Schenck W., Moeller R. Learning

visuomotor transformations for gaze-control and
grasping. Biological Cybernetics, vol. 93, 2005. p.
119-130.
8. Hurvich C.M., Tsai C-L. Regression and Time
Series Model Selection in Small Samples.
Biometrika, 76, 1989. p. 297-307.
9. Jekabsons G. Ensembling Adaptively Constructed
Polynomial Regression Models, International
Journal of Intelligent Systems and Technologies
(IJIST). vol. 3, no. 2, 2008. p. 56-61.
(http://www.waset.org/ijist/v3/v3-2-11.pdf)
10. Jekabsons G., Lavendels J. A Heuristic Approach
for Surrogate Modelling. Proceedings of Conference
Applied
Information
and
Communication
Technologies, Jelgava, Latvia, 2008. p. 11-20.
11. Jekabsons G., Lavendels J. A Heuristic Approach
for Surrogate Modelling of Electro-technical
Systems. Proceedings of International Conference
Electrical and Control Technologies 2008,
Kaunas, 2008. p. 62-67.
12. Jekabsons G., Lavendels J. Polynomial Regression
Modelling using Adaptive Construction of Basis
Functions. Proceedings of IADIS International
Conference, Applied Computing 2008, Algarve,
Portugal, 2008. p. 269-276.
13. Kalnins K., Eglitis E., Jekabsons G., Rikards R.
Metamodels for Optimum Design of Laser Welded
Sandwich Structures. Scientific Proceedings of
International Conference on Welded Structures,
Design, Fabrication, and Economy 2008, Miskolc,
Hungary, 2008. p. 119-126.
14. Kalnins K., Ozolins O., Jekabsons G., Metamodels
in Design of GFRP Composite Stiffened Deck
Structure. Proceedings of 7th ASMO-UK/ISSMO
International Conference on Engineering Design
Optimization, Association for Structural and
Multidisciplinary Optimization in the UK, Bath,
UK, 2008, 11 p. (in press).
15. Kohavi R. A Study of Cross-Validation and

Bootstrap for Accuracy Estimation and Model
Selection. Proceedings of the Fourteenth
International Joint Conference on Artificial
Intelligence, Morgan Kaufmann, San Mateo, CA,
1995. p.1137-1145.
16. Loughrey J., Cunningham P. Overfitting in
Wrapper-based Feature Subset Selection: the Harder
You Try the Worse It Gets. 24th SGAI International
Conference on Innovative Techniques and
Applications of Artificial Intelligence, 2004. p. 3343.
17. Martin J.D., Simpson T.W., Use of Kriging Models
to Approximate Deterministic Computer Models.
AIAA Journal, vol. 43, no. 4, 2005. p. 853-863.
18. Matveev A. Development of Methods, Algorithms
and Software for Optimal Design of Switched
Reluctance Drives, PhD thesis. Eindhoven
University of Technology, Eindhoven, The
Netherlands, 2006.
19. Opitz D., Maclin R. Popular Ensemble Methods: An
Empirical Study. Journal of Artificial Intelligence
Research, vol. 11, 1999. p. 169-198.
20. Pudil P., Ferri F.J., Novovicova J., Kittler J. Floating
search methods for feature selection with
nonmonotonic criterion functions. Proceedings of
the International Conference on Pattern Recognition,
vol. 2, IEEE, Los Alamitos, CA, 1994. p. 279-283.
21. Pudil P., Novovicova J., Kittler J. Floating Search
Methods in Feature Selection. Pattern Recognition
Letters, 15, 1994. p. 1119-1125.
22. Rawlings J.O. Applied Regression Analysis: A
Research Tool, 2nd ed. Wadsworth & Brooks/Cole,
Pacific Grove, CA, 1998.
23. Reunanen J. Search Strategies. Feature Extraction:
Foundations and Applications, Guyon I., Gunn S.,
Nikravesh M., Zadeh L.A. (eds.), Springer, 2006. p.
119-137.
98

CATEGORIZING SEQUENCES OF LARYNGEAL DATA FOR

DECISION SUPPORT
Adas GELINIS , Evaldas VAIIUKYNAS , Edgaras KELERTAS , Marija BAAUSKIEN ,
Antanas VERIKAS , , Virgilijus LOZA , Aurelija VEGIEN
* Kaunas University of Technology, Lithuania; ** Halmstad University, Sweden; *** Kaunas University of
Medicine, Lithuania
Abstract: This paper is concerned with kernel-based

techniques for categorizing laryngeal disorders based
on information extracted from sequences of laryngeal
colour images. The features used to characterize
a laryngeal image are given by the kernel principal
components computed using the N -vector of the
3-D colour histogram. The least squares support
vector machine (LS-SVM) is designed for categorizing
an image sequence into the healthy, nodular and
diffuse classes. The kernel function employed by
the SVM classifier is defined over a pair of matrices,
rather than over a pair of vectors. An encouraging
classification performance was obtained when testing
the developed tools on data recorded during routine
laryngeal videostroboscopy.
Keywords:
Larynx pathology, Image sequence,
Classification, Support vector machine.
1.
cation accuracy of over 95% was achieved when categorizing the images into one healthy and two pathological (nodular and diffuse) classes. Fig. 1 presents
characteristic examples from the three decision classes
considered, namely, nodular, diffuse, and healthy. As
can be seen from Fig. 1, examples representing the
different classes are rather different. However, it is
worth noting that due to the large variety of appearance of vocal fold mass lesions, the classification task
can sometimes be difficult to solve even for a trained
physician [10].
Fig. 1. Images from the nodular (left), diffuse (middle),

and healthy (right) classes
Introduction
Laryngeal still images, image sequences, voice signal,

and patients questionnaire data can be considered
as the main information sources to characterize human larynx. Nowadays, automated analysis of voice
is increasingly used for detecting and screening laryngeal pathologies [13]. It was demonstrated that even
telephone-based voice may lend itself for screening laryngeal disorders [1].
There were very few attempts to create systems
for automated analysis of still colour laryngeal images. In [4], a technique for automated categorization of manually marked suspect lesions into healthy
and diseased classes was presented. The categorization is based on textural features extracted from cooccurrence matrices [5, 6] computed from manually
marked areas of vocal fold images, taken by a CCD
camera. The classification accuracy of 81.4% was reported when testing the technique on a very small set
of 35 images. A set of 785 colour laryngeal images has
been used in studies presented in [79]. The classifi-
When categorizing the same set of images into seven

classes (one healthy and six pathological), the classification accuracy of over 80% was reported [11]. Image
texture, distribution of colour, and geometry of edges
of vocal folds are the types of features used for the
categorization. It was found that colour is amongst
the most discriminative types of features.
1.1.
Videostroboscopy
Laryngeal videostroboscopy is used extensively for

analysis of vocal folds and in the clinical practice
for diagnosing voice disorders [12]. For example,
videostroboscopy is a well-established technique for
measuring the glottal gap or examining the glottic
closure [13]. Videostroboscopy is one of the standard
methods used to examine moving objects. Flashing
light is used to illuminate an object in stroboscopy.
When the flashes are synchronized with the vocal fold
vibrations, a stationary view of the vocal folds is obtained.
However, the single-flash-timing laryngeal videostroboscopy has a limitation that it is effective only when
This study was supported by COST Action 2103

Advanced Voice Function Assessment.
99
vocal fold vibrations exhibit only one single fundamental frequency. Multiple tones (fundamental frequencies) may be recorded in the case of some diseases, such as polyps, nodules, and cysts [14]. In such
cases, a clear view of the vibrating vocal folds can
not be obtained with the single-flash-timing laryngeal
videostroboscopy. A multiple-flash-timing laryngeal
videostroboscopy technique was proposed by Deguchi
et al. [14], to deal with such cases. Multiple light
emitting diodes are used as illumination sources.
In [15] image sequences recorded with the stroboscopy
system have been used to measure the glottic angle and the angular velocities of vocal fold abduction and adduction. The authors point out that semiautomated edge tracking would be an important improvement of the technique.
It is worth mentioning that not only edge tracking
but also other tasks usually carried out when analyzing video data need automated or semi-automated
procedures. Decision making is one of such tasks. In
clinical practice, decision making is quite often based
on subjective evaluation of video data. Quantitative
measures of motion, colour distribution and geometry
of vocal folds can provide objective information and
be useful in medical treatment planning and greatly
facilitate tracing progress over time.
The long-term goal of this work is a decision support
system for diagnostics of laryngeal diseases. A voice
signal, sequences of colour vocal folds images obtained
from videostroboscopy, and questionnaire data [16]
are the information sources to be used in the analysis. This paper is concerned with automated characterization of image sequences obtained from laryngeal
videostroboscopy into a healthy class and two classes
of disorders.
2.
colour images of vocal folds can be extracted [8,17,18].

Features characterizing distribution of image colour is
used in this study.
The approximately uniform L a b colour space [19]
was employed to represent colours. We characterize
the colour content of an image by the probability distribution of the colour represented by the 3D colour
histogram of N = 4096 (161616) bins and consider
the histogram as an N -vector. Most of bins of the histograms were empty or almost empty. Therefore, to
reduce the number of components of the N -vector,
the histograms built from a set of training images
were summed up and the N -vector components corresponding to the bins containing less than N hits in
the summed histogram were left aside. Hereby, when
using N = 10 we were left with 733 bins a
vector of measurements with 733 components.
Having a vector of measurements , the feature vector
x is computed in the following way. We assume that
is a kernel [20] and is a mapping of onto the feature space F , such that ( i , j ) = h( i ), ( j )i,
e i)
where h, i stands for the inner product. Let (
denote the centered data point in the feature space F
M
X
e i ) := ( i ) 1
( i )
(1)
(
M i=1
with M being the number of data points. The features x are then given by the kernel principal components (kPC) computed as projections of the centered
e
-pattern ()
onto the eigenvectors
v=
e i)
i (
(2)
i=1
e i ), (
e j )i of the
of the covariance matrix Kij = h(
centered data points, where the expansion coefficients
i of the eigenvector are found from the eigenvalue
problem
= K,
(3)
The data
The medical task considered in this paper concerns laryngeal colour image sequences, based automated categorization of laryngeal disorders into three decision
classes: healthy and two pathological classes, namely
diffuse and nodular mass lesions of vocal folds [7].
Mass lesions of vocal folds could be categorized into
six classes namely, polypus, papillomata, carcinoma,
cysts, keratosis, and nodules. This categorization is
based on clinical signs and histological structure of
the mass lesions of vocal folds. We distinguished
two groups of mass lesions of vocal folds i.e. nodular
lesions (localized thickenings)nodules, polyps, and
cystsand diffuse lesionspapillomata, hyperplastic
laryngitis with keratosis, and carcinoma.
The data have been recorded at the Department
of Otolaryngology, Kaunas University of Medicine,
Lithuania. The image sequences were acquired during
routine videostroboscopy. The duration of one image
sequence was equal to 4s. The resolution of 720 576
pixels was used to record the image sequences.
3.
M
X
where the solutions are normalized by requiring

h, i = 1. Thus, the feature x is given by
e
x = hv, ()i
=
M
X
e i ), ()i.
e
i h(
(4)
i=1
The optimal number of components (features) used is

determined experimentally.
4.
The classifier
Depending on the definition of the optimization problem, several forms of support vector machine (SVM)
can be distinguished, for example, 1-norm or 2-norm
SVM. Assuming that (x) is the non-linear mapping
of the data point x into the new space, the 1-norm
soft margin SVM can be constructed by solving the
following minimization problem:
M
X
1
i
(5)
min wT w +
w,b, 2
i=1
subject to
Features
yi (hw, (xi )i + b) 1 i ,
i 0,
i = 1, ..., M ,
Various types of features characterizing colour, texture, and geometry of biological structures seen in
100
(6)
where w is the weight vector, yi = 1 is the desired

output (1), M is the number of training data points,
hi stands for the inner product, i are the slack variables, b is the threshold, and is the regularization
constant controlling the trade-off between the margin
and the slack variables. The discriminant function for
a new data point x is given by:
X

M
f (x) = H
i yj k(x, xi ) + b ,
(7)
between the spaces and derive a positive definite kernel based on that concept. The "QR" factorization of
the matrices (A, B) and the kernel Gram-Schmidt orthogonalization process are used to derive the kernel.
Applying the "QR" factorization the matrices (A, B)
can be written as A = QA RA and B = QB RB , where
Q is an orthonormal basis and R is an upper-diagonal
matrix of size M M of the Gram-Schmidt coefficients representing the columns of the original matrix
in the new basis. The principal angles cos(i ) are
given by the singular values i of the matrix QTA QB ,
cos(i ) = i , i = 1, ..., M . It was shown that
i=1
where k(x, xi ) stands for the kernel and the Heaviside

function H[y(x)] = 1, if y(x) 0 and H[y(x)] = 1
otherwise. The optimal values i , b of the parameters i and b are found during training.
4.1.
(A, B) = det(QTA QB )2 =
1X 2
1 T
w w+
e
2
2 i=1 i
5.
i = 1, ..., M.
5.1.
Experimental setup
Data were available from 30 patients. Amongst those,

10 patients belong to the nodular class, 12 to the diffuse class and 8 to the healthy class. There were 100
of image frames in one image sequence. However, only
10 image frames were used to estimate the kernel defined over a pair of matrices. Due to the small number of data points available for the experiments, the
leave-one-out test [23] has been used to estimate the
classification accuracy. The data used were normalized to zero mean and variance one. The polynomial
kernel of degree q = 1, 2 and 3 has been used to extract the kernel principal components. The Gaussian
kernel has been used to estimate the kernel defined
over a pair of matrices.
(9)
The main difference between the LS-SVM and SVM is

the equality constraints (Eq.(9)) used in LS-SVM instead of unequally constraints defined by Eq.(6). Due
to the equality constraints, the optimal parameter values can be found by solving a set of linear equations,
instead of quadratic programming applied in the case
of SVM. The solution is given by [21]

b
0
yT
0
=
,
(10)
1
y Z+ I
1
where Zij = yi yj (xi , xj ), I is the identity matrix,
1 = [11 , ..., 1M ], y = [y1 , ..., yM ], and = [1 , ..., M ].
Since an SVM is a binary classifier while the task is
to distinguish between three classes, the one-againstone scheme is used to make the categorization in this
work.
4.2.
Experimental investigations
(8)
subject to
yi (hw, (xi )i + b) = 1 ei ,
(11)
is a positive define kernel [22]. We use this kernel

in our work. The algorithm for evaluating the kernel without explicit computation of QA and QB can
be found in [22]. Only inner-products between the
columns of A and the columns of B are used.
Suykens and Vandewalle [21] have introduced a least

squares version of the SVM classifier (LS-SVM). We
use this type of SVM in this work. Parameters of
the LS-SVM are estimated by solving the following
optimization problem:
w,b,e
cos(i )2
i=1
Least squares SVM
min
M
Y
5.2.
Results
Table 1, Fig. 2, and Fig. 3 summarize results of the

tests. In Table 1, shown is the number of kernel principal components used to characterize colour of one
image frame along with the test data set classification
accuracy.
Kernel function
For (xi , xj ), one usually uses the linear: xTi xj , Gaussian: exp{||xi xj ||2 /} or polynomial: (xTi xj +1)d
kernel. The kernel is defined over a pair of vectors.
In this work, classification is based on a set of vectors
rather than on a single vector. For example, a sequence of images is recorded from a patient. Each
image is represented by a feature vector. Feature
vectors are then collected into a matrix (each vector constitutes a matrix column) and used to make
a decision. Therefore, a kernel function utilized by
the LS-SVM classifier is defined over a pair of matrices (A, B) rather than over a pair of vectors. A
positive definite kernel of such type has been recently
proposed by Wolf and Shashua [22]. The authors use
the principal angles between the two column spaces
defined by the matrices (A, B) to assess the matching
Table 1. The test set data classification accuracy

5
7
8
9
10
11
13
# of kPC
Accuracy, % 33.0 44.9 69.8 85.7 78.3 72.1 63.3
The graph presented in Fig. 2 plots the correct classification rate of the test set data as a function of the
percentage of the data variance accounted for by the
number of the kernel principal directions used. As
can be seen from Fig. 2, the percentage of the data
variance accounted for by the optimal number of the
components is close to 90. The graph presented in
Fig. 3 relates classification accuracy the regularization constant , and the number of the kernel principal components used.
101
90
7.
Accuracy, %
80
70
8.
60
50
9.
40
30
10
12
14
16
Number of kPC
10.
Fig. 2.
The correct classification rate of the test set
data as a function of the percentage of the data variance
accounted for by the number of the kernel principal
components used
11.
Accuracy, %
100
80
12.
60
40
13.
20
15
20
10
15
10
0
14.
Number of kPC
Fig. 3. The classification accuracy as a function of the

regularization constant and the number of the kernel
principal components used
6.
15.
Conclusions
16.
Sequences of colour laryngeal images were categorized

into the healthy, nodular and diffuse classes using an
LS-SVM with a kernel defined over a pair of matrices.
The relatively high classification accuracy obtained in
the experimental tests encourages further studies in
this area.
17.
References
1.
2.
3.
4.
5.
6.
18.
R. J. Moran, R. B. Reilly, P. de Chazal, P. D. Lacy,

Telephony-based voice pathology assessment using
automated speech analysis, IEEE Trans Biomedical
Engineering 53 (3) (2006) 468477.
J. I. Godino-Llorente, P. Gomez-Vilda, Automatic
detection of voice impairments by means of shortterm cepstral parameters and neural network based
detectors, IEEE Trans Biomedical Engineering 51 (2)
(2004) 380384.
A. Gelzinis, A. Verikas, M. Bacauskiene, Automated
speech analysis applied to laryngeal disease categorization, Computer Methods and Programs in
Biomedicine 91 (1) (2008) 3647.
J. F. R. Ilgner, C. Palm, A. G. Schutz, K. Spitzer,
M. Westhofen, T. M. Lehmann, Colour texture
analysis for quantitative laryngoscopy, Acta OtoLaryngologica 123 (6) (2003) 730734.
R. M. Haralick, K. Shanmugam, I. Dinstein, Textural
features for image classification, IEEE Trans System,
Man and Cybernetics 3 (6) (1973) 610621.
A. Gelzinis, A. Verikas, M. Bacauskiene, Increasing
the discrimination power of the co-occurrence matrix-
19.
20.
21.
22.
23.
102
based features, Pattern Recognition 40 (9) (2007)

23672372.
A. Verikas, A. Gelzinis, M. Bacauskiene, V. Uloza,
Towards a computer-aided diagnosis system for vocal
cord diseases, Artificial Intelligence in Medicine 36 (1)
(2006) 7184.
A. Verikas, A. Gelzinis, D. Valincius, M. Bacauskiene,
V. Uloza, Multiple feature sets based categorization
of laryngeal images, Computer Methods and Programs in Biomedicine 85 (3) (2007) 257266.
M. Bacauskiene, A. Verikas, A. Gelzinis, D. Valincius, A feature selection technique for generation of
classification committees and its application to categorization of laryngeal images, Pattern Recognition
42 (5) (2009) 645654.
P. J. P. Poels, F. I. C. R. S. de Jong, H. K. Schutte,
Consistency of the preoperative and intraoperative
diagnosis of benign vocal fold lesions, Journal of Voice
17 (3) (2003) 425433.
A. Verikas, A. Gelzinis, M. Bacauskiene, V. Uloza, Integrating global and local analysis of colour, texture
and geometrical information for categorizing laryngeal images, International Journal of Pattern Recognition and Artificial Intelligence 20 (8) (2006) 1187
1205.
P. S. Popolo, I. R. Titze, Qualification of a quantitative laryngeal imaging system using videostroboscopy
and videokymography, Annals of Otology Rhinology
and Laryngology 117 (6) (2008) 404412.
H. Rihkanen, P. Reijonen, S. Lehikoinen-Soderlund,
E. R. Lauri, Videostroboscopic assessment of unilateral vocal fold paralysis after augmentation with
autologous fascia, European Archives of Oto-RhinoLaryngology 261 (2004) 177183.
S. Deguchi, Y. Ishimaru, S. Washio, Preliminary evaluation of stroboscopy system using multiple light
sources for observation of pathological vocal fold oscillatory pattern, Annals of Otology Rhinology and
Laryngology 116 (9) (2007) 687694.
S. H. Dailey, J. Kobler, R. E. Hillman, K. Tangrom,
E. Thananart, M. Mauri, S. M. Zeitels, Endoscopic
measurement of vocal fold movement during adduction and abduction, Laryngoscope 115 (1) (2005)
178183.
M. Kaseta, Using the patients questionnaire
data to screen laryngeal disorders, Computers in Biology and Medicine 39 (2) (2009)
doi:10.1016/j.compbiomed.2008.11.008.
A. Verikas, A. Gelzinis, M. Bacauskiene, D. Valincius,
V. Uloza, A kernel-based approach to categorizing
laryngeal images, Computerized Medical Imaging and
Graphics 31 (8) (2007) 587594.
Intelligent vocal cord image analysis for categorizing
laryngeal diseases, in: M. Ali, F. Esposito (Eds.),
Lecture Notes in Artificial Intelligence, Vol. 3533,
Springer-Verlag Heidelberg, 2005, pp. 6978.
A. Verikas, K. Malmqvist, M. Bacauskiene,
L. Bergman, Monitoring the de-inking process
through neural network-based colour image analysis,
Neural Computing & Applications 9 (2) (2000)
142151.
J. Shawe-Taylor, N. Cristianini, Kernel Methods for
Pattern Analysis, Cambridge University Press, Cambridge, UK, 2004.
J. A. K. Suykens, J. Vandewalle, Least squares support vector machine classifiers, Neural Processing
Letters 9 (1999) 293300.
L. Wolf, A. Shashua, Learning over sets using kernel principal angles, Journal of Machine Learning Research 4 (2003) 913931.
A. Verikas, M. Bacauskiene, Using artificial neural
networks for process and system modeling, Chemometrics and Intelligent Laboratory Systems 67 (2003)
187191.

SYSTEM PARAMETERS IDENTIFICATION IN THE STATE SPACE

Leonas BALAEVIIUS, Kstas RIMKUS
Department of Control Technology, Kaunas University of Technology, Lithuania
x Ax Bu, x(0) x 0
Abstract: This article presents the identification of the

system parameters in the state space, by using an
interface
between
Matlab/Simulink
and
the
programmable logic controller (PLC). The effectiveness
of the estimation of the parameters was tested by
calculating the arithmetic mean of the response error
between the responses of the real system and the system
model with an observer into the input in the real time. It
is shown that the Matlab/Simulink software could be
used for identifying the system parameters in real time
and setting the model parameters in the PLC.
(1)
y Cx
Here x is an n-dimensional state vector, u is an rdimensional control vector, y is an m-dimensional

output vector, A is an mn system matrix, B is an nr
control matrix and C is an mn output matrix, x(0) is the
state of a system at time t0=0. Assuming that u is a
continuous time function and according to the
eigenvalues of matrix A, the system is stable. Also let us
assume that the object is thoroughly observed and
completely under control.
The problem is that the parameters of the object - A, B,
C along with the state vector x, have to be evaluated by
using only the input and output data of the real system.
This can be done with the help of the observer [1]
Keywords: Programmable logic controllers, Matlab,

Simulink, identification.
1. Introduction
One of the main problems of control is the identification
of the parameters of the control object. The control
designer must have the most exact possible
mathematical description of the object. The evaluation
of the precision of the control system becomes relevant
only when the system works under unforeseen
conditions. This way, the initial conditions may differ
from the calculated ones, whereas the control and
measurement signals may be distorted by noises that
were not evaluated during the design of the system.
That is why it is important to have as much information
on the controlled object as possible while designing the
control. In this article, the problem of the identification
of the parameters of the object in the state space is
researched. For this, a link between Matlab/Simulink
and the PLC is used. The PLC processes the control and
measurement signals and the design of the object model
is executed by Matlab and is based on received
parameters. The adequacy of the object model with an
observer for a real object is tested in Matlab in real
time.
x A (t ) x B (t )u K (t )( y y )
y [0 1]x
(2)
Here x is the estimated state vector, y is the estimated

output vector, A is the estimated system matrix, B is
the estimated control matrix, K is the gain matrix of the
state observer. The overall system structure of the
estimation of the object parameters is shown in Fig. 1.
Programmable logic controller
Plant
A , B
Estimation
Matlab/Simulink
State
Observer
2. State observer
The real system can be described in the state space:

Fig. 1. System structure of the estimation of the object
parameters
103
The response error e (hereinafter response error)

between the responses of the real system and the system
model with an observer into the input is calculated
e y y
0 2
2
A
, B 0, C 0 1
0.154 0.154

(3)
C1
The object is connected to the PLC Modicon 140 CPU

113 03, which measures the input and output signals of
the object. A Wonderware Modbus I/O server and
Matlab/Simulink software are installed on a PC. The
Wonderware Modbus I/O server is configured as
written in [2]. The s function blocks are used for the
exchange of data between Matlab/Simulink and the
Wonderware Modbus I/O server. These blocks are
responsible for the plant model simulation in real time.
The program created in Matlab estimates the system
parameters A and B .
The estimation of the object parameters is conducted in
such order:
1. The order of the object is chosen and the initial
values of matrixes A , B and C are indicated. The
R2
1 n
e(i)
n i 1
R4
y
R1
R3
Fig. 2. The scheme of the second-order object (R1 = 50 k;

R2 = 100 k; R3 = 100 k; R4 = 100 k; C1 = 5F;
C2 = 65 F)
It is indicated that the elements of matrixes A and

B are evaluated. By using the values of the input and
output signals of the object the evaluated values of
matrixes A and B are acquired.
initial values of matrix A are chosen so that the system

would be stable.
2. The unknown (to be estimated) elements of matrixes
A and/or B are indicated.
3. The interval of data reading from the PLC is
indicated 100 ms.
4. Based on the values of the input and output signals
the values of the elements of matrixes A and B are
calculated [1].
5. Based on the evaluated description of the system in
the state space, adaptive observers gains are chosen by
the trial and error method using simulations in order to
achieve a good rate of convergence. Small gains may
result in slow convergent rate, whereas large gains may
make the differential equations "stiff" and difficult to
solve numerically on a digital computer [1].
6. The adequacy of the system model response to the
real object response to the same input signal is verified
by calculating the arithmetic mean of the response error.
AME
C2
0 1.9681
4.1032
A
, B 0.0498
0.0291 0.0194
The values of the gain matrix of the observer are:

K 4956.6 0.01943T
These values continue to remain unchanged.

The responses of the object and the second-order state
space model with an observer to the input signal of the
object are presented in Fig. 3.
(4)
Here AME is arithmetic mean of the response error, n is

the number of points of the calculated response error.
Fig. 3. The responses of the object and the second-order state

space model with an observer to the input signal u of the
object
3. Second-order object parameters estimation
The created Matlab program will be verified by

evaluating the parameters of the second-order object
(see Fig. 2), connected to the PLC, in state space.
A second-order state space model with an observer will
evaluate the parameters of the object. The following
values for the initial values of matrixes A , B and C
The variation of the response error e is presented in

Fig. 4.
The AME value is 1.62 10 2 .
There has been a research on what influence the
selection and evaluation of the initial values of
matrix A , when the matrixes B and C are unchanged,
are chosen:
have on the arithmetic mean of the response error (see

Table 1).
104
4. Third-order object parameters estimation
The parameters of the third order object (see Fig. 5),

connected to the PLC, will be evaluated in the state
space.
C1
C2
R2
u
R1
C3
R4
R3
R6
R5
Fig. 5. The scheme of the third-order object (R1 = 1M;

R2 = 1 M; R3 = 1 M; R4 = 1 M; R5 = 1 M;
R6 = 1 M; C1 = 5 F; C2 = 1 F; C3 = 5 F)
A third-order state space model with an observer will

evaluate the parameters of the object. The following
values are the initial values for matrixes A , B and C :
Fig. 4. Response error to the input signal of the object

Table 1. The influence of the selection of the initial values of
matrix A to the AME
a ii , i 1,2
-20; -1
3.0254 10 4
-200; -10
3.5039 10 4
-0.2; -0.01
1.6916 10 4
-0.2; -0.2
0
0
0.3
1

A 0
0.2
0 , B 0, C 0 0 1
0
0
0.5
0
AME
2.538 10
It is indicated that only the elements in the diagonals of

the matrix A are evaluated. By using the values of the
input and output signals of the object the evaluated
matrix A is acquired.
-20; -20
4.6 10 4
-36;-36 after additionally

selecting the remaining
elements from matrix A
3.9136 10 4
0
0.2
0.19547
A
0
0.2
0
0.145
0
0.898
After that a research has been conducted on what

influence the selection of the initial values of matrix B
has on the value of arithmetic mean of the response
error, when matrix

K 910.6911 181.8901 181.8901T
36.204 0
A
3
0

The responses of the object and the third-order state
space model with an observer to the input signal of the
object are obtained. The variation of the response error
e is presented in Fig. 6.
And the elements of matrix B and the eigenvalues of

matrix A are evaluated (see Table 2).
The AME value is 1.3555 10 4 .

matrix A , when matrixes B and C are unchanged,

matrix B to the AME
bi , i 1,2
5; 2
3.3043 10 5
0.1; 5
2.8067 10 5
0.2; 0.2
3.0379 10 5
20; 20
4.0738 10 5
AME
have on arithmetic mean of the response error (see

Table 3), when the object is approximated by a third
order state space model with an observer.
There will be an evaluation on the parameters of the
object with a second-order state space model with an
observer.
Initial values of the system poles and the control matrix

influence the accuracy of the system.
105
Table 4), when the object is approximated by a second

order state space model with an observer.
Fig. 6. Response error the input signal of the object
Fig. 7. Response error to the input signal of the object
matrix A to the AME
matrix A to the AME
4.252 10 4
a ii , i 1,2
-10; -5
1.9856 10 4
-0.3; -0.2; -0.5
1.3555 10 4
-1; -0.5
9.6398 10 5
-10; -5; -20
2.2967 10 4
1.33 10 4
-10; -5; -20 after additionally

selecting the remaining
elements from matrix A
1.0642 10 4
-1; -0.5 after additionally

selecting the remaining elements
from matrix A
a ii , i 1,2,3
-3; -0.2; -4
AME
AME
6. Conclusions
The following values are the initial values for

matrixes A , B and C :
It can be stated that the initial values of matrix A

should be chosen in a canonical form to make the
system stable. The accuracy of the system model is
mostly influenced by the initial values of the system
poles and the control matrix.
There was no significant increase in the AME value
after approximating a real object with a lower-order
state space model with an observer.
Such observed behaviour is valid in the analyzed case,
and the AME could be higher in the case of a highorder object if its parameters are different from the ones
chosen in this article.
Unknown-order real object approximation could be
started from a lower-order state space model with an
observer and the order of the model should be increased
until the desired AME value is obtained.
10 0 1
A
, B , C 0 1
5
0
0
It is indicated that only the elements of matrix A are

evaluated. By using the values of the input and output
signals of the object the evaluated matrix A is obtained.
14.739 3.1817
A
31.343 8.9445

K 910.9611 181.8901T
7. References

The variation of the response error e is presented in
1. Ioannou P. and Sun J. Robust Adaptive Control.

Prentice-Hall, Inc., Upper Saddle River, NJ, 1995.
825 p.
2. Balaeviius L., Dervinis G. and ioinys M.
Communications between programmable logical
controllers and Matlab/Simulink. Proceedings of
Fig. 7. The AME value is 1.9856 10 4 .

matrix A , when matrixes B and C are unchanged,
have on arithmetic mean of the response error (see
106

MODEL OF CURRENT SPACE VECTOR CONTROL

Robertas JANICKAS, Algirdas SMILGEVIIUS
Abstract: space vector control Simulink model is

developed and investigated. Physical phenomena
as to establish a magnetic field in the rotor and cause the

rotor to align with the field flux. Then, by properly
controlling the currents producing stator field the torque,
developed by motor and the speed are controlled
Such technique, described above is called field-oriented
control. Two reference frames shown in Fig. 1, Parks
and Clarkes are used to simplify mathematical model of
induction motor.
For required greater torque, greater iq value is needed.
For greater rotor flux, greater id is regarded. Knowing
these two values, it is possible to solve for the vector
that represents developed torque of a motor. By feeding
id, iq, by command values from controller, result values
are generated that may be used to correct the motion of
position, velocity, or torque in the device.
occurring in AC motor during mentioned control usage

and results are discussed.
Keywords: vector control, space vector control, motor
investigation, indirect vector control.
1. Introduction
Induction motor rotation speed depends on three phase
grid frequency and motor pole number. If motor is
powered from standard grid, its synchronous speed is
constant. However, very often motor rotation speed shall
be controlled in accordance with the process
requirements and for energy saving purpose.
Yet, a single ability to control speed only is not enough.
For precision measurements and specific process
requirements it is necessary to maintain precision
rotation speed when the load changes or the system is
influenced by disturbances.
Europe market investigations show that frequency
converters market is constantly growing. This is
conditioned by competitive prices, in the long term such
devices development expand their application areas for
the best process control and energy saving.
Implementing one popular contemporary science idea as
self-diagnostic ability without usage additional sensors
by using soft sensor technique may be implemented
both in frequency converter and in motor control system.
Due to this, more investigations and studies has been
carried out [1, 2, 3]. Simulation is safe and not an
expensive investigation alternative that is applicable for
investigation of extreme conditions and for developing
soft sensor models creating. This paper deals with
elaborated Simulink model and simulation results at
different inertia of the system and speed reverse as well
as tracking the reference signal.
Fig. 1. Currents vectors
3. Indirect field orientation control

For very low-speed operations and for positioning
control, the use of flux sensing that relies on integration
which has a tendency to drift may not be acceptable. A
commonly-used alternative is indirect field orientation,
which does not rely on the measurement of the air gap
2. Vector control principle

In a three-phase induction motor, the currents are
controlled in each of the phase windings in such a way
107
flux. Torque can be controlled by regulating,

speed,
e r .
iqse and slip
It has been shown that when properly oriented

equal to zero and
Rotor flux can be controlled by
idse . Subscript denotes reference frame, rotating with
e*
rr` iqs
e r ` e* .
Lr p ids
synchronous speed. Given some desired level of rotor
`*r ,
the desired value of necessary
*
2
idse* may be
obtained from [4, 5]:
`*r
The desired torque of
`
r m
`
`
r
r
rL
idse* .
r L p
Teme*
(1)
Teme* at the given value of rotor

3 P Lm `* e*
r ids .
2 2 L`r
(2)
i as*
i qse *
Teme*
(3)
The above conditions, if satisfied, ensure the decoupling

of the rotor voltage equations to what extent this
decoupling is actually achieved will depend on the
accuracy of motor parameters used.
Developed indirect field-oriented control scheme for a
current controlled PWM induction motor drive is
presented in Fig. 2. The field orientation, , is the sum
of the rotor angle given by position sensor, r, and the
angle, 2, obtained by integrating the slip speed [6, 7, 8,
9, 10].
flux is calculated as:
*
rm
thus, the slip speed
relation can also be written as:
maintaining
flux,
`dre Lmidse*
idr`e is
i bs*
`e
dr
i qse * , i dse *
i cs*
e*
ds
2*
2*
Fig. 2. Indirect field-oriented control of a current regulated PWM inverter induction motor drive
If orthogonal outputs of the form cosr and sinr are

available from the shaft encoder, the values of cos and
sin can be generated from the following trigonometric
identities [4]:
cos cos( r 2 ) cos r cos 2 sin r sin 2 ;

sin sin( r 2 ) sin r cos 2 cos r sin 2 .
4. Simulation of transients
Simulation results witch were got from Fig. 3 are shown
in Fig 4. The greatest current value at starting reaches
110 A, but after motor starting, steady state currents
amplitudes are approximately 25 A at motor starting
without load. The torque change low is from 0.5 s
suddenly grows tile nominal, from 1 s its goes down tile
haft of nominal torque of machine, next from 1.25 s its
grows tile nominal and after 1.5 s its drop tile zero.
Response to that function can be seen in the second
graph. During the load torque changing, motor windings
currents change also. They repeat load changing
behaviour.
(4)
Simulation model is created by using one of MATLAB

program package SIMULINK. In this program space
vector controlled induction motor model is developed
by using standard function blocks (Fig. 3) [5].
108
Voltage, V
Fig. 3. SIMULINK model of vector controlled drive2

200
100
0
-100
-200
Tem, Nm
100
50
0
-50
IPsist, Wb
150
100
50
Currents, A
0
100
50
0
-50
0
0,2
0,4
0,6
0,8
1
1,2
Time, s
1,4
1,6
1,8
Fig. 4. Transients of motor variables in vector control system
Fig. 5. Transients with increased 10 times motor inertia
Fig. 5 presents starting transients with increased 10

times inertia. Torque, developed by motor, presented in
second graph differs from the torque in Fig. 4 by its
greater value at the initial time of transients.
At this situation rotor flux magnitude curve changes
radically, particularly in start-up instant it increases by
50 Wb.
For motor acceleration with larger inertia energy
consumption from grid is increased. Later for steady
state rotation, less energy is used.
Due to each time when load torque or motor set point

changes the influence to PWM frequency is made.
Voltage amplitude dependence on frequency is clearly
monitored by changing speed set point.
From Fig. 6 a one control advantage is monitored
during the speed changes a current value is limited and
all speed changes may be performed by changing
frequency when current amplitude value is
approximately 80 A.
The difference between the speed set point and the
rotation speed significantly depends on motor inertia.
109
6. Conclusions
Simulink model of current space vector control is
created.
Acquired simulation results allow making these
conclusions:
1. Developed motor torque shows that during the startup a slight overshoot exists. After the system start-up,
the applied torque is changed by proper low for
investigation of changes of motor.
2. The current composes constant part conditioned by
motor resistances and variable part conditioned by
motor load.
3. The speed error at motor starting due to static friction
is insignificant.
4. PWM space vector control way allows using by order
slower microprocessors practically without loss
converter output signal quality.
7. References
1. Bierke S. Enchanced control of an alternating current
motor using fuzzy logic and a TMS320 digital signal
processor, Texas Instruments, 1996. 524 p.
2. Valentine R., Motor control electronics handbook
[CD-ROM] 1998.
3. Kliuchev I.V. Theory of electrical drives. (In Russian)
Moscow: Energoatomizdat, 1985. 253 p.
4. Holts J. Pulsewidth modulation - a survey, IEEE,
1992, 13 p.
5. Caron J.P., Hautier J.P. Modelling and Control of
Induction Machine. Technip Edition, France, 1995, 312
p.
6. Field Orientated Control of 3-Phase AC-Motors.
Application report, Texas Instruments, USA, 1998, 473
p.
7. Hazzab A., Bousserhane I.K., Kamli M., Rahli M.
New Adaptive fuzzy PI-Sliding Mode Controller for
Induction Machine Speed Control. Third IEEE
International Conference on Systems, Signals & Devices
SSD'05, Tunisia, 2005. 125 p.
8. Hazzab A., Bousserhane I.K., Kamli M., Rahli M.
Design of fuzzy sliding mode controller by genetic
algorithms for induction machine speed control. Third
IEEE International Conference on Systems, Signals &
Devices SSD'05, Tunisia, 2005. 56 p.
9. Lorenz R.D., Lawson D.B. Simplified A Approach to
Continuous On-Line Tuning of Field-Oriented Induction
Machine Drives. IEEE Trans. On Industry application,
Vol.26, Issue 3, May/June 1990. 425 p.
10. Geleeviius V., Kriinas K., Kubilius V. Control
systems of electrical drives. (In Lithuanian) Vilnius:
Mokslas, 190. 360 p.
Fig. 6. Curves with changed speed reference function

and reversing
From Fig. 8 and 7 rotation speed lag from set point

signal is monitored. Speed lag from set point determined
by motor inertia and control system existing
disturbances as load torque changes.
Fig. 7. Speed reference and speed response in current

space vector control system
Fig. 8. Motor speed curves at ten times increased inertia
110

PARAMETER ESTIMATION OF INDUCTION MOTOR AT LOCKED ROTOR

Andrius PETROVAS, Roma RINKEVIIEN, Saulius LISAUSKAS
the secondary winding. Symmetrical induction motor

can by replaced by equivalent two-phase transformer.
Magnetic coupling does not exist between equivalent
two-phase machine orthogonal windings; therefore each
of them can be analyzed separately. System of similar
windings is shown in Fig. 1.
Abstract: Design of control systems of electrical drives

deals with motor parameters whose assessment is
important and complex problem. Proposed method
allows estimating induction motor parameters on the
base of transients of motor with locked rotor.
Keywords: induction motor, transients, parameter
estimation.
1. Introduction
Advanced sensorless control of electric drives deals with
computer models of motors. Mathematical model of
direct current motor is simpler than induction motor,
therefore problems of their parameters estimation are
solved also simpler [1, 2].
Development and investigation of modern induction
drives requires using models of induction motors. Just
more or less exact parameters give sufficient operation
of a model. Some of parameters, such as resistances of
stator windings, can be measured quite easy.
Nevertheless some of them require a lot of preparations.
Therefore estimation of induction motor parameters is
state of the art in modern electric drives.
For estimation of induction motor parameters usually
two conventional experiments should be made: i. e. no
load and locked rotor experiments. The first one allows
calculating stator parameters, the second one rotor
parameters. The stator resistance may be measured
through a d.c. voltage test with two phases in series.
Calculation of other equivalent circuit parameters using
motor catalogue data is presented in [3]. All these
parameters are calculated from steady-state currents and
voltages [4, 5].
The paper deals with estimation of motor parameters
just from results of locked rotor experiments. The
method differs from conventional one, using steady-state
values of parameters by exploring all information for
estimation from transient process.
Fig. 1. Equivalent circuit of induction motor with locked rotor
The circuit, presented in Fig. 1 is described by set of

differential equations in this way:
di2
di1
L1 dt M 12 dt u1 i1 R1 ,
M di1 L di2 i R ,
2
2 2
12 dt
dt
(1)
where u1 is stator supply voltage, L1 and L2 are

inductances of stator and rotor windings, R1 and R2 are
resistances of stator and rotor windings, i1 and i2
currents in the stator and rotor windings.
Set of equations (1) is the set of first order linear
differential equations; therefore its solution can be
obtained by analytical way. Canceling current of rotor
winding i2 , the equation, relating stator current i1 and
voltage u 2 looks like this:
2. Mathematical model of induction motor

Induction motor with locked rotor can be assumed as
three-phase transformer with short circuit, placed across
111
d 2 i1 R1 L2 R2 L1 di1
R1 R2
i1
dt 2
L1 L2 M 122 dt L1 L2 M 122
Constants C1 and C2, entering (7), are calculated from

initial conditions. According to the first commutation
law i1(0) =0, the set of equations gets a form:
(2)
L2
du1
R2
u1 .
2
L1 L2 M 12 dt
L1 L2 M 122
Solution of differential equation (2) consists of natural

and forced solutions. Homogeneous equation is:
d 2i1
dt
R1 L2 R2 L1 di1
R1 R2
i 0.
2 dt
2 1
L1 L2 M12
L1 L2 M12
(3)
1,2
3. Identification of the system
(4)
Estimation of induction motor parameter from

transients results at locked rotor corresponds to
identification of parameters of the system presented in
Fig. 1. At transients experimentally are recorded values
of voltage u1n and current i1m, where m = 1, 2, 3, ... N. N
is number of measured values. Task of identification
should be solved successfully, if the proper values of
parameters A, B, C1, C2, 1 and 2 of expression (7)
could be found, matching with minimal error
experimentally obtained expression i1m with analytical
expression i1. Therefore objective function can be
determined by such Euclidean norm:
where 1,2 are roots of characteristic equation.

Analysis of square root (4) shows, that discriminant sign
is always positive, therefore 1, 2 R and solution of (3)
has the form:
i1h = C1exp(1t) + C2exp(2t),
(5)
where i1h is general solution of homogeneous differential

equation (3), C1 and C2 constants defined from initial
conditions.
Solution of differential equation (3) is found as sum of
general solution i1h of homogeneous differential
equation (3) and particular solution i1n of nonhomogeneous differential equation (2) which depends
on input voltage u1 . At u1=Umaxsin(t) the particular
solution is found as:
i1n = Acos(t) + Bsin(t),
i i12m i12 tm ; m 1, 2, , N , (10)

where i1(tm) according to (7) at the moment of time tm
analytically calculated value of current i1.
Parameters A, B, C1, C2, 1 and 2 are related by (8) and
(9) expressions. Expression (9) shows that C1 and C2
can be found from coefficients A, B, 1 and 2.
Therefore parameters A, B, 1 and 2 are independent,
and they can be varied by optimizing algorithm.
Coefficients of characteristic equation are denoted as:
(6)
where A and B are constants found from initial

conditions. i1n is particular solution of non-homogeneous
equation.
The general solution of (2) has the form:
i1 = i1h + i1n = C1exp(1t ) + C2 exp(2t ) +
A U L
max 2 .
B U max R2
R1 L2 L1 R2
R1 R2 L1 L2 M
2
2
12
c R1 R2 / L1 L2 M .
2
12
At known 1 and 2 coefficients b and c can be obtained

by Vite's formulas.
Additional notations are introduced:
Free constants A and B, entering (7), are obtained by

solution of system, presented below:
R1 R2 2 L1 L2 M 122
R1 L2 L1 R2
b R1 L2 R2 L1 / L1 L2 M 122 ,
(7)
+ Acos( t) + Bsin( t).
(9)
Expressions (8) and (9) relate coefficients A, B, C1 and

C2 of solution (7) with coefficients of differential
equation. By experimental measurement values of i1 and
determining of coefficients A, B, C1 and C2 according to
that, the inverse task can be solved, i. e., according to
expressions (8) and (9) both coefficients of differential
equation (2) and parameters of the circuit, shown in
Fig. 1 can be found. This is the base of proposed and
elaborated method.
The roots of characteristic equation have the form:
R1 L2 R2 L1
2
L L M 12
1 1 2

2
2 RL R L
4 R1 R2
1 2
2 1

2
2
L1 L2 M 12
L1 L2 M 12
1 C1 A
.
2 C 2 B
/ L L
d L2 / L1 L2 M 122 ,
e R2
2
12
Unknown variables d, e, C1 and C2 are found by solving

set of equations at any iteration:
(8)
112
U m 0
Um
0
0
0
0
0
0
0
1
0 d A c 2 Bb

0 e Ab B c 2
1 C1
A

2 C 2
B
Fig. 2 indicates the good matching between

experimental and approximating current curves just in
the end of transient time.
. (11)
0.15
0.1
The equation (7) is non-linear; therefore numerical

methods of optimization should be used. For this
purpose Nelder-Mead optimization algorithm is chosen.
This method does not require any derivative information
and is widely used to solve parameter estimation, where
the function values are uncertain or subject to noise [6].
Because simplicity and popularity, it is implemented in
many mathematical programs like Matlab and others.
However other optimization methods could be used
also.
During the search of minimal value of objective function
vector x = [A, B, 1, 2] is being varied. Any iteration
begins with calculation of coefficients b and c of
characteristic equation. For this purpose the Vite's
formulas are applied.
1 + 2 = b, 1 2 = c.
0.05
Current [A]
-0.05
-0.1
-0.15
-0.2
-0.25
-0.3
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
Time [s]
Fig. 2. Transients of current i2 (1000 points)
The thicker line in the figure represents analytically

obtained curve, approximating the experimental one.
Dependence of error against time is presented in Fig. 3.
Error reaches its maximum value at the beginning of
transient and begins to reduce.
(12)
After that the values C1, C2 are obtained from set of

equations (11). All coefficients A, B, C1, C2, 1 and 2
are used to calculate the values of i1. At the end of
iteration the calculated values of i1 is compared with
experimentally obtained values i1n. For comparison of
both values the Euclidean norm (10) is used.
The input parameter X of the program is matrix of five
vertices X = [x1; x2; ; x5], where xR4. Other input
parameters , and are related to transformation of
simplex: to reflection, to contraction, to
expansion. The standard values, used in most
implementations, are =1, =0,5 and =2.
Conventional Nelder-Mead algorithm is committed for
search of termless extreme. In this case two of used
variable parameters are roots of characteristic
polynomial; they should be negative. Therefore
conventional Nelder-Mead algorithm is supplemented
by instructions, forbidding simplex vertices with 1>0 or
2>0. These actions are important just at simplex
reflection and expansion operations. Obtained simplex
vertices are checked for constraints, and if they do not
meet those, i. e. if 1>0 or 2>0, simplex is reduced for
giving new vertices to meet constraints.
0.06
0.05
0.04
300 points
Absolute error [A]
0.03
0.02
0.01
-0.01
1000 points
-0.02
-0.03
-0.04
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
Time [s]
Fig. 3. Error of approximation
In the Table 1 experimentally determined and real

parameter values are presented.
Table 1. Real and experimentally determined values
4. Comparison experimental and approximated

results
Parameter
1 [1/s]
2 [1/s]
A [A]
B [A]
C1 [A]
C2 [A]
All equations were derived with assumption, that sine

wave voltage was applied to the windings with initial
phase, equal to zero. The equipment allowing switching
voltage to windings at the instant when it crosses zero,
was constructed and implemented for experiment.
Proposed method cancels requirement to measure
instantaneous value of voltage. Solution of this type
allows refusing expensive voltage sensors.
1000 pt.
-18.1
-290
0.133
0.022
0.028
0.1
300 pt.
-18.95
-45.8
0.151
0.01
0.025
0.045
Real value
-15.2
-31.7
0.116
0.02
0.028
0.05
Results show the greatest error at determinations of

decaying component and it occurs for the fastest
decaying term C2exp(2t). Thus can be concluded, that
proposed method is suitable to find harmonic
113
components of expression (7), and just coefficients A

and B are estimated with the greater accuracy. Relative
errors of estimation are:
0.133 0.116
0.022 0.02
0.02
1. Proposed method allows determining roots of

characteristic equation and coefficients of general
solution from experimental data.
2. Approximation of experimental measurement results
is nonlinear problem requiring using discrete
optimization methods.
3. Proposed method on the base of Euclidean norm as
objective function allows approximating of solution of
harmonic components with error not exceeding 15
and is suitable for the system. It is not suitable for fast
decaying exponential components due to great error.
4. Roots of characteristic equation, describing fast
decaying components can be estimated with error, not
exceeding 50 , but accuracy of determining of
harmonic
coefficients
reduces
and
general
approximation error increases.
100 % 14.6 %,
0.116
6. Conclusions
100 % 10 %.
Accuracy can be increased by cancelling of transients

points in the end of the process and in the same way
reducing influence of harmonic components. Suppose
the time interval 0<t<0.3 s is chosen (300 points of
measurement), then determined roots of characteristic
equation get values 1 = 18.2 s-1 and 2= 45.8 s-1. In
this case the relative error of roots estimation is:
18.96 15.2
15.2
45.8 31.7
31.7
100% 24.7 %,
7. References
1. Rinkeviien R., Petrovas A. Direct current drive with

observer. Elektronika ir Elektrotechnika. Electronics
and Electrical Engineering. 2008 Nr.5 (85). p. 53-56.
2. Rinkeviien R., Petrovas A. Model of direct current
drive with reduced order observer. 2nd International
Conference on Electrical and control technologies,
May 08-09, 2008. ECT2008: proceedings of the
3rd international conference on electrical and control
technologies. p. 150153.
3. Boldea I., Nasar S.A. The induction machine
handbook, CRC press, 2001. 950 p.
4. Chiasson J. Modelling and high performance of
electric machines. Willey Interscience, 2005.
736 p.
5. Schrder D. Elektrische Antriebe Regelung von
Antriebssystemen. Springer Verlag, 2001. 1172 p.
6. Singiresu S. Rao. Engineering Optimization: Theory
and Practice, 3rd Edition. Wiley-Interscience, 1996.
920 p.
100 % 44.5%.
Nevertheless the accuracy of determining parameters A

and B is reduced by reducing number of measured
points. Current transient is shown in Fig. 4. The
experiment was performed by taking 300 points of
measurement.
0.15
0.1
0.05
Current [A]
-0.05
-0.1
-0.15
-0.2
-0.25
-0.3
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
Time [s]
Fig. 4. Transients of current i2 (300 points)
Approximation of current transient in the beginning of

the process has evident great accuracy, but the error
increases in the end of process. It can be seen from
transient curve, shown in Fig. 5, where experiment
includes 300 points.
114

MODELLING OF AMMONIA PRODUCTION PROCESS

Giedrius OBERAUSKAS, Vytautas GALVANAUSKAS
Abstract: In this paper the detailed first principles

mathematical model of industrial ammonia production
process is elaborated. Mass balance equations and
chemical reaction rate equations of the process are
presented and analyzed. Computer simulation tests
were performed using software tools created in Matlab
program environment. The modelling results were
compared with the experimental data and discussed.
signals lead to even more distorted estimates of the

specific reaction rates.
Thus, one needs a robust and effective process
technique for models in chemical engineering while
dealing with complex chemical processes that cannot
be modelled with the necessary precision using only
simple engineering correlations and mechanistic
models.
Keywords: Mass balance, reaction rates, modelling,

ammonia production process.
2. Ammonia production process description

Modelling of ammonia production process will be
performed using mass balances of the four distinct
phases, i. e. secondary reforming, shift conversion,
CO2 removal, and methanation. The technological
scheme of the process is presented in Fig. 1.
1. Introduction
In the context of this paper modelling of ammonia
production process are representations of chemical
process incorporating mass balance systems for key
components, and mechanistic sub-models for
modelling of absolute reaction rates.
Due to complex reaction systems in this process, like in
all biotechnological and chemical processes [1, 2, 3], it
is not possible to describe the process in details using
only mechanistic models. Thus it is necessary to use
methods that are able to fill some gaps in special
chemical or engineering knowledge. Further
expansions of the data and knowledge driven
approaches used in chemistry engineering are mass
balances systems. In the recently published literature,
some interesting examples show, that hybrid
combination of artificial neural networks, mechanistic
kinetics and mass balance equations can lead to
considerable advantages [1, 2, 3, 4]. Though, in
chemistry one must work with many data of off-line or
quasi-off-line types, and for this reason the above
mentioned data must be generated using interpolation
techniques. Regrettably, it may be challenged not only
by some artificial disturbances, but also by accuracy of
such interpolation methods. Moreover, in chemical
technology one is most often interested in reaction
rates, because they define the behaviour of the
conversions in chemical processes. Rates have to be
determined from the measured or estimated
concentrations by differentiation. In this respect noisy
2.1 Secondary reforming

In the secondary reformer an internal combustion of
the part of the gas and the process air takes place which
also provides the nitrogen for the final synthesis gas.
2H2 + O2 2H2O+Q,
(1)
CH4+O2 CO2+2H2+Q.
(2)
In the conventional reforming process the degree of

primary reforming is adjusted so that the air supplied to
the secondary reformer meets both the heat balance and
the stoichiometric synthesis gas requirement. The
process gas is mixed with the air in a burner and then
passed over a secondary reformer catalyst [6].
2.2 Shift conversion
Shift conversion consists of two levels. First level is
average CO conversion and the second level is low CO
conversion. At both levels CO parts are removed from
process gas based on chemical equilibrium (3). The
process gas from the secondary reformer is converted
in the shift section according to the reaction:
CO+H2OCO2+H2+ 41.0 [kJ/mol].
115
(3)
Fig. 1. Technological scheme of ammonia production process [7]
the component can change due to the feeding of the

corresponding component into the reactor and due to
the dilution effect. Differential equation of the mass
balance for component Ciout is
2.3 CO2 removal

CO2 removal process consists of absorber and stripper.
Process gas from shift conversion is transferred to the
absorber according to the chemical equilibriums (45):
CO2 + H2O H2CO3 ,
d (Ciout )
F
F
qi Ci Ciin in Ciout in ,
dt
V
V
(4)
[R2NCH3]+H2CO3 [R2NHCH3]++[HCO3]-, (5)
where Ciout concentration of i-th component, [mol/l];

Ciin concentration of i-th component in the flow,
[mol/l], qi Ci i-th absolute reaction rate, [mol/(ls)];
where R HOCH2CH2 group.
V volume, [l]; Fin mass flow into the reaction

volume, [l/s].
The CO2 is removed in a chemical or physical

absorption process. The solvents used in chemical
absorption processes are aqueous amine solutions. The
solvent is regenerated using by stripping.
According to (17) chemical equilibriums and data

from ammonia technological rules of procedure [7] the
mass balances equations were elaborated.
Argon and nitrogen are inert gases, and for this reason
they do not play a significant role in mass balances of
the described chemical process.
The amounts of all participating components in a group
of reactions can be expressed in terms of a number of
key components equal to the number of independent
stoichiometric relations. The independent rate
equations will then involve only those key components.
For multiple reactions, the procedure for finding the
concentrations of all participating components starts by
assuming that the reactions proceed consecutively.
Intermediate concentrations are identified by
subscripts. The resulting concentration from a
particular reaction is the starting concentration for the
next reaction in the series. The final value carries no
subscript [6]. The compositions of the excess
components will be expressible in terms of the key
components. The absolute reaction rate equations for
key components can be found as follows:
2.4 Methanation
The small amounts of CO and CO2, remaining in the
synthesis gas, are poisonous for the ammonia synthesis
catalyst and must be removed by conversion to CH4 in
the methanator. The composition of the gas leaving the
methanator follows the chemical equilibriums (6) and
(7):
CO+ 3H2 CH4+H2O + 206.3 [kJ/mol],
(6)
CO2+4H2 CH4+2H2O+165.1 [kJ/mol].
(7)
(8)
3. Mass balance equations of ammonia production

process
Mass balance equations of the substances are based on
mass preservation law, which means that part of
particular concentration variation can be caused by
chemical reactions, during which the elements of
chemical reaction combine with each other to form a
more complex product or decomposed into smaller
compounds or elements. Additionally, the amount of
k1
aA bB cC ,
116
(9)
where a and b empirical coefficients of chemical

equilibriums; K and k0 coefficients of proportionality;
E activation energy, [J]; R universal gas constant,
[J/(K mol)].
k2
aA cC dD bB,
(10)
where a, b, c, and d are empirical coefficients of
chemical equilibriums; A, B, C, and D components of
chemical equilibriums; k1 and k2 constants of reaction
rates.
The absolute reaction rate for components in (9) and
(10) are:
rA ak1 Aa B b ak2 Aa C c ,
(11)
rB bk1 Aa Bb bk2 AaC c ,
(12)
rC ck1 AaC c ck2 Aa C c ,
(13)
rD dk2 AaC c ,
(14)
When presumably accurate data deviate from linearity

as stated by the last equation, the reaction is believed to
have a complex mechanism [6].
Usually, absolute reaction rate of chemical process is
not constant and depends on various factors, like
process temperature, pleasure, activation of catalyst,
etc.
Coefficient of proportionality is necessary in order to
taking into account not only the process temperature
but also the process pressure, activation of catalyst and
other important process parameters, which have
influence on absolute reaction rate.
During the modelling mass balances of ammonia
production process, the data from the ammonia
technological rules of procedure [7] was used as
reference. The data dimensions, like concentrations,
flows and volumes, that do not agree with dimensions
used in modelling of mass balances, are given in the
text. E. g., in mass balances dimension of concentration
is [mol/l], although in ammonia technological rules of
procedure it is expressed in volumetric percents the
same as the values measured in the technological
process. Concentration values were converted to the
necessary dimension of concentration taking into
account the process temperature and pressure (see
Table 1):
where k1 and k2 constants of reaction rates for (9) and

(10) chemical equilibriums.
The equations (11-14) have to be solved numerically
for A, B, C, and D as time functions. Alternatively,
differential equations can be written and solved
directly for the participating components as functions
of time, thus avoiding the use of stoichiometric
balances, although these are really involved in the
formulation of the differential equations [6].
The constant of reaction rates taking into account the
temperature effect on the process has an influence on
the specific rate. The Arrhenius equation relates the
absolute rate to the absolute temperature for chemical
equilibrium (9):
(15)
E
b
) K exp(a ),
k k0 exp(
RT
T
nRT
,
p
(16)
where n amount of substance, [mol]; T process

temperature, [K]; p absolute pressure, [Pa].
Table 1. Data of the ammonia production process

Second reforming
Parameters
Temperature, [oC]
Pressure, [kPa]
Flow Fi, [l/s]
CCO2i , [mol/l]
CCOi, [mol/l]
CH2i, [mol/l]
CCH4i, [mol/l]
CH2Oi,[mol/l]
CO2i, [mol/l]
1(in)
492.00
3115.6
17242
4.99e-5
0.0521
0.0917
2(in)
830.00
3066.5
60676
0.0281
0.027
0.196
0.0321
0.2042
-
CO2 conversion
Shift conversion
3(in)
350.45
2970.4
82376
0.0241
0.0416
0.1845
8.264e-4
0.1804
-
4(in)
199.63
2868.4
84032
0.0715
0.0133
0.279
1.067e-3
0.2055
-
Mass balance and reaction rate equations of ammonia

production process for several chemical elements were
created in accordance with the approaches described
above. According to chemical equilibriums (12), mass
balances equations for the second reforming of
ammonia production process (see structural scheme in
Fig. 2) for the following components are:
d (CCO 2 )
F
F
rCO 2 (CCO 2in1 CCO 2 ) 1 (CCO 2in 2 CCO 2 ) 2 ,
dt
V1
V1
5(in)
65.01
2711.5
60199
0.1684
1.93e-3
0.5819
2.31e-3
0.0108
-
6(in)
50.00
2611.5
77.92
0.2273
-
Methanation
7(in)
50.00
2611.5
151.25
0.2273
-
8(in)
295.00
2518.3
49120
1.59e-4
1.27e-3
0.3901
1.43e-3
2.97e-3
-
d (C H 2 )
F
rH 2 (C H 2 in 2 CH 2 ) 2 ,
dt
V1
d (CCH 4 )
F
rCH 4 (CCH 4in 2 CCH 4 ) 2 ,
dt
V1
(19)
d (CH 2O )
F
F
rH 2O (CH 2Oin1 CH 2O ) 1 (CH 2Oin 2 CH 2O ) 2 ,
dt
V1
V1
(20)
d (CO 2 )
F
rO 2 (CO 2in1 CO 2 ) 1 ,
dt
V1
(17)
117
(18)
(21)
CCO 2 , CH 2 , CCH 4 , CH 2 O , and CO 2 modelled component

concentrations in gas/steam flow after chemical
reaction in the second reforming, [mol/l]; V1 volume
of the reactor, [l].
In the equations (1721) instead of qi C1 one uses
absolute reaction rate ri, because formally they are
equal to each other but the latter is more convenient to
use in the process model analyzed.
where F1 process air flow to the second reforming,

[l/s]; F2 gas/steam flow to the second reforming,
CCO 2 in1 , CH 2 Oin1 ,
and CO 2 in1
component
[l/s];
concentrations in the process air flow ( F1 ) to the
CCO 2 in 2 , CH 2in 2 ,
second
reforming,
[mol/l];
CCH 4 in 2 , and CH 2 Oin 2 component concentrations in
gas/steam flow ( F2 ) to the second reforming, [mol/l];
Fig. 2. Structural scheme of ammonia production process model
The absolute reaction

components are:
rates
of
the
rCO 2 k1CCH 4CO 2 ,
modelled
(22)
rH 2 2k1CCH 4CO 2 k2CH 2CO0.52 ,
(23)
rCH 4 k1CCH 4CO 2 ,
(24)
rH 2O k2CH 2CO0.52 ,
rO 2 k1CCH 4CO 2 0.5k2CH 2CO0.52 ,
d (CCO 2 )
F
rCO 2 (CCO 2in 3 CCO 2 ) 3 ,
dt
V2
(25)
(26)
(27)
d (CCO )
F
rCO (CCOin 3 CCO ) 3 ,
dt
V2
(28)
F
d (C H 2 )
rH 2 (C H 2in 3 C H 2 ) 3 ,
dt
V2
(29)
d (CH 2O )
F
rH 2O (CH 2Oin 3 CH 2O ) 3 ,
dt
V2
(30)
where F3 gas/steam flow to the average temperature

convertor, [l/s]; V2 volume of the average
CCO 2 in 3 , CCOin 3 ,
temperature CO convertor, [l];
CH 2 in 3 , and CH 2 Oin 3 component concentrations in
gas/steam flow ( F3 ) to the average temperature
convertor, [mol/l]. The other parameters of the mass
balances equations are the same as in the second
reforming process described above.
Absolute reaction rates for the components are:
where rCO 2 , rH 2 , rCH 4 , rHO 2 , and rO 2 absolute reaction rates

of component concentrations in the second reforming,
[mol/(ls)]; k1 reaction constant of chemical
equilibrium (2); k2 reaction constant of chemical
C H 2O , and CO 2
equilibrium (1); CCO 2 , CH 2 , CCH 4 ,
concentrations of the modelled component in gas/steam
flow after chemical reaction in the second reforming,
[mol/l].
The reactor has incoming two flows (see structural
scheme in Fig. 2): process air and steam/gas mixture.
The mass balances equations of the reacting chemical
components are (1721), and the absolute reaction
rates are (2226).
Mass balance equations of shift conversion for the
components (see structural scheme in Fig. 2) and
chemical equilibrium (3) for average temperature
conversion of CO are:
rCO 2 kCCO C H 2O ,
(31)
rCO kCcoCH 2O ,
(32)
rH 2 kCcoC H 2O ,
(33)
rH 2O kCCOCH 2O ,
(34)
where k reaction constant of chemical equilibrium

(3).
According to Fig. 1, there are two types of CO
convertors, i. e. average and low temperature
118
parameters of the mass balances equations are the same

as in the second reforming process described above.
convertors. Mass balance equations (2730) are valid

for the average temperature convertor. The chemical
reaction (3) is the same in both convertors, except of
process conditions, such as process temperature and
pressure.
Mass balances of the shift conversion for the
components (see structural scheme in Fig. 2) based on
the chemical equilibrium (3) for low temperature
conversion of CO are:
rCO 2 k1CCO 2C H 2O ,
(45)
rH 2O k1CCO 2CH 2O ,
(46)
where k1 reaction constant of chemical equilibrium

(4).
Mass balances equations of methanation process for
the components (see structural scheme in Fig. 2) based
on chemical equilibriums (67) are:
d (CCO 2 )
F
rCO 2 (CCO 2in 4 CCO 2 ) 4 ,
dt
V3
(35)
d (CCO )
F
rCO (CCOin 4 CCO ) 4 ,
dt
V3
(36)
d (C H 2 )
F
rH 2 (C H 2in 4 C H 2 ) 4 ,
dt
V3
d (CCO 2 )
F
rCO 2 (CCO 2 in 8 CCO 2 ) 8 ,
dt
V5
(47)
(37)
(48)
(38)
d (CCO )
F
rCO (CCOin8 CCO ) 8 ,
dt
V5
d (CH 2O )
F
rH 2O (CH 2Oin 4 CH 2O ) 4 ,
dt
V3
where F4 gas/steam flow to the low temperature

convertor, [l/s], V3 volume of the low temperature
CCO 2 in 4 , CCOin 4 , CH 2 in 4 , CH 2 Oin 4
CO convertor, [l];
component concentrations in gas/steam flow ( F4 ) to
the low temperature convertor, [mol/l]. The other
parameters of the mass balances equations are the same
as in the second reforming process described above.
rCO 2 kCCO C H 2O ,
(39)
rCO kCcoCH 2O ,
(40)
rH 2 kCcoC H 2O ,
(41)
rH 2O kCCOCH 2O ,
(42)
F
F
(CCO 2 in 6 CCO 2 ) 6 (CCO 2in 7 CCO 2 ) 7 ,
V4
V4
d (CH 2O )
F
rH 2O (C H 2Oin5 C H 2O ) 5 ,
dt
V4
(49)
d (CCH 4 )
F
rCH 4 (CCH 4in8 CCH 4 ) 8 ,
dt
V5
(50)
d (CH 2O )
F
rH 2O (CH 2Oin8 CH 2O ) 8 ,
dt
V5
(51)
where F8 gas/steam flow to the methanator, [l/s];

V5 volume of the methanator, [l]; CCO 2 in 8 , CCOin 8 , CH 2in 8 ,
CCH 4 in 8 , and CH 2 Oin 8 component concentrations in
gas/steam flow ( F8 ) to the methanation process,
[mol/l]. The other parameters of the mass balances
equations are the same as in the second reforming
process described above.
Absolute reaction rates for the components:
where k reaction constant of chemical equilibrium

(3).
Mass balance equations of CO2 conversion process for
the components (see structural scheme in Fig. 2) based
on the chemical equilibrium (4) are:
d (CCO 2 )
F
rCO 2 (CCO 2 in 5 CCO 2 ) 5
dt
V4
F
d (C H 2 )
rH 2 (C H 2in 8 C H 2 ) 8 ,
dt
V5
(43)
rCO 2 k2Cco 2C H4 2 ,
(52)
rCO k1CcoCH3 2 ,
(53)
rH 2 3k1CcoCH3 2 4k2CCO 2CH4 2 ,
(54)
rCH 4 k1CCO C H3 2 k2CCO 2C H4 2 ,
(55)
rH 2O k1CCOCH3 2 2k2CCO 2CH4 2 ,
(56)
where k1 reaction constant of chemical equilibrium

(6); k2 reaction constant of chemical equilibrium (7).
(44)
4. Software implementation of the model
where F5 gas/steam flow to the absorber, [l/s]; F6

solvent flow to the absorber, [l/s]; F7 solvent flow to
the absorber, [l/s]; CCO 2 in 5 , CH 2 Oin 5 component
concentrations in gas/steam flow ( F5 ) to the absorber,
[mol/l]; CCO 2 in 6 component concentrations in solvent
flow ( F6 ) to the absorber, [mol/l]; CCO 2 in 7 component
concentrations in solvent flow ( F7 ) to the absorber,
[mol/l]; V3 volume of absorber, [l]. The other
The process model was implemented using the

elements from the standard user-defined libraries for
Matlab/Simulink environment together with the
embedded Matlab functions. The created user-defined
libraries consist of the expressions for absolute
reactions rates, mass balance equation systems.
Additionally, part of the software tool was
programmed using m files.
119
the measured values are influenced by the disturbances

and errors in the control systems and instruments, also
by the procedures of measurement unit conversion
(from [v/v %] to [mol/l]). Generally, concentrations are
modelled with good accuracy as compared with values
from ammonia technological rules of procedure.
Modelling results also depend on the model
parameters coefficients of proportionality. The
optimal values of the model parameters in sense of the
minimal modelling error were calculated using NelderMead optimization algorithm.
5. Modelling results
The most important concentration components of
ammonia production process are shown in Fig. 3.
Concentrations of the components were chosen for
illustration based on the importance to the ammonia
production process and available measurement points
in the plant. The model was created taking into account
the data provided by ammonia technological rules of
procedure (see Table 1) and the modelling results were
compared with the reference values. The results, i. e.
H2 concentration after second reforming
8
6
4
CH4 conc.,[mol/l]
-3
x 10
time,[s]
CH4 concentrations after methanator
3
2
-4
0.16
0.14
0.12
0.1
3
4
5
time,[s]
CO concentration after low temp. CO converter
3
time,[s]
0.011
0.01
0.009
5
modelled
0
-4
9.7
2
CO concentration after average temp. CO converter
x 10
9.75
0.012
CO2 conc.,[mol/l]
9.8
time,[s]
x 10
CO conc.,[mol/l]
CH4 concentrations after second reforming

H2 conc.,[mol/l]
x 10
CO conc.n,[mol/l]
CH4 conc.,[mol/l]
-4
10
3
4
time,[s]
CO2 concentration after CO2 convertor
3
2
1
0
measured
tech.reg.
time,[s]
Fig. 3. Concentrations in the ammonia production process
6. Conclusions and further work

Mass balances of ammonia production process were
presented. The mass balance and absolute chemical
reaction rate equations of ammonia production process
are analyzed and presented. The computer simulation
tests were performed using software tools created in
Matlab program environment. The modelling results
were compared with the experimental data and
discussed.
Further the extensive sensitivity analysis should be
performed in order to determine the most important
influencing parameters. Then the ANNs for the
modelling of the reaction rates should be created,
trained and validated. Depending on the identification
results the appropriate model-based control strategy
should be elaborated.
2.
3.
4.
5.
7. References
1. Galvanauskas V., Georgieva P. and Feyo de
Azevedo S. Dynamic Optimisation of Industrial
Sugal Crystallization Process based on a Hybrid
(Mechanistic+ANN) Model. Proceedings of IEEE
World Congress on Computational Intelligence,
6.
7.
120
IJCNN 2006, July 16-21, 2006, Vancouver,

Canada, Vol. 6. p. 5035-5042.
Galvanauskas V., Simutis R. and Lbbert A.
Hybrid process models for process optimisation,
monitoring and control. Bioprocess and Biosystems
Engineering, 2004, 26(6). p. 393-400.
Schubert J., Simutis R., Dors M., Havlik I. and
Lbbert A. Bioprocess optimization and control:
Application of hybrid modelling, J. Biotechn.,
1994, 35. p. 51-68.
Galvanauskas V., Simutis R. and Leviauskas D.
Application of evolutionary computing for hybrid
model based optimization of biochemical
processes. WSEAS Transactions on Information
Science and Applications, Issue 5, Vol. 2, May
2005. p. 507-511.
European Fertilizer Manufacturers Association,
Best Available Techniques for Pollution Prevention
and Control in the European Fertilizer Industry,
Production of Ammonia, 2000. p. 7-21.
Perry R.H., Green D.W. and Moloney J.O.
Chemical Engineers Handbook. 7th edition,
McGraw-Hill Professional, 1997. 2640 p.
Ammonia technological rules of procedure (AB
Achema)

MATHEMATICAL MODEL FOR ADAPTIVE CONTROL OF

DISSOLVED OXYGEN CONCENTRATION IN
BIOLOGICAL WASTEWATER TREATMENT PROCESS
Kstutis JONELIS, Kstutis BRAZAUSKAS, Donatas LEVIAUSKAS
Abstract: Mathematical model has been developed for

modelling dynamical behaviour of the dissolved oxygen
concentration in an industrial scale biological
wastewater treatment process. Parameters of the process
model are indentified using monitoring data from
wastewater treatment plant. The model identification
and validation results are presented.
0.8
DOC, mg/l
0.7
Keywords:
mathematical
model,
identification,
dissolved oxygen concentration, biological wastewater
treatment process.
1.
DOC
Set point
0.9
0.6
0.5
0.4
0.3
0.2
0.1
0
Introduction
10
20
30
40
50
60
70
Time, h
Fig. 1. Performance of the DOC control system with ordinary

PI controller under varying technological conditions (data
from industrial wastewater treatment plant)
An accurate control of dissolved oxygen concentration

(DOC) at optimal level during the biological waste
water treatment process is a relevant technological
problem due significant energy expenses for aeration.
However, automatic control of DOC is complicated by
nonlinearity and time-varying dynamical characteristics
of the controlled process. The traditional linear
controllers (PI, PID) with constant parameters are not
adequate to ensure an accurate control under process
state variations [1].
Currently, various control methods are developed to
cope with process nonlinearity and state variations that
are based on using process state models to predict the
process dynamics [2-4].
The DOC control problem is live in the Kaunas
wastewater treatment plant, as the ordinary automatic
DOC set-point control system demonstrates poor
performance under varying technological conditions
(Figure 1).
In this work, the state model is developed for modelling
dynamical behaviour of DOC in the biological
wastewater treatment process. The mathematical model
is identified using monitoring data from wastewater
treatment plant. The model is intended for design of the
DOC adaptive control system.
2.
Technological process
Technological scheme of the wastewater treatment

process is shown in Figure 2. Incoming wastewater flow
Fin with substrate concentration Sin through the screens,
comminutor and grit chamber falls into the distribution
chambers. Subject to the position of bypass gate, the
flow is proportioned in two directions: one portion goes
through primary sedimentation (Fsed) tank, the other straight to the distribution chamber and the biological
treatment tanks (Fbypass). The flow distribution
proportion is evaluated by coefficient ksed that shows the
percentage of the flow through the primary
sedimentation, depending on the motorized bypass gate
position g (0% - closed, 100% - fully opened). This
relationship is estimated by interpolation of three points
given in the documentation of the distribution chamber
construction (Table 1).
Table 1. Bypass gate position and ksed points
Bypass gate position g, %

0
28.8
100
121
ksed, %
100
50
0

Fig. 2. Technological scheme of wastewater treatment process
The relationship
polynomial:
is
approximated
by
ksed = 0.01034g2-2.034g+10.
biological treatment tanks. As the methanol was not

added during the process investigation period, the
substrate concentration Sbio in the biological treatment
tanks inflowing wastewater was estimated using the
following formula:
quadratic
(1)
The wastewater flow through primary sedimentation

tank is
Fsed
k sed
Fin .
100
S bio
(2)
where Fret is flow rate of returned sludge from the final

settling tank with substrate concentration Sret. The
excessive sludge with the flow rate Fexc is removed for
digestion.
The airflow Fair, comes from the blower station to four
biological tanks and the air flow rate is evaluated by
measuring power consumption P of compressors.
Referring to technical documentation of the plant the air
flow rate is estimated by the following linear
relationship:
The sludge flow Fsl from primary sedimentation tank is

very low, compared to the wastewater flow (~1%), so it
is not taken into account in the following calculations.
Depending on the wastewater flow through primary
sedimentation tank, substrate concentration in the
outflow decreases from 15% (at high flow rates) to 40%
(at low flow rates). The percentage of the substrate
concentration decrease in the outflow of sedimentation
tank is evaluated by the linear relationship
kdec = 0.0043Fsed + 40,
(3)
Fair = 0.05455P 1200,
where kdec, is the percentage of substrate concentration

decrease in the wastewater outflow, %; Fsed is the flow
rate through sedimentation tank, m3/h.
With the estimated value of kdec, the substrate
concentration in the outflow of sedimentation tank is
estimated using formula:
S sed S in
k dec
S in .
100
k
k
sin Fin sed Fin ssed sed Fin sret Fret

100
100

, (5)
Fin Fret
(6)
where P is the power consumed by compressors.

In the presented investigation, monitoring data from one
of biological treatments tanks was used. Airflow to the
selected tank is estimated taking into account position of
airflow valves to each tank:
Fair 4 Fair
(4)
4
,
1 2 3 4
(7)
where Fair4 is airflow to the selected tank (4th), m3/h; v1,

v2, v3, v4 are valve positions to corresponding tanks, %
The two wastewater flows are mixed in the distribution

chamber. If the total flow outreaches 5855 m3/h, the
overflow Fov is directed straight to the outlet passing the
biological tanks and the final settling tank. Some
methanol is added (Fmet) if necessary for better
biological treatment. The mixed wastewater flow Fbio
with the substrate concentration Sbio falls into four
122
3.
stable, the term N is identified as parameter * .

Preliminary values of parameters * , , max , K, Yx/s,
Mathematical model of DOC in biological

wastewater treatment process
Yx/o are taken from [6-9]. The parameters values

minimizing the objective function (12) are identified
using stochastic search algorithm (evolutionary
programming, [10]). The computation program is
realized using the Matlab tools.
The initial values of state variables and the identified
values of the model (8)-(11) parameters are presented in
Table 2.
The state model of the waste water biological treatment

process is based on the mass balances for dissolved
oxygen and substrate [1, 5]:
F
dc
k L a (csatur c) (cin c) bio OUR,
V
dt
max s
sbio s
ds
X
Fbio ,
dt
Yx / s K s
V
max s
OUR
Yx / o K s
X,
k L a N Q ,
(8)
(9)
Table 2. Initial values of state variables and identified values

of the model
(10)
Initial values of state variables

Variable
Unit
c
mg/l
s
mg/l
Identified values of the model
*
h-1
max
(11)
where c is the dissolved oxygen concentration, kLa is

coefficient of mass transfer from gas to liquid phase,
csatur is saturated value of DOC, cin is DOC in the inlet
wastewater, Fbio is the inlet flow rate of wastewater, V is
volume of aeration tank, OUR is oxygen uptake rate, s is
resultant substrate concentration, max is maximum
specific growth rate of biomass, K is half saturation
coefficient, Yx/s is biomass/substrate yield coefficient,
Yx/o is biomass/oxygen yield coefficient, X is biomass
concentration, particular amount of substrate to keep
biomass alive, Sbio is substrate concentration in inlet
wastewater, N is stirring speed, Q is air flow rate,
, , are parameters.
4.
mg/l
mg/mg
mg/mg
mg/l
-
K
Yx/s
Yx/o
X0
k
Value
0.16
32.28
5.09
0.138
0.10
35.36
2.80
0.052
140
-0.0006
The observation data for model identification and model

prediction of DOC are presented in Figures 3-6.
Model parameters identification
1020
Air flow
As the only state variable c is directly measured (the

state variable s is estimated by the mathematical
model), an objective of the model parameters
identification procedure is to minimize the squared error
1000
980
m3/h
960
940
S cmod el cdata 2 dt min, (12)
900
where
cmod el are model predictions of DOC, cdata are

the DOC observation data of real process, T is time
880
860
interval of observation.
For integration of the model equations (8), (9) along the
observation time interval, the discrete observation data
of technological parameters Fin , Fret , Q and c were
840
0.05
0.1
0.15
0.2
Time, h
0.25
0.3
0.35
0.4
Fig. 3 Inlet air flow rate

1000
interpolated using the Matlab tools (the liner Lagrangian

interpolation). Initial value of directly unmeasured state
variable st 0 is estimated by the process expert, the
Inlet flow
Returned flow
Total flow
900
800
700
sbio is estimated by
Total flow
m3/h
inflowing substrate concentration
920
formula (5). Biomass concentration X was estimated by

linear relationship
600
500
400
X X 0 Fret k ,
(13)
Inlet flow
300
200
where X 0 and k are parameters to be identified.

Model parameter cin is directly measured, csatur is
physical constant, V is constructive parameter. As the
stirring speed in the investigated process is kept up
100
Returned flow
0
0.05
0.1
0.15
0.2
Time, h
0.25
0.3
0.35
Fig. 4. Inlet, returned and total wastewater flow rates
123
0.4
176
concentration in the wastewater biological treatment

process under varying operating conditions. The model
identification and validation results prove suitability of
the model to describe and predict dynamical features of
the process. The model is further intended for on-line
adaptation of controller parameters in the dissolved
oxygen automatic control system in order to increase
performance of the system.
Estimated inlet substrate consentration S bio in wastewater
175
174
mg/l
173
172
6.
171
1.
Leviauskas D. An Algorithm for Adaptive Control

of Dissolved Oxygen Concentration in Batch
Culture. Biotechnology Techniques, 1995, Vol. 9,
n.2, p. 85-90.
2. Bastin, G., Dochain, D. On-Line Estimation and
Adaptive Control of Bioreactor. Elsevier, New
York, 1990.
3. Lindberg, C. F. Multivariable Modeling and
Control of an Activated Sludge Process. Wat. Sci.
Tech., 37, 149, 1998.
4. Yoo C.K., Cho J.H., Kwak H. J., Choi S. K., Chun
H.D., Lee I. Closed-Loop Identification and
Control for Dissolved Oxygen Concentration in the
Full-Scale Coke Wastewater Treatment Plant. Wat.
Sci. Tech., 2001.
5. Staniskis J., Levisauskas D., Simutis R.,
Viesturs U., Kristapsons M. Automation of
Biotechnological Processes: Automatic Monitoring,
Optimization and Control. (In Russian), Riga,
Zinatne, 1992. 348 p.
6. Eswaramoorthi S., Dhanapal K., Chauhan D.D.
Application of membrane bioreactor for textile
wastewater treatment: Pilot plant process modeling
and scale-up. New Deli, India, 2003.
7. Daigger G.T., Buttz J.A. Upgrading wastewater
treatment plants. Second edition. Lancaster,
Pennsylvania, 1998.
8. Repyt J., Simutis R. Mathematical model of
biological
wastewater
treatment
process.
Informacins technologijos ir valdymas. 2001,
No. 4(21), p. 21-30.
9. http://www.ensic.inpl-nancy.fr/COSTWWTP
10. Porto V.W., Fogel D.B., Fogel L.J.. Alternative
neural networks training methods. IEEE Expert,
1995, p. 16-22.
170
169
0
0.05
0.1
0.15
0.2
Time, h
0.25
0.3
0.35
0.4
Fig. 5. Estimated substrate concentration in the biological

treatment tank inflowing wastewater
0.3
Observed DOC
Predicted DOC
DOC, mg/l
0.25
0.2
0.15
0.1
0.05
0.05
0.1
0.15
0.2
Time, h
0.25
0.3
0.35
0.4
Fig. 6. Observed and model predicted DOC
In Figure 7 a-b the model validation results are

presented for prediction the DOC from the process
monitoring data that were not used for model
identification.
Fig. 7a. Model validation
DOC, mg/l
0.16
Observed DOC
Model predicted DOC
0.14
0.12
0.1
0.08
0.05
0.1
0.15
0.2
0.25
Time, h
Fig. 7b. Model validation
0.3
0.35
0.4
DOC, mg/l
0.25
Observed DOC
Model predicted DOC
0.2
0.15
0.1
0.05
0.05
0.1
0.15
0.2
Time, h
0.25
0.3
0.35
0.4
Fig. 7. Model validation results
The results presented in Fig. 6, 7 a-b demonstrate that

the identified model satisfactory predicts the DOC
variation under varying technological conditions of the
biological wastewater treatment process.
5.
References
Conclusions
In this work, the mathematical model is developed for

modelling alternation of the dissolved oxygen
124

ADAPTIVE CONTROL SYSTEM FOR DISSOLVED OXYGEN CONCENTRATION

IN KAUNAS WASTEWATER TREATMENT PLANT
Jolanta REPYT*, Ieva EBATORIEN*, Andr WENZEL**
* Kaunas University of Technology (KTU), Lithuania
** University of Applied Sciences Schmalkalden (fhS), Germany
in a quantity that requires the biological processes

[1, 3, 7].
Under this scheme aerobic and anoxic tanks combined.
Nitrification and denitrification is achieved in
alternating aeration intensity.
Aeration intensity varies cyclically recurring set of two
periods, etc. and creating the conditions for full
nitrification take place in aeration intensity increased by
0,5-1 mgO2/l, the oxygen concentration to maintain.
Then reduce the intensity of aeration, and maintaining
0,1-0,5 mgO2/l, the oxygen concentration, allowing for
the full denitrification occur [1].
Organic matter, phosphorus and nitrogen compounds are
disposed from the sewage in Kaunas waste water
treatment plant. Activated sludge, separated by
secondary settler, is returned to the anaerobic chambers
of the bioreactor. Technological process in the
bioreactor is controlled by dissolved oxygen,
temperature, pH, and the active sludge concentration
sensors. The excess sludge is drained and used for
biogas producing [3].
Abstract: The APROMOCS project researchers

developed Kaunas waste water treatment plant
biological wastewater treatment process mathematical
model, with the goal to develop control system for strict
dissolved oxygen concentration requirements. In the
next stage was developed adaptive control system for
dissolved oxygen concentration. This article describes
the implementation of the biological wastewater
treatment process model and control system into
software SIMULINK (MATLAB). The adaptive control
system for dissolved oxygen concentration is a part of
the EUREKA-Research-Project E!3692- APROMOCS.
Keywords:
Biological
wastewater
treatment
mathematical modeling; adaptive control system.
1. Introduction
SYMBIO technology is implemented in new Kaunas
wastewater treatment plant. Danish firm "BioBalance"
has been patented this technology in the United States
Patent Office on 1996 [1]. This is one of the modern
nitrogen and phosphorus removal technologies [2].
Biological nitrogen removal from wastewater based on
nitrification and denitrification processes are taking
place in biological treatment plants. The aim of this
technology is to assess the nitrification and
denitrification rates and aeration regimes effect to the
biological nitrogen removal [4].
Typically, the nitrogen removal scheme combined with
the phosphorus removal schemes. One of them is the
SYMBIO technology in which the nitrogen is removed.
Nitrification and denitrification area is not specifically
excluded in the bioreactor, and the biological removal of
phosphorous against nitrification and (or) denitrification
areas are equipped with anaerobic zone. The technology
is based on the fact that the new technique allows the
monitoring of the biological control of the process
directly. The control objective is to supply a minimum
quantity of air to ensure a continuous supply of oxygen
2. Object Description
Two new settlers, methanol dosing station, blower room,
4 bioreactors, and distribution chamber with the pump
station were built in Kaunas waste water treatment plant
in 2008 [8].
Wastewater from the primary settlers is transported to
the distribution chamber. Here, wastewater is mixed
with the circulated sludge and dosed methanol. Sewage
and sludge mixture circulation is divided into 4 channels
for sewage and sludge mixture circulation of the
biological treatment tanks (bioreactors).
Kaunas waste water treatment plant has 4 biological
treatment tanks, where the mixture stirred 24 mixers
ensuring adequate mixing of sewage. Each mixer is
equipped in the bottom of the tanks. Effluent at the
origin of organic material will be removed biologically
active sediment phase, seeks to phosphorus, nitrogen
and carbon degradation.
125
The main control parameter at this stage is the dissolved

oxygen concentration. Currently disposed control
system does not ensure the necessary quality parameters.
It is necessary to maintain the oxygen concentration is
0.15 mg/l, but it is much higher or lower. Oxygen and
air valve position is shown in Figure 1.
W0 ( s ) =
0.0018
e 12600 s , mg/l/%
3.5962 103 s + 1
(1)
Disolved oxygen concentration, mg/l
0.1
0.09
0.08
0.07
0.06
0.05
0.04
0.03
Experimental data
Modeling data
Filter data
0.02
0.01
0
10
12
14
Time, h
Fig. 2. Object reaction curve from the first experiment
The object transfer function (II model) from the second

experiment:
Fig. 1. Example of the dissolved oxygen concentration control

in Kaunas waste water treatment plant. (1 air valve position
(0-100 %), 2 set point (0.15 mg/l), 3 measured dissolved
oxygen concentration (0-2 mg/l))
W0 ( s ) =
There is a modern process control system with

technological
process
visualization
(SCADA)
WinCC v6.2 (Siemens) and controllers SIMATIC S7400 (Siemens) in Kaunas waste water treatment plant.
WinCC is used to visualize the process. WinCC allows
to the operator to monitor the process. The process is
graphically displaying on the screen. The image is
updated, a change in time. This program also allows you
to control the process. This program show in details all
the process and put in place control, data collection and
storage functions. Experiments data of the technological
process were recorded in the archiving system every 15
seconds.
Experimental data for dissolved oxygen concentration
mathematical model identification were collected during
two months. Experimental data was selected only when
the controller was turned off and the valve was opened
or closed in step manually. Total experiment time, the
dissolved oxygen concentration measurement and the
adjustment has been used by one meter.
All the experimental data were selected in the three
experimental space (Fig. 2 Fig. 4), of which the object
was made up of mathematical models and identification
of them parameters.
Before approximation, experimental data was filtered by
moving average filter. Area method was selected for
approximation of experimental data [9].
Object reaction curve approximation was carried out in
the area method with software MATLAB. Object was
described by the first row transfer function with
transport delay.
The object transfer function (I model) from the first
experiment:
0.0029
e 12000 s , mg/l/%
4.9796 103 s + 1
(2)
Dissolved oxygen concentration, mg/l
0.2
Experimental data
Modeling data
Filter data
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
10
12
14
Time, h
Fig. 3. Object reaction curve from the second experiment
0.35
Experimental data
Modeling data
Filter data
0.3
0.25
0.2
0.15
0.1
0.05
10
12
Time, h
Fig. 4. Object reaction curve from the third experiment
126
14
The object transfer function (III model) from the third

experiment:
0.965 T pr 0.855
( )
K pr pr
T pr
I=
0.0039
W0 ( s ) =
e 12240 s , mg/l/%
6.9547 103 s + 1
(3)
= 0.01179
pr
0.796 0.147
T pr
3. Dissolved Oxygen Concentration Control System
0.965 T pr 0.855
D=
( )
0.308 T pr
K pr pr
Were selected PI and PID controllers for process

controlling. Regulator parameters were calculated by
direct quality criterion of the minimum integrated
absolute error with the time weighted (ITAE) method
[5]. The object transfer function with biggest gain
coefficient (III mathematical model) was used for
regulators parameters estimation on the basis of classic
control theory.
PI regulator parameters were calculated by these
equations [9]:
(7)
pr
T pr
0.929
(8)
= 5.527 105
The dissolved oxygen control system was realized in the
software SIMULINK (MATLAB).
The control systems with different regulators (PI and
PID) reaction to process interference (Fig. 5) and set
point change (Fig. 6) were presented and analyzed.
I=
pr
1.03 0.165
T pr
(4)
= 0.01179
(5)
0.24
0.22
I model
II model
III model
0.24
I model
II model
III model
0.25
0.18
0.16
0.24
PI
0.23
0.22
0.21
0.2
20
40
60
80
I model
II 120
model
III model
100
Time, h
0.19
0.18
0.17
0.16
20
40
60
80
100
120
Time, h
0.22
Fig. 5. Control systems reactions to set point change
0.2
The PI controller and the PID controller works well

enough, when the object is described by the third
mathematical model, but not enough to ensure quality
control, when object is described by the first or the
second mathematical model. Therefore, it is appropriate
to use adaptive dissolved oxygen concentration
regulator.
Classic automatic control system is designed on the
assumption that the process is linear in working point
environment. From experiment data (Fig. 2 Fig. 4) we
see, that our object is not linear.
One way to solve control problems associated with the
linearization is to use the design procedure, known as
gain scheduling [6, 10]. Gain scheduling control system
parameters are replaced, depending on the assessed realtime operation conditions.
Controller design procedure is divided into two phases.
The first stage establishes local coordination parameters
of the linear regulator, which are optimal in the
individual points of stationary states environment.
In the next stage, localized controller parameters are
interpolated and the result is obtained the global
I model
II model
III model
0.18
0.16
20
40
60
80
100
120
Time, h
0.18
0.16
0.005
0.2
0.15
0.2
PID
0.22
0.26
0.26
0.24
0.586 T pr 0.916
( )
K pr pr
T pr
0.26
0.586 T pr 0.916
P=
( )
= 89.53
K pr pr
0.01
0.015
0.02
0.025
0.03
0.035
Time, h
Fig. 5. Control systems reactions to process interference
PID regulator parameters were calculated by these

equations [9]:
P=
0.965 T pr 0.855
( )
= 152.6
K pr pr
(6)
127
regulator. The global regulator parameters are adjusted

automatically, depending on the status of the process.
One of the variables easily measured in real time and
used to determine the point of stationary states. This
variable may be the regulator output signal, the status of
the process variable or external variable, if it
unambiguously reflects the dynamic changes in
controlled process parameters [6].
To determine the object mathematical model parameters
dependencies, were analyzed all the measured
parameters, such as flow, temperature, air pressure, etc.
It was found that the object mathematical model
parameters depend on the solids concentration Tss .
K o = 0.0015 Tss 0.0229 Tss + 0.0864
Object time constant dependence on the total solids

materials concentration Tss :
2
To = 5451.65 Tss 86496.43 Tss + 346668.62 (10)

Table 2. PI regulator parameters dependence on the Tss
Tss , g/l
7.99
8.44
8.72
Table 1. Transfer function parameters dependence on the Tss

Transfer function, mg/l/%
0.0018
W0 ( s ) =
e 12600 s
3.5962 103 s + 1
0,0029
W0 ( s ) =
e 12000 s
4,9796 103 s + 1
0.0039
W0 ( s ) =
e 12240 s
6.9547 103 s + 1
(9)
Tss , g/l
106.00
88.66
89.53
0.0138
0.0111
0.0095
Table 3. PID regulator parameters dependence on the Tss
7.99
Tss , g/l
7.99
8.44
8.72
8.44
188.1
154.2
152.6
0.01547
0.01346
0.01179
D
6.501105
5.455105
5.527105
8.72
3. Adaptive Dissolved Oxygen Concentration
Control System in the software SIMULINK
(MATLAB)
Ko, mg/l/%
Simulator of the adaptive dissolved oxygen control

system is presented in Figure 8.
The simulator functional block "Object Parameters" is
calculating object transfer function parameters in
accordance with 9 10 equation. Next, the functional
block, "Regulator Parameters" is calculating the new
regulator parameters according to the recalculated
object transfer function values.
Fig. 6. Object gain coefficient dependence on the total solids

materials concentration Tss
ko
7.99
Tss
Tssfcn
Differential
Equation
Editor
to
Object
Param eters
Process
Interference
Out1
Object
Set Point
0.15
P
K
ko
fcn
to
Transport
Delay
Scope
u
I
D
D
PID
Regulator
Parameters
Fig. 8. Adaptive dissolved oxygen control system realized in

the software SIMULINK (MATLAB)
Fig. 7. Object time constant dependence on the total solids
materials concentration Tss
The analysis of the dissolved oxygen control systems

reactions to process interference and set point change
(Fig. 9 and Fig. 10), it was found that the control system
Object gain coefficient dependence on the total solids

materials concentration Tss :
128
0.19
0.18
0.22
0.21
0.2
0.24
Adaptive PI regulator
PI regulator
0.22
0.2
0.18
PI regulator
0.16
PI regulator
0.19
20
40
60
80
100
120
Time, h
0.18
0.17
0.16
0.15
0.17
20
40
60
80
100
120
0.21
0.2
0.19
0.18
0.22
0.21
0.2
0.19
0.18
0.24
Adaptive PID regulator

PID regulator
0.22
0.2
T =8.72 g/l

ss
PID regulator
0.18
0.16
20
Adaptive0 PID regulator
PID regulator
0.17
40
60
80
100
120
Time, h
Tss=8.44 g/l
0.16
0.15
20
40
0.17
60
80
100
120
Time, h
Tss=7.99 g/l
0.16
0.15
Time, h
0.23
0.26
20
40
60
80
100
120
Time, h
0.16
20
40
60
80
100
Fig. 11. Control systems (adaptive PID regulator and classic

PID regulator) reactions to process interference
120
Time, h
Fig. 9. Control systems (adaptive PI regulator and classic PI

regulator) reactions to process interference
0.26
0.25
0.24
0.21
0.2
0.19
0.18
0.17
0.16
0.15
0.25
0.24
PI regulator
0.23
0.25
0.22
0.21
0.2
0.19
0.18
20
0.17
40
0.16
0.15
0.22
0.23
0.24
0.22
0.2
0.18
0.16
0.24
0
0.23
0.22
PI regulator
80
100
120regulator
Adaptive
PI
0.2
Time, h
PI regulator
0.21
60
40
60
80
100
0.23

PID regulator
0.22
0.25
0.24
0.21
0.2
0.19
20
0.18
0.17
0.16
0.15
0.19
0.18
20
0.17
0.25
120
0.15
0.23
0.2
0.24
0.26
0.21
with the adaptive PI controller control faster the process

than the control system with classic PI controller.
0.24
0.22
40
0.21
60
80
100
120
Time, h
0.2
0.19
0.18
20

60 PID regulator
80
100
120
Time, h
40
0.17
0.16
0.15
Time, h

PID regulator
0.23
20
40
60
80
100
120
Time, h
0.16
0.15
20
40
60
80
100
Fig. 12. Control systems (adaptive PID regulator and classic

PID regulator) reactions to set point change
120
Time, h
Fig. 10. Control systems (adaptive PI regulator and classic PI

regulator) reactions to set point change

(Fig. 13 and Fig. 14), it was found that the control
system with the adaptive PID controller control faster
the process than the control system with adaptive PI
controller.

(Fig. 11 and Fig. 12), it was found that the control
system with the adaptive PID controller control faster
the process than the control system with classic PID
controller.
129
0.2
0.19
0.18
0.21
0.24
0.23
0.22
0.21
0.2
Tss=8.72 g/l
0.24
0.22
Adaptive PIDAdaptive
regulatorPID regulator
0.2
0.18
0.16
Adaptive
PI regulator
0.19
0.18
20
40
60
80
100
120
Time, h
0.17
Tss=8.44 g/l
0.16
0.15
0.17
20
40
60
80
100
120
Time, h
5. Acknowledgment
Tss=7.99 g/l
0.16
0.15
mathematical model, but not enough to ensure quality

control, when object is described by first or second
mathematical model. Therefore, it is appropriate to use
adaptive dissolved oxygen concentration regulator.
3. The analysis of the dissolved oxygen control systems
(Fig. 9 Fig. 14), it was found that the control system
with the adaptive PID controller control faster the
process than the control system with adaptive PI
controller.
4. This dissolved oxygen control system with adaptive
PID regulator could be installed in Kaunas waste water
treatment plant control system.
0.26
20
40
60
80
100
This research is related to the Eureka project E!3692

APROMOCS supported by Agency for International
Science and Technology Development Programmes in
Lithuania.
120
Time, h
Fig. 13. Control systems (adaptive PI regulator and adaptive

PID regulator) reactions to process interference
1. Sewage Treatment Process. SymBio process of U. S.

Pat. No. 6743362 B1., Jun.1, 2004
2. Methods and aparatus for biological treatment of
aqueous waste. U. S. Pat. No. 2002/0148779 A1.,
Oct. 17, 2002
3. Carlsson B., Lindberg C.F. Some control strategies
for the activated sludge process, Systems and
Control Group, Uppsala University, Oct 96, Rev Oct
97, Aug 98, 21p.
4. http://www.enviroquip.com/tech_transfer_db//
Symbio/technical_resources/technical_papers/
Indutrial Wastwater 2002 SymBio Paper.pdf.
5. Ogunnaike B.A., Ray W.H. Process Dynamics,
Modeling and Control. Oxford University Press,
1994, 1260 p.
6. Hautala J., Yli-Kotila H. Parameter adaptive control:
predictive control. Oulu, 1993.
7. http://www.kaunovandenys.lt/lt/?id=178
8. http://www.kaunovandenys.lt/lt/aplinksauga/es/
9. Aleksa V., Galvanauskas V. Technological processes
automation and control. (In Lithuanian), Kaunas,
Technologija, 2008. 284 p.
10. Leviauskas D. Adjustment of automatic control
systems. Vilnius, 2008, (in Lithuanian).
0.24
0.2
0.18
0.16
0.25
0.24
0.23
0.22
0.21
0.2
0.19
20
0.18
0.17
0.16
0.15
0.22
6. References
0.26
0.25
0.24
0.23
0.22
60 Adaptive
80 PI regulator
100
120
PID regulator
Time,Adaptive
h
0.21
40
0.2
0.19
0.18
20
0.17
40
60
Adaptive
PID regulator
80
100
120
Time, h
0.16
0.15
20
40
60
80
100
120
Time, h
Fig. 14. Control systems (adaptive PI regulator and adaptive

PID regulator) reactions to set point change
4. Conclusions
1. To determine the object mathematical model

parameters dependencies, was analyzed all the measured
parameters, such as flow, temperature, air pressure, etc.
It was found that the object mathematical model
parameters depend on the solids concentration Tss .
2. The classic PI controller and the PID controller
works well enough, when the object describes the III-rd
130

ANALYSIS OF EFFICIENCY OF A PUMP SET OPERATING

IN AN IRRIGATION SYSTEM
Elbieta SZYCHTA, Leszek SZYCHTA, Radosaw FIGURA
weather conditions, and soil type. In order to distribute

water to rotors, a pipeline system of several dozen
kilometres is constructed all over an irrigated area [9].
Uniform water supply to the entire irrigated area is
ensured by hundreds of rotors, divided into k sections.
One to several pieces of rotors, depending on their flow
rate, are situated in each section [9]. Sum total of flow
rates Qi of the equipment located in one section may not
exceed the maximum flow rate of the pumping station at
nominal head of the rotor, in consideration of the head
differences in the field and of the hydraulic losses on the
path from the pumping station to a working section.
Combining rotors of various operating characteristics
results in sections of differing flow rates Qi, with those
of maximum Qmax and minimum flow rate Qmin.
It is assumed that ith section can operate correctly if the
head in the station's suction branch is Hti. The value of
Hti is the sum total of static head Hsti, considering land
topologies and nominal operating pressure of the
equipment in ith section and hydraulic losses Hi
resulting from hydraulic resistance of the pipeline [3],
and is calculated as follows:
Abstract: The paper presents selected properties of

irrigation system and the influence of pump set control
system on pumping efficiency for its system. The
method of calculating pumping efficiency was shown.
The variation of pump set efficiency as function of
pumps flow rate was analyzed. The method of
calculating pumps rate of flow for the most energysaving control system was presented.
Keywords: irrigation system, pump set, pumping
efficiency, energy-saving control system.
1. Introduction
Countereffective utilization of energy adversely affects
the natural environment. Energy efficiency of
operational processes, increased by developments in
power electronics and emergence of modern control
systems [5,7], reduces demand for energy of end users
and has beneficial effects on climate phenomena,
mitigates noxious emissions, improves health condition,
and reduces energy consumption of the gross national
product.
Water pump stations in Poland, consuming
approximately 2,000 GWh per annum, constitute a
group of electricity users [7]. They comprise pump
stations for irrigation systems. Irrigation systems in
intensive farming and sports and leisure facilities are
necessary elements, while minimized operating costs
boost profits and return on investment. Operating costs
of irrigation systems are related to such factors as
repairs, maintenance, and electricity consumption and
dependent on a control system of a pump set, among
other things. Selection of an appropriate control system
for a pump station is a major element which should be
based on a preliminary analysis of a client's
requirements.
H ti H sti H i ,
(1)
H sti H ni H sti ,
(2)
where: Hni nominal head for ith section at Qi,

Hsti static losses.
The equations (1) and (2) define the required head at the
station's suction branch. Depending on the value of head
at the suction branch, the head of the pump set Hdi is
expressed:
H di H ti H s ,
(3)
where: Hs inlet water head.
2. Selected properties of irrigation systems
In most extensive irrigation systems, water collected by

a pump station comes from open reservoirs, it is
therefore assumed that Hs=0, and the pump set's head
Hdi equals Hti. The varied static Hsti for the individual
Irrigation systems in Poland must supply several

hundred thousand m3 of water a year [10]. The
quantities of water may vary and depend on the crops,
131
more effective than cascade control. There are cases of

pump set control with smooth regulation of pump
rotational speeds, however, where water pumping
efficiency falls below 20%. To minimize energy
consumption of pump sets, pumping efficiency should
be maintained at the maximum level possible. Measures
have been taken to find a control system for a pump set
that would maintain its efficiency at the highest level.
To establish the maximum efficiency of a pump set,
variability of the efficiency function needs to be tested
as dependent on such parameters as pump set's head and
flow rate. This article analyses variations of the pump
set's efficiency as a function of flow rate of pumps in an
irrigation system's pump station.
sections and diverse Hi cause that the head at the

station's suction branch, Hti, should range between Htmin
and Htmax (1). Systems controlling pump stations in most
rotors maintain a constant head of Htmax at the suction
branch. Those sections for which pipeline hydraulic
losses, Hi, and static head Hsti, are the greatest operate
correctly. In such cases, head Hzi supplied to ith section
is expressed:
H zi H t max H sti H i .
(4)
The control system mentioned above makes it necessary

to install, upstream of sections, elements that provide
for choking the head from Hzi down to Hni, which
ensures correct operation of a section. Properties of an
irrigation system may be illustrated as a family of pump
H=f(Q)-curves of a pumping system e.g. Fig.1 described
with the equation (1). The maximum number of such
curves depends on the number of sections k in an
irrigation system.
3.1. Control of a pump set using one frequency

converter
Systems controlling a pump set using one frequency

converter effect shifts of the working point of only one
pump unit (guiding pump). The working points of the
remaining pump units in a pump set adjust to set heads
at a station's suction branch Hti and the current water
distribution Qi. A unit of regulated rotational speed
maintains a set Hti. By smoothly varying rotations of a
pump unit, the guiding pump's flow rate Q1 changes.
H
n=nN
Htrma
Htri
Htmax
Hstma
Htrmin
Hti
Htri -
3.2. Control of a pump set using multiple frequency

converters
(Qi, Hti)
Hsti
Hstmi
Qmin
Qi
Qmax
Systems controlling a pump set using multiple

frequency converters (Multi Motor Drive - MMD)
require frequency converters to supply power to each
pump unit included in a pump set. This control system
provides for random variation of working points of the
individual pump units (Qzi, Hti), while adapting heads at
the station's suction branch to variable distribution of
the pumping system.
The converter selects working points of the particular
pump units on the basis of numerical calculations as per
the instructions in the converter's internal program to
maximize total efficiency of the pump set.
Fig.1. Pump H=f(Q)-curves of irrigation system and pump's

H=f(Q)-curve at rotational speed n=nN.
To simplify the graphic representation of a pump

H=f(Q)-curve, the set is assumed to comprise a single
pump. Fig.1 plots the pump's H=f(Q)-curve at rotational
speed n=nN and pumping system's curves for selected
sections. The pumping system's curves for the initial
(i=1) and final (i=k) section are presented, where static
heads become lowest (Hstmin) and greatest (Hstmax),
respectively. The pumping system's curve for a selected
ith section is plotted as well. A point located in the
pumping system's curve for ith section, coordinates (Qi,
Hti), is assumed to graphically represent the operation of
a single section. Heads at the pump station's collector
Hti, water requirement Qi, and operating times for the
particular sections ti enable an accurate determination of
distribution characteristics of an irrigation system. This
helps to calculate variations of water pumping
efficiency coefficient in irrigation systems and to
develop a control system that minimizes energy
consumption of a pump station.
4. Efficiency of a pump set during operation in

irrigation systems
The ultimate purpose of a system control operation of a

pump set is to maintain a rated head at the pump
station's collector with variable distribution of the
pumping system. The controller should select rotational
speeds of pump units in such a way that the overall
efficiency of a pump set c is the greatest possible while
maintaining the parameters set by the user.
4.1. Analyzing variability of the pump set's efficiency
The overall efficiency of a pump set c depends on

efficiencies and flow rates of working pump units. The
efficiency of a pump unit zi depends on the efficiency
of the frequency converter fi, motor si, and pump
employed pi [3,4,7].
3. Effect of pump set control system on pumping

efficiency
Cascade and converter control are most commonly

employed in irrigation systems. Results of analysis and
research prove that application of converter control is
zi fi si pi .
132
(5)
Assuming that the rotational speed n of the

asynchronous squirrel-cage motor in the pump unit is
within the range from 0.5nN to nN (where nN is the
motor's rated speed), the efficiencies of both the motor
and the frequency converter are constant. Therefore, the
overall efficiency of the pump unit zi can be formulated
as:
(6)
zi C pi .
speed, it is necessary to find a flow rate Q1

corresponding to Qzi [7]. On the basis of (10) and (13), a
flow rate Q1 dependence is obtained [7]:
Q1
(7)
C fi si .
Qm
.
Q zi
(8)
n zi
nN
i 1
zi
To determine the efficiency of ith pump pi the method

described in [7] has been employed. The efficiency of
the pump varies as the flow rate Qzi, head Hti and the
pump's rotational speed nzi change. The efficiency
=f(Q) for nominal rotational speed nN is characterized
by quadratic equations of varying factors depending on
the pump's flow rate:
2 a 2 Q b2 Q
2
for
Q 0.5Q N
for
Q 0.5Q N .
(14)
Q
1 zi
Q1
0, 09
(15)
Based on (5), (8), (14) and (15), a dependence of the

overall efficiency c of a pump set on efficiencies of the
individual pumps Qzi in the set is obtained:
where: m number of pump units operating in a pump set.
1 a1Q 2 b1Q c1
0, 09
pi 1
The dependence 1(Q1) is determined according to (9).

Employing the value of 1 and the original power
function of the pump [3,4], the pumps efficiency pi is
calculated:
The efficiency of the pump unit c is expressed in [7]:

c
H zad
A 2 4 H 0 B
Q zi2
H zad
2 B
Q zi2
c f Q z1 , Q z 2 ...Q zm .
(16)
Equation (16) is an implicit function of multiple

variables and determination of its maximum value is
complicated. It is possible, though, to analyze variability
of c for an assumed pump set and set heads at the
station's suction branch Hzad and a momentary
distribution Qm.
(9)
where: a1, b1, c1, a2, b2 factors of the quadratic equations

QN nominal flow rate of pump unit
4.2. Analyzing variability of a pump set's efficiency

as a function of its flow rate
The pump H=f(Q)-curve for the nominal rotational

speed nN is also described by means of a quadratic
equation [7]:
(10)
H ti H 0 AQ BQ 2 .
Operation of a set is simulated which comprises three

CR-32-6 pumps with characteristics of a pumping
station distribution corresponding to distributions in
irrigation systems e.g. Table 1.
Table 1. The system flow rate table
where: H0 static head at the station's suction branch

A, B factors of the quadratic equations
Hti [m]
Qm [m3/h]
Hti [m]
Qm [m3/h]
Hti [m]
Qm [m3/h]
The pump's operation at variable rotational speed

changes its flow rate and head. According to the theory
of probability, Qzi is assumed to vary in accordance with
[3,7]:
Q zi
n
zi .
Q1
nN
(11)
(12)
A relation between the pump's head and flow rate at

variable rotational speed is derived from (11) and (12):
Q
H ti
zi
H1
Q1
70
85
70
80
70
75
70
70
70
65
70
60
70
50
70
40
70
30
65
90
65
85
65
80
65
75
65
70
65
65
65
60
65
50
65
40
65
30
60
90
60
85
60
80
60
75
60
70
60
65
60
60
60
50
60
40
60
30
Analysis of variability of the pump set's efficiency is

based on the following assumptions:
pumping station's flow rate ranges from 30 to 90
m3/h [1,2,9],
the analysis is conducted in respect of three heads
at the station's suction branch: 60, 65 and 70 m
pumps flow rate ranges from 0 to 38 m3/h [8],
variation of the pumps' flow rate Qm = 0.1 m3/h,
Calculations for these cases are executed in an MS
Excel spreadsheet. Fig.2 and Fig.3 illustrates variability
of a pump set's efficiency c as dependent on the flow
rates of three pumps at constant overall flow rate of the
pump set Qm = 80 m3/h and constant head Hti = 65 m.
Variations of the pump set's efficiency were of a similar
nature in all the three cases introduced in the Table 1.
The results represent an area including the maximum.
The pump's head depends on the rotational speed

squared:
n
H ti
zi
H1
nN
70
90
(13)
To define the pump's efficiency for the working point

(Qzi, Hti) at a rotational speed lower than the rated
133
The most outlying point corresponds to the maximum

efficiency of the pump set cmax. Based on flow rates of
pump units read from the diagram, Qz1 and Qz2, and a
known overall flow rate of the pump set Qm, flow rate of
a third pump set can be determined:
Q z 3 Qm Q z1 Q z 2 .
According to the theory of probability (11), it is

established that pumps operating at identical flow rate
Qzi and at the same head at the suction branch Hti have
identical rotational speeds:
n zi n z1 n z 2 n z 3 .
(17)
(19)
5. Conclusion
Increased coefficient of a pump set's efficiency c

translates into more cost effective pumping. By
applying the proposed method of selecting pump
rotational speeds, a maximum efficiency of a pump set
can be achieved at set output parameters of a pump
station. It is therefore possible to apply extreme control
of a water pump station for irrigation systems.
Continuous operation at maximum practicable
efficiency of a pump set reduces electricity
consumption, operating costs, and greenhouse gas
emissions resulting from generation of electricity.
Fitting all pump units in a set with frequency converters
and PLC programmable controllers helps to select a
rotational speed for each individual pump unit, provides
for smooth startup and turn-off of the particular pump
units. This solution helps eliminate water hammers that
arise when pump sets are turned on directly at the mains
voltage. This significantly reduces fault rates of
pumping systems, minimizes repair and maintenance
costs of the system equipment.
cmax
Fig.2. Variability of a pump set's efficiency c - 3-D view
6. References
cmax
1. Figura R., Szychta L. Energy Consumption of Water

Pumping in Irrigation of Golf Courses. XI
International Conference TRANSCOMP 2007,
Zakopane.
2. Figura R., Szychta L. Cost Effective Water Pumping
in Extensive Sprinkling Plants. Pomps, Set Pomps
8/2008.
3. Jdral W. Impeller Pumps. PWN, Warszawa 2001.
4. Jdral W. Impeller Centrifugal Pumps. Publications
of the Technical University of Warsaw, 1996.
5. Luft M., Cio R.: Increase of Accuracy of
Measurement Signals Reading From Analog
Measuring Transducers, Silesian University of
Technology Scientific Publications 2005,Transport
vol. 59 (1691), Gliwice 2005.
6. Polish Standard PN-EN 127-03. PKN, 2004.
7. Szychta L. Principles of Selecting Control Systems
for Water Pumping Stations. Publications of the
Technical University of Radom 2006,
8. Technical documentation of CR/CRE Grundfos
pump, July 2006.
9. Rain Bird Deutschland GmbH System Analysis &
Documentation of Golf Course GaufeldenNebringen 2003.
10. Archive records of rainfall in central Poland
http://www.imgw.pl/wl/internet/zz/pogoda/tempsred
.html, 30.07.2007.
Fig.3. Variability of a pump set's efficiency c - top view
A careful analysis of the results leads to the conclusion

that a maximum efficiency of the pump set cmax obtains
for only one combination of pumps' flow rates. The
calculations show that the maximum efficiency of the
pump set cmax can be obtained when the pumps operate
with identical flow rate Qzi. Flow rates of the pumps can
be calculated:
Q
(18)
Qzi Qz1 Qz 2 ... Qzm m .
m
If the overall flow rate of the set diminishes, the number
of working pumps is reduced. This enables to maintain
the pump set's operation at an efficiency reaching its
maximum for set input parameters of the pumping
station. As a result, an extreme regulation system [5]
can be applied, maintaining efficiency of the pump set
at a maximum.
134

DIFFERENT SIMULATION MODELS IN EXEMPLARY EMPLOYMENT

FOR PUMP AGGREGATE PIPELINE SYSTEMS
Anna KOZIOROWSKA, Jacek BARTMAN
University of Rzeszw, Institute of Technology, Poland
Abstract: The paper presents the application of

differential equations in modelling and simulation tests
of water-supply network pumps drive system together
with the powered water-supply pipelines. Mathematical
models of PWM converter, squirrel-cage induction
motor, and pump together with a pipeline, allow to carry
out computer simulations of the whole drive system
consisting of the above-listed elements. Simulations of
the system of such complexity pose definitely more
difficult problem than simulation of each of the
elements performed separately. In the following,
mathematical models of individual elements and
selected, simulation tests of the system are presented.
Experience data
Real
system
Software
Reality
modelling
Simulation
Model
Fig. 1. Modelling and simulation scheme [1]
Keywords: modelling, simulation, pumps drive system.
Simulation models can have different form and it is

possible to present them as:
algebraic and function equations;
characteristic package;
neuron models;
fuzzy models;
etc.
1. Types of simulation models

Modern science uses research for pulling down cost of
project design and exploitation of real system.
Simulations allow to get image of system behaviour in
different situations - often extreme - however, adequacy
of gotten results is for reality derivative accuracy of
accepted model.
So, modelling has creation of object for task, which
owns chosen attributes of originals. The basic purpose is
to present most important devoted from the point of
view of led analysis - attribute of original object and
omitting the less important feature. Determination of its
purpose is an important element of modelling negligible features can be very important from the point
of view of another analysis in a different context. It is
possible to attempt building model taking into
consideration all features of the original - then we
would get universal model. Unfortunately, that is not
always possible, but most often completely purposeless.
The recent development of the information technology
has caused a situation where a major group of models
undergo computer simulations (so-called simulation
models). They allow to simulate in an easy way
extreme, sometimes unacceptable conditions of work of
systems.
Real object
Definition of essentials
and inessential for
analyze object features
Separation
compact
subobjects
Ideological schema
Model
description
Analyses of sub-objects
determination of type
of model
Simulation model
Fig. 2. Periods of development of simulation model
Selection of models form to employ it for definite

description of system or phenomenon is the
135
consequence of our knowledge about the object.

Models written down in the form of algebraic and
function equations are characterised by grate accuracy
and they are universal, unfortunately, they require
acquaintance of parameter of object, but accuracy of
gotten solution depends not only on description but
from equation solving manner. Packages of
characteristics describe concrete object, they are not
universal and they are gotten most often as an
experiment.
Neuron models are built as a result of artificial teaching
of neuron network and they take advantage them for
generalizing gotten knowledge ability. There important
advantage is lack of parameters, however, they are not
universal
they describe
concrete
object
(phenomenon).
Fuzzy models have character descriptivelyexpert.
They do not require acquaintance of parameter of object
directly but experimental acquaintance of its behaviour
is indispensable.
It is possible to divide process of modelling into several
periods (Fig. 2).
Firstly, the ideological scheme of modelled object
should be constructed, provided that all the particles
have been previously distinguished. It will be necessary
to carry analysis of the problem what kind of simulation
model will be proper for description of chosen element.
In final period it is necessary to describe simulation
models of individual elements and define manner of
exchange of data among them.
a)
b)
Fig. 3. Water-supply system: a) part of real water-supply

system, b) scheme of ideological system
3. Mathematical model of PWM rectifier

The rectifier consists of power circuit, three-phase diode
rectifier, DC circuit and three-phase voltage inverter.
[4, 7]. There are accepted simplified assumption in the
model:
transistor starts up immediately;
transistor is replaced by connector which resistance
during operation corresponds to resistance of
conductive transistor;
switched off transistor does not carry electric
current.
2. Idea of water-supply system model

Components of classical water-supply systems, in view
of fulfilled functions, have differentiated technological
structure, proper technical parameters and hydraulic
characteristics with their local conditionality of modes
of work subsequent. All of water-supply system
elements match different behaviour, dependent on fulfil
function and moment extortions in the form of fate
conscription of water.
It is possible to differ in construction of mathematical
model of water-supply system match from the point of
view of its different hydraulic character, two parts of
warehouse elements: static and dynamic. Static part of
model includes structure of network with valves and
pumps. Dynamic part of model presents reaction of
system on variable conscription of water in time.
The paper presents the model of multipump system with
one pump adjusted and affixed pumps unadjusted [2].
The system consists of pump system includes induction
motor and pump, frequency converter with control
system and pipeline (Fig. 3). Complexity of system and
miscellaneous character of its individual component
determines utilization of different method of modelling
of different components of system:
induction motor has been described with the
aid of match of differential equations,
frequency convertor with the aid of algebraic
equations,
pumps are described by characteristics
pipeline by neuron model.
Fig. 4. Waveforms of switching functions sA(t) for sinusoidal

modulating wave SPWM
Receiver phase-to-phase voltage equations have a form:
1 1 0 s A (t)
u AB (t)
u (t) U 0 1 1 s (t) ,
d
B
BC
1 0 1 s C (t)
u CA (t)
(1)
where U d 2U ac .
These switching functions were used to control
transistors in simulation model of PWM rectifier.
136
4. Induction motor model
5. Pump characteristic
Application of A-models to description of induction

motor is used in water-supply network pumps drive
system. Finally, the following system of equations was
adopted for simulation tests of pumps drive system.
Fundamental equations of the motor:
Pump operation basis on differential pressure

generation. The characteristics of pump are shown at
Fig. 5. The diagram shows set of curves for constant
speed of motor-pump unit. Speed of pump is changing
during the operation and pump pass from one curve to
the other, keeping the forcing pressure constant (in
assumed limits of pressure changes).
dI
1
A( U RI ) ,
dt K
(2)
where:
K = r + s + m;
m inverse of magnetising inductance;
s, r inverse of stator and rotor inductances;
s r
0
0
s r m
s r ;
0
s r m
0
A
s r
0
r s m
0
s r
r
s
m
U u sa
u sb
Fig. 5. An example set of pressure p vs. flow Q

characteristics of a pump
0 0 T ;
There is assumed reduced model of pipeline to load the

unit induction motor pump. There are disregard
effects of cavitations and water propagation in pipeline.
usa = Um sin(2ft + );
usb = Um sin(2f t + + 120o);
I i sa i sb ira i rb T
6. The pipeline model
1
1
3 2rsa rsb 3 rsc rsb
1
1
2rsb rsc
R rsc rsa
3
3
0
0
0
0
A neuron model of pipeline is presented in the paper. It

was indispensable to make measurements in real object
and using them as teaching data for neuron network [6].
The multilayered neutral network learned with back
propagation rule has been used in the modelling
process. Neutral network was built from 3 neurons in
input layer, 4 neurons in implicit layer and 1 neuron in
output layer (Fig. 6).
0 0 ;
rr 0
0 rr
0
rsa, rsb, rsc resistance values of stator phases;

rr rotor resistance value;
isa ira ira
m
r
;
i sb irb irb

r
m
0
0
0
0

2
.

3
3

2
3
3
Fig. 6. Scheme of neuron network taken advantage as a model

of pipeline
7. Algorithm of calculations
The equation of motion can be written down as:

d p
Te T ,
dt
J
Te
3 pira isb irb isa

,
There is necessary to formulate a mathematical model

of system and to make computer simulations for
interesting dynamical states of system to make an
analysis of water pump drive system and pipeline
system operation.
The system consists of few elements and there is
necessary to formulate an algorithm of calculations. The
algorithm establishes sequence of carrying out
individual operations and decides about connecting and
disconnecting an additional motor, fed from power
network [1, 7].
(3)
(4)
where:
T() mechanical moment;
p number of pole pairs;
J moment of inertia;
Te electromagnetic moment.
137
instantaneous current values to values admissible in

given power supply system.
8. Simulation results
The models presented before were used to simulate

work of water-supply system. There were observed
parameters important from the point of view of whole
exploitation system:
a) water pressure - important parameter for water
receivers (pressure stabilization is required and
such task is required from system);
b) rotary speed of pump system;
c) electric current taken by asynchronous motor with
adjusted speed - parameter important from the
point of view of exploitation of energy system.
9. Conclusions
There are generally used the circuit models in

mathematical description of drive systems, which
consist of induction motor, frequency converter and
control system.
They enable description with assistance of operational
transmittance (linear systems) or differential equation. It
makes the time of simulation shorter giving grate
accuracy in engineer practice [3, 5].
Mathematical models of frequency converter, induction
motor, pump and pipeline let us carry simulation tests of
all drive systems consisting of the aforementioned
elements [2, 3, 7]. Simulation of such complicated
system presents harder problem than running the
simulation of each element separately.
The differential equations representing the whole
system are those of the stiff type, therefore equations
describing individual elements were solved using
different integration steps in order to accelerate the
calculation process. The obtained results confirm that
such approach is acceptable. Computer simulations are
modern tool which let us analyse varied systems
operation. There is important that there is possible to
make lots of tests without system damage. There can be
analysed critical states of system and system work with
damaged elements. It is grate importance in power
electronic control system and it lets decrease the tests
costs.
Fig. 7. Results of simulation test concerning - pressure
10. References
1.
2.
3.
Fig. 8. Results of simulation test motor speed
4.
5.
6.
7.
8.
Fig. 9. Results of simulation test motor current
Received results show, that model behaves in

researched state of work according to expectations and
to adequate manner for behaviour of real system. From
the operational point of view it is important to keep
quasi-constant delivery pressure and to limit
138
Gutenbaun J. Mathematical modeling of systems

(in Polish). Exit, Warsaw, 2003.
www.hydro-vacuum.com.pl, February 2009.
Kulesza A., Koodziej H. Systems of pumps control
(in Polish). Pumps and pump stations 2/2001.
p. 4751.
Sobczyski D. Highspeed drive system with
induction motor fed from PWM rectifier (in
Polish). Rzeszow, 2004.
Tunia H., Barlik R. Theory of converters (in
Polish). Warsaw, 2003.
Duch
W.,
Korbicz
J.,
Rutkowski
L.,
Tadeusiewicz R. Neural Networks, vol. 6. Exit,
Warsaw, 2000.
Koziorowska A, Bartman J., Buczek K. Forcing
pressure simulations in water supply system during
water requirement changing. Procedings of 2nd
Conference on Intelligent Knowledge Systems,
Istambul, Turkey, 2005. p. 97-101.
Kulbik M. Computer Simulation and On-Site
Examination of Municipal Water Supply System: a
Monograph (in Polish). Gdansk, 2004.

APPLICATION OF SIMULATION TOOLS FOR

A FERTILIZER GRANULATION CIRCUIT
Gediminas VALIULIS*,**, Rimvydas SIMUTIS*

**
Electrical Engineering Department, iauliai University, Lithuania
simulator available. The increasing capabilities of

computer hardware and software ensure the
incorporation of complex knowledge (models)
represented by differential and algebraic equations,
measured process data, process experts information,
etc. But to be of use for the day-to-day work of the
engineer these models have to become more user
friendly, than the one that the scientist is dealing with
[2]. The simulator developed according to the extended
modelling approach aims at giving a better insight in
granulation process features and can therefore be useful
in control and staff training.
Abstract: The paper presents the extended modelling

and simulation approach to granulation in a continuous
drum granulator-dryer circuit for fertilizer production.
The model is based on granule coating phenomena and
involves the basic process parameters, features and
equipment. For a better model performance, information
from the process sensors and the experience of process
experts has been applied as the extension to the main
model. GrowSim simulation tool has been used to
simulate some processes frequent in the real plant and
test the granulation model. The simulator proposed is
intended to ensure a better process understanding, staff
training, and evaluation of control strategies.
2. Main process details and model development

Keywords: granulation, modelling, simulation, control.
2.1. Brief circuit description
1. Introduction
Drum granulation is a particle size enlargement process
often obtained by spraying a liquid binder or slurry onto
fine particles as they are agitated in a rotary drum [8]. A
commercial continuous granulation circuit for
granulated diammonium phosphate fertilizer (formed by
the reaction of phosphoric acid and ammonia)
production consists of: a pipe reactor, spray nozzle
system, drum granulator-dryer, granule classifier
(screens), crusher and nuclei feed system (Fig. 1).
Many continuous granulation plants operate well below

design capacity, suffering from high recycle rates and
even periodic instabilities [8]. The main reasons are
related to raw material properties, process equipment
and control problems. The process control still depends
on the experience and skills of process operators,
namely experts.
Diagnostic systems show the potential to apply systems
engineering approaches to complex operational problems
such that operators are well informed, are able to quickly
diagnose abnormal conditions, test quickly possible
solutions via detailed simulations and then proceed to
apply corrective actions [5]. Mathematical modelling and
simulation can improve dealing with materials, energy
and information through better understanding of
underlying mechanisms in the system. The trade-off
between the necessary accuracy and resulting
complexity becomes increasingly important when the
nonlinear and multivariable behaviour must be taken
into account [3].
A numb of scientific works deal with the modelling of
granulation mechanisms. However, only a few are
focused on practical implementation with a user friendly
Fig. 1. Typical drum granulation-drying circuit utilized in the

diammonium phosphate (DAP) production industry
139
Particle circulation is achieved by the action of the

rotating drum and lifters. Granules are cycled many
times through the spray zone and the liquid layer
attached is pre-dried before the particles return to the
spray zone again. After some residence time granules
leave the drum and the classification procedure is
performed. They are split into three fractions: small,
marketable and large. Small, crushed large and even
marketable granules are fed back to the drum as external
nuclei.
A more detailed description of granulation circuit and
granulation mechanisms can be found in [6].
the granulation process model development can be found

in [7].
2.3. Modelling assumptions
Obviously an extremely rigorous model that includes
every phenomenon down to microscopic detail would be
so complex that it would take a long time to develop and
might be impractical to solve, even on the latest
supercomputers [4]. An engineering compromise
between a rigorous description and getting an answer
that is good enough is always required [8].
Basic modelling assumptions:
granule shape is spherical;
each granule in the granulation circuit is analyzed;
stochastic nature of the process is estimated;
preferred growth is by layering;
granule agglomeration is an unacceptable mode of
operation;
growth rate is a function of initial granule size, slurry
flow rate, temperature inside the granulator, granule
position in the drum, number of particles in the
granule bed;
granule moisture evaporation inside granulatordryer;
mechanical attrition of granules inside the granulatordrier, defined by attrition function;
presumable nucleation (formation of new seeds)
during slurry spraying;
external classification of granules into three fractions
(undersize, marketable and oversize), defined by
classification function;
external crushing of oversize granules, defined by
grinding function;
residence and transportation delays in the plant;
internal and external feed of seeds (nuclei for new
granules).
2.2. Modelling Basis

The main requirements for granulated fertilizer are
related to granule size, shape and chemical composition,
moisture content, mechanical strength, etc. A number of
different process parameters and material properties, as
well as control actions, influence the granulation process
(Fig. 2). This creates a strong nonlinear system exposed
on different time scales that is difficult to predict just by
intuition. Detailed modelling is needed in order to
describe the coexisting processes and the transients of
the process state variables crrtly [2].
2.4. Main process variables

Table 1 presents the main input (manipulative) and
output variable set available to the operator for process
state prediction and control.
Fig. 2. Representative interfaces between process engineering
and a typical granulation plant [1]
Table 1. Main process input and output variables
The model presented is essentially based on

fundamental conservation principles and it partially
takes into account equipment properties and the
stochastic nature of the process. The mechanistic part
incorporates the understanding of physics and
underlying mechanisms (e.g. mass and energy balances,
growth kinetics). The empirical part uses raw and/or
filtered process sensors data, their storage, retrieval and
parameter identification techniques in addition to the
mechanistic (white box) model.
The expert component involves the process experts
recommendations, which are of great value due to the
lack of other knowledge mentioned above. The survey of
Input variables
Output variables
Slurry flow rate
Granule mass throughput from

granulator
Slurry viscosity, density

and moisture content
Granule
size
distribution
parameters
Recycle feed flow rate
Granule moisture content
Natural gas flow rate
Granule mechanical properties
Slurry N:P ratio
Granule N:P ratio
NH3(gas):NH3(liquid) ratio
However, some process variables related to unique

material and equipment properties can not be evaluated
and controlled directly and instantly. In such a situation,
the process model can provide information about
140
important process states, such as recycle size flow rate

and distribution, drum system jamming factor, granule
moisture content, size evolution of single granule inside
the granulator-dryer. This information can help to
predict future process states and prevent abnormal
situations, which can initiate process stoppage and loss
of productivity.
disturbances, faults, malfunctions. The manual or

automatic process control modes are available. The
operator can take a challenge to manage the process
manually or leave all or part of the job to PI or Fuzzy
controller. The simulation time can be easily adjusted if
the computing resources are sufficient. Simulation can
be paused at any moment, to make it possible to weigh
ones decisions. The main process parameters are
logged and can be observed during the simulation or
even later. The simulator is provided with routine to
compare the simulated and real measured process data.
Some experiments have been carried out for model and
simulator validation purposes.
3. Simulator
A new GrowSim simulation tool for drum granulationdrying process is under development. The simulator is
based on the extended process model and is intended to
be used for advanced process control and knowledge
acquisition. Using powerful MATLAB software the tool
simulates the physical process combining mass and heat
transfer, including major units such as a drum-granulator
dryer, classifier, crusher, transportation system. The
graphical user interface (Fig. 3) has been built to mimic
the process control environment available to the process
operator in the real plant, with important additional
information provided.
The simulation environment is composed of sections
where the operator can change process parameters,
observe the current or past states, get some advice on
how the process should be controlled by the skilled
operator. The simulator allows to introduce common
4. Results
The most common description of particles in means of
their size is given by particle size distribution (PSD).
Strong evidence of normal granule size distribution can
be observed in Fig. 6, where the measured initial nuclei
and granules out of granulator distribution data is fitted
to the cumulative normal distribution curve with
parameters defined.
Some simulation scenarios have been carried out to
check the model and simulator using the same data as in
Fig. 4. The results obtained are in fair accordance.
Fig. 3. Graphical user interface of GrowSim simulator
141
The decrease of slurry flow (the recycle flow rate was

kept constant) initiates the reduction of granule mass
flow from the granulator as well as the decrease of
median granule size. The latter is proved by the decrease
of oversize and increase of undersize granule fraction. In
this situation the number of marketable granules is
almost the same. Compared to the measured data, these
results are similar.
1
Measured distribution (initial seeds)
Measured distribution (granules from granulator)
Fitted Normal distribution (=2.95 mm; =0.79 mm)
Fitted Normal distribution (=2.75 mm; =0.67 mm)
0.9
0.8
0.7
F(x)
0.6
0.5
0.4
0.3
0.2
6. Conclusions
0.1
0
0
A DAP fertilizer granulation circuit has been modelled

using basic physical principles such as growth, mass and
heat transfer. The model has been implemented and
simulation executed using GrowSim simulator in the
MATLAB environment.
Some model validation procedures have been carried
out and the results obtained are in fair accordance with
the plant measured data.
The proposed model shows good potential for
representing the behaviour of the granulation plant and
hence can be advised for dynamic simulation leading to
improved granulation circuit control and operator
training.
Granule Diameter, mm
Fig. 4. Normal cumulative size distributions of initial seed

flow and granule flow out of granulator
Granules in, Empirical cdf
Percentage
Simulated
Measured
0.8
0.6
0.4
0.2
0
0
0.4
Percentage
Particle Size, mm
Histogram
0.3
0.2
7. Acknowledgment
0.1
0
0
0.5
1.5
2.5
3.5
4.5
This study is supported by the doctoral grant from the

Lithuanian State Science and Studies Foundation.
Particle Size, mm
Fig. 5. Initial seed cumulative size distribution and histogram
8. References
Granules out, Empirical cdf

Percentage
1
0.8
1. Ennis B.J. Agglomeration and Size Enlargement:

Session summary paper, Powder Technology, vol.
88, 1996. p. 203-225.
2. Ihlow M., Drechsler J., Peglow M., et al. A New
Comprehensive Model and Simulation Package for
Fluidized Bed Spray Granulation Processes.
Chemical Engineering and Technology, vol. 27,
2004. p. 1139-1143.
3. Juso E.K. Modelling and Simulation with Intelligent
Methods. Sim-Serv, 2004, 17 p.
4. Luyben W.L. Process Modeling, Simulation, and
Control For Chemical Engineers. Mc Graw-Hill,
1999. 724 p.
5. Salmon A.D., Hounslow M.J., Seville J.P.K.
Handbook of Powder Technology, vol. 11.
Granulation. Elsevier, 2007. 1390 p.
6. Valiulis G., Simutis R. Analysis of Control Methods
for Granulation Process. Proceedings of the 1st
International Conference on Electrical and Control
Technologies (ECT-2006), Kaunas, 2006. p. 496600.
7. Valiulis G., Simutis R. An Extended Modelling
Approach to the Drum Granulation Processes.
Proceedings of the 3rd International Conference on
Electrical and Control Technologies (ECT-2008),
Kaunas, 2008. p. 122-127.
8. Wang F.Y, Cameron I.T. Review and future
directions in the modelling and control of continuous
drum granulation. Powder Technology vol. 124,
2002. p. 238253.
0.6
0.4
0.2
0
0
Particle Size, mm
Histogram
0.4
0.3
0.2
0.1
0
0
Particle Size, mm
Fig. 6. Granules out of granulator cumulative size distribution

and histogram
Another simulation experiment has been performed on
decreasing slurry flow rate by 2/3 (Fig. 7).
SlurryFlow
Granule Mass
0.9
d0.5-2, %
0.8
Normalized Parameters
Flow,out
d2-4, %
d4-8, %
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
10
15
20
25
30
35
40
45
50
Time, min
Fig. 7. Impact of the change of slurry flow rate (simulation)
142

INNOVATIVE MOISTURE CONTROL METHOD FOR CONCRETE MIXING

SYSTEMS
Gintautas LENGVINAS, Vytautas DEKSNYS
Department of Electronics and Measurement Systems, Kaunas University of Technology, Lithuania
Abstract: Moisture control in concrete mixing systems

is usually done using microwave moisture sensors.
These sensors must be adapted for every type of mixture
separately. We propose a new material type adaptive
real-time moisture control method, which is based on
transfer function time-domain analysis, used in
conjunction with artificial neural network (ANN).
There are two sources of water in concrete: wet

aggregates and water added directly to the mix.
Today, the moisture of the mix is often controlled using
microwave moisture sensors [1]. These sensors include
an HF signal generator and resonator. Resonant
frequency and resonator quality depends on dielectric
permittivity and loss tangent of the surrounding
substance. Water ratio changes these parameters for the
given mixture. However, dielectric properties of
different materials change in different ways.
Consequently, the sensor parameter sets have to be
calibrated for each type of material, in order to reliably
monitor the dielectric properties of that material.
Furthermore, this calibration is a very complicated and
time consuming operation, that must be performed for
every type of material or mix used in the manufacturing
process. This is the main disadvantage of microwave
moisture meters.
Keywords: ANN, time domain, transfer function, real

time, moisture control, aggregates, concrete, hollow
core slabs.
1.
Introduction
Concrete is a construction material composed of cement

(usually Portland cement) as well as other cementitious
materials such as fly ash and slag cement, aggregate
(generally a coarse aggregate such as gravel, limestone,
or granite, and a fine-grained aggregate such as sand),
water, and chemical admixtures. Concrete solidifies and
hardens after mixing with water due to a chemical
process known as hydration. Water reacts with cement,
bonding the other components together, eventually
creating a stone-like material. Concrete is being used to
make pavements, foundations, motorways and bridges,
parking structures, brick walls and footings for gates,
fences and poles.
Aggregates, cement and water must be mixed to get a
uniform mixture, suitable for moulding.
Workability is the ability to fill the form properly with a
fresh concrete mix, with the desired work and without
reducing the concrete's quality. Workability depends on
water, aggregates (shape and size distribution),
cementitious ratio and age. The quality of the product
depends heavily on concrete workability. Product made
of too wet concrete collapses immediately during the
moulding process. On the other hand, if concrete is too
dry, the resulting product will be porous and difficult to
mould. The main factor contributing to poor workability
of concrete is unstable water/cement ratio. If the amount
of cement and aggregates in the mix is stable, the only
influence on final concrete workability is water content.
2.
Moisture measurement and control in automatic

concrete mixing systems
In practice, moisture sensors are not calibrated for the

particular recipe of concrete. This recipe indicates the
relative moisture of the mixture and the target value
must also be set to a relative value. When using a
moisture sensor, the concrete mixing algorithm consists
of five phases. During the first phase (Fig 1.),
aggregates, cement and a coarse amount of water
(nearly 90% of the amount required) are added to the
mix. During the second phase, the first mixing operation
is performed. Depending on the mixer type, it takes 1030 seconds. The goal of the first phase is to get a
homogenous concrete mix. The third phase is the
water/cement ratio correction phase. Water is added
according to the moisture sensor readings. When target
moisture value is reached, final mixing phase starts. It
can take 15 150 seconds. This time depends on mixer
type and water cement (w/c) ratio in the mix. If it is low,
mixing time is long, and if w/c ratio is high, final mix
time goes shorter. The fifth phase, is the mixeremptying phase.
143
Typically, relative moisture of the mix in the mixer

changes as shown in Fig. 1.
7
Relative moisture
dielectric in our transmission line was used testing

material. The gap between ground plane and signal
electrodes was filled with it [2].
Two popular materials (stone 5-11 and sand 0-2) were
used for the experiments. Crushed stone is the roughest
material it is useful for moisture content control,
because in coarsegrained materials maximum available
moisture is too small to influence the overall mix
moisture. Maximum available moisture of crushed stone
is 1-2 %. Sand 0-2mm is the most fine-grained material
used in concrete mixtures. It can accumulate up to 15-18
% of water.
Target moisture
value
5
4
3
3
Mixing phase
1
0
0
20
40
60
80
t,s
100
120
140
160
Table 1 represents mixed materials and their proportions

used for our experiments. The first and the last materials
are the compounds of pure sand and pure crushed stone
respectively. The other materials are mixes of these two
using different ratios.
180
Fig. 1. Moisture changes during mixing process
Fig. 2 represents real data collected in the local facility

producing hollow core slabs.
120
90
60
30
Material
number
1
2
3
4
5
Added water, l
Mixing time, s
150
Table 1. Materials used in experiments
6 7 8 9 10 11 12 13 14 15 16
Experiment number
Fig. 2. Mixing times and water amount, added to the mixer
As shown in Fig. 2, the mixing time depends on the

water amount, which is added to reach the target
moisture. Fig. 1 and Fig. 2 show that most of the mixing
time is spent on water/cement ratio correction.
So, in order to reduce the mixing time and get the best
workability of concrete, a new, material and mixture
independent moisture measurement and control method
is required.
3.
Material type independent moisture control

method
We present a new, material type independent moisture

measurement method. The structure of our moisture
measurement system with artificial neural network for
material type detection is shown in Fig. 3.
Short Pulse
Generator
Transmission
line under test
Signal
reading
circuit
100
75
50
25
0
Crushed stone 511 mm,

%
0
25
50
75
100
Water was added to the materials, in order to get seven

different moisture ratios (0%, 1.8%, 3.7%, 5.4%, 7.1%,
8.6% and 10.2%) for measurements.
Having 5 material types and 7 moisture values for each
type, 35 measurements were made in one experiment
session. To get more reliable data we repeated
experiment a second time. So, a total of 70 curves were
collected. The power spectrum was taken as an
informative parameter [3].
To find the relationship between material type and
power spectrum curve the artificial neural network
technology was used.
An artificial neural network (ANN) is a collection of
simple interconnected analogue signal processors. The
purpose of an ANN is to provide a mathematical
structure that can be trained to map a set of inputs to a
corresponding set of outputs. Fig. 4 illustrates an ANN
used in this.
This ANN has three layers: the input, hidden and output
layers. Each layer consists of nodes or neurons. Each
node has sigmoid activation function, associated with it.
Each interconnection between the nodes has a weight
associated with it. The nodes in the hidden layers and
the sum of output layers are the weighted inputs from
sending nodes and applying this net input to the
activation function. Applying the inputs and computing
the output from the various nodes, activations and
interconnection weights determine the output of the
network.
Sand
0-2 mm, %
ANN
Fig. 3. Structure of the moisture sensor with ANN
A coplanar strip-line was used as a sensor. It was tuned

to dielectric permittivity of = 6 and electric line length
of l = , when the operating frequency is 1 GHz. As
144
As it is shown in
Fig. 5, the possibility to identify the type of material
was found. Most of the samples were recognized. Only
a few samples were assigned to wrong material type. It
shows the possibility to use this method in real time
moisture control systems.
5.
Using transfer function analysis and artificial neural

network technology for material type detection makes
possible material and mixture independent moisture
measurement.
Single sensor, material independent technology for real
time moisture control systems enables fast integration of
moisture meters without individual calibration for every
material or mixture used in technological process of
concrete. It enables to reduce the mixing time and
enhance a quality compare to classical moisture control
systems.
Fig. 4. Artificial Neural Network
Ten power spectrum values of different frequencies

were taken as input data. The type of material was the
learning target (Table 2). Target constants in the ANN
output value range of -1 to +1 were acquired for
different materials.
Neuron number was chosen accordingly to input layer
size.
Table 2. Output values for ANN
Input data
Material No. 1 (7 different moisture values)
Learning
result
-1
-0.5
0.5
Material N. 5 (7 different moisture values)
4.
6. References
1.
2.
Results
3.
We have learned network using a collection of data with

7 different moisture points for every material. The
generated network was tested using the same data used
in the learning process, and the data collected during
independent experiment with the same materials.
Number of sample is set on x-axis. 1-7 material No. 1,
8-14 material No. 2, 15-21 material No. 3, 22-28
material No. 4, 29-35 material No. 5. On y-axis we have
set type of material (M1 M5).
4.
M5
Material
M4
M3
M2
M1
14
21
28
Conclusions
35
Sample Nr
Fig. 5. ANN Tested with independent data
145
Okumava S. Microwave Technology for moisture

measurement. Subsurface Sensing Technologies
and Applications, Vol. 1., Springer Netherlands,
2000, No. 2. p. 208-210.
Deksnys V., Lengvinas G.. Humidity Evaluation
Method in Building Materials Using Time Domain
Reflectometry. (In Lithuanian), Matavimai.
Kaunas, Technologija, 2004, Nr.2(30). p. 38-40.
Lengvinas G., Deksnys V., Ragauskas A.
Technology For Aggregates Classification in
Concrete Compounds. Electronics and Electrical
Engineering, Kaunas, Technologija, 2006, No.
2(90). p. 61-65.
Daschner F., Knochel R. Microwave moisture
sensor with nonlinear processing using artificial
neural networks. Silicon 01 sensors for industry.
Conference Rosemont. - Illinois, USA, 5-7
November, 2001.

THE INDEPENDENT FUZZY CONTROL WITH CORRECTION OF TWO

DIMENSIONAL PROCESS
Gediminas LIAUIUS, Vytautas KAMINSKAS
combined air outlet channel at the end. A fan (7) is at

the bottom of each air tank, which pumps the air from
the combined air channel and makes an air pressure in
the channel. The special corks of channels are forced to
move depending on the sizes of generated air pressures.
The air tanks at each channel's bottom are necessary for
the stabilization of air pressure oscillation. Additionally,
air tanks connection with its designated fan provides a
nonlinearity feature to control system. There are four
special air feeding vents in the combined air channel by
which may be adjusted the size of air inlet to air
channels. The air channels dependence of each other is
decided by the number of opened air feeding vents since
fans are pumping the air from the combined air channel.
The more air feeding vents are closed the greater
dependence between air channels exist since the lack of
air necessary for both fans. Each air channel also has an
ultrasound distance sensor (5) which is operating on the
ultrasound echo principle. They operate as the distance
between the sensors and the air channel corks measuring
equipments. The control target of the process is the
maintenance of corks in desirable heights of air
channels.
The input signals of the plant are the air fans voltage
values, where 0V means that the fan is switched off,
10V - the fan is running at the maximum. They are
controlled by direct current signals. The intermediate
values of voltage effects the power of fans
disproportionately.
The output signals of ultrasound distance sensors are
also in 0-10V range where 0V corresponds to 0.2 metres
range between the sensor and the cork, 10V - the
distance of 1.2 metres. Since the moment the fan is
switched off, the cork is about 1.1 metres far from its
designated sensor, but the sensor measures that the
distance is about 0.9 metres.
The input/output controller BK9000 is used to read the
data from ultrasound distance sensors and to send the
fans data to the plant. The BK9000 controller is
configured and controlled by TwinCat software. The
communication between the input/output controller and
the software is reached via computers network and is
Abstract: In this paper is analyzed the independent

fuzzy control of two input and two output process with
additional control signals' correction. The control signal
is additionally corrected by estimating other channel's
control error.
The experimental analysis was performed to air pressure
plant which is a nonlinear and dynamic process. The
results of fuzzy control with correction are compared
with PID control results.
Keywords: Fuzzy control, PID control, two input two
output process, air pressure plant, nonlinear, dynamic.
1. Introduction
Lots of operational control systems in industry are
multidimensional processes. Such control systems are
more complex comparing with single dimensional
processes. The multidimensional processes commonly
are dynamic, nonlinear systems with dominance of
interactions between input and output variables. The
routine of multidimensional processes is that there is not
enough information for designing the appropriate
preciseness mathematical models.
One of the most powerful techniques for nonmathematical model based, dynamic and nonlinear
system control is fuzzy control. The synthesis and tuning
process of fuzzy controllers are mostly time consuming,
but commonly guarantee desirable control results.
The experimental analysis was performed to the
multidimensional process called an air pressure plant,
which is a nonlinear and dynamic system.
2. The structure of control system
The scheme of physical control components of the plant
is shown in Fig. 1. The plant consists of four main
elements: the combined air channel (1), three air tanks
(2), three air channels (3) and the combined air outlet
channel (4). The combined air channel is connected with
three air tanks which are connected with air channels,
respectively. The air channels are coupled with the
146
used the TCP/IP protocol to interchange data. The

components module communicates with the BK9000
controller to directly control the air pressure plant. In
this module are implemented the fans controllers which
control the plant with analogue signals.
tune the controller gains by them are necessary to

possess the information (characteristics) about the
transient process which is ambiguous in air pressure
plant. Accordingly, the gains were tuned by using a
random search when the standard deviation of channels
control error (4) was as tuning criterion. The same tuned
PID controllers were synthesized to both air channels of
air pressure plant. The PID controllers were designed to
control just the running intensity of appropriate fan not
considering to the possible interactions between
different air channels. This type of implementation was
called as an independent PID control to the plant.
Fig. 1. The scheme of physical control components of air

pressure plant
Since the experimental analysis was performed to two

air channels of the plant, therefore the process was
reduced to a two dimensional system. The control
scheme of air pressure plant as a two dimensional
process is demonstrated in Fig. 2, where x i (t ), u i (t ),
y i (t ) when i 1,2 are reference, control and output

signals at t time moment of the system, respectively,
W11 , W22 , W12 , W21 - transfer functions of the two
Fig. 3. The scheme of independent PID control of two

dimensional process
The results of PID control (Fig. 8 and Fig. 10)

acknowledged that the process was a nonlinear, dynamic
system where existed interactions between different air
channels.
As control results showed, the process dynamics had
depended on the corks moving direction and moving
proportion (the change size of reference signal). The
greater corks moving proportion the greater dynamics in
channels acted. A similar situation with the moving
direction but the dynamics was more significant than the
cork was moving down. Especially, there is a huge
dynamical features difference between two situations:
the cork was moving up by more than 15 centimetres
and the cork was moving down by more than 15
centimetres. The results of PID control were suitable
when the corks were moving up, but were not when
moving down, especially when the corks were moving
down by more than 15 centimetres. At this situation, the
results were extremely unsatisfying since first of all the
corks were moving down till reached the channels
bottoms and after that achieved its reference signals.
Such a poor PID control results with static tuned PID
controllers for all sets of reference signals meant the
necessity of the synthesis of adaptive PID control or
applying another control methods to air pressure plant.
The detailed control results of PID are demonstrated in
the section of experimental results.
dimensional process.
Fig. 2. The control scheme of air pressure plant as a two

dimensional process
The air pressure plant was controlled with two different

control types: PID and fuzzy. The principles of the
synthesis of both control types are analyzed in next
sections.
3. PID control
PID control is a useful as an initial control technique to
almost everyone control system since the synthesis and
the tuning process of it is commonly straightforward.
When the studied control system is a complex one, this
methodology gives an ideal opportunity to reveal the
main control challenges to researcher. The abovementioned principle of control systems analysis was
used to identify the main control problems of air
pressure plant.
The scheme of the synthesis of PID control of two
dimensional process is demonstrated in Fig. 3.
PID controllers commonly are tuned using ZieglerNichols, Kappa-Tau and Takahashi methods. In order to
4. Fuzzy control
The results of static tuned PID control of two
dimensional process showed that this control method
was not suitable for the plant. Consequently, we needed
another control method which could control the plant in
suitable manner. One of possible control techniques to
air pressure plant is a fuzzy control.
147
Fuzzy control provides a formal methodology for

representing, manipulation and implementing a humans
heuristic knowledge about how to control a system [2].
The general scheme of fuzzy control architecture is
demonstrated in Fig. 4 where the fuzzy controller is
embedded in a closed-loop control system. The fuzzy
controller consists of four main components: the
fuzzification and defuzzification interfaces, the rule-base
and the inference mechanism. The rule-base holds the
knowledge, in the form of a set of rules of how best to
control the system [2]. The inference mechanism
evaluates which control rules are relevant (active) at the
current time and then decides what the control signal to
the plant should be. The fuzzification interface simply
modifies the inputs so that they can be interpreted and
compared to the rules in the rule-base and the
defuzzification interface converts the conclusions
reached by the interface mechanism into the control
signal to the plant [2].
Fig. 5. The scheme of independent fuzzy control with

correction of two dimensional process
The formula of moving average is:

f (ei ,t , N ) ei ,t
1
N
i, j
j t N 1
(2)
where: N - a sample size of moving average, ei ,t - an N

sized vector of control errors of ith air channel until
time moment t, i 1,2 - an appropriate air channel.
If the cork was below its reference signal depending on
the control error of another channels cork position the
primarily generated control signal must be strengthen,
otherwise if the cork was above weaken. This control
signals correction module having information about the
control error of each air channel produced two control
signals correction values for both air channels.
Fig. 4. The scheme of fuzzy control architecture [2]
The fuzzy controller may be interpreted as an artificial

decision maker that operates in a real time closed-loop
system [2]. First of all, the fuzzy controller aggregates
the process output data y (t ), then compares it with the
4.1. The synthesis of independent fuzzy controllers
x(t ) and by that information decides

what the control signal u (t ) of the process should be in
reference signal
Having a general idea about the independent fuzzy

control with correction of two dimensional process its
time to look at the synthesis of those fuzzy controllers.
Fuzzy control fuzzification interface operates linguistic
variables, fuzzy sets and membership functions in order
to transfer the fuzzy controller inputs to suitable data
format that is necessary for comparing with the rules in
the rule-base. Linguistic variables simply are the inputs
of fuzzy controller. Each linguistic variable is
characterized by fuzzy sets which in order are
characterized by membership functions.
Each of independent fuzzy controllers used reference
signal and control error as linguistic variables. Each
linguistic variable was characterized by 9 fuzzy sets.
There were used the triangular form of membership
functions of fuzzy sets. The detailed scheme of linguistic
variables, fuzzy sets and membership functions of
independent fuzzy controllers is demonstrated in Fig. 6.
The fuzzy sets of linguistic variable called Reference
signal were selected without considerable constraints,
lots of fuzzy sets concentrating in the middle heights of
the channel in order to accurately control them. The two
fuzzy sets named 15 and 75 were chosen a bit far
from the main fuzzy sets concentration to ensure
desirable control results when the corks were at the
bottom or at the top, otherwise if the concentration of all
fuzzy sets were in the middle heights area the control
order to ensure that the performance objectives will be

met.
Considering to revealed problems of PID control and the
existence of interactions between different air channels
there was constructed the control scheme of independent
fuzzy control with correction of the plant (Fig. 5). The
main objective of independent fuzzy control with
correction of two dimensional process was to improve
the control results of PID. Additionally, fuzzy control
had to consider to interactions between the air channels.
In Fig. 5 Fuzzy1 and Fuzzy2 controls were generating
the independent primal control signals of fans as in PID
control scheme (Fig. 3) but additionally to them was
implemented a special fuzzy control module that
corrected both (Fuzzy1 and Fuzzy2 generated) control
signals by estimating the interactions between channels.
The interactions were estimated by calculating a moving
average of the last 50 values of an appropriate channels
control error. The control error e(t ) of the channel is
calculated by equation:
e(t ) y (t ) x(t ).
(1)
148
results could be unsuitable since the dynamical features

were different where the corks were situated at the
bottom or at the top. The fuzzy sets of another linguistic
variable named Error were selected more precisely
since the control results of it depended significantly.
Three fuzzy sets (with negative values) were set to
control when the corks were moving up and five fuzzy
sets when moving down. Five fuzzy sets were sacred to
moving down in order to control such situations when
PID controllers were not able to control or the control
results were not desirable (the corks were moving down
till reached the channels bottoms and after that achieved
its reference signals). In these situations control signals
had to be changed very slightly since the corks were
moving down with additional power free fall
acceleration. Especially two positive fuzzy sets 10 and
30 drastically slowed down the effect of free fall
acceleration when the reference signals of corks were
changed more than 10 centimetres. The positive fuzzy
sets values corresponded to the corks moving down
direction since the corks positions in the channels were
calculated not from the ultrasound distance sensors but
from the bottom of each channel. Such an inverse
distance calculations simplified visual estimation of
actual corks positions in the channels.
The formula of centre-average is:

R
i 1
( y)
( y)
(3)
i 1
where: bi - the centre of the ith fuzzy set, R - the

number of rules in rule-base, ( y ) - the membership
functions value (the appropriate fuzzy sets weight to the
final fuzzy control solution), y - the output of fuzzy
controller.
4.2. The synthesis of control signals correction
module
The synthesis of fuzzy control module that corrected
primarily produced control signals was similar to the
synthesis of independent fuzzy controllers. The main
differences were in designing the fuzzification interface,
the rule-base and that the module produced two output
values (control signals corrections) instead of one.
The fuzzification interface was designed by two
linguistic variables Control error e1 (t ) and Control
error e 2 (t ) . These variables were the appropriate air
channels control errors average of the last 50 values. A
moving average was used to calculate them since it
helped to estimate the actual control error of the channel
more precisely. Each input linguistic variable was
characterized by five fuzzy sets -15, -5, 0, 3 and
10 as demonstrated in Fig. 7. The fuzzy sets of each
linguistic variable were chosen asymmetrically since the
dynamics when the corks were moving down or up were
different. Five fuzzy sets of each linguistic variable was
the sufficient number in order to estimate the
interactions between air channels.
Fig. 6. Input linguistic variables, fuzzy sets and membership

functions of Fuzzy1 and Fuzzy2 controls
The rule-base of each independent fuzzy controller was

designed separately since the dynamics of air channels
after detailed analysis were different. The rules in rulebase of independent fuzzy controllers were described by
IF Reference signal AND Error THEN Fans voltage
clauses. Each rule-base was composed of 81 rules since
there were 2 input linguistic variables with 9 fuzzy sets
in each.
The solutions linguistic variable was produced by minmax algorithm and the final solution (control signal of
the plant) was extracted by centre-average method.
The min-max algorithm works as follows: for each
rule is discovered the least (minimum) fuzzy sets truth
weight of IF part which is provided to the appropriate
(THEN parts assigned) solution fuzzy set. If there was
assigned a weight to the appropriate fuzzy set earlier,
then has to be taken the bigger one (maximum). The
final solution of fuzzy controller may be extracted, after
the synthesis of solution linguistic variable with fuzzy
sets weights assigned to it.
Fig. 7. Input linguistic variables, fuzzy sets and membership

functions of the module of control signals correction (Fuzzy
control)
Since the dynamics of air channels were a bit different,

two separate rule-bases were designed in order to
produce two output signals corrections. The rules of the
first rule-base were described by IF Control error
e1 (t ) AND Control error e 2 (t ) THEN Fans1 voltage
correction clauses. The rules of the second one: IF
Control error e 2 (t ) AND Control error e1 (t ) THEN
149
Fans2 voltage correction clauses. Each rule-base was

composed of 25 rules.
There was used the same methodology to produce the
solutions linguistic variables and the final controllers
outputs as in independent fuzzy controllers but the
correction module were generated two output signals
instead of one.
changed from 75 to 60. The length of each task mode

was 50 seconds. The statistical control results also were
analyzed by two lettered modes: A analyzing the entire
task mode, B analyzing just the portion of task mode
when the output signal of air channel got steady. The
control error ( ) was calculated by formula:
5. Experimental results
The experiment was performed with different sets of
reference signals, but the main control challenges
existed when the reference signals were changed
significantly, especially when they were reduced.
Therefore the experiment was concentrated to those
problems. In the section about the structure of control
system was mentioned that the air channels dependence
of each other was decided by the number of opened air
feeding vents. The experiment was performed with such
a fixed number of opened air feeding vents that ensured
the necessary amount of air to channels fans since this
configuration of opened air feeding vents guaranteed a
certain amount of interactions between air channels and
the corks were able to reach the reference signal values
physically.
In Fig. 8 and Fig. 9 are demonstrated the results of PID
and fuzzy controls, respectively, when the reference
signals for both air channels are 0, then changes to 75
and later to 60, the abbreviation of such a reference
signal configuration is marked as 0->75->60.
1
N
( y(t ) x(t ))
(4)
t 1
where N - the sample size.
Fig. 9. Fuzzy control results, reference signal 0->75->60

Table 1. The statistical results of PID control, reference signal
0->75->60
Control
<air channel>
Analysis mode:
Reference signal:
Sample size ( N ):
Control error ( ):
Reference signal:
Sample size ( N ):
Control error ( ):
PID <1>
A
PID <2>
0 -> 75
50
28.771
50
4.48
26
50
1.095
28.263
75 -> 60
31
50
1.13
4.11
30
0.549
38
0.44
Table 2. The statistical results of fuzzy control, reference

signal 0->75->60
Control
<air channel>
Analysis mode:
Reference signal:
Sample size ( N ):
Control error ( ):
Reference signal:
Sample size ( N ):
Control error ( ):
Fig. 8. PID control results, reference signal 0->75->60
The parameters of PID controllers were static for all

reference signals. This implied that the control signals of
channels were produced symmetrically not depending on
the corks moving direction that was a crucial factor why
the PID control results were not desirable. Since the
fuzzy control additionally considered the corks moving
direction the control results were significantly improved.
The statistical control results of PID and fuzzy when the
reference signals were 0->75->60 are shown in Table
1 and Table 2, respectively.
The analysis of statistical control results was
accomplished to each task mode separately. A task
mode in statistical analysis was called the 75->60
notation, which meant that the reference signal was
Fuzzy <1>
A
Fuzzy <2>
0 -> 75
50
21.535
50
3.025
32
50
1.084
20.938
75 -> 60
43
50
0.991
2.731
33
1.589
43
0.782
The output signals of air channels using fuzzy control

get in steady-state faster than PID control. The statistical
results of control errors when analyzing the A mode are
also better in fuzzy control, but the B mode results are
ambiguous.
The control results that are demonstrated in Fig. 10 and
Fig. 11 when the configuration is 0->75->40 reveal
150
the plants control strength of fuzzy control comparing

with PID.
The tuning parameters of PID control (Fig. 10) were
selected for each task mode separately since the usage of
fixed parameters as in Fig. 8 produced the insufficient
output signals (the mentions situation when the corks
were moving down till reached the channels bottoms and
after that reached its reference signals). These control
results shows the necessity of the synthesis of adaptive
PID control instead of static parameters to all task
modes. Conversely, fuzzy control (Fig. 11) results are
more desirable even if the time necessary to achieve the
steady-states of output signals takes more time.
Table 3. The statistical results of PID control, reference signal

0->75->40
Control
<air channel>
Analysis mode:
Reference signal:
Sample size ( N ):
Control error ( ):
Reference signal:
Sample size ( N ):
Control error ( ):
PID <1>
A
PID <2>
0 -> 75
50
23.975
50
8.056
16
50
0.299
23.401
75 -> 40
50
6.435
21
1.041
28
1.652
Table 4. The statistical results of fuzzy control, reference

signal 0->75->40
Control
<air channel>
Analysis mode:
Reference signal:
Sample size ( N ):
Control error ( ):
Reference signal:
Sample size ( N ):
Control error ( ):
Fuzzy <1>
A
Fuzzy <2>
0 -> 75
50
21.395
50
9.527
35
50
0.72
19.91
75 -> 40
32
50
0.954
10.044
28
0.438
35
0.636
6. Conclusions
The analysis of experimental results proves that fuzzy
control is more effective control method to air pressure
plant than PID. The implementation of humans
heuristic information about how to control a system,
enables to estimate the critical system control problems
(the dynamic and nonlinearities of system, interactions
between the channels) better than PID which directly
guides to better system control. The crucial factors of
successful plants control are the estimation of corks
moving directions and proportions. The synthesis of
adaptive PID control could be one of possible
replacements to static tuned PID in order to control the
plant with PID-based control.
Fig. 10. PID control results, reference signal 0->75->40
7. References
1. lstrm K.J., Hgglund T.H. PID controllers:
Theory, Design and Tuning. North Carolina, 1995.
343 p.
2. Passino K., Yurkovich S. Fuzzy Control. California,
1998. p. 9-10.
3. Kovaic Z., Bogdan S. Fuzzy Controller Design
Theory and Applications. Portland, 2005. 416 p.
4. Abonyi J. Fuzzy Model Identification for Control.
Boston, 2003. 273 p..
5. Albertos P., Sala A. Multivariable Control Systems.
London, 2004. p. 35-38.
Fig. 11. Fuzzy control results, reference signal 0->75->40
The statistical control results of PID and fuzzy control

when the configuration is 0->75->40 are shown in
Table 3 and Table 4, respectively. The output signals of
PID control not achieve or achieve later its steady-states
when the appropriate fuzzy control outputs signals. The
control errors of fuzzy are better in all task modes even
in the first channels task mode 0->75 when the output
signal is in its steady-state since the interval of the
output steady-state values is longer in fuzzy control than
PID - 35 opposite to 16.
151

INVESTIGATION OF ADAPTIVE pH CONTROL SYSTEM

Vytautas GALVANAUSKAS
Abstract: The paper deals with pH control quality

issues in fed-batch biochemical processes. First
principles mathematical model of the process was
elaborated. An adaptive pH control system based on
gain scheduling approach was proposed. Simulation
tests were carried out. Significant increase in control
quality as compared to a standard PI control system was
shown.
2. Mathematical model of the process

As an example of a control object, a laboratory scale
fed-batch biochemical process was investigated. The
control of pH level of cultivation medium is
implemented using closed loop control with a PID
controller. The pH signal is measured by means of an
in-situ probe that returns standard electrical signal to the
controller. The latter calculates the value of deviation
signal and defines appropriate control action. The
output of the controller is set to pulse mode and is
connected to peristaltic pumps which feed alkali or acid
solution into the bioreactor depending on the polarity
and value of the deviation signal. The controller allows
implementing either standard or adaptive PID control
algorithm.
The concentration of hydrogen-ions (which is
straightforwardly related to pH, [2]) in a fed-batch
cultivation process can be modelled by taking into
account influence of bacterial growth, addition of acid
and alkali solutions during pH control and dilution
effects:
Keywords: adaptive control system, gain scheduling,

pH control, fed-batch biochemical process.
1. Introduction
The main application fields of the pH control systems
are the production processes in chemical and
biochemical industry. The typical processes in this field
become more and more complex, and the control quality
requirements are relatively high. Modern biochemical
processes are very complex and difficult to control [1].
The dynamics of such processes is nonlinear and timevarying. The problem is even more difficult since there
are few reliable online measurement methods for such
important biotechnological quantities like biomass
concentration [1]. Therefore, often application of
complex and time consuming mathematical models for
off-line optimization, indirect state estimation and/or
optimal online control are required. On the other hand,
control systems should use mathematical models as
simple as possible in order to avoid high computational
load and numerical problems. This makes the task of
high quality control even more complicated.
High quality control of pH level is difficult because of
many reasons, some of which are: very strong
nonlinearity of biotechnological processes, titration
curves and pH measurement itself, high sensitivity of
the microorganisms even to small temporary deviations
of pH level of the cultivation media, and signal drift of
the pH sensors [2, 3].
There are various approaches for high quality pH
control shown in the literature. Nevertheless, most of
them suffer from the already described drawbacks [3].
Therefore, it is important to elaborate simple, robust and
easy to implement methods for pH control.
dCH
(1 x 2 x)
dt
FpH (CH0 CH ) Fs CH
,
V
V
(1)
where:
CH and C H0 concentration of hydrogen-ions in the
cultivation medium and in the feeding solution,
respectively, [mol/l]; the latter concentration can differ
from the theoretically calculated and is subjected to
model based identification;
x biomass concentration in the cultivation medium,
[g/l];
specific biomass growth rate, [1/h];
FpH flow of the alkali solution for pH control, [l/h];
Fs flow of the feeding solution, [l/h];
V cultivation medium volume, [l];
1 , 2 model parameters to be identified.
152
The first summand in parenthesis on the right hand side

of the differential equation (1) is related to biomass
growth and maintenance, respectively. During the
metabolic reactions in the microbial culture excessive
production of H ions takes places, and the addition of
alkali solution is necessary in order to maintain constant
pH level which is optimal for the growth of bacteria
and/or production of the target product.
The second summand takes into account the influence
of the flow of the alkali solution for pH control. The last
summand accounts for the dilution effect of the feeding
solution.
The initial value C (0) is equal to 107 [mol/l], and this
linearised around the working point. Under the

equilibrium conditions around the working point the
following equations are valid:
dCH
0,
(4)
(1 2 ) xV *
.
(CH0 CH )
(5)
dt
FpH
level corresponds to pH 7.
The biomass growth in the fed-batch process can be
modelled by means of the differential equation
F FpH
dx
x s
x.
dt
V
(2)
The initial value x (0) is determined after inoculation,

and is known at the beginning of each cultivation
process.
Another key quantity of the process is the oxygen
uptake rate, OUR. It can be evaluated from the equation
taking into account oxygen demand for the biomass
growth and maintenance:
OUR 1 xV 2 xV ,
Fig. 1. General structure of the adaptive pH control system
After the linearization of the dynamic equation for C

H
around the working point one gets the following
equation:
d (CH )
dt
(3)
where 1 , 2 model parameters to be identified.

The cultivation process under consideration consists of
two distinct phases (biomass growth and product
formation phase), in which these coefficients have
different values that should be identified separately.
All the variables except of x can be measured online
during the process. Each time moment the biomass
concentration x is obtained applying numerical
integration of (2) in real time. In a real system the value
of OUR can be calculated instantly from the off-gas
composition and other measurement data obtained
online, and is derived and calculated from (3) using
the most recent value of x. Then the calculated value of
is applied in the next integration step solving (2), i. e.
calculating next value of x.
FpH
V
(CH )
CH0 CH
V
FpH .
(6)
Multiplication of both sides of (6) by V leads to:

FpH
CH0 CH
V d ( CH )
C
)
FpH .
H
FpH
dt
FpH
(7)
Equation (7) is a first order differential equation, where

the object input is change of alkali feeding rate FpH ,
and the object output is change of concentration C .
H
The structure of the corresponding process transfer
function (in Laplace transform) around the working
point is
W ( s)
3. Mathematical model of the adaptive control

system
C H ( s )
FpH ( s )
K0
,
T0 s 1
(8)
and the coefficients of the transfer function (8) are as

follows:
After the creation of the process model (1-3), and the

identification of the unknown model parameters, a
mathematical model of the adaptive pH control system
can be elaborated. The general structure of the adaptive
pH control system investigated is shown in the Fig. 1.
In the controllers input the deviation between the set
point and the actual value is recalculated from pH to
C .
K0
CH0 CH
FpH
T0
(CH0 CH ) 2
(1 2 ) xV *
(CH0 CH )
V
.
FpH (1 2 ) x
(9)
(10)
In order to investigate, which of the controllers

parameters should be adapted online, the equation (1) is
Unfortunately, the process model has no clearly defined

delay. Hence, while tuning the controller parameters it
153
is not possible to apply the empirical tuning rules

involving delay.
It can be proved (e.g., [5]) that for such type of the
control object model the derivative part of the PID
controller is not able to additionally improve the control
quality as compared to a control system based on PI
controller. Hence, in further investigations PI controller
will be used.
In order to investigate the functional relationships
between the controller tuning parameters (Kr, Ti) and
the process variables ( C , FpH , , V, x) it is of
H
In the investigated case Ziegler-Nichols rule (sensitivity

approach, [5]) was used to calculate the initial values of
Kr and Ti parameters.
In order to adapt the Ti online one cannot use the x
values because in the analyzed case it is not possible to
measure biomass concentration online. Nevertheless, it
is possible to use any variable that is closely correlated
with x and can be calculated or measured online, e. g.,
OUR. As it can be seen from (3) and (10), OUR is
linearly correlated to the denominator of the tuning rule
for Ti. Therefore, the following empirical equation was
elaborated and used for adaptation of Ti:
advantage to make use of the internal model control

tuning rules [5] and to simplify them taking into
account (9-10).
The final expressions for Kr and Ti show that only Ti
depends on x, which is the main disturbance during the
control, while Kr depends on V. Because in the analyzed
case the latter variable varies within the narrow range
during the process, its influence on Kr can be neglected
and the parameter Kr can be excluded from the
adaptation rules and kept constant during the control of
the process.
Ti
1
,
2 OUR
(11)
where 1 , 2 empirical coefficients, that were chosen

during the design process.
4. Results and discussion
The model of the adaptive pH control system was built
using Matlab/Simulink (Mathworks, Inc.) environment.
General view of the model is presented in Fig. 2.
PID discr
f(u)
Error
Fcn1
pH SP
kr
Ti
kr
Un
Td
Td
Saturation Fb
Tdiscr
Tdiscr
mu
Tsp
Ti
fcn
Fb
Fb2 offline
w
Ti const / adapt
Fb1 online
Fb offline
x
1
1.e-7
20/3600s+1
pH sensor pH sensor T ransfer Fcn
Delay
C(H+)(0)
t
C(H+)
mu
fcn
-log10(u(1))
Fs
dx/dt
Saturation C(H+)
C(H+)->pH
pH
Fcn3
-6.e-5
f(u)
Fs1
OH-
pH err %
Fs
6.e-5
Switch
Base-Acid
Process x, C(H+), V
Fcn4
f(u)
H+
C(H+) err %
10
Clock
mu set
t
Ti adapt
fcn
mu
rH
OUR fcn
Fcn2
mu
f(u)
T ransport
Delay OUR
OUR1
H+ production rate
OUR
Fig. 2. General view of the adaptive pH control system model
The model consists of discrete controller, and

functional blocks for modelling of (1-3).
Additionally, the blocks for transfer function of the
pH sensor were introduced. The parameters of the
transfer function were defined in advance. Also the
block for disturbance (specific biomass growth
rate, ) modelling was used.
The simulation was performed under the starting

conditions similar to the experimental ones. Also
the values of model parameters for simulation were
chosen close to those typically observed for this
type of the cultivation process. The set point for pH
was set to 7.0. Despite that the controller was set to
work in PI mode, the tracking error was increasing
154
due
to
exponentially
growing
biomass
concentration that has acted as a disturbance. The
simulation results (see Fig. 3) show that the
tracking error of the adaptive PI controller was
significantly lower than the one in the case of a PI
controller with constant tuning parameters.
At the end of the process the adaptive controller
still was able to keep pH value of 6.999 as
compared to 6.96 in the case of the control system
based on the standard PI controller.
introduced disturbance was of the magnitude lower.

This proves once more the effectiveness of the
proposed control approach.
Fig. 4. The evolution of Ti value during the process

control using the adaptive control system
5. Conclusions
The proposed adaptive control algorithm proved to
be a robust and stable solution ensuring high
quality pH control in the analyzed biotechnological
fed-batch process. It can become a real alternative
for standard PID control. Further practical
investigations should be carried out in order to
identify model parameters from the experimental
data and to practically implement the proposed
adaptive approach into a real control system.
Fig. 3. Tracking quality of the adaptive (black line) and

standard (grey line) control systems
The tuning parameters of the latter were chosen

taking into account the dynamics of the control
object at the beginning of the process. If trying to
reduce the tracking error at the end of the process
one tunes the controller parameters on the basis of
the dynamics at the end of the process and uses
these values during the whole process it can lead to
instability of the control system. Hence, the only
solution is to gradually adapt the value of Ti. The
evolution of the Ti according to (11) during the
adaptive pH control process is shown in the Fig. 4.
During the 11.5 [h] and 99.5 [h] of the process
the disturbance in form of stepwise reduced
biomass specific growth rate (both times from 0.5
to 0.1 [1/h], and vice versa) was introduced. This is
a typical case in practice when substrate or oxygen
supply is disrupted due to malfunction of the
equipment. In order to keep the desired pH level a
rapid action of the pH controller is required.
The simulation results (see Fig. 3) show that in the
case of adaptive control the overshoot was about
30 % smaller, and the tracking error during the
6. References
1. Sonnleitner B.
Instrumentation
of
Biotechnological Processes. Advances in
Biochemical Engineering/ Biotechnology, 2000,
66. p. 1-64.
2. Carr-Brion K.G. (Ed.) Measurement and
Control in Bioprocessing. Elsevier Science
Publishers, London, 1991. 262 p.
3. McMillan G.K., Cameron R.A. Advanced pH
Measurement and Control. ISA, 2005. 294 p.
4. Ljung L. System Identification: Theory for the
User, 2nd Edition. Prentice Hall, New Jersey,
1999. 672 p.
5. Levine W.S. (Ed.) The Control Handbook.
IEEE/CRC Press, 1996. 1566 p.
155

BUILDING OF SECURITIES VALUATION IT SYSTEM

USING MULTI-AGENTS APPROACH
Andrius JURGUTIS*,**, Rimvydas SIMUTIS*
* Department of Process Control, Kaunas University of Technology, Lithuania
** AB Bankas Finasta, Lithuania
it. Speed of decision-making process gains a huge

significance.
The investment management is sufficiently complicated
and complex task. It requires a high competence of
financial assessment, quick orientation in continuous
stream of information and 24 hours work a day. It is
unlikely that it is reasonable for one person. The
efficiency of investment management team is
questionable, too. There is therefore a great need to
develop intelligent information systems that could help
the investor. The traditional investment advisory or
investment management companies employ teams of
various professionals: economists, statistics, financial
brokers, IT professionals and ect. Some of them are
gathering the information, the others are analysing it
and making the conclusions, the third group is making
the future predictions. The last group is summarizing
and making the final decisions. We believe that such
model of companies operations is not appropriate in the
twenty-first century. It doesnt meet the investors needs
and company owners' interests. Tremendous
competition between the investment management
companies promotes the optimisation of operating costs.
As one of the possible solutions we offer to automate a
part of the functions performed by professionals: to
create and develop an intelligent information system or
robot consultant. Such IT system would be useful to the
investors and investment management companies. First,
it would allow increasing the profitability of the
company. Secondly, it would provide services to
investors at lower costs (such as lower commission for
stock trading) and enable reduce the management fee of
various investment funds. The cost reduction has a
direct positive impact on the ultimate profitability of the
customer portfolio. Third, the automated system
accelerates the decision-making process, which is very
important in dynamic market. Based on practice, we
could say that the high investment returns reaches those
who are first to receive, detect and correctly evaluate the
information, and first to make the investments
decisions. Abundance of information sources and
Abstract: Application of software agents and multiagents systems for decision making are the new
approach to the capital markets. The flow of
information is increasing rapidly and is extremely
important to evaluate the news on time and make the
right decisions. The papers propose the basic structure
of the multi-agent system that will function in real time
dynamic environment in capital markets. This IT system
will be able to suggest investment solutions taking into
consideration variable characteristics of individual
investors. IT system will acquire information from and
monitor changes to stock reporting databases, interpret
stock information, and predict the near future of an
investment. Broad spectrum of securities valuation
methods will be employed and profile of the user will
be incorporated in the decision making process. A
multi-agent system approach is ideal in cases that
require the coordination of several component tasks
across distributed and flexible networks of information
sources and expertise. The paper will present the basic
structure of Securities valuation IT system using multiagents approach.
Keywords: software agents, multi-agents systems,
securities valuation.
1. Introduction
Deep knowledge of fundamental, technical or other
securities valuation methodology is not enough for
successful trading in financial markets. Since trading
moved to electronic space, new challenges have
emerged. First, financial markets have become global.
On one hand, the investing in any region of the world
has become very simply. On the other hand, there is a
need to know the specifics of the region adapt to them
and learn to work with. Second, the financial markets
have become dynamic. The abundance of information
could quickly change the direction of the markets.
Therefore, there is a need to continuously track and
interpret the information and make decisions based on
156
surroundings and implements the task. In our case we

will use the software agents in following order: the
primary task to make a recommendation to investor to
buy or sell the security will decompose into smaller
tasks like information gathering or applying a certain
securities evaluation approach. Independently acting
software agent will execute each task in separate
manner. In order to achieve the different task we will
use various types of software agents. In part 2 of the
article, we will review the properties of software agents.
In part 3 we will analyze the types of agents used in
resolving the tasks. In part 4 well propose the structure
of securities evaluation IT system. At the end of the
article we will present how the IT system should
function.
formats complicates and prolongs the perception of

information and its processing time. Information
sources vary, some of them occur, while others
disappear, therefore is always desirable to know about
these changes. The proposed investment robotconsultant would be able to interpret the information,
compare and update it. Fourth, the automated advisory
system would ensure greater confidentiality of
investors. The investor would interact with IT advisory
system
instead
of
human-consultant,
which
autonomously and anonymously will seek his client's
objectives. The human consultant may unacceptable
distribute the clients ideas to other investors.
Accordingly, such IT system could carry out preventive
work in market manipulation and detect misleading
information. The system would prevent the spread of
false information. Fifth, the intelligent advisory system
would be able to adapt to changing objectives of
investor. In this way, the customer should be always
surrounded with attention and customised individual
proposals. At the same time, this will be a leap in
quality of customer service standards and new approach
to sales of investment ideas and products. Sixth, the
robotconsultant will posses the best experience of
various fields professionals thus avoiding possible
human errors (such as emotions, fear, etc.). Seventhly,
IT system containing all the listed properties is likely to
increase the number of correct decisions as compared to
traditional consulting firm. We could assume that
company offering such instruments to investors will
gain a significant popularity. Eight, if robot-consultant
would be constructed it will possibly increase a number
of people interested in investing. At the same time, it
enables investors to learn about the investment world
quickly. We believe that robot-consultant possessing all
the listed properties would become one of the main
investor tools.
The problem is how to construct the IT system capable
to evaluate various financial parameters and securities.
Portfolio management and investment decision-making
process is complicated and complex task. The
complication is found in searching data, choosing the
financial instruments or assessment methods and
ultimately, drawing the conclusions. The complexity
means that there is a need to coordinate a number of
related processes such as information collection,
performance assessment, market surveillance, etc.
Therefore, the recommendation to investor includes
resolving of several tasks, the control and management
of individual tasks [19, 20]. We like others believe that
it is beneficial to create such system using the software
agents and, more precisely, to develop the multi-agent
systems [1, 2, 3, 4, 5, 6]. Already there are some
examples of such systems, as Warren, MASST [12].
These IT systems are more focused on information
collection and processing. The adaptation to user needs
and investment decision-making is less important.
Software agents are the new tools of artificial
intelligence and software engineering. An autonomous
software agent is the system (a computer program)
which existing in surrounding environment (for
example internet) or being a part of it executes the
prepared plan at the same time monitoring these
2. Agents properties
The software agents in our IT system will possess the
following basic properties [7, 8]:
Autonomy: the ability of the agent to act
without any external interference. The agent
must be able to make its own choices and make
its own decisions.
Communication skills: the agent must be able
to establish relation with environment. The
most important is the interaction between
similar agents and the user.
Reactivity: surrounding environment is
typically dynamic and requires the adaptation
ability from the agent.
Mental notions: during his life cycle, the agent
is learning from different situations. Thus,
through the memories and through the
interaction with others agents, the agent is able
to gain its own knowledge.
Persistence: the mobile agents life is usually
longer than the length of his basic tasks. Thus,
it is created to satisfy one of the following
criteria:
1. To finish after having done the basic task;
2. To finish after having exhausted the
assigned internal resources.
Social ability: the agent exists among the other
agents, thus it must be able to exchange the
information in order to achieve its assigned
goals or help the other agents.
Mobility: a mobile agent has the inner attitude
to vary its communication partners. It is able to
interact with both a similar agent and a user.
Vitality: in order to achieve its goal, the agent
must cope with anomalous situation that could
create a state of instability that could damage it
irreparably. An agent who has a strong vitality
is able to solve the most adverse situations it
meets.
Pro-activity: a mobile agent is able to start new
interactions with other agents and coordinate
various agents activities stimulating their
certain responses.
Truthfulness: the agents use the information
received during the exchange and not viewed
157
under direct control inspection entities, so they

are expected to respect truthfulness.
Benevolence: the agent must act in good will
and must not perform any acts contrary to the
users will or to the other agents interests.
4. Information system architecture

In the process of information system design for
assessment of securities, we shall take a deeper look at
development of better investment proposition
generation process in consideration of individual needs,
goals and suspense of investor. In other words, the
structure of information system, namely the part of
system responsible for investment proposition design,
must safeguard the safety of system itself and proposals
should correspond to the needs and goals of investor as
well as coincide with natural behaviour of investor.
From the point of view of us as investors, the system
architecture could be double. With reference to
investment solutions, investors could be divided into
two groups. First would be conservative investors
avoiding risk and requiring more time for final decision
making. The named group appeals on several securities
assessment methods before decision making, namely
use of both the technical and fundamental analysis,
consultations with experts (brokers, analysts) and insert
some other methods. Second group is considered as
aggressive, risk unfearful and spontaneous decision
makers or in other words profiteers. Usually, such
investors make use of securities assessment methods
they are most fond of, for example: only technical
analysis or only fundamental analysis. If they have time
in their hands, than such investors verify investment
solutions with other methods too, in other words
check if other method satisfies their solution. Therefore,
information system must firstly determine investors
type a job performed by Interface agent.
Interface agent determines investors type and later
revises decision through an interactive questionnaire for
the investor. Questionnaire consists of two parts:
general and special. General part consists of some
common information about the investor: age, education,
income, expenditure, investment experience, risk
perception, tolerance for risk, etc. Special part
introduces investor to securities assessment methods
realized in the system, investor receives control
questions for ascertaining methods mostly trust-worthy
for him/her or investor indicates which methods system
must use to perform securities evaluation and finally
present proposal. Suppose every method is evaluated by
investor by giving points. If investor gives approximate
number of points to several methods, in this case
information system generates the proposal by appealing
on those methods. Influence of methods could be equal
for the final decision or respectively the authority of
method for final investment proposal making could be
evaluated by the number of points given to that method.
Information system structure of this type is presented in
Fig. 1. Interface agent informs solution agent about
the methods (together with method parameters, e.g. beta
ratio, Sharpe ratio, etc.) to be used by the latter in
modelling the investment proposal. Solution agent
launches agents of appropriate methods through
coordinating
agent
3. Types of agents
The securities evaluation IT system will consist of 3
types of agents:
o Interface agent,
o Information agent,
o Co-operating agent (due the functional clarity it
will be named the Task agent in the further
article.)
The system could have more agents, too [10, 11].
The Interface agent is the mediator between the user
(investor) and the securities evaluation IT system [17,
18]. Interface agent receives the information from the
user and accepts its target, then transmits it to other
system elements. On the contrary, it presents the
systems generated answers to the user in an
understandable form.
The main functions of the Interface agent would be the
following: to collect and analyze the information about
the user and share it with other IT system agents, to
submit the system generated responses to the user, to
request for additional information and if it is necessary
to inform the user. Therefore, the IT systems user
interacts with interface agent only, no other agents are
involved.
The main function of the Information agent is to provide
the information to the Task agents [10, 15]. For
example, in order to learn the fundamental value of
securities, the information agent needs to gather the
financial data about the issuer using the Internet data
search. Then the Information agent transmits the
information to the Task agent, which is responsible for
fundamental analysis. The Information agents could
vary and perform different information gathering
strategies, use different sources of information.
Information agents distinguish their selves with proactivity feature. They are able to generate events in the
surrounding environment and to start new interactions
with other agents instead of waiting for assignment of
the data search.
Task agents perform a lot of work in shaping the
recommendation to the user. Each Task agent has
certain knowledge about certain security evaluation
approach. Depending on users request, which is
expressed through the interface agent, IT system may
take different methods and their combinations. When
performing the tasks, the agents are cooperating with
others thus communication skills and social ability is
the main property. Task agents are more complex than
the interface and information agents.
The security evaluation IT system obtains more
possibilities when agents are able to use the mental
notions. It is likely to have a more efficient IT system
when it is able to accumulate the history of
recommendations and to select the best decisions [18].
158
Fig. 1. Horizontal Structure of Information System

the declining sequence of points scored. In this way a
vertical top-down method architecture is determined for
every individual investor reflecting his/her behaviour on
the market. Fig. 2 pictures vertical structure of
information system. Solution (Reasoner) agent calls
major method agent through a coordinating agent.
Major method agent presents the selected securities
through coordinating agent to solution agent. Solution
agent evaluates if theres enough data to safe solution
making and if not calls next method agent by the list
of importance through coordinating agent and the
process repeats. Finally, solution agent provides
proposal to investor.
Such type of information system structure should
safeguard a kick-start of securities selection and repeat
actualization of natural investors behaviour. We may
presume that with knowledge of investors statistics, i.e.
by assumption that all investors can be described by
finite number, multi-agent of vertical structure,
prognosis of the market reaction to speculative (future)
events could be made and with this knowledge
investment solutions could be formed.
From time to time Interface agent must renew investors
data, namely record changes of investors suspenses,
goals, etc. It could be done by provoking investor with
securities proposal uncomfortable to him/her and
observe the reaction. Another way could be to offer an
overfilled questionnaire. Interface agent must also
secure feedback about proposals provided by
information system. Problem is the feedback receipt is
most
purposeful
after
investment
method agents do their work in an asynchronical and

independent way. Expert agent searches for securities
corresponding investors needs at other expert agents or
expert people. Fundamental analysis agent hunts for and
evaluates securities with fundamental analysis methods,
while respectively technical analysis agent makes
assessment of securities through fundamental analysis
tools. Market agent evaluates securities through market
news [13, 14, 15]. Risk agent seeks for and selects
securities through risk level (according to Sharpe ratio,
beta ration, etc.). In case of finding appropriate
securities by appropriate method, every method agent
informs coordinating agent and the latter checks if that
security was earlier evaluated by other methods and in
case not other method agents are called in
additionally. In this way, system can have several
agents of the same method and accelerate presentation
of investment proposal. Method agents collect
information with information agents help. Finally,
coordinating agent collects assessment of all methods
required by the investor and presents them to solution
agent eventually delivering solution to investor. Our
title for this system architecture is Horizontal Structure
of Information System.
If one of securities assessment methods got extremely
larger number of points than any other method, then we
may call investor a profiteer with one securities
assessment method he/she is most fond of and which
must be referred to by the system in proposal
presentation. Then, Interface agent informs solution
agent about the major method and the sequence of
importance for the remaining methods is determined by
159
Fig. 2. Vertical Structure of Information System

5. Conclusions
proposal presentation when it is possible to say if

investment proposal served the purpose or would have
served the purpose even if not accepted by investor.
The problem could be solved by adding a virtual
portfolio formed according to all information system
proposals alongside the investors portfolio. Virtual
portfolio would be automatically supplemented or
reduced with financial means according to information
systems proposals. Investor himself could observe the
virtual portfolio beside his/her own.
The generated securities assessment information system
would constantly watch the situation, composition of
investors portfolio and in case of vital events on the
market possibly strongly influencing the change of
investors portfolio, would inform investor at once by email or SMS when he/she is disconnected from the
system. After some time, information system learns
decision making satisfying investors goals and
corresponding his/her behaviour, and investor can allow
the system to make deals through use of sales agent.
Securities assessment information system shall be built
in JADE platform base [16]. Agents perform tasks by
communicating in ACL (agent communication
language) protocol, in our opinion a more functional
agent protocol than KQML. ACL communication
protocol conforms to the up-to-date FIPA standard.
Programming agents and multi-agent systems they

design are a perfect tool for building financial means
assessment information system. Securities evaluation
and their portfolio management are a complicated and
integrated task requiring combination of independent
and different methods of securities assessment and
disperse information resources they need. Programming
agents are able to unite all the necessary elements into
one and complicated multi-agent system. Research with
already working systems such us WARREN, MASST
testify it and our hope is that our system orientated to
investors behaviour shall also answer the purpose.
With possession of securities assessment information
system structure we shall continue our work by
gradually realizing separate system parts with JADE
platforms help.
6. References
1. Cohen P. R. and Levesque H. J. Intention=choice +
commitment. In Proceedings of AAAI-87, Seattle,
WA., 1987. p. 410-415.
2. Kan Lang. Newsweeder: Learning to filter netnews.
Proceedings of the Twelfth International Conference
160
13. Sycara K., Giampapa J. and Seo Young-Woo.

Financial news analysis for intelligent portfolio
management. The robotics institute, Carnegie
Mellon University, 2004. 17 p.
14. Sycara K., Giampapa J. and Seo Young-Woo. Text
classification for intelligent portfolio management.
The robotics institute, Carnegie Mellon University,
2002. 14 p.
15. Sycara K., Giampapa J. and Seo Young-Woo.
Designing behaviors for information agents. The
robotics institute, Carnegie Mellon University, 1996.
15 p.
16. Bellifemine F., Caire G. and Greenwood D.
Developing multi-agent systems with JADE. UK,
2008. 286 p.
17. Maes P. Agents that reduce work and information
overload. Communications of the ACM, 37(7),
1994. p. 30-40.
18. Mitchell T., Caruana R., Freitag D., McDermott J.
and Zabowski D. Experience with a learning
personal assistant. Communications of the ACM,
37(7), 1994. p. 81-91.
19. Oates T., Nagendra Prasad M. V. and Lesser V.
R. Cooperative information gathering: A
distributed problem solving approach. Technical
Report, Department of Computer Science,
University of Massachusetts, 1994. p. 94-66.
20. Sycara K. and Zeng D. Coordination of multiple
intelligent software agents. The robotics institute,
Carnegie Mellon University, 1996. 31 p.
on Machine Learning, Tahoe City, California, USA

1995. p. 331-339.
3. Rao A. S. and Georgeff M. P. A model-theoretic
approach to the verifcation of situated reasoning
systems. In Proceedings of I J C A I -93,
Chambery, France, 1993. p. 318-324.
4. Shoham
Y.
Agent-oriented
programming.
Artifcial Intelligence, 1993. p. 51-92.
5. Sycara K. and Zeng D. Towards an intelligent
electronic secretary. In Proceedings of the CIKM94, Gaithersburg, Maryland, 1994. p. 239-260.
6. Wooldridge M. and Jennings N. R. Intelligent
agents: Theory and practice. The Knowledge
Engineering Review, 10(2), 1995. p. 115-152.
7. Editor Genco A. Mobile Agents: principles of
operation and application. UK, WIT Press, 2008.
271 p.
8. Wooldridge M. An introduction to multiagent
systems. UK, 2006. 348 p.
9. Franklin S. and Greasser. It is an agent, or just a
program? : a taxonomy for autonomous agents.
Third international workshop on agent theories,
architecture, and languages. Springer-Verlag, 1996.
12 p.
10. Nwana H. S. Software agents: an overview.
Knowledge Engineering Review, Cambridge
University Press, 1996. p. 1-40.
11. Wooldridge M., Sycara K. and Jennings N. R. A
roadmap of agent research and development.
Autonomous agents ant multi-agent systems,
Boston, 1998. p. 7-38.
12. Sycara K., Decker K., Williamson M., Zeng D. and
Pannu A. Distributed intelligent agents. The robotics
institute, Carnegie Mellon University, 1996. 32 p.
161

MODELLING OF ATM CASH TURNOVER

Vaidas KAZLAUSKAS*, Vygandas VAITKUS**
*Scientific service company GTV, **Kaunas University of Technology, Lithuania
Abstract: This paper presents a model for money

turnover prediction in automatic teller machine (ATM).
This approach is based on an artificial neural network to
forecast a daily money demand for ATM. At first the
investigation of model variables was preceded. Then
multilayer perceptrone was used for prediction of
money turnover. The prediction accuracy was tested on
real historical money demand data by comparing ANN
model prediction with Nave prediction hypothesis.
The paper consists of 7 sections. In section 2 data

analysis is presented. In section 3 model identification
procedure is defined. Modelling results are discussed in
section 4. Comparison of ANN model prediction and
Nave prediction hypothesis is presented and discussed
in section 5. Finally future works and conclusions are
presented in sections 6 and 7.
Keywords: ATM forecast of money demand, multilayer

perceptron, Nave prediction hypothesis.
For the project purpose a daily cash turnover from 2139

automated taller machines were received. First step was
to analyze these data. For further investigation it was
selected only 1213 machines which have the data
accumulated in at least one year period. After discussion
with human experts potential inputs were identified:
2. Data analysis
1. Introduction
Automatic taller machines (ATM) are equipment for
money supply for consumers. More than one thousand
ATMs are in Lithuania and over one billion in whole of
the world. So it is not a difficult to imagine what
amount of cash flow is needed for loading and ensuring
of ATMs correct action. At present the analysis and
forecast of money demand is made by human expert. He
is analyzing money flow with his own empirical model
witch estimate input parameters like money demand few
days ago, how much days till holidays and others.
Depending on strategies the ATM loading is executed
daily or weekly by fixed size of money load.
The main objective is to make load of ATM exactly
what it needs to be because money has a cost depending
on interest rate.
Studies of ATM turnover simulation and prediction
made in recent years were stimulated by private
companies and were unpublished. The main attention in
current publications in ATM field is concentrated to
transaction security. Recently investigation in ATM
turnover modelling field was performed in Kaunas
University of Technology [1, 2]. The main goal of the
research was to optimize ATM cash upload. It was
formed and identified an artificial neural network model
for each ATM and forecasted demand of cash for
specified period of time. These articles focus on an
optimization routine and wasnt briefly discussed
significant variables, ANN model structure and
parameters identification. This article tries to eliminate
these deficiencies.
Week day number,

Month number of the year,
Number of the month day,
the average of money turnover within a few
previous days,
Number of days before the holidays and after,
And others.
Fig. 1. Real and mean cash flow of the all ATMs dependence
on week day number
162
MATLAB software was used for this research.

Feedforward backpropagation neural network structure
used for modelling is presented in figure 4.
Fig. 4. Neural network structure used
The network has input layer, one hidden layer, with a

different number of neurons and one output neuron.
Hidden layer neurons have hyperbolic tangent sigmoid
transfer function, and output linear transfer function.
Fig. 2. Real and mean cash flow of the all ATMs dependence
on month
TFout
2
1.
1 e2in
(1)
ANN training was produced by Levenberg-Marquardt

training algorithm. Training on the training data set
continues as long the training reduces the network's
error on the validation vectors. After the network
increase error on validation data, training is stopped.
This technique automatically avoids the problem of
over-fitting.
Before training procedure data for modelling was
normalized by the rule that the standard deviation equal
to 1, and average to 0.
4. Analysis of modelling results
Fig. 3. Mean cash flow of the all ATMs dependence on
holiday day. Red line and day number 6 mark a holiday
Investigation of modelling accuracy depending on input

vector is presented in table 1. Neural network model has
ten neurons in hidden layer.
After data pre-processing some figures was made in

order to ensure that inputs identified by human experts
really have an influence to cash turnover. The cash
turnover, depending on few selected features is
presented above in the figures 1, 2, and 3.
In columns of figures 1 and 2 is presented a daily
turnover of all ATMs for one day of the week or month.
Figures showed a certain dependencies between the
features and money turnover, and it would be rational to
use them for model identification.
Table 1. Modelling results
Inputs
[D M MD M7D]
[D M MD M5D]
[D M MD C1]
[D M MD C1 C2]
[D M MD]
[D M MD H]
[D M MD M5D C1]
[D M MD M5D C1 C2]
[D M MD C1 C2 C3]
3. Model identification
Artificial neural networks were selected as proper
technique for nonlinear modelling. Neural networks
were successfully used in many areas such as pattern
recognition, classification, nonlinear modelling and
others [3-7]. It is also convenient to use ANN when
necessary to adapt the model settings or select an
appropriate model structure. Requirements for model:
Mean
prediction
error of 5
learning
restarts,
MSE
2.1671*105
2.7676*105
6.7330*105
3.5555*105
2.3275*105
2.6056*105
2.3444*105
2.6756*105
2.4678*105
Best of 5
learning
restarts,
MSE
1.8137*105
1.8214*105
1.8473*105
1.9098*105
1.7813*105
2.1436*105
1.8689*105
1.9264*105
1.9393*105
For each ATM 5 network learning restarts were

executed and mean squared error performance value
calculated on testing data set.
Input variables were labelled:
D Day of the week number,
M Months number,
To give best possible cash forecasting

accuracy,
To be as small structures as possible for fast
learning, or adaptation to new data procedure.
163
The prediction performance was estimated by sum

absolute error of each ATM on testing data. It was
computed for all 1213 ATMs. The result is presented in
table 2 and figure 6.
MD Months day number,

C1 Cash turnover day ago,
C2 Cash turnover two days ago,
C3 Cash turnover three days ago,
H Number of days till holiday,
M7D mean of cash turnover during last 7
days,
M5D mean of cash turnover during last 5
days.
Table 2. Comparison of to two cash flow prediction strategies
Nave prediction
hypothesis
9.5504*104
Prediction
error
As it is seen from table 1 it was tested different input

sets for cash turnover modelling. The best result was
received by a model with three inputs day of the week
number, month number and day of the month number.
Looking at the network outputs and real cash turnover
data, it could be seen that the model cant predict the
sudden changes in the money circulation. Models
behaviour is like a human experts. It is trying to predict
the potential demand, but random changes are
unpredictable (Fig.5).
ANN model
prediction
5.8679*104
x 10
2.2
naive prediction hypothesis

ANN prediction
1.8
1.6
S u m a b s o lu te e rro r
1.4
1.2
0.8
0.6
0.4
0.2
630
640
650
660
670
680
690
700
710
720
730
ATM number
Fig. 6. Comparison of Nave prediction hypothesis and ANN

prediction error
Experimental results have proved the efficiency of ANN

model proposed. The prediction error of the ANN
model was significantly smaller.
6. Future works
Fig. 5. Randomness in cash flow data
Made abundant experimental studies showed that the

model can accurately predict cash turnover in certain
time intervals (Fig. 7). In other apparently an error
increases. It was also investigated that model is very
accurate for some of ATM, and for some of them
inaccurate. The results of this study let us to make
conclusion that a strong correlation between the input
variables with the turnover of money is only in certain
moments.
This indicates that the data is very noisy. We have tried

to reduce the impact of random shocks to the turnover
data by moving average filter with different moving
window widths. Filtering of the input data didnt make a
significant influence to model prediction result.
5. ANN prediction comparison with Nave prediction
hypothesis
The question arises whether effective model is
established and identified? Whether it can compete with
a human expert who is predicting ATM network cash
circulation? We have not had the opportunity to
compare the human-expert and ANN accuracy. To
ensure the efficiency of the ANN model it was
compared with Nave prediction hypothesis. The nave
prediction hypothesis took today's cash flow as the best
estimate of tomorrow's cash flow. It can be expressed as
follows:
y t 1 yt ,
(2)
where y t 1 is the predicted value of the next period; yt is

the actual value of current period.
Fig. 7. Identified periods of small and big prediction error
164
This may be because the input variables magnitude

varies over time and sometimes the model with other
input variables could be significantly more accurate. If
there is a big randomness in cash turnover, there is no
ways to increase model accuracy. If not a random
disturbance is guilty, it is necessary for further research
what set of inputs for model in one or other period is
more appropriate.
There will be a question of when the model needs to go
to the next set of inputs? One way is to identify the error
threshold of the model and if it is above the limit move
to the next set. Another way could be to calculate the
model by several sets of inputs and to use this one
which gives the best result.
Of course some questions about the calculation
resources and the time required performing them,
especially if the calculations are carried out in real time
will arise. All these questions will be analyzed in the
next phase of our research.
8. References
1. Simutis R., Dilijonas D., Bastina L. Cash demand
forecasting for ATM using neural networks and
support vector regression algorithms. In: Selected
papers of EurOPT-2008: the 20th International
Conference "Euro Mini Conference on Continuous
Optimization and Knowledge-Based Technologies",
Vilnius, Technika, 2008. p. 416421.
2. Simutis R., Dilijonas D., Bastina L., Friman J.,
Drobinov P. Optimization of cash management for
ATM network. Information technology and control,
2007, vol. 36, p. 117-121.
3. Gupta M.M., Jin L., Homma N. Static and dynamic
neural networks: from fundamentals to advanced
theory. John Wiley & Sons, 2003. 700 p.
4. Gnoth S., Jenzsch M., Simutis R., Lbbert A.,
Control of cultivation processes for recombinant
protein production: a review. Bioprocess and
Biosystems Engineering, 2008, vol. 31, no. 1. p. 2139.
5. Simutis R. Exploratory analysis of biochemical
processes using hybrid modeling methods,
Proceedings of the First European Symposium
Principles of Data Mining and Knowledge
Discovery Trondheim, Norway, June 24-27, 1997.
p. 200-210.
6. Wei H., Shouyang W., Lean Y., Yukun B., Lin W. A
new computational method of input selection for
stock market forecasting with neural networks.
Proceedings of International Conference ICCS 2006,
Part IV, 2006. p. 308-315.
7. Huang W., Lai K.K., Nakamori Y., Wang S.Y.
Forecasting foreign exchange rates with artificial
neural networks: a review. International Journal of
Information Technology & Decision Making, 3(1),
2004. p.145-165.
7. Conclusions
Computational method for ATM cash flow forecasting
using artificial neural network was proposed. Influence
of various input sets to model accuracy was
investigated. The best input set was used for model
identification. ANN model prediction efficiency was
estimated by comparison with Nave prediction
hypothesis. The error of ANN model was about 39%
smaller. The result is hopeful, but further experimental
investigation about practical implementation and more
accurate cash turnover prediction is necessary.
Acknowledgments
This research work has been supported by means of EU
structural funds (Project No. BPD2004-ERPF-3.17-0606/0045).
165

LITHUANIAN ENERGY AFTER 2009

Arvydas GALINIS
thermal power plants, that use fossil fuel, do. The

fraction of electricity produced at Ignalina NPP made
more than 71% in electricity energy balance in 20052008. It is worth noting that electricity production from
renewable energy sources increased from 460 GWh to
593 GWh during 2005-2007, however their share in
Lithuanian electricity balance increased only slightly
from 3.9% in 2005 to 4.7% in 2007. Until 2015 the
expected capacities of the renewable power plants will
satisfy only approximately 10-12% of all electricity
energy demanded. New large electricity generators at
Lithuanian TPP or at other power plants will not appear
in the market earlier than in 2012-2013. Until this time
small-scale electricity generators (mainly gas based) can
emerge only either in industrial or central district
heating companies, but their impact on electricity and
capacity balances will remain small. Considering to this,
it is expedient to examine the worst case, when the
increasing electricity demand is met only with at the
moment existing power plants and electricity import for
several reasons is not possible. Here some explicitness
should be carried: in many cases the produced
electricity depends on capacity installed. Electricity
production at hydro power plants is an exception,
because water flow of Lithuanian rivers limits the
electricity production. Thus it is more relevant to
analyse capacity balance than energy balance. The 1
table presents the worst cases capacity balance under
basic demand scenario.
The presented data show that after closure of the second
unit of the Ignalina NPP, the generating capacity of
power plants will reduce from 4599 MW till 3466 MW,
meanwhile demanded capacity of consumers will
increase from 1917 MW in 2008 to 2471 MW in 2015.
It means that capacity supplied to the transmission
network is approximately 1000 MW larger than the
demand; therefore there shouldnt be any problems.
However, 2x150 MW units of the Lithuanian TPP are
of low technical state and cant reliable work. It is
considered, that this 287 MW capacity is mothballed. In
addition, it is required to keep a reserve capacity in the
system that is equal to capacity of the largest unit. In the
case of breakdown of any unit at any power plant this
reserve capacity should start operation. Thus emergency
Abstract: Government of the Lithuanian Republic

committed to close the second unit of the Ignalina NP
plant at the end of 2009. As a consequence Lithuanian
Thermal Power Plant (Lithuanian TPP) will become the
main electricity energy source. Although the provision
of installed capacity at power plants is sufficient until
2014-2015, however a moderate probability exist, that
under the particular circumstances a capacity deficit will
be evident in 2010 and therefore part of consumers
might be switched off. Nevertheless the main problem
would still be economical and the scale of it would
depend on the prices of fossil fuel.
Keywords: electricity, capacity, balance, production,
costs, price.
1. Introduction
Lithuanian Government by signing the Treaty of the
Accession to European Union committed to close the
first unit of the Ignalina NPP at the end of 2004 and the
second one at the end of this year. However these
decisions were accepted with reference both to technical
and economical analysis that was performed until 2002,
and to information about global energy evolution
trends, energy sector development tendencies, and main
provisions in EU member states at that time.
Nevertheless the situation in this field had essentially
changed during the several recent years. Therefore it is
advisable to examine, what situation will emerge in
Lithuanian power sector after 2009, whether Lithuania
itself, without the assistance of other countries could
generate all required quantity of electricity. Other very
important questions are the following: are there enough
capacities and how this will respond both to operational
cost and to the price of energy supplied?
2. Electricity generation and capacity balances
Even after the closure of the first unit, Ignalina NPP
remained the main source for electricity generation in
the country. There werent significant changes because
this power plant produces electricity cheaper than
166
reserve of the system would reach 289 MW. For a

normal operation of the system, balancing reserve is
also required. The purpose of this reserve is to
compensate short-run electricity demand fluctuations.
Its quantity depends on system load and increases from
102 MW in 2008 to 131 MW in 2015. Wind power
plants are also not reliable power resource, because
there arent any guarantees that at the moment of
maximum demand the wind will blow and rotate
generators. Consequently wind power plants cant be
involved into the capacity balance of systems peak
load. Kruonis hydro pump storage plant (Kruonis
HPSP) in some rare cases cannot also be used for
system maximum demand satisfaction. Suppose, before
the moment of system maximum load, accident
occurred in the system and turbines of Kruonis HPSP
were used as fast reserve. The turbines could consume
water from the upper reservoir and therefore at the
moment of system maximum demand there wouldnt be
water, therefore Kruonis HPSP would not produce
electricity. Thus, data presented in table 1 show that in
the case electricity is not imported, 176 MW capacity
deficit shall appear in 2010. If no generating units are
built until 2015, the deficit will increase till 525 MW.
Under conditions of high demand growth scenario,
capacity deficit in 2010 and 2015 correspondingly will
be 227 MW and 675 MW.
As it was mentioned, Kruonis HPSP cannot participate

in covering systems maximum demand only in rare
accidental cases. In normal operating regime, a part of
its capacities can be used. Lets analyse which part of
its capacity can be used. Capacity of plant with limited
output is utilised more efficiently, if it operates during
system peak load. Kaunas HPP and Kruonis (HPSP) are
limited energy output power plants. A part of Kaunas
HPP capacities work in basic regime, because it is
necessary to keep sanitarian flow of water below the
power plant. Usually one unit of power plant works in
basic regime. Thus 75.6 MW of Kaunas HPP is
assigned to the peak load. Below Kaunas HPP there is a
zone for Kruonis HPSPP. Assuming that Kruonis
HPSPP is used for optimal weekly operating regime we
can estimate that to the systems peak load in 2010 it
can contribute by 440 MW and by 540 MW in 2015.
The results showed that in the case when new power
plants are not built, but technical features of all existing
power plants are maximally exploited, the demand of
Lithuanian consumers with a considerable probability
can be satisfied till the end of 2015 under the basic
demand scenario. Under the high demand growth
scenario it is expected that capacity deficit will start to
emerge in 2014 and in 2015 capacity deficit can reach
135 MW. Although small, however there is a
probability that because of Kruonis HPSP inability to
work, there could be some capacity shortages even in
2010. In this case some consumers should be switched
off. If new electricity generating units were built at
Lithuanian TPP or somewhere else, the problem of
capacity deficit would be postponed several years
forward.
The information presented above took into account only
technical aspects of functioning of Lithuanian electricity
energy system. However economical questions are even
more important. Perspective fuel prices have the major
impact on economical aspects of power system
operation.
1 table. Capacity balance of Lithuanian power system under

basic demand scenario, MW
Demand. Basic scenario

Generating capacity (without
own needs)
Thermal power plants total
Lithuanian TPP
Vilnius CHP-3
Kaunas CHP
Petrasiunai CHP
Mazeikiai CHP
Klaipeda CHP
Panevezys CHP
Industrial PP total
Hydro power plants and
hydro pump storage plant
Biofuel PP total
Wind PP
Not used capacity
Mothballed capacity
Emergency reserve
Balancing reserve
Wind PP
Kruonis HPSP
Capacity used for demand
satisfaction
Capacity
surplus(+)/shortage(-)
2008
1917
2010
2144
2015
2471
4599
3466
3466
1183
2425
1722
344
161
7
148
2425
1722
344
161
7
148
9
2425
1722
344
161
7
148
9
34
34
34
73
73
73
887
887
837
29
52
2384
287
1183
102
52
29
52
1498
287
289
110
52
760
29
52
1519
287
289
131
52
760
760
1968
1947
2214
-176
-525
3. Long-run fuel prices

Oil prices in global market have a significant impact on
the prices of other fuel types. For this reason it is
important to take a look at tendencies of oil prices.
Approximately from 1986 to 2004 oil prices fluctuated
in a range of 20-30 $/barrel. Later oil price started to
grow and in July 2008 it reached 145 $/barrel.
Thereafter oil prices suddenly fell down and they were
less than 40 $/barrel in the beginning of February 2009.
The survey of literature shows that at the moment there
is an absolute uncertainty about future oil price and
therefore price of other fuels. This was excellently
illustrated by data of USA Department of Energy that
were presented in December 2008. The boundaries of
future oil prices are very wide. Prices fluctuate from 50
$/barrel to 170-200 $/barrel. According to the main
scenario the oil price will be near 90 $/barrel during the
period 2010 - 2015. This 90 $/barrel oil price level was
chosen for further analysis of consequences of the
closure of the second unit of Ignalina NPP. With
167
reference to this, prices for other fuels that can be used

in Lithuanian power plants are evaluated and presented
in 2 table.
3 scenario. Conditions of scenario 2 are valid, however

during two months in winter natural gas from Russia is
not supplied with exception of Kaunas CHP.
4 scenario. Natural gas is supplied without any limits
on quantity and regime; however since 2010 there are
no any opportunity to import electricity.
5 scenario. Natural gas is supplied according conditions
described in 2 scenario, however since 2010 there are
no opportunity to import electricity.
6 scenario. Conditions of natural gas supply are the
same as in scenario 3, however since 2010 there are no
opportunity to import electricity.
Additionally the following assumptions have been taken
into consideration: modern 400 MW combined cycle
gas turbine unit at Lithuanian TPP and 320 MW
combined cycle gas turbine unit at Kaunas CHP will
start operation in 2012; the first unit of new nuclear
power plant will come into the market since 2018.
Operation of all power plants correspond to the least
cost criteria.
2 table. Prices of the main energy sources since 2010
Fuel
Lt/t
Lt/ toe
Price of
import,
production
1130
1183
Price
at
Lithuanian
power
plants
1191
1247
Lt/t
Lt/ toe
Lt/th.m3
Lt/ toe
Lt/m3
Lt/toe
Lt/t
Lt/ toe
Lt/t
Lt/ toe
2105
2151
1000
1250
140
837
1100
1152
1200
1256
2145
2099
1080
1350
Unit
Fuel oil S2,5%

produced
AB
Mazeikiu nafta
Diesel 0,2
Natural gas
Wood
Imported fuel oil
S3,5%
Imported fuel oil
S1,0%
1161
1216
1261
1320
5. The main results of the analysis

The performed analysis of power sector development
confirmed the hypothesis that Lithuanian TPP will
become the main source of electricity generation in
Lithuania after the closure of the second unit of the
Ignalina NPP. In the case when there are no additional
limits on electricity and gas import (1 scenario), the
existing units of the Lithuanian TPP will produce
approximately 50% of electricity in 2010-2011. About
26% of all electricity will be produced at CHPs, 7%
will be produced from renewable energy sources and
17% of all electricity will be imported. The role of
existing units of the Lithuanian TPP in 2010-2011
would increase even more if there were no opportunity
to import electricity from neighbouring countries. In
this case these units would have to produce 57-59% of
all electricity needed. Combined heat and power plants,
that also use imported fuel, would produce 35-37% of
all electricity and power plants running on renewable
energy sources would generate about 7% of electricity.
The most difficult period for Lithuanian electricity
consumers will be 2010-2011, when existing units of
Lithuanian TPP will produce the greatest part (59%) of
electricity. It is worth noting that the efficiency of these
units is about 38%. Therefore it is very important not
later than in 2012 to start operation of new modern units
both at Lithuanian TPP (400 MW) and at Kaunas CHP
(320 MW). These units could produce about 5 TWh of
electricity. This will make 40% of electricity needed. In
addition, these units, in comparison with existing ones
of the Lithuanian TPP, for the same quantity of
electricity generated would require about 30% less
natural gas. Therefore a share of existing units (300
MW) in electricity production balance would not
exceed 20% in 2012-2015. However, their role will
remain important for reserve capacity of the Lithuanian
power system, especially in the case when there are
some interruptions in supply of natural gas from Russia.
In the first scenario it is calculated that in 2010-2011
the existing units of Lithuania TPP will consume about
If oil prices were high, oil refineries would be oriented

towards oil deep processing, the opportunities to
purchase fuel oil at reasonable price would decrease and
price of natural gas for Lithuanian power plants
(including transportation costs) would reach 1065-1080
Lt per 1000 m3. During winter price of natural gas can
be 78 Lt per m3 higher if they are supplied to Lithuanian
power plants not directly but from Latvian underground
gas storage. The price of fuel oil depends on both fuel
price and sulphur content. Thus when oil price is about
90 $ per barrel, the price of low sulphur (<1%) fuel oil
will be about 1260 Lt per ton in 2010. The price of high
sulphur fuel oil (till 3.5%) will be about 100 Lt per ton
lower than the price of low sulphur fuel oil.
4. Analysed scenarios
In order to estimate consequences of the closure of the
second unit of the Ignalina NPP, six main scenarios
were analysed. The assumptions made are as follows:
1 scenario. There are no limits for natural gas supply
through the pipeline Minsk-Vilnius, e.g. natural gas is
received when it is needed and as much, as it is needed.
The guaranteed quantity of electricity imported is about
300 MW (approximately 2 TWh) from Russia and
Finland at any time; since 2014 300 MW will be
available from Poland at any time in summer and 300
MW at night in winter; 500 MW in summer and in off
peak load in winter, but only 100 MW during peak load
in winter will be available from 2016. Since 2016 500
MW will be also available from Sweden at any time.
2 scenario. There are no quantity limits on natural gas
supply from Russia but supply regime is base load
(constant). Assumptions regarding electricity import are
the same as for scenario 1.
168
1400 ktoe of natural gas and 320 ktoe of fuel oil. Since
2012 a great role will fall on new combined cycle gas
turbines at Lithuanian and Kaunas power plants.
Therefore natural gas demand at Lithuanian TPP will
decrease till 760-860 ktoe in 2015, from which 430 ktoe
will fall on the new unit of the Lithuanian TPP that will
produce 2.8 TWh of electricity.
If natural gas supply through pipeline Minsk-Vilnius
will be lost, lack of natural gas could be displaced by
supply from Latvian underground gas storage. In
extreme case when electricity import is not possible and
gas supply Minsk-Vilnius is interrupted for two winter
months Lithuanian power plants from Latvian gas
storage will require 300 ktoe in 2010 and 380 ktoe in
2015. One third of this gas (130 ktoe in 2010 and 112
ktoe in 2015) will be consumed by units of Lithuanian
TPP, that have no flue gas desulphurisation equipments.
In this a case electricity to Lithuanian consumers could
be provided without any limits and there were no
environmental violations on SO2 emissions. If natural
gas supply from Latvian gas storage to Lithuanian
power plants werent higher than 120 million m3 (about
100 ktoe), then fuel oil and other liquid fuel would be
fired and this would lead to environmental violations on
SO2 emissions or to limitation of electricity
consumption.
In 2010 nuclear fuel in Lithuania will be changed to
expensive fossil fuel. Mainly because of increased
expenditures for fuel procurement, the costs of power
plants operation will greatly increased 2.6-3.1 times
and will make about 3.0-3.7 billion Lt. This would have
an impact on electricity price, which will increase.
Average annual electricity production price in
Lithuanian power system in 2010-2011 will increase till
31 Lct/kWh (or 89.8 Euro/MWh). When new combined
cycle gas turbine power units start operation, the
average annual electricity production price would
decrease till 26-27 Lct/kWh (or 75.3-78.2 Euro/MWh).
However this price would be 3 times higher than the
price of Ignalina NPP. This electricity production price
will be much higher than electricity price in
neighbouring countries. For example, it is forecasted
that electricity production price in Scandinavian market
will range from 30-45 Euro/MWh in 2010-2015, e.g. it
would be 2 times lower than in Lithuania. In the
beginning of February 2009 the average daily electricity
production price in Scandinavian market was
approximately 40.8 Euro MWh, in Poland and Germany
correspondingly 44 Euro MWh and 50.7 Euro/MWh.
Contrary to these prices the production price of
electricity in Europe-Ural price zone in Russia was
equal to only 12.6 Euro/MWh in January 2009.
The presented information about electricity prices in
neighbouring markets illustrates the importance of
connection of Lithuanian power system with Poland and
Sweden systems. After connection of these systems,
Lithuania will have ability to choose electricity
suppliers and therefore will avoid price intrusion, which
is substantial if there is the only link with one country.
It is worth to note that electricity price to final
consumers is not equal to the price of electricity
production. Electricity transmission, distribution, supply
prices, value added tax should be also taken into

account. Thus the average electricity tariff for
consumers that receive electricity from high voltage
(transmission) network could reach 36.67 Lct/kWh. The
costumers connected to medium voltage network should
pay a price of 43.96 Lct/kWh and 55.05 Lct/kWh if
electricity is supplied from low voltage network.
Electricity price for households can increase till 56.55
Lct/kWh if the value added tax is included. These
electricity prices are calculated assuming upper price
cap values for different services approved by National
Price and Energy Control Commission for 2009. If
electricity transmission, distribution and other
companies for their services will apply lower prices
than foreseen upper cap, electricity price for consumers
will be correspondingly lower. Final electricity
consumers can also expect to pay lower prices, if oil
price in global market will be lower than prices
employed in this analysis, e.g. if oil price is lower than
90 $/barrel. However if oil price is higher,
correspondingly electricity price will be higher too.
6. Conclusions
1.
If 1500 MW of Lithuanian PP could reliably
operate, Lithuania after the closure of the second unit of
the Ignalina NPP would not meet with considerable
technical capacity shortage problems till 2014-2015
even in the case if new generating units were not built
and there werent any electricity import during peak
load time.
2.
If oil price in global market will approximately
be 90-100 $ barrel, then electricity generation costs
would increase more than 3 times.
3.
Electricity tariff for final consumers that receive
electricity from transmission and distribution networks
will correspondingly be 36.7 Lct/kWh and 55 Lct/kWh.
Electricity tariff for household consumers will reach
56.6 Lct per kWh.
7. References
1. Annual Energy Outlook 2009. Early Release. Energy
Information Administration, December 17, 2008.
http://www.eia.doe.gov/oiaf/aeo/pdf/aeo2009_presentati
on.pdf
169

ANALYSIS OF LATVIAN POWER SYSTEM DEVELOPMENT IN THE

KURZEME REGION USING MULTICRITERION OPTIMIZATION
Anatoly MAHNITKO, Alexander GAVRILOV
Riga Technical University, Power Engineering Institute, Latvia
Abstract: Latvian power system is in deficit. According
to prognoses of the Latvenergo for 2015 the most
important question will be where and from whom to buy
electric power. What power and what type of energy
resources power plants we need to build and where
should they be located all of this is widely discussed
in different articles.
The energy department of Latvian ministry of economy
has examined some scenarios of power system
development. In the 2010 the deficit will be already
doubled, and to the 2020 the deficit is assumed to be 2.5
times more then produced power.
In the work it is attempted to analyze development of
Latvian power system in the Kurzeme region. Some
variants of high voltage network development are
considered.
2. Actuality of Latvian power system development

In our days power of Latvia power plants is not enough
to cover demand in any moment, because the mode of
large hydroelectric power plants depends on river
Daugava water level. A state power supply lies with a
base mode of Latvian and neighboring countries power
plants. After Ignalina AES closing in the year 2009, to
prevent continued overload modes in neighboring
countries transmission networks it will be necessary
substantially limit power transfers from Russia. It is
possible that after the year 2010 the accessible
generated power in the Baltic region strongly reduces.
As in Latvian power system is real base power deficit it
is necessary to reconstruct or to build power plants. We
should increase the general set power above 800MW
and ability to operate in the base mode, reconstructing
Riga TEC-2 (400MW) and building power plant in
Kurzeme (400MW) [2, 3].
While analyzing the possible coal power plant placing
variants in Kurzeme as the most perspective are selected
locations in Liepaja or Ventspils (Fig. 1). Each of the
examined variants has advantages and disadvantages.
Liepaja is a larger power demand center and in the
future it will grow up. In the Liepaja variant power
delivery in transmission network will be provided by
building new 330 kV substation in Liepaja and 330 kV
overhead line to substation "Grobina" (about 15 km).
On the other hand in the Ventspils variant it is necessary
to construct new 330 kV substation and 330 kV
overhead line (119 km) to substation "Grobina".
Ventspils, by the side of Liepaja, has a number of
advantages. Ventspils port can provide enter of noice
class ships whole year round. There are 9 deepwater
piers in the port district of the liquid shiploads with
depth between 11.5 and 17.5 m. Necessary coal
amounts for power plant the port can provide already.
To decrease ships unloading time and expensis of
conveyers it is foreseen to build temporary towertype
coal storages [3].
In Liepaja port there are essentially additional means for
coal terminal building and so it can provide service of
ships with a less carrying capacity. In Ventspils, in
contradistinction to Liepaja, is also better developed
Keywords: power system, high voltage line, annual

expenses, environment protection, criterions of
evaluation.
1. Introduction
It is practically difficult to accumulate electric power,
therefore power production, distribution and demand
balance is a very important question in the economy of
any state. Without required adequate networks and
guidance of the systems infrastructure it is important to
provide necessary power generation, to meet
requirements in primary energy and its delivery systems.
Currently power industry development should be
analyzed in a close context with economy and
environment aspects, because still growing demand of
energy resources, available fossil energy resources
decrease and necessity to take care on quality of
environment have pulled out notices of long-term
development principles. Those principles consideration
and implementation provides stable, long-term
development of economic, business competitiveness
increase
and
employment
encouragement,
simultaneously saving quality of environment [1-3].
170
junction that can provide coal delivery by railway. Coal

transport to power plant is planned using closed
conveyer gallery.
Electric power market in Baltic region is opened for all
commercial consumers, but, at that moment, only
formally, because there is no concluded a treaty for
supply direct from sources of generation to end
consumers. Relatively small market of Baltic states is
isolated from other European Union (ES) markets.
Physically the only possibility of trading for each of
Baltic states is with one market participant in Russia
(IterRAO) [1].
One of the most important question company
Latvenergo gets busy at the present moment is Latvia
Sweden connection development. For the Baltic states
connection with Sweden is very important to promote
electric power market development and to provide
region with missing powers. Although, due to the 350
MW cable line (CL) between Estonia and Finland Baltic
states now can buy electric power in the Scandinavian
electric power exchange Nord Pool Spot, cable power
is too small to form valuable market with Western
Europe. For Sweden connection building in Kurzeme
the most decent place is Ventspils region, where cable
line will be shorter and that will promote 330kV
network development in Kurzeme instead of Liepaja
[4].
Power system development primary purposes are: to

provide qualitative and safety power supply at low
expenses, and also minimally affect environment.
Therefore choosing optimal development strategy there
were pulled out such criterions:
C total annual expenses;
AR undelivered amount of electric power in the year

due to emergency failures;
Cek annual expenses to prevent harmful influence on
the environment.
3. Annual expenses of development strategies
Let us examine the annual expenses of transmission
lines. Specific capital investments in 330 kV overhead
lines are K 0 = 3.5 105 /km, but sea cable expenses are
K 0 kab = 1 10 6 /km [4]. Then capital investments in line

sections are, :
K j = K0 l j ,
(1)
Capital expenditures could be defined accepting interest

rate i = 10%, [5]:
CK j =
According to the mentioned reasons there are examined

such Latvian power system in region Kurzeme
development strategies:
1 330 kV OL Ventspils Grobina (119 km) building
and port improvement in Ventspils (coal power
plant building in Ventspils);
2 330 kV OL Liepaja Grobina (15 km) building
and port reconstruction in Liepaja (coal power
plant building in Liepaja);
3 CL Stockholm Ventspils (300 km) building.
The geographical reflection of variants is shown in
Fig. 1.
i
K j .
100
(2)
In constant operational costs calculations it was

accepted that:
for OL ( U nom 35 kV ) pa = 2.5% and p = 1.5% ;
for
sea
CL
( U nom = 110 330 kV ) pa = 3%
and
p = 2% .
Then constant operational costs are, /year:
CEk 1 =
100
( pa + p ) .
(3)
j =1
330 kV OL will be constructed using 2xAC-300/39

mark wires. Wire specific resistance is R0 = 0.049 .
Active power losses due to corona are:
STOCKHOLM
Pk j =
Pk 0 max + Pk 0 min
2
lj ,
(4)
where Pk 0 max active power losses due to corona in

bad whether, kW/km;
Pk 0 min active power losses due to corona in
good whether, kW/km.
Total active power losses in overhead lines at maximum
demand mode:
PL j = 3 I 2 R0 l j + Pk j ,
where I =
Fig.1. Latvian power system development variants in the

region Kurzeme
171
P
3 U n cos
demand current, A.
(5)
P3 f = I 2 R0 l (3 + 11 + 12 + 1V ) + 3 Piz , (14)
For CL Stockholm Ventspils will be used XLPE cable

with cross-section 2000 mm2.
Power losses in wire conductor [6]:
Pvad = I R ;
R = R [1 + 1.7 ( ys + yt )] ,
2
where 11 first outside phase metal screen losses

coefficient that observe metal screen active
resistance and proximal cables reactive resistance
( 11 = 0.53 0.68 );
(6)
(7)
12 second outside phase metal screen losses

coefficient ( 12 = 0.35 0.44 );
1V central phase metal screen losses coefficient
( 1V = 0.22 0.36 ).
where R conductor resistance to alternating current

at 90C, / m ;
ys surface effect factor;
yt closeness effect factor;
R conductor resistance to direct current at
90C, / m :
)]
R = R0 1 + 20 p 20 ,
Total active power losses in CL:
PKL = Pvad + P3 f .
(8)
Transmission network parameters are taken from

Latvenergo homepage [7]: = 0.043 103 /MWh;
1000
conductor resistance at 20C;
where R0 =
F
conductor
specific
conductivity
= 5.27 103 /MW.

Maximum losses hours count, h:
Cu = 32 m mm 2 ;
20 = 0.00393 1 C temperature coefficient for
= (0.124 + Tmax 104 ) 2 8760 ,
copper;
p = 90C maximum allowed temperature for
xs2 =
8 f ks
10 7 ,
R
CEm L j = PL j ( + ) .
(9)
(10)
CE j = CEkj + CEmL j .
;
2
x
1.18
d dz
d dz
yt =
0.312
+
4
4
s
192 + 0.8 xt s
192 + 0.8 x 4 + 0.27
8 f kt
xt2 =
10 7 ,
R
(18)
Total annual expenses for lines, :
f = 50 Hz frequency.
Closeness effect factor:
2
(17)
Total operational expenses, :
where k s surface effect coefficient;
4
t
(16)
where Tmax maximum time of usage ( Tmax = 5000 h).

Variable operational costs for lines, :
cables with plastic insulation.

Surface effect factor:
xs4
ys =
;
192 + 0.8 xs4
(15)
C j = CK j + CE j .
(19)
Total strategies 1 and 2 annual expenses taking into

account capital expenditures for port improvement:
(11)
C j = C j +
(12)
i
KP j ,
100
(20)
where K P j port reconstruction capital investments in
where kt closeness effect coefficient;

s distance between wire axes ( s = 204 mm );
Ventspils and Liepaja. As it was told above

Ventspils port reconstruction include temporary
towertype
coal
storages
building
6
( K P1 = 1.5 10 ). It is necessary to build coal
d dz wire diameter with screen ( d dz = 102 mm ).

Active power losses in cable insulation, kW:
terminal in Liepaja ( K P 2 = 25 106 ).

2
Piz = U nom
C0 tg l j ,
It was assumed that every year demand increases by

1.5%. Thereby there were gotten annual expenses for
every strategy 1 , 2 and 3 (Table 1 3), where Kd
demand increase coefficient.
(13)
where C0 cable specific capacity ( C0 = 0.23 F / km );

angular frequency;
tg insulation dialectical losses angel tangent.
For 330 kV cables with XLPE insulation
tg = 0.0001 0.0002 .
Power losses in cable outside metal screen, kW:
172
Table 1. Annual expenses for strategy 1
Using the accepted demand increase coefficient (1.5%

per year) there were gotten results about undelivered
power for each strategy if line is disconnected (table 4),
where AR 1 , AR 2 , AR 3 undelivered power,
C 1 ,
Kd
CEmL1,
CE1,
C1,
1
1.015
1.03
1.045
1.06
962715
990036
1017765
1045901
1074443
2628715
2656036
2683765
2711901
2740443
6793715
6821036
6848765
6876901
6905443
6943715
6971036
6998765
7026901
7055443
accordingly, for strategy 1 , 2 , 3 .

Table 4. Undelivered power for strategies 1 , 2 and 3 if
line disconnects
Kd
Kd
1
1.015
1.03
1.045
1.06
1
1.015
1.03
1.045
1.06
C 2 ,
CEmL2,
CE2,
C2,
121351
124795
128290
131836
135434
331351
334795
338290
341836
345434
856351
859795
863290
866836
870434
3356351
3359795
3363290
3366836
3370434
AR 1 ,
AR 2 ,
AR 3 ,
MWh
1500000
1522500
1545338
1568518
1592045
MWh
1160.25
1177.65
1195.32
1213.25
1231.45
MWh
146.25
148.89
151.27
153.70
156.16
5. Damage to the environment
Alienated territory area [6], m2:
Kd
1
1.015
1.03
1.045
1.06
CEmL3,
356249
360516
364845
369239
373696
S j = a l j ,
C 3 ,
CE3,
15356249
15360516
15364845
15369239
15373696
where lj transmission line terrestrial length, m. For CL

Stockholm Ventspils terrestrial section length is
20 km;
a trace width, m. 330 kV OL trace width is 68
m; CL tunnel width ac = 6 m.
Damage to the environment for the alienated areas, :
45356249
45360516
45364845
45369239
45373696
Y j = S j cY ,
4. Loses for national economy if line breaks off
d ,sp kab = 0.00072 discon./yearcount. Then for every

strategy transmission lines emergency switching off
parameter in the year:
CKY j =
OL emergency disconnection duration (330 kV OL

emergency switching off average duration 1.3 103
year/discon.) is, h:
Td , L = 1.3 10 8760 = 11.4 h .
CY j = S j y ,
d j =
(22)
Cec j = C KY j + CY j .
.
(28)
(29)
(23)
Assuming that annual cost increases by 5% due to
inflation there were gotten annual damage expenses for
strategies 1 , 2 and 3 , accordingly, Cec 1 , Cec 2
Undelivered by customers energy during the year (Pmax

= 300 MW; Tmax = 5000 h), MWh:
AR j = d j Pmax Tmax .
(27)
where y region annual specific costs for ecology (in

this work it was accepted y = 0.5 /m2).
Total annual costs for the damage to the environment
and its renovation, :
Emergency disconnection outage probability in the year:
d j Td , L
i
Yj .
100
Annual installment for the damaged nature renovation,

:
(21)
(26)
where cY damage to the environment specific price,

/m2 (in this work it was accepted cY = 10 /m2).
Capital expenditure for damage to the environment if
interest rate is i = 10%, :
Emergency switching off load flow parameter in the

year [5] for the singlecircuit 330 kV OL is
d ,sp = 0.005 discon./yearcount, but for CL -
d j = d ,sp l j .
(25)
and Cec 3 (table 5).
(24)
173
Table 5. Damage to the environment for strategies 1 , 2
Table 7. Imprecise relations possession function
and 3
Year
Kd
Kd
0.15309
0.07400
1.015
0.15368
0.07407
1.03
0.15428
0.07414
Cec 1 ,
Cec 2 ,
Cec 3 ,
1552500
180000
1578375
183000
1.045
0.15488
0.07421
0.15550
0.07428
12316500
1.015
12519722
R1 C , S
1.03
12712681
1604250
186000
1.06
1.045
12911798
1630125
189000
0.12605
0.27925
1.06
13106809
1656000
192000
1.015
0.12643
0.28009
R2
~
AR , S
1.03
0.12655
0.28037
6. Summarization of results and optimal strategy

detection
1.045
0.12668
0.28065
1.06
0.12681
0.28093
Perspective demand possessive function was gotten

using expert judgement method (table 6), where S
0.12605
0.01461
expert judgement rationed possessive function; Kp

demand increase coefficients.
R3 Cec , S
Table 6. Perspective demand possessive function
S
Kp
0
1
0.28
1.015
0.51
1.03
0.15
1.045
0.06
1.06
annual expenses; AR undelivered amount of electric

power in the year due to emergency failures; Cec
annual expenses to prevent harmful influence on the
environment.
~
Total annual expenses C , undelivered amount of
~
electric power AR and annual expenses to prevent
~
harmful influence on the environment Cek indirect
~
groups are demand power S indirect function groups,
~ ~ ~
that form indirect relations R1 , R2 , R3 with possessive
0.12607
0.01462
0.12619
0.01463
1.045
0.12625
0.01464
1.06
0.12635
0.01465
0.15428
0.07414
0.51
AR
0.51
0.12655
0.28037
Cec
0.51
0.12619
0.01463
Table 8 cells form matrix

A = [aij ] , i = 1, 2, 3; j = 1, 2, 3 .
As it is known from [9], matrix A saddle point
aij is
defined using the condition max min aij . Saddle point is

i
functions R~ , R~ , R~ . There are shown functions

2
1.03
Table 8. Target function values
There are summarized selected strategies 1 , 2 and

3 partial criterions in the tables 1-7, where C total
1.015
appropriate row minimum element and appropriate

column maximal element. In this case matrix A row
minimum elements are:
normalized
values
using
the
expression
R (K , S ) = K (S )i K (S )max in the table 7, where K
corresponding partial criterion denomination [8].
Indirect target possessive functions (table 8) were gotten
using expressions:
0.07414; 0.12655; 0.01463.

Matrix A columns maximum elements are:
C = max[min ( R~1 , S~ )] ;
0.51; 0.12655; 0.51.
Ar = max[min ( R~2 , S~ )] ;
It could be concluded that
Cec = max[min ( R~3 , S~ )] .
max min aij = 0.12655 .

i
Element a22 is saddle point, that corresponds with

strategy 2 330 kV OL Liepaja Grobina (15 km)
building and port reconstruction in Liepaja.
174
3.
7. Conclusions
The result of proposed algorithm of Latvian power system

development in region Kurzeme allow to get practical
justification from the point of exercise solution
multicriterion analysis.
4.
5.
8. References
1.
2.
http://www.em.gov.lv // Enertikas attstbas

pamatnostdnes 2007.-2016.gadam
Mahitko A., Gavrilovs A., Gerhards J.
//
:
,
,
.
III
-
13-16 2008. 2, III
-
, Krievija, Ekaterinburga, 13. 16.
oktobris,
2008.
274277.
lpp.
6.
7.
8.
9.
175
http://www.em.gov.lv // Elektroenerijas raoanas

jaunu
bzes
jaudu
ievieanas
scenrijs.
Informatvais ziojums.
Zviedris I. PSO redzes lok gan bzes jaudas, gan
jauni starpsavienojumi un droba // Energoforums.
Enertikas nozares jaunumu apskats, diskusijas,
analze. 2008. Nr. 4 (14) 8. 9. lpp.
Krins Z., Oeinikova I. Elektroenertisko
uzmumu vadbas pamati. R.: RTU
izdevniecba, 2007. 160 lpp.

/ . ... .:
, 2005. 320 .
www.latvenergo.lv
..

//
. 2003. 3 14 22 .
.. . ,
, 1975. 320 .

BALANCE MANAGEMENT IN THE BALTIC COUNTRIES AND POSSIBILITIES

FOR HARMONIZATION
Aura PARAIT*, Mindaugas KRAKAUSKAS *, Giedrius RADVILA**, Vidmantas BALTAKIS***
*Lithuanian Energy Institute, Lithuania; **Lietuvos Energija AB, Lithuania; ***Sprendim erdv, Lithuania
balance management and contains two different

functions:
Balance regulation,
Imbalance settlement.
Providing balance regulation the TSO in each country
activates regulation for keeping the frequency stable
and within the required rage in the network. Regulation
is based on a number of different reserves, such as:
Automatic
- Frequency controlled normal operation
reserve,
- Frequency controlled disturbance reserve,
- Voltage controlled disturbance reserve.
Manual
- Fast active forecast reserve,
- Fast active disturbance reserve,
- Slow active disturbance reserve.
Most of these reserves are activated automatically,
whereas others are activated manually according the
TSO commands. These reserves can be provided by
special power plants which are not in ownership of the
TSO, particularly, after unbundling. Therefore,
appropriate procurement schemes are needed to activate
these reserves. In other words, activation of reserves
under such schemes means power generated or not
generated according the TSOs requirement, and it is
called regulated power. The EU directive on internal
electricity market in article 11 states that balancing
should be organized transparently, no discriminatorily
and on market based principles [2].
Regulating power market (RPM). The regulating power
market is a tool for TSO's to keep balance between total
generation and consumption of power in real time. The
participants in the RPM are producers and consumers
who can regulate their generation or consumption on
request from TSO's. RPM is real-time market covering
operation within the specific hour. The main function of
the RPM is to provide power regulation to counteract
imbalances related to day-ahead operation planned. The
TSO alone makes up the demand side of this market and
approved participants on the supply side, including both
electricity producers and consumers. If the vendors
Abstract: An overview of the balancing mechanisms in

Baltic countries and the ways of harmonization is given
in the paper. The role of Russia in the balancing process
is presented as well. Lithuania has created the balancing
mechanism based on the hourly regulating power
market. It would be worth Latvia and Estonia to join
this market and to develop it into common reserve
market. The common market would lead to better usage
and lower costs of the reserves. And the most important
outcome could be the first step to the common Baltic
electricity market. As it is shown in the paper, this
balancing mechanism is very close to the Nordic
system.
Keywords: transmission system operator, reserve,
balance regulation, regulating power market, balancing
power market.
1. Introduction
Each power system is always in balance; generation is
always equal to demand, including technological looses.
On an AC system, this rule is maintained in real time by
the frequency, deviation of which from nominal (50 or
60 Hz) represents the difference between the 'required
demand' and 'delivered demand'. The frequency has to
be kept within strict limits to avoid system degradation;
usually +/- 1% for normal operation. If the frequency
deviates beyond 2%, automatic disconnection and
measures are necessary to arrest the slide and avoid
collapse [1]. Imbalances in the physical trade on the
spot market must be levelled out in order to maintain
the balance between production and consumption, and
to maintain power grid stability. Totalling the deviations
from bid volumes at the spot market yields a net
imbalance for that hour in the system as a whole. The
transmission system operator (TSO) is the body
responsible for securing the system functioning in a
region. Within its region, the TSO controls and
manages the grid. The TSO is in need of a mechanism
for balancing the amount of power fed into and taken
out of the system at all times. This mechanism is called
176
offers or buyers bids on the spot market are not

fulfilled physically, the regulating market comes into
force to keep system in balance.
Balancing power market (BPM). Participants in this
market, in many cases, are passive, they are consumers
or generators, and they are not enabled to adjust their
consumption or generation in real time during delivery
period according the TSO instructions. They are the
main reason why imbalances occur. The basic principle
for settling imbalances is that participants causing or
contributing to the imbalance will pay their share of the
costs for re-establishing the balance in the system.
Imbalances in other words mean power deviation from
generation or consumptions plans and it is named as
balance power. Balance power is always automatically
purchased or sold by the TSO. Balancing power market
is linked to the RPM and handles participants
imbalances recorded during the previous 24-hour period
of operation. The TSO alone acts on the supply side to
settle imbalances. Participants with the imbalances on
the spot market are price takers on the balance market.
The Regulating power market and the Balancing power
market may be regarded as one entity, where the TSO
acts as an important intermediary or facilitator between
the supply and demand of regulating power. Regulating
Power and Balancing Market is an important tool for
managing the balance and stability of the system. If the
grid is congested, the market breaks up into different
price area markets, and equilibrium must be established
in the each area.
The problem in the future to be sold is related to the fact
that electricity markets under globalization process tend
to integrate in the single market when system operation
will remain under national TSOs control, so
harmonization issues in the balance management
process across countries have an important role to
guarantee competitive and secure global electricity
markets.
Therefore it is important to discuss about balancing
problems not only in advanced market countries, but
also about possibilities to adopt their experience in such
emerging electricity markets as Check, Croatia, Estonia,
Latvia, Lithuania, and a.c.t. [3,4,5]. These opportunities
are discussed in more detail for Baltic countries,
particularly, Lithuania. Additional requirement for
harmonization of electricity market between Nordic and
Baltic countries is related with the new interconnection
between Estonia and Finland [6], and planed
interconnection between Lithuania and Sweden [7].
Nordpool Spot AS has already made significant efforts
to expand its market towards the Baltic countries as the
separate price area, called Estlink [9,10]. The
establishment of Estlink price area will facilitate the
creation of common Baltic electricity market. The
creation of viable and credible common Baltic States
power market would secure the conditions for free trade
in electricity as well as higher level of energy security.
Viable and transparent market regulations based on
Nordic experience should create conditions for the
establishment of the transparent common Baltic
electricity market.
2. Regulating and Balancing Power Trading

In Lithuania
Only TSO of Lithuania has right to trade regulating and
balancing power with other power systems TSOs or
open balance electricity suppliers in Baltic. The internal
market of Lithuania is determined by the physical
territory of the country. The main players of regulating
and balancing powers are TSO, balance providers,
generators and open balance supplier. The main
activities of TSO for managing the system balance are
represented by the following actions, as show in
Table 1.
Table 1. TSO actions
Time line
Before the
delivery day
Before
operating hour
During
operating hour
After
operating hour
Actions
Calculates transmission capacity limits,
losses and forecasts consumption
Receives information on the volumes of
fixed transactions. Receives regulation
power bids
Monitors national balance, regulates if
needed and registers regulating energy
transactions
Collects measurement data and balance
information
Calculates the imbalance for each hour
According the law, all market players has mandatory to

provide the bids for regulation power. The regulation
power auction is organized by TSO and is done
according Trading rules at auction. The auction of
regulating power is part of common trading system in
Lithuanian power market governed by Market operator
and TSO. Legal framework of Lithuanian power market
defines the rules of electricity trading. After gate
closure of spot market, each producer must submit
available capacity for reserves and regulation power
auction. The available capacity for RPM is defined as
the difference between maximum available generation
capacity and used capacity for commercial
commitments defined by spot market transactions.
Firstly, the TSO uses available capacity for ensuring
emergence and stand-by reserves. The remaining
capacity is submitted to the regulation power auction
which is organized by TSO.
Each market participant submits the following
information:
Available maximum and minimum generating
capacity - possible technical minimum and
maximum of power stations operation.
Capacity price for 1 MW.
Energy price for 1 MWh of used capacities by
TSO.
Only after signing the regulation-balancing contract, a
market participant may submit proposals for system
services i.e. sell reserved capacity and participate in
regulating power auction.
TSO must ensure the required amount of reserve
capacity for the internal system. Reserve capacities are
used in lower price priority order from those production
sources, which meet technical requirements for
177
providing the ancillary services.. However TSO has

right to, select reserved capacity sources without taking
into account the electricity price priority principle. For
example, Lithuanian power plant, being strategic in
respect to capacity security, has a priority for
maintaining operative and stand-by reserve since it has
a sufficient capacity potential after 12 hours to change
the capacities of Kruonis hydro pumped storage power
plant in case of emergency disconnection of Ignalina
nuclear power plant.
Regulating power auction is single-sided, i.e. auction
participants submit bids or offers for regulation and
TSO acts as the single buyer or seller of regulation
power. Ensuring that regulation in the system will be
carried out under most optimal prices, TSO groups the
proposals according the low price priority principle
Performing system regulation by compensating the
deficit which occurs during the operation hour, TSO
purchases power from the producer, who suggested the
lower regulating power price, and when there is a
surplus in the system, he sells power with the highest
price, which being submitted by the producer. It means,
that TSO may reduce his own price, and when there is a
surplus in the system sells power at the highest price,
proposed by the producer, which means that he may
reduce his own production volume. When there are
congestions or bottlenecks in transmission grid,
regulation may be performed without following the
price priority principle, however activate regulation
proposals, taking into account technical possibilities and
reliability of the system. This alternative provides more
opportunities for TSO. It can operatively and
technically coordinate and control system in emergency
cases, when regulation resources must be forecasted for
several hours or a day forward. Activating several
proposals, a common regulating power price in a certain
hour is calculated as weighted average price of all
regulation power transactions prices.
The maximum regulating power price is restricted by
the open balance supply price, defined, according to the
TSO balancing agreement with the open balance
provider from Russia. If the regulating power price is
higher than open balance provider price, the regulation
is not performed in the Lithuanian power system, and
the open balance supply ensures the needed electricity
which is purchased from the Russia. Only due to
congestions and restrictions of system reliability margin
the regulation must be performed internally and
regulation power bids or offers may be activated,
whenever the price is higher. The average regulation
power price is basis for the setting the system balancing
electricity price.
There are two options of maintaining the system
balance in Lithuania. In the first case, TSO activates
regulation up bids and buys electricity from
regulation power auction or open balance supplier. The
aggregated transactions allow to set the balancing
power price for producers and market participants, who
participate in the hourly trading and are under balance.
The balancing power price defined as the weighted
average price of all transactions made by TSO
multiplied by the coefficient which is determined by the

regulator, i.e. 1.2.
In the case of surplus, regulation down is done by
TSO and purchasing price of balancing power is
formed analogically, however the price is reduced by
multiplying the regulation down and open balance
supplier weighted average price by coefficient 0.8.
If regulation was not performed in the system, price of
balancing power is calculated by multiplying the price
of open balance supplier electricity price by mentioned
coefficient appropriately.
A unique method of balance settlement for consumers
or suppliers who have the consumers is adopted in
Lithuania. The regulation costs are included into
supplementary power price, which is settled and defined
by Market operator to the costumers after the month.
All consumers share costs of imbalances. The applied
method would have to stimulate balance responsible
parties to become balance provider or choose balance
provider and, therefore, participate in hourly trading.
But for the time being the situation does not changes
very much and the implemented method sends the week
signals of balance responsibility for consumers.. The
TSO has cheep regulating power sources in his own
hydro pumped storage power plant and hydro power
plant and rather cheep balancing power from Russia.
The avoidance of additional costs for administration of
hourly trading and balancing responsibility, leads the
situation, when suppliers agree to share costs by
ignoring their own contribution. The public supplier or
the last resort supplier gains the biggest benefit from
such situation. Lithuanian balancing mechanism,
particularly, the regulation power auction is open for
Latvian and Estonian players such as:
Generators,
Suppliers,
Eligible customers,
TSOs.
The regulation power auction is open for all market
players across the Baltic States. The participants should
fulfil the following requirements:
Regulation power agreement with Lithuanian
TSO,
Information providing - biding and offering
hour by hour,
The bids must be submitted electronically no
later than 120 minutes before the beginning of
the operational hour,
Minimum power requirement for bidding or
offering- 5MW,
The bid or offer must state the power (+/MW), price (euro/MWh),
It must be possible to carry out the regulation,
which has been bid, to its full power in 30
minutes,
Balancing responsibility.
Lithuania has made attempt toward single Baltic market
for reserves and regulating power.
178
energy and also to ensure the TSOs justified

profitability), but in the Latvian system, the coefficients
for the balance energy purchase and delivery for market
participants are fixed and published quarterly on the
Latvenergos website. On the other hand, in the
Lithuanian system, the balancing energy price is a
market-based price.
In the Estonian system, the grid tariff includes costs of
withholding
the emergency reserve capacity on
annually basis and the price of balancing electricity
includes costs of activating the aforementioned
emergency capacity and keeping/activating the
regulating capacity [13]. The grid tariff covers
emergency reserves capacity and energy and regulating
capacity in the Latvian system [14]. Lithuanian grid
tariffs includes costs for providing the emergency
reserve for Ignalina nuclear power plant, but there is no
additional costs related to the regulation or other kind of
capacity withholding as in other Baltic countries. There
are differences of the balance management system in
each Baltic States. Significant differences exist on legal
basis for promoting common electricity market.
Lithuanian market is fully opened, according the EU
directive, by issuing the following legal acts, as shown
in the Table 2.
3. Differences in Balance Management across Baltic

States
Baltic counties are in progress on creation of common
electricity market in the region. The integration of
Baltic region in to the European electricity system will
require the implementation of adequate organized
market principles and harmonization of existing
mechanisms, particularly related to the capacity
allocation, congestion and balance management. TSOs
have published the common approaches and differences
of balance management in the Baltic countries.
Latvian and Estonian TSOs are responsible for national
balance settlement. In both systems, Market participants
must have an open balance agreement with an open
supplier to provide balancing in their responsibility
area. The TSO performs a balance settlement for the
market participants connected directly to the grid. In
Estonia, the TSO settles the balance power between
TSO and balance providers. Estonian TSO has
responsibility to provide the system services for TSOs
of neighbouring countries. Similarly, in Lithuania, the
TSO settles the balancing energy between TSO and
other neighbouring countries, but in addition, the
regulation energy as well.
Then, in Latvia, the TSO is responsible for performing
the balance settlement for the traders who have been
delegated by the generators/consumers.. Latvian TSO
also performs a balance settlement for the DSO, which
in their respective area provides a balance service.
Lithuanian TSO is responsible of handling the national
balancing function for all market players as the sole
open balance supplier for internal maker participants
and neighbouring countries TSO. The TSOs provide
uniform, non-discriminatory and competitive conditions
for all market participants.
Lithuanian TSO has implemented organized RPM
principles based on Nordic experience within the own
control area. The mandatory regulation power auction is
organized for market participants as the tool for TSO to
provide the regulation and balancing services for the
power system.
The number of final balance settlements in the Estonian
system is 1, as generation and consumption are viewed
together. In the Latvian system, if the generator and
consumer is under one trader, they are also viewed as 1
in the settlement. In Lithuania, the volumes and prices
of balancing energy are estimated by the Market
Operator separately for consumption and generation. At
the moment, there are 4 Balance Providers in Estonia
and 1 in Latvia. All market participants of regulating
power market are balance providers in Lithuania. The
sole open balance provider sets the price for balancing
electricity in Estonia and Latvia. In Lithuanian power
system the reference RPM price is defined by
combining the regulation power auction price and open
supplier price. Here is where the systems differ quite a
lot. In Estonia, the price of balance energy in the system
has to cover the justified expenses of TSO (the justified
expenses incurring in the purchase of regulating
capacity, in connection with balance determination and
in connection with the purchase and sale of balancing
Table 2. Legal steps to improve electricity market in

Lithuania [15]
Legal steps for markets development
Law on the reorganization of vertical
integrated electricity company JSC
Lietuvos energija
The 1st edition of Law on electricity
Trading Rules of Electricity
Beginning of the electricity trading
under market conditions in wholesale
electricity market
Auction Trading Rules
The 2nd edition of Law on electricity
Date
May 18, 2000
July 19, 2000
December 18,
2001
January 1, 2002
April 18, 2003
June 1, 2004
Existing aspects of Estonian electricity market opening,

mentioned in [16], could be summarized as follows.
Customers with consumption over 40 GWh per year
become eligible customers. Only eligible customers
have the right to choose a supplier. They have also the
right to apply for an import license, which is issued by
the Electricity Market Regulator. At present we have in
our system 13 eligible costumers who consume about
16.5 % of energy in Estonia. Starting from January
2009, 35% of market must be opened, but the decision
is not made at the moment. The opening of market is
planned to be completed to the full extent by the year
2013. Non-eligible customers can purchase their
electricity from the grid company they are physically
connected to or from the seller named by that grid
company.
The regulation tariffs applied in the Estonian power
system are main obstacle to facilitate the real market
opening and implement Nordpool Estelink price area
project [8]. Main consumers are not allowed to buy
electricity from the market and they are not uses the
179
eligibility rights to change the suppliers until the

regulated tariffs are lower the market price in Estonia or
external markets.
this purpose Lithuanian TSO has signed the balancing

power supply agreement with the representative of
Russia. According this agreement Lithuanian deviation
from planed balance could not exceed 50 MW in both
directions. The limits are determined taking into
account systems reliability and capacities of
interconnected lines. Lithuania with other Baltic
countries attempt to minimize the imbalance flows with
Russia by implementing and improving internal
balancing principles, but unfortunately with lack of
cooperation. Another reason for hindrance of aggregate
balance responsibility, that separate imbalances in
different Baltic state are important for Russia from
technical view of point.
4. Russias Role in the Balancing Baltic Power

System
Lithuanian power system, as well as power systems of
others Baltic States, is a part of UPS/IPS synchronous
area. The Russian UPS/IPS system has significant
influence and plays the main role for balancing the
Baltic region power system. Physically all imbalances
in Baltic region are handled by Russian UPS/IPS
system. Practically the open balance supply service is
provided by one Russian balance provider. Estonian
TSO differently from the Lithuanian one signed this
open balance agreement not with Russia, but with the
Latvian TSO [17]. But finally both these countries are
balanced by Russian open balance provider due to
Latvian open supply agreement.. Frequency is
maintained centrally in Russia, i.e. only a few large
generators perform primary frequency regulation in the
whole system. The insensibility zone of other generators
is determined to be rather big so that small frequency
deviations from the determined 50 Hz would not impact
the planned operating regime of generators. Time
deviation of such system is usually higher and such
strict requirements are not applied to it as in
decentralized frequency control interconnected systems:
UCTE or Nordel.
No unified system and balancing services market, which
would enable on equal and non-discrimination
principles to participate producers in regulating and
balancing power markets or provide others system
services is developed in Russia. Russia has huge
resources, which enable to ensure the quality of power
in allowable limits, without involving small parallel
power systems.
But they are themselves financially responsible
(according bilateral contracts) for the balancing service,
provided by Russia. Since Lithuanian power system is
part of Russian power system and operates under
parallel operation agreement coordinated by Russia
TSO, it is not participating in the primary balance
maintenance.
Though all generators installed in Lithuania have
technical possibilities to ensure frequency standards in
accordance with the requirements, they cannot be used
in the primary frequency regulation due to particularly
small impact on Russian power system. But, on another
hand, Russia itself may strongly affect Baltic power
systems. Lack or surplus of generation in Russia may
influence frequency changes, it cause the changes in
Lithuanian or other Baltic states generating capacities in
that direction to compensate these frequency changes.
But these changes are destroying the Lithuanian balance
between planed production and consumption. Thus
production and consumption balance is maintained by
keeping defined power flow level on the interconnected
lines between defined limits. These limits are
determined by TSO after evaluating Lithuanian
impacts value to the Russian and Baltic systems. For
5. What is the most important in harmonization

process?
In the year 2001 there was an attempt to harmonize a
system services market by implementing the common
Baltic reserve and regulation power market. The
common project was not successful and experienced
total failure due to different legal environment in
different Baltic States.
The Nordic BPM shows a practical example of such
cooperation with different approaches across Nordic
countries. They established single market only for
regulating power; it means a single market of all kind
reserves. At the same time, trading schemes of balance
power remain according national rules in each country.
Several years experience shows that such reforms fully
respect responsibilities of the Nordic TSOs. Thinking
about new harmonization attempts in the Baltic region,
the Nordic experience should be taken in to account
more than only regulating power market was
harmonized after 2002. Changes in operation of
reserves related with legal basis of power systems,
when changes in balance power trading usually relates
with more wide legal base included consumers right and
protection or even constitutions issues. Furthermore
single balance responsibility of Baltic States to the
Russia is not acceptable from Russia side. From this
point of view, Lithuanian role is important, because
only Lithuania has developed a regulating power market
in the Baltic region under unique principles of
organized market and based on Nordic experience.
5. Conclusions
After unbundling power system, balance management
process becomes not only technical but commercial
problem as well. Regulating power market and Balance
power market shall be established.
Trading mechanism of regulating power in Lithuania is
close to the Nordic model. The main difference is that
price averaging procedure is included in Lithuania to
minimise balancing cost.
Trading by regulating power and balancing power has
significant differences across Baltic States, but
differences related with regulating power could be easy
harmonized.
180
7.
It is not acceptable for Russia to treat Baltic States as a

single balance unit.
There is no big obstacle to harmonize reserves
operation and trading of regulating power between three
Baltic States using Lithuanian experience.
Furthermore, it is no need to start to harmonise total
balance management system; it is enough to establish
integrated regulating power market.
It is very important to assure balance responsibility in
Baltics power markets in a way to enable the
functioning of the common Baltic market while keeping
technical integrity of the system in a decentralized way.
The harmonised balance management mechanism
would be beneficial for all market players who are
currently active and the ones yet preparing to become
active in the Baltic region.
The common Baltic balancing mechanism is important
precondition for creating the common day-ahead market
in Baltic States.
8.
9.
10.
11.
12.
6. References
1.
2.
3.
4.
5.
6.
13.
Future Power Systems 1 - the Balance principle,

Frequency and the Grid Submitted by Stephen
Browning on Wed, 2007-11-28 12:48.
http://www.leonardo-energy.org/drupal/node/2422
Directive 2003/54/EC of the European Parliament
and of the council concerning common rules for
the internal market in electricity and repealing
Directive 96/92/EC.
Rules on balancing the electric power system in
Croatian.
www.hep.hr/ops/en/documents/RulesOnBalancing
.pdf
Agreement on the access to the balancing market
with regulating energy in Check Republic.
http://www.ote-cr.cz/the-participants/registrationprocedure/files-registrationprocedure/W_Agreement_bez_revizi_080118.pdf
Official internet site of Lithuanian power
company. www.lpc.lt
Baltic-Finish
submarine
cable.
http://www.nordicenergylink.com/index.php?id=2
9
14.
15.
16.
181
Baltrel
annual
report
2007.
http://www.baltrel.org/annual_report_2007.pdf
Nord Pool Spot Baltic project No.97/2008.
http://www.nordpoolspot.com/en/Market_Informat
ion/Exchange-information/No972008-Nord-PoolSpot-Baltic-project-ends-31-December-2008/
No.55/2008 Introduction of implicit auction via
the Estlink cable by 1 July 2009 at the earliest .
ion/Exchange-information/No522008Introduction-of-implicit-auction-via-the-Estlinkcable-by-1-July-2009-at-the-earliest/
No.27/2007 Joint Baltic-Nordic project launched.
ion/Press-releases-list/No272007-Joint-BalticNordic-project-launched/
The Nordic Power Exchange and the Nordic
Model for a Liberalised Power Market
http://www.nordpoolspot.com/upload/Nordic%20p
ower%20market/Nordic%20power%20market.pdf
Sweden grid company official internet site.
www.svk.se
Electrical Power System Code. The Act of the
Government of Estonian Republic No.184,June.
http://www.mkm.ee/failid/The_Grid_Code.doc
Latvian grid code.
Legal acts about electricity market in Lithuanian.
http://www.ukmin.lt/lt/veiklos_kryptys/energetika
/elektra/
Official internet site of Estonian Power system.
http://www.pohivork.ee/index.php?id=311&L=1
Balancing principles for Latvia and Estonia, Arnis
Staltmanis Nowember 2006. Vilnius, Mini forum.
http://www.energyegulators.eu/portal/page/portal/EER_HOME/EER
_INITIATIVES/ERI/Baltic/Meetings1/SG_meetin
gs/3rd_Baltic%20SG/DD/A._Staltmanis_Balanci
ng.ppt#330,1,Slide 1

INTEGRATION OF TRACKING STANDARD INTO ELECTRICITY MARKETS

OF BALTIC STATES
Inga KONSTANTINAVIIT, Dalius TARVYDAS
Europe Programme, has investigated the feasibility

of a harmonized standard for tracking electricity in
Europe [2]. The major benefit of such tracking standard
is that electricity generation attributes can easily be
accounted, problems of multiple counting of attributes
can be avoided and cross-border trade of attributes can
be facilitated. The E-TRACK standard has been
designed in such way as to support implementation of
European and national policies.
The term standard is used in an informal way. It
indicates the set of rules, which can be applied by EU
countries in order to implement coordinated electricity
tracking system.
Abstract: Electricity supply usually is characterized in

physical terms, such as voltage, frequency and etc.
However, in relation to energy policy and consumer
interest there is a lot of additional information, which
currently becomes very interesting for electricity market
stakeholders and for the final consumer. This additional
information is related to generation of electricity
(primary fuel sources, the related emissions, the
technology used, etc.). The E-TRACK project has
successfully developed a blueprint for a European
electricity tracking standard. Principles of this standard
have been already taken over by several countries.
However, tracking systems used in Europe are still far
from being coordinated. This article provides an
overview of proposed tracking standard for electricity
and analysis of its integration into electricity markets
of Baltic States.
2. European standard for electricity tracking

The main principle of E-TRACK standard is that
information should be separated from physical energy at
the point of generation. Therefore the tracking system
should exist separately from electricity flows.
Electricity tracking system must:
provide meaningful information to the users, e.g.
enabling consumer choice based on disclosure or
facilitating support;
provide accurate results;
be robust against errors and fraud by the actors
involved in the system;
be compatible with the existing economic,
socioeconomic,
regulatory
and
legislative
framework conditions;
be cost-efficient, by providing the services required
at reasonable cost;
be flexible enough to adapt to changing framework
conditions.
The policies which use tracking results are called
schemes. According to the use of tracking results three
types of schemes are allocated: disclosure, quantitative
targets and support.
These different schemes define the information, which
need to be tracked. This information is called the
attributes, which usually is related with electricity
generation. The fuel source coal used by fossil fuel
Keywords: electricity tracking, attributes, schemes,

domains, certificates.
1. Introduction
The purpose of tracking scheme is to establish an
independent, comprehensive accounting mechanism for
generation attributes and to assign these characteristics
to the electricity consumed by final customers [1].
Systems for the allocation of generation attributes to
electricity suppliers and consumers respectively are
required in order to facilitate a variety of policies on the
European level. Taking into account these requirements
three major uses of tracking can be distinguished:
Electricity disclosure (labeling) of information to
final consumers;
Verification of compliance with quantitative targets,
e.g. indicative targets for green electricity for 2010
or 2020;
Management of public support to certain types of
electricity generation, e.g. green electricity or
electricity produced at high efficiency CHP.
The E-TRACK project, partially supported by the
European Commission through the Intelligent Energy
182
power plant is one of the attributes, which will be used,

in the tracking system for disclosure purposes. Another
attribute could be the specific CO2 emissions from this
power plant.
In order to reduce the barriers for cross-border tracking
of attributes, all countries should agree on common
(minimum) set of information. Tracking information
based on the minimum list of attributes should be
accepted within all countries, which are using proposed
standard.
Within E-TRACK standard all tracking activities should
be organized in domain. A domain is defined as a
geographic region for the purposes of a policy scheme.
It is recommended to have one domain per country.
Issuing Body should be appointed as the main operator
of the tracking system in the domain. A common hub
should connect all domains.
The E-TRACK standard distinguishes two generic
options for tracking: explicit and implicit tracking.
Explicit tracking is based on a mechanism, which
creates a link between generation and consumption of
electricity. It can be based on contracts and on
certificates. In the case of contract tracking, all
generation attributes are allocated to consumers based
on bilateral contracts. De-linked tracking is in the case
of certificates, i.e. it is possible to allocate attributes
from generation to consumers independently from
physical electricity market using certificates. Issue of
each certificate is based on corresponding volumes and
attributes of electricity generation. Certificates can be
transferred independently for physical electricity. The
attributes represented in each certificate are used by
redeeming of certificate. After redemption certificates
are removed from the market.
Implicit tracking is based on default set of attributes,
this is meaning that statistical data or averages from a
group of power plants is used as attributes for a certain
volume of electricity.
The E-TRACK standard requires all domains to provide
facilities for explicit and also for implicit tracking of the
attributes of electricity generation. Implicit tracking
should be introduced in order to reduce the burden of
tracking for those parts of the market, where explicit
tracking information is not available. However the ETRACK standard aims to reduce the share of implicit
tracking, i.e. implicit information should be used only
where explicit evidence is not available.
Technically the E-TRACK standard requires to set up a
registry system for handling of attributes. Registries
will allow tracking the ownership of attributes and will
support transfers of ownership within the registry and to
other registries under the E-TRACK standard. A
registry should provide as much open access as possible
to provide transparency and support the accuracy of
reported information. This suggests to have a webenabled registry, which will be the most efficient
solution offering a suitable broad coverage.
There are already a number of operation registries in
different EU countries, some of which are handling
Guarantees of Origin (GO) and other certificates. Most
of these registries are connected to the European Energy
Certificate System (EECS). This system is the most
comprehensive certificate system for electricity in

Europe. However, almost none of them handle all forms
of electricity sources yet. But it would not be necessary
to build an entirely new registry infrastructure, only will
be necessary to develop existing registries further.
Explicit tracking attributes will be recoded in the
registry as transferable electronic certificate. The life
cycle of certificate consist of three steps: issuing,
transfer and redemption.
Issue of certificate will be based on information on
electricity production devices, which will be held
permanently in the registry, and additional information
about certain generation episodes.
The issuing of certificate is voluntary for power plant
owner and operator. They can decide whether to register
their power plants in the registry or not. After
registration, they can decide for each generation episode
to issue certificates or not. Issuing of certificates should
be normally based on full megawatt hours (MWh).
Rounding up of part units of some generation episode
should not be used because this discriminates small
power plants. Any remaining part units should be
forwarded into following generation episode.
Only current owner can use the attributes represented in
a certificate. The seller must initiate transfers of
certificates within the E-TRACK standard.
In order to make use of a certificate for one or several
schemes, the owner must redeem the certificate. After
redemption certificate will be removed from the market.
The owner must specify for which scheme the
redemption is made for and in case of disclosure, for
which retailer and for which electricity product the
attributes are redeemed. After redemption, the registry
operator will produce a redemption statement, which
gives proof of the volume of electricity generation and
attributes.
Almost in all EU countries GO for RES-E and CHP-E
are implemented as tracking certificates, which can be
used for disclosure purposes. Therefore besides GO,
certificates can be issued for any type of electricity
production. This means that all GO should be integrated
into a comprehensive system of tracking certificates.
However, GO for RES-E and CHP-E retain their
specific legal status.
In relation to GO, implicit tracking should not just be
based on statistics of electricity production in certain
region. This is important because explicit tracking, i.e.
GO or support systems for RES-E or CHP-E, will cover
part of the overall production in each domain. Therefore
in order to avoid double counting of attributes should be
used, so called, a Residual Mix, which corrects the
generation statistics in a certain geographic region by all
attributes. The Residual Mix represents all attributes in
a certain region, which have not been allocated to final
consumption of electricity based on explicit tracking.
In order to calculate Residual Mix, the lifetime of
disclosure certificates must be limited. Only after the
end of this lifetime, the Residual Mix calculation can be
started. Any certificates, which have been issued, but
not redeemed until the end of their lifetime, will expire
and their attributes will become part of the Residual
Mix.
183
GO for RES-E adopted on 14 October 2005 by

Regulation of Ministry of Economy No. 4-346
Rules on Issue of GO of Electricity produced
from Renewable Energy Sources,
GO for CHP-E adopted on 19 May 2008 by
Regulation of Ministry of Economy No. 4-206
Rules on Issue of GO of Electricity produced
from High-efficiency Cogeneration.
According to these regulations Issuing Body of GO is
Transmission System Operator (TSO). TSO is
responsible for database administration, registration of
new participants, issue of GO in response to request,
reporting about RES-E production and about issued,
transferred and used GO. But GO was not issued yet
because there was no request for it.
Legislation regarding disclosure of generation attributes
is not adopted in Lithuania.
Major part of electricity consumed in Lithuania is
produced locally, therefore information about primary
energy use for electricity production is easily available
from several sources, including department of statistics.
CO2 emissions and nuclear waste also can be obtained,
but it isnt readily available. Despite information
availability it isnt currently presented to final
consumers even on the TSO, DSO or suppliers
websites.
The Residual Mix for a domain will be calculated based

on following procedure:
Attributes from all power generation in the
domain
Attributes from all disclosure certificates issued
in the domain
Attributes from power generation covered by
GO or support systems
+
Attributes from expired disclosure certificates
(this includes the certificates which have been
imported, but have not been used)
+/- An adjustment of the attribute volume in the
domain with a superior European attribute mix
=
Residual Mix in the domain
The Residual Mix is then allocated to all electricity sold
to final consumers in this region, for which no explicit
attributes information is available. But in general
explicit tracking should be used where possible.
One of the major purposes of the tracking standard is to
make explicit and implicit transfers between domains
possible, i.e. certificates issued in one domain can be
transferred to another domain and can be redeemed
there for one or several schemes.
The objective of the E-TRACK standard is to support
the coordination between the tracking systems used
across Europe. It is not aiming for one uniform tracking
system in all countries, but if individual tracking
systems will be developed in compliance with standard,
these systems will deliver a reliable and cost-efficient
service to the electricity industry, consumers,
governments and regulators.
3.2. Latvia
Latvia already has in place legislation for
implementation of GO for RES-E and CHP-E:
GO for RES-E adopted on 24 July 2007 by
Regulation of Cabinet of Ministers No. 503
Regulations on Electricity Generation from
Renewable Energy Sources,
GO for CHP-E adopted on 6 November 2006 by
Regulation of Cabinet of Ministers No. 503
Regulations Regarding Electricity Production in
Cogeneration.
Ministry of Economy. Rules of GO issuing and the
status with regard to the registry are still unclear.
Therefore GO was not issued yet.
Legislation regarding disclosure of generation attributes
is not adopted in Latvia.
Latvia is net electricity importer. It produces only
around 2/3 of electricity consumed. Major part of
electricity import comes from Lithuania and Estonia,
some from Russia and Scandinavia (Finland). Latvian
TSO, Latvenergo, provides information about local
electricity generation and amounts of imported
electricity. Information about CO2 or nuclear waste is
not available. Currently Latvia doesnt have plans on
introduction of disclosure.
3. Currently implemented electricity tracking

systems in Baltic States
As it was earlier mentioned several EU policies require
to account for certain attributes of electricity generation.
Directive 2003/54/EC requires that electricity suppliers
should provide information to its consumers on primary
energy sources usage, CO2 emissions and nuclear
waste. This information should be specified in the bills
or in promotional materials and should be provided to
each consumer. Currently implementation of these
reguirements in the Baltic States is still lagging behind.
In general, information according to the requirements of
this directive is readily available, but this information is
still is not provided the to the final consumers.
Moreover, monopolistic suppliers, with control of
distribution networks, dominate in electricity markets of
all three countries; therefore there is no real competition
among suppliers to the final consumers. Lack of
competition in electricity markets is the main reason
why Baltic countries are behind the required policy
implementation.
Situation regarding GO for RES-E and CHP-E in all
three countries are similar.
3.3. Estonia
Estonia already has in place legislation
3.1. Lithuania
Lithuania already has in place legislation
for
184
for
GO for RES-E entered into force in January 2005

by amendment of the Electricity Market Act (11
February 2003);
GO for CHP-E entered into force in February
2007 by Amendment of the Electricity Market Act
(11 February 2003).
Estonian TSO. At the request of a producer TSO should
issue to the producer GO. But currently nobody has
applied for GO yet.
Estonia has legislation on disclosure since January 2005
according to the amendment of the Electricity Market
Act (December 2004).
Major part of electricity consumed in Estonia is
produced locally; therefore information about primary
energy use for electricity production is available.
Estonia is the only Baltic country that provides paper
bills for electricity to its consumers. Limited
information of electricity generation sources is available
in these bills, but no information about CO2 is provided.
there is no driving force for electricity tracking or value

added electricity products creation.
Consumers, the main driving force in some EU15
countries, arent very concerned about primary sources
of electricity they use. Only small number of household
consumers expressed willingness to buy green
electricity products and pay extra. There are no strong
consumer organizations in the region that could
advocate preferences for green electricity. Preferences
of industrial consumers are based mainly on price and
reliability of supply. Industrial consumers understand
the importance of eco-friendly electricity production,
but in general are not willing to increase production
costs by buying green electricity.
Regulators of Baltic countries expressed interest in the
possible additional costs that implementation of the
tracking system could impose. They are not entirely
sure, if consumers will benefit from knowledge of
primary energy sources used or pollution created by
production of electricity they consumed.
Electricity suppliers dont see immediate benefits
neither from informing the consumers nor from creating
differentiated electricity products. Marginal number of
consumers that are currently interested in it cant justify
the cost of creating such products. Moreover, the
general (country wide) information about primary
energy sources used are freely available, therefore
interested costumers, even now could find all
information required by the directive.
Decision makers in general agree that EU legislation
should be implemented locally, but they also dont see
any immediate benefits for any of the involved parties.
4. Proposals for integration of tracking standard in

Baltic States
Seeking to participate in common electricity attributes
trading activities across Europe it is reasonable to
introduce E-TRACK standard in Baltic States. Creating
a common domain for these purposes would be
preferable.
All three Baltic counties have already implemented GO
for RES-E and CHP-E systems with database. Power
plants that are currently not covered by GO are regular,
usually big and producing only one type of energy
(electricity). Adding them to the certificate system will
cause less effort and will bear only marginal extra costs.
Based of this we can assume, that introduction of ETRACK standard in Baltic States could be done at
minimal cost. It would allow not only to provide correct
information to electricity consumers, but also enables
participation in EU trade of generation attribute.
Currently there are very limited grid connections to
outside of Baltic States (except from Russia), therefore
major electricity exchanges goes only inside the region.
Creating common domain for three countries, not only
reduce the E-TRACK standard implementation cost, but
also will make intra-region trade less complicated.
In other EU countries the most popular mean of
disclosure is providing information on the bill or with
the bill. In Baltic States only Estonia provides paper
bills for electricity consumers, therefore sending
personal information to each electricity consumer will
cause extra costs. It leave only to mainstream
information media: newspapers (which will also add
extra cost) or electricity supply companys website,
where information could be placed.
6. Conclusions
1. Implementation of the proposed E-TRACK standard
could be beneficial for all three Baltic States
accommodating integration into EU electricity and
its attributes markets.
2. Taking into account current level of available
information E-TRACK standard could be
implemented at the relatively low cost.
3. Based on current electricity market structure of
Baltic States it is preferable to have one domain for
all three countries.
4. Despite current cautions of all stakeholders groups
implementation of E-TRACK standard could have
positive impact on RES-E electricity and overall
market liberalization.
7. References
1. Konstantinaviciute I., Tarvydas D. Tracking
Systems for Electricity. Energy 2007, Nr. 3. P. 33
39.
2. A European Standard for the Tracking of Electricity.
Final report from the E-TRACK Project. - August
2007. 114 p.
3. The E-TRACK Standard August 2007, 22 p.
www.e-track-project.org
5. Position of stakeholders
From provided discussions with major stakeholders and
from questionnaires it is evident that in Baltic States
185

TENDENCIES OF NUCLEAR FUEL UTILIZATION MODES

Gediminas ADLYS, Alvydas JOTAUTIS, Ramnas NAUJOKAITIS Pranas KANAPECKAS
Abstract: The nuclear fuel burn-up in power reactors

have important influence on efficiency of nuclear power
plant and on amount of radioactive waste. It depends on
reactor type and on fuel enrichment. During the last
thirty years fuel burn-up in power reactors was
increased 2-3 times depending on reactor type. Physical
limiting factor of the fuel burn-up in thermal reactors is
increasing amount of the new produced neutron
absorbing fission products during the reactor operation.
The strongest neutron absorbers among fission products
in nuclear fuel of typical PWR reactor have been
calculated using codes APOLLO1 and PEPIN1. This
paper presents an overview of the nuclear fuel origin for
fusion and fission reactors, thermal and fast reactors,
fuel conversion and breeding and trends in the evolution
of nuclear fuel burn-up utilization. Possibilities to apply
modern practices for nuclear fuel utilization are
discussed.
2. Nuclear fusion and fission reactors

Fusion is a very promising future energy option, which
is characterized by almost unlimited fuel reserves,
favorable
safety
features
and
environmental
sustainability.
Most promising nuclear reaction to be used for
generation of fusion power is fusion of deuterium 12 H
and tritium 13 H : 12 H 13 H 24He 10 n 17,6 MeV . This
amount of energy corresponds to the recoil energy of
helium nuclei and neutron. The energy of 17.6 MeV per
fusion process in this reaction is distributed between the
neutron and the helium nucleus as 4/5 and 1/5. It
corresponds to 14 MeV and 3.6 MeV, respectively.
The biggest part of the reaction energy (14 MeV) is
neutron kinetic energy to be used as heat source. To use
the energy released by moderating of neutrons for
electric power generation is an actual challenge.
Fusion technologies are in the experimental stage. The
aim of International Thermonuclear Experimental
reactor ITER is to prove the scientific and technical
viability of fusion energy and to test the simultaneous
ant integrated operation of the technologies needed for a
fusion reactor. One of main objectives is to achieve 500
MW fusion power for 300 seconds. Upon the expected
success of the ITER, it is planned to build the DEMO demonstration nuclear fission power plant. The size of
ITER reactor coincides approximately with the size of
1000-1500 MWe power commercial fission reactors.
Later on, it is planned to build PROTO prototype of
commercial fusion reactor. However, according to the
prognosis it will be very difficult to commission the
first commercial-sized fusion reactor before 2050.
Almost all operating and new built nuclear reactors are
nuclear fission reactors at present time.
The largest amount of the nuclear energy released
during fission event is represented by the kinetic energy
of the recoiling fission fragments. Contrary to the fusion
reaction released neutral neutron, the fission fragments
are heavy charged particles. The range of these particles
in the fuel element is a fraction of a millimeter.
Therefore the recoil energy is effectively deposited as a
heat at the point of fission. Kinetic energy of prompt
Keywords: nuclear reactors, fuel, fuel burn up, fuel

enrichment, fuel conversion, fuel utilization.
1. Introduction
Nuclear energy is released by the splitting or merging
together of the nuclei of atoms. The conversion of one
element into another is called transmutation process.
The conversion of nuclear mass to energy is
corresponding
to
mass-energy
equivalence
formula: E m c 2 , in which c is a speed of light in
a vacuum. The actual mass of an atomic nucleus is not
the sum of the masses of nucleons (protons and
neutrons) of which it is composed. This mass difference
is called mass defect. In this case the energy E is the
binding energy of the nucleons. Any process that results
in nuclides being converted to other nuclides with the
higher binding energy per nucleon will result in the
conversion of mass into energy. The fusion of light
atomic mass nuclides (e.g. hydrogen) is the basis for
fusion reactor technologies. The splitting of very high
atomic mass nuclides is the basis of the fission process
for the release of nuclear energy and for fission reactor
technologies.
186
and delayed neutrons and gamma radiation is deposited

in the surrounding material. The fission energy released
in material depends on heavy nuclei. Uranium and
plutonium are the main isotopes used as a nuclear fuel
in fission reactors. Releases of the total fission energy
for the most important heavy nuclides in nuclear fuel
calculated for one single fission event are presented in
Fig. 1. It could be derived that elementary fission
energy increases with the increasing mass number of
nuclides.
The values of nucleus fission cross sections for thermal

and fission neutrons for most important heavy nuclides
are presented in Table 1. For some nuclides
235
239
241
), the fission probability is
(e.g. 233
92 U , 92 U , 94 Pu, 94 Pu
sufficient enough and a chain reaction is possible when
the bombarding neutron has very low kinetic energy.
Such nuclides are referred to as fissile. Operating under
these circumstances reactors are thermal reactors.
Table 1. Fission cross sections, , for heavy nuclides
Nuclide
Thermal neutron
(0,025 eV) , b
U235
584,4
U238
1,17710-7
Pu238
17.89
Pu239
747,4
Pu240
0,0588
Pu241
1012
Pu242
2,55710-3
Am241
600,4
Am242
2100
1 barn ( 1b 10 24 cm 2 )
Fission energy, MeV
210
205
200
66
195
190
Am242
Am241
Pu242
Pu241
Pu240
Pu239
U238
U235
185
Fig.1. Fission energy of the main radionuclides in

nuclear fuel
In thermal reactors a moderator is integrated into reactor

core to reduce the neutron energy from the MeV region
to thermal energies. Neutron moderation is achieved by
successive elastic collisions with light nuclei.
Moderator is low atomic mass material: hydrogen as a
part of light (ordinary) water, deuterium as a part of
heavy water, carbon as graphite. Ideal in this respect is
hydrogen, except that it can absorb neutrons. Heavy
water and carbon are other good moderators because of
their small neutron absorption probability.
240
242
are non-fissile in respect of
Nuclides 238
92U , 94 Pu , 94 Pu
Such the fission energy and chain of such nuclear

fission reactions is physical bases of nuclear fission
reactors.
3. Features of thermal and fast neutron reactors
The advantage of nuclear chain reaction is that a
number of prompt neutrons are produced for each
bombarding neutron during the fission process of heavy
nuclei. The average neutron number released per fission
depends on bombarding neutron energies, thermal (T)
or fast (F) and on heavy nuclide (Fig.2). This number
increases when fission is induced by fast neutrons.
thermal neutrons but can undergo fission induced by

fast neutrons.
Reactors designed to operate with fast neutrons are
called fast-neutron reactors or fast reactors. The average
fission neutron energy is of the order of 0.5 MeV to 1.0
MeV. The fission threshold for 238U is 1.0 MeV. There
is no need to slow down neutrons in the fast reactors;
thus there is no need to integrate moderator. The
nuclear fuel composition depends on reactor type, on
moderator and coolant physical features (Table 2). In
some reactors types, one substance serves both
functions: to assist in neutron slowing and to remove
the fission heat. Others involve one material for
moderator and another for coolant. Moderators are light
water, heavy water, graphite, and beryllium. Coolants
are light water, carbon dioxide, helium, liquid metal.
The condition of the coolant serves as further
identification. The pressurized water reactor (PWR)
provides high-temperature water to a heat exchanger
that generates steam, while the boiling water reactor
(PWR) supplies steam directly. RBMK-1500 type
reactor operating in Ignalina nuclear power plant is light
water BWR reactor with graphite moderator. In hightemperature gas cooled reactors HTGR gas is used as a
coolant, but in fast neutron reactors (FBR) liquid metal
is used at present time (Table 2). Liquid metal
3.2
3.0
2.8
2.6
2.4
2.2
Pu242(F)
Pu241(F)
Pu241(T)
Pu240(F)
Pu239(F)
Pu239(T)
U238(F)
U235(F)
2.0
U235(T)
Number of neutrons
Fission neutron
, b
1,235
0,3284
1.994
1,800
1,357
1,648
1,127
0,2296
1,756
Fig.2. The average neutron number per fission

Thermal neutrons are characterized by a nominal energy
of 0.0253 eV. The prompt fission neutrons have an
energy spectrum in the range 0-10 MeV with the
average energy of 1.5-2 MeV.
The probability of an interaction of neutron with
nucleus is referred to as the neutron cross section, .
187
transports heat very efficiently and only lightly

moderates the neutrons from fission. Sodium is used as
the most common form of liquid metal for these
reactors.
CANDU, heavy water was used for both functions:

moderation and cooling. The CANDU-9 has been
developed on the base of the existing design.
High-temperature gas-cooled reactors (HTGR or HTR)
are graphite moderated, helium cooled reactors.
High temperature reactors HTR can potentially use
thorium-based fuels, consisting of uranium with
thorium, plutonium with thorium.
A Pebble Bed Modular Reactor (PBMR) is advanced
version of high temperature reactors (HTR). It utilizes a
helium cooled and graphite-moderated nuclear core as a
heat source.
Table 2. Characteristics of nuclear power reactors.
Reactor
type
PWR
CANDU
BWR
HTGR
FBR
Fuel
Moderator
Coolant
UO2
enriched U235
Natural UO2
(0.7 % U235
UO2
enriched U235
UO2
( 8-19 %)
UO2/PuO2
( 16-20 %)
Light
water
Heavy
water
Light
water
graphite
Light
water
Heavy
water
Light
water
Gas
none
Liquid
metal
4. Fuel conversion and breeding

If a neutron is absorbed by uranium 238U, then the
nucleus after two successive decays is converted
into plutonium 239Pu, which is fissile. Because of
subsequent neutron captures plutonium isotopes 240Pu,
241
Pu, 242Pu are formed. Plutonium isotopes 239Pu and
241
Pu are fissile in the thermal reactor.
Production of plutonium isotopes was evaluated for the
nuclear fuel of typical PWR reactor (enrichment 3.3 %
of 235U) and for RBMK-1500 reactor (enrichment 2.4 %
of 235U + Er) fuel. The evaluation was performed using
APOLLO1 neutron code. The actinides 232, 234-236, 238U,
236, 238 - 242
Pu, 237Np, 241-243Am and 242-244Cm were used
as an input data for the performed calculations. The
composition of plutonium isotopes in typical PWR
reactor fuel after 40 MWd/kg burn-up is shown in the
Fig. 5. and in the case of RBMK-1500 reactor fuel after
20 MWd/kg burn up - in Fig.6. Production of these
isotopes is an important factor for reactors economics.
The ratio of the rate of creation of new fissile isotopes
to the rate of destruction of fissile isotopes is defined as
conversion ratio. When this ratio is > 1, it is called
breeding ratio. The conversion ratio for the light
water PWR and BWR reactors is 0.6, for heavy water
reactor CANDU - 08. The breeding ratio for fast
breeder reactors can vary from 1.06 to 1.60.
The distribution of reactors types in the world is shown

in Fig.3. Figure 4 presents a number of power reactors
which are under construction or are planned to start
operation until 2016. Baltic-1 (Russia) 1200 MWe
PWR type reactor is planned to be built by Energoatom.
Number of Units
300
250
200
150
100
50
0
PWR
BWR
CANDU
HTGR
FBR
Fig.3. Distribution of operating reactors by type
60
0.6
Composition, %
Number of Units
50
40
30
20
10
0.5
0.4
0.3
0.2
0.1
0
Pu238
Pu239
Pu240
Pu241
Pu242
0
PWR
BWR
PHWR
FBR
HTR
Fig. 5. Composition of plutonium isotopes in spent

nuclear fuel of typical PWR (40 MWd/kg burn up)
Fig.4. Power reactors under construction (or almost so)

Pressurized light water reactors are dominant type of
reactors.
The Canadian firm, Atomic Energy of Canada Limited
(AECL), has developed a modified pressurized heavy
water reactor PHWR (the ACR series) which would
only use heavy water as a moderator. Light water would
cool this reactor. In earliest version of PHWR reactors,
188
235
U with high burn up, extending the fuel life by about

three times and reducing high-level waste volume
accordingly. PHWRs have been popular in several
countries because they use less expensive natural (not
enriched) uranium fuels, and can be built and operated
at competitive costs.
The fuel for high temperature reactors HTR is in the
form of particles. Each has a kernel of uranium
oxycarbide, with the uranium enriched up to 9 % 235U.
This is surrounded by layers of carbon and silicon
carbide, giving a containment for fission products
which is stable to 2000 C.
A Pebble Bed Modular Reactor (PBMR) fuel consists
of approximately 10 %enriched uranium triple-coated
isotropic particles contained in a molded graphite
sphere. A coated particle comprises a kernel of uranium
dioxide surrounded by four ceramic layers. A fuel
sphere consists of some 15,000 particles and has a
diameter of 60 mm. The total mass of a fuel sphere is
210 g and uranium mass is 9 g. During normal
operation the PBMR core contains a load of 456,000
fuel spheres [7]. Fuel pebbles, having 5-6 %
enrichment, reaches burn-up of 80 MWd/kg. Target of
burn-up is 200 MWd/kg.
It is expected that a particle keeps inside all fission
products during all accident conditions, and hence
requires no separate containment building.
Composition, %
1.E+00
1.E- 01
1.E-02
1.E-03
1.E-04
1.E-05
1.E-06
1.E-07
1.E-08
Pu236
Pu238
Pu239
Pu240
Pu241
Pu242
Fig. 6. Composition of plutonium isotopes in spent

nuclear fuels of RBMK-1500 reactor (20 MWd/kg)
5. Nuclear fuel burn up
The economics of nuclear power is strongly affected by
the efficiency of fuel utilization to produce power. The
amount of produced power is related to the fuel burn
up. The most commonly used measure of fuel burn up is
the fission energy release per unit mass of fuel. It is
characterized as the fission energy release in megawattdays divided by the total mass of fuel nuclei existing in
the initial loading.
Fuel burn up depends on fuel composition and changes
during reactor operation. For thermal reactors typical
fuel burn up is from 30 MWd/kg to 50 MWd/kg. Fuel
burn up in fast reactors is projected from 100 MWd/kg
to 150 MWd/kg. A burn up of 3 MWd/kg to 6 MWd/kg
is attained when natural uranium fuel is used. The
higher fuel burn up produces more actinides and fission
products with the large thermal neutron cross sections.
Higher discharged burn-up of light water reactor fuel
leads to a more economical fuel cycle and reduced
amounts of spent nuclear fuel [6]. To reach higher burnups higher fuel enrichment is necessary. At present
time fuel enrichment approaches the 5 % 235U level.
7. Re-enriched and reprocessed nuclear fuel

After a burn-up of 50 MWd/kg spent nuclear fuel
contains about 0.7 % 235U and 0.6% 236U. Such
reprocessed uranium can be re-enriched. In addition,
also fresh enriched 235U can be added to attain a
necessary mixture. By such a mode low enriched
uranium fuel with 5.52% 235U, 3.0% 236U and 91.48%
238
U can reach a discharge burn-up of 60 MWd/kg [8].
Furthermore, the substantial quantities of plutonium
remain in the spent fuel removed from the reactor (Fig.5
and Fig.6). This plutonium may be recovered by the
chemical reprocessing of the spent nuclear fuel, and
may be reused after mixing it with uranium in the form
of mixed fuel oxide (MOX). Modern thermal reactors
can be loaded by MOX fuel for 0-100 %.
6. Nuclear fuel composition

The use of thermal nuclear reactors for energy
production is based on the ability of certain fissile
nuclei, such as uranium-235, plutonium-239 and
plutonium-241, to maintain fission chain reactions.
Pressurized water reactors (PWR) and boiling water
reactors (BWR) use light water as a coolant and
moderator. Due to the neutron absorption by hydrogen,
it is necessary to use uranium fuel enriched to about 3%
of 235U, when hydrogen is thought as a moderator.
Raising the uranium 235U content the possibility of
fission increases. If light water is replaced by heavy
water as a moderator, the reactor can be operated using
natural uranium as it is in the case of CANDU reactor.
Natural uranium contains about 0.7 % of 235U and
99.3% of 238U.
The CANDU-9 has flexible fuel requirements ranging
from natural uranium through slightly-enriched
uranium, recovered uranium from reprocessing spent
PWR fuel, mixed oxide fuel, direct use of spent PWR
fuel, to thorium. The advanced CANDU reactor ACR700 will run on low-enriched uranium (about 1.5-2.0%
Composition, %
100
10
0.1
U235
Pu238
Pu239
Pu240
Fig.7. Composition of MOX fuel
189
Pu241
Pu242
Am241
At the same time the number of neutrons produced per

fission is 25% more than from uranium (Fig. 2), and
this means that there are enough neutrons to maintain
the chain reaction and also to convert 238U into fissile
plutonium. The use of liquid sodium as coolant avoids
any neutron moderation and provides a very efficient
heat transfer medium. So, the fast reactor during its
operation can produce plutonium. It can operate as fast
breeder reactor FBR.
The FBR was originally conceived to extend the world's
uranium resources, and could do this by a factor of
about 60. But until now and in nearest future too it is
enough natural uranium for producing nuclear fuel and
fast reactors are not competitive. Another factor which
blocks the FBR development is changed utilization of
plutonium separated from reprocessing used light water
reactor fuel. It was originally envisaged for FBRs, but
now is used as mixed oxide (MOX) fuel in conventional
reactors. The economics of FBRs depends on the value
of the plutonium fuel which is bred, relative to the cost
of fresh uranium.
Fast neutron reactors have a high power density and are
normally cooled by liquid metal: sodium, lead, or leadbismuth. Liquid metal is characterized with high
conductivity and boiling point and no moderating
effect.
An international task force is developing six nuclear
reactor technologies for deployment between 2020 and
2030. Four are fast neutron reactors (Table 3). In the
table: GCFR gas cooled fast reactors, LCR leadcooled reactors, MSR molten-salt reactors, SCRsodium-cooled reactors, SWCR- supercritical watercooled reactors, VHTGR- very high temperature gas
reactors
Plutonium can be used more efficiently by reducing the

quantity of uranium or by altering the moderating ratio.
The latter approach has been adopted in the Advanced
Plutonium Assembly (APA) in which the neutron
moderation has been increased to enable a more
efficient use of plutonium in a PWR than it is in the
case with MOX recycling. Plutonium, produced by
conversion of 238U in a thermal reactor, after separating
procedure could be used as the fuel in the fast breeder
reactor. So, the breeder reactor can complement the
thermal reactor.
18000
Contribution, %
16000
14000
12000
10000
8000
PWR
6000
RBMK
4000
2000
0
0
10
20
30
40
50
60
Burnup, MWd/kg
Fig. 8. Relative contribution of fission products to

neutron absorption in nuclear fuel
The fission process leads to the reduction of the number
of fuel nuclei and a big variety of actinides (Fig. 5 and
Fig. 6) and fission products are produced, many of
which have a large neutron capture cross sections.
The buildup of fission product poisons in the fuel
eventually leads to the loss of the efficiency. Figure 8
represents the total relative contribution of all fission
products to neutron absorption in the fuel of PWR and
RBMK reactors respectively. Presented distributions
were obtained from the modeling results, which were
performed using the French code APOLLO1.
9. Worldwide marketed reactors

At present time reactors suppliers in North America,
Japan, Europe, Russia and South Africa have a set of
new nuclear reactor designs at advanced stages of
planning while others are at the research and
development stage.
In the Figure 9 there are number of contracts for nuclear
power reactor construction or completion projects
around the world to 2008 shown.
8. Problems of fast fission reactors

The fast reactor has no moderator and relies on fast
neutrons alone to cause fission. This process is less
efficient than using slow neutrons.
Coolant
Fuel
GCFR
LCR
MSR
fast
fast
Epithermal
fast
helium
Pb-Bi
Fluoride
salts
sodium
Thermal
or fast
thermal
water
U238
U238
UF in
salt
U238,
MOX
UO2
helium
UO2
SCR
SWCR
VHTR
Power,
MWe
288
50-1200
1000
1501500
1500
20
18
16
14
12
10
8
6
4
2
0
EV
To A
s
G hib
E/
a
H
ita
ch
R
i
us
si
Ko an MH
I
In
re
an du
s
tr
I
C
hi ndu y
ne
st
y1
se
0
in
In
du
di
s
an
t
i n ry
du
st
ry
AE
C
L
Neutron
AR
Type
Number of contracts
Table 3. Generation IV nuclear reactors
250
As its basic fuel a fast reactor uses plutonium or

relatively highly-enriched uranium (about 20% 235U).
Fig. 9. Contracts around the world
190
Asia will see most of the nuclear plant construction,

with over 45 nuclear reactors being built or planned for
the near future. The Western companies (AREVA,
Toshiba, Ge/Hitachi, MHI and AECL) account for
about one quarter of the currently ongoing reactor
construction projects.
LWR can be introduction of longer fuel cycles, partial

reload of fuel with UO2/PuO2 mixed oxide assemblies,
introduction of low leakage core loading patters, etc.
Many utilities are increasing the initial fuel enrichment
usually to more than 4% 235U (Fig. 10) and are usually
burning the fuel to leave about 0.5% 235U at the end of
fuel life.
Due to the higher enrichment and the higher burn up the
fuels remains in the reactor longer time.
Table 4 provides a list of reactors that are currently

being marketed worldwide.
Table 4. A list of marketed reactors
Name
1
2
3
4
5
6
7
ABWR
APR-1400
AP-600
AP-1000
ESBWR
APWR
US-APWR/EUAPWR
ATMEA
EPR
SWR 1000
Advanced BWR
AES-92
(VVER-1000)
AES-2006
(VVER-1200)
VVER-1500
CANDU-9
Enhanced
CANDU-6
ACR-700
ACR-1000
PBMR
GT-MHR
8
9
10
11
12
13
14
15
16
17
18
19
20
Power,
MWel
1400
600
1000
1520
1538
1700
Available
for
construction
Yes
From 2010
Yes
Yes
From 2010-2011
Yes
From 2012
1100
1600-1750
1200-1290
1500
1000
Not yet
Yes
Yes
Not yet
Yes
1150-1200
Yes
1500
925-1300
750
Not yet
Yes
Yes
700
1080-1200
165
285
Not yet
Not yet
After 2013
Not yet
48
Burnup, MWd/kg
No
46
44
42
40
38
36
3.2
3.4
3.6
3.8
Enrichm ent, %
Fig.10. Worldwide trends of enrichment and burn up in

the period of 1990 to 2010
11. References
1. W.M. Stacey. Nuclear reactor physics. John Wiley
and Sons, Inc. 2007. 735 p.
2. A.R. Foster, R.L. Wright, Basic nuclear engineering.
Boston. 1993. 611 p.
3. D. Bodansky. Nuclear energy. Principles, practices
and prospects. Springer, 2004. 701 p.
4. H-D Berger, High Fuel Burn-ups and related safety
consideration, Presentation at 2006 FJOHSS, 23
August 1 September, 2006.
5. WNA information on Uranium Markets,
The type of reactors provided in the Table 2 are: boiling

water reactor BWR (1, 5, 10, 11), pressurized water
reactor PWR (2-4, 6-9, 12-14), heavy water reactor
HWR (15-18) and high temperature gas cooled reactor
HT-GCR (19, 20). The vendors of the reactor
construction are: Areva (8-10), AECL (15-18), Eskom
(19, 20), General Electric (1, 5), Gidropress (12-14),
Hitachi (1, 5), Mitsubishi (6, 7), MHI (8), South Korea
(2), Toshiba (1, 11), Westinghouse (3, 4).
www.world-nuclear.org
6. W.Goll, Ch.Hellwig, P.B.Hoffmann, W.Sauser,

J.Spino, C.T> Walker. UO2 Fuel Behavior at Rod
Burn-ups up to 105 MWd/kg HM. International
Journal for Nuclear Power, 2007 No 2, 95-102 p
7. G.Brahler, K.Buttner, K. Froschauer, W. Kress.
PBMR fuel sphere production facility project
Pelindaba, South Africa. International Journal for
Nuclear Power, 2006 No 11, 701-704 p
8. C.H.M. Broeders, G. Kessler. The generation of
denatured reactor plutonium by different options of
the fuel cycle. International Journal for Nuclear
Power, 2006 No 11, 711-716 p
9. S.Niesen. Do we live a nuclear renaissance?
International Journal for Nuclear Power, 2008 No 5,
317-319 p.
10. Trends in nuclear fuel utilization

The majority of the reactors in operation worldwide are
light water reactors (LWRs), both of pressurized or
boiling water (PWR or BWR) types.
For a given energy output the mass of the fresh fuel
loaded and the mass of the spent fuel discharged are
inversely proportional to the average discharge burn up
[4]. For an individual country, the optimum of the
achieved burn up with respect to the fuel cycle costs
may vary, depending on the specific technical, licensing
and economic conditions, but the gain of a higher
amount of energy form the same amount of fuel with
increased burn up clearly indicates a strong economic
benefit. Other ways to improve fuel utilization in a
191

NUCLEAR ENERGY MIX IN THE BALTIC SEE REGION

Gediminas ADLYS*, Olegs LINKEVIS**, Saulius RAILA*, Rta ADLYT*, Helmut ALT ***
*Kaunas University of Technology, Lithuania; **Latvenergo, Latvia;
***FH Aachen Fachbereich Elektrotechnik, Germany
Abstract: Nuclear energy is a key component of lowcarbon and secures energy supply policy in the context
of the increasing demand of energy, depletion of fossil
resources, increasing energy prices, growing energy
dependence and greenhouse emission.
This article presents the overview of the electricity
generation mix in the Baltic Sea region. It is based on
the analysis of the nuclear energy contribution to the
total electricity generation in different countries.
3.0
Generation, TWh/a
2.5
Keywords: electricity generation, energy mix, nuclear

power, interconnections.
100
80
60
40
20
2001-2007
1996-2000
1991-1995
1986-1990
1981-1985
2006
2002
1998
1994
1990
1986
1982
1978
From 1990 to 2006, world capacity rose by 44 GWe due

to net addition of new plants and uprating some
established ones. The relative contributions to this
increase were: new constructions - 36%, uprating -7%
and availability increase - 57%. The increase over the
last five years (210 TWh) is equal to the output from 30
large new nuclear plants.
Nuclear energy takes important place in electricity
generation around the world. As of 2005, nuclear power
provided 6.3% of the world's energy and 15% of the
world's electricity.
At present time 44 reactors with a net electrical capacity
41.5 GWe are under construction in 14 countries. Most
of them are being built in Russia, China, India and
Korea.
120
1976-1980
1974
1970
Fig. 2. Nuclear electricity production worldwide
140
1971-1975
1.0
0.0
As of 2008 nuclear generation capacity in the world

represented around 393 GW. There are 439 nuclear
power reactors in operation around the world, operating
in 31 countries. Most of the nuclear power plants which
are operating until now were built between 1970 and
1990 (Fig.1).
1965-1970
1.5
0.5
1. Introduction
Number of Reactors
2.0
2. Nuclear energy tendencies in European Union

Fig. 1. Number of installed nuclear power reactors
versus time
New nuclear power plants have been planned or are

already being built across Europe: two units at Belene in
Bulgaria to be built by 2014 and 2015, the third and
fourth units of the nuclear power plant at Mochovce,
southern Slovakia to be constructed by 2013 and
Despite the fact that after this period there was a sharp
decrease in the number of nuclear power reactors put
into service, nuclears share of global electricity
generation all time was increasing (Fig.2).
192
169,891 MWe in operation in Europe and 14 units with

12,817 MWe were under construction in five countries.
Cernavoda NPPs units 3 and 4 to be completed by

2014 and 2015 in Romania [1].
The construction of a European pressurized water
reactor (EPR) will begin at Penly/France, on the
Channel coast near Dieppe, in 2012. It will be
connected to the European energy grid in 2017. Penly
has already two operating 1300 MW PWRs. The new
EPR will be the second unit of its kind in France, after
Flammanville, and the third in Europe after Olkiluoto,
Finland. The government of UK has announced the
building of a new fleet of reactors (around 8) and it has
already developed its action plan for the construction. In
Finland, three companies (TVO, Fortum and
Fennovoima) are seeking for the potential building of
three reactors in total. The Baltic States agreed to jointly
build a new Visaginas NPP in Lithuania. The Polish
government adopted a Resolution on nuclear power,
which sets the conditions to construct one or two NPPs
by 2020.
On 24 February 2009 an agreement between France and
Italy was signed which paves the way for the building
of nuclear power plants in Italy, using French
technology and the participation of Italian Electricity
Utility in the construction of NPPs in France. Italy
phased-out its nuclear power plants after being forced to
abandon its nuclear energy programme following a
referendum in 1987. Therefore, Italy is the world's
largest electricity importer and electricity prices in Italy
are 45% higher than the European average. In the
framework of the Italian-French agreement a study of
the feasibility of constructing at least four EPR units is
planned. The first EPR reactor in Italy should come into
operation by 2018.
4. Energy mix in Baltic States

The power systems in the three Baltic countries differ
considerably. The Estonian power system is purely
thermal; the Latvian system is based on hydro and
thermal technologies, and the Lithuanian system is
dominated by nuclear power production [2]. The big
differences in the power systems create potential
benefits to be gained from electricity trade between the
countries [3].
In 1990, Estonia, Latvia and Lithuania produced more
than 50 TWh of the electrical power in total (Fig. 4). In
2006 it was only 27 TWh [4].
Electricity generation, TWh
60
Estonia
Lithuania
50
Latvia
Total
40
30
20
10
0
1990
1995
2000
2005
Fig.4. Electricity generation in Baltic States

The drop down in electricity consumption after 1990
mainly caused by the economical recession and
restructurization of the economy was the main reason
for the decrease of the generated power capacity. A
change in energy export capacities to neighboring
countries (Kaliningrad district, Russia and Belarus) was
the second reason.
The numbers of the annual net export of electricity from
each Baltic country during the time period from 1990 to
2006 are provided in Fig.5.
3. Nuclear energy situation in European Union

Nuclear energy provides 30% of the electricity from the
total energy production of the European Union.
Many European countries are relying on nuclear energy.
In terms of electricity generated by nuclear energy in
2007, France holds the top position with a share of
76.9% followed by Lithuania with 64.4 %, Slovakian
Republic and Belgium with 54% and Sweden with
46.1%. (Fig. 3).
8
70
60
50
Net export, TWh
Nuclear share, %
80
40
30
20
10
Czech Rep.
Bulgaria
Hungary
Switzerland
Slovenia
Sweden
Belgium
Slovakia
Lithuania
France
-2
-4
Estonia
Latvia
Lithuania
Total
-6
Fig.3. European countries with nuclear share in

electricity generation > 30 % in 2007
1992
1996
2000
2004
Fig. 5. Net export of electricity in the Baltic States
As of November 2008, there were 197 nuclear power

plant units with an installed electric net capacity of
193
2008
Hydro power represents about two-third of the installed

power capacity Latvia (Fig.8). The three major
hydroelectric plants of Latvia are set on the Daugava
River: Kegums Hydro Electrical Station (HES),
Plavinas HES and Riga HES. Kegums HES consists of
two hydroelectric plants with at total capacity of 264.1
MW. Plavinas HES consists of 10 hydroelectric units
and was put into operation in 1968 with a total capacity
of 825 MW. Plavinas is the largest hydroelectric plant
in the Baltic States by its capacity.
Negative numbers indicate a net import of electricity.

Latvia is a net importer of electricity while Estonia and
Lithuania are net exporters of electricity.
The total net export out of the three Baltic countries was
ranging from 0.4 to 6.8 TWh per year in the period from
1999 to 2003. These numbers include an import from
Russia and export from Lithuania to Belarus and
Kaliningrad. The energy balance from year 2003 is
illustrated in Fig.6.
3.5
120
Production
100
2.5
Electricity, MWh
Power, TWh
3.0
2.0
1.5
1.0
0.5
Export
80
60
40
0.0
LT-KGD
BY-LT
LT-LV
LV-RU
EE-LV
RU-EE
EE-RU
20
Fig. 6. Annual electricity exchange between Baltic and

neighboring countries
0
0
10
20
30
40
50
Weeks
In October 1998, an agreement on co-operation in the

energy sector between the three Baltic States was
signed, and a Baltic Energy Strategy was approved by
the Baltic Council of Ministers. In 1999, the strategy
stipulated the creation of the Common Baltic Electricity
Market as a long-term energy policy goal.
The share of different kinds of energy in The Baltic
States energy system is shown in Fig.7.
Fig. 8. Annual fluctuations of power production by

Daugava HPP
In 2004 Lithuania accounted for the largest power
generation in the three countries, i.e. 48-60%. The
power generation in Lithuania is highly dependent on
the Ignalina nuclear power plant, which will be closed at
the end of 2009.
60
Hydro
Nuclear
Coal
Gas
5. Energy mix in Nordic Countries
50
The Nordic (Sweden, Finland, Norway) power system is

highly integrated in terms of physical connections and
system stability. The Nordic electricity market and the
exchange of electricity with neighboring countries
(Fig.9) are of crucial importance for Swedens
electricity supply [5]. Therefore the project to join
Sweden and Baltic States electricity systems by
underwater cable and to form common Baltic See
countries energy market is of high priority.
Share, %
40
30
20
10
0
1990
1994
1998
2002
2006
6
Exchange, TWh
Fig.7. Coal, nuclear, gas and hydro share in Baltic

States energy system
The Estonian electricity sector is characterized by the
large share of oil shale power. Thermal power stations
dominate in the Estonian electricity generation. Two
largest plants are Eesti PP and Balti PP. Estonia
generates approx. 30% of the total power generation in
the three countries.
Latvia accounts for 11-18% of the power generation in
the Baltic States, which is mainly produced by hydro
power and the CHP plants in Riga.
Import
Export
4
6
3
2
1
0
Denmark
Finland
Norway
Poland
Germany
Fig.9. Electricity exchange between Sweden and

neighboring countries in 2006
194
Electricity generation in Sweden is dominated by CO2free hydro and nuclear power (Fig.10).
Table 1. Operating nuclear power plants in Germany

Nuclear
power
plant
Biblis A
Biblis B
Brokdorf
Brunsbttel
Emsland
Grafenrheinfeld
Grohnde
Gundremmingen B
Gundremmingen C
Isar-1
Isar-2
Krmmel
Neckar-1
Neckar-2
Philippsburg-1
Philippsburg-2
Unterweser
70
Nuclear
Hydro
Coal
Gas
60
Schare, %
50
40
30
20
10
0
1990
1994
1998
2002
2006
Fig.10. Electricity mix in Sweden

The Swedish government announced recently, that it
intends to reverse the countrys long-standing ban on
nuclear energy and allow the building of new nuclear
reactors to gradually replace its existing nuclear fleet.
The nuclear phase-out policy was brought in following
a national referendum in 1980.
The contribution of the nuclear energy is nearly 30% of
the total energy production in Finland
NC,
MWe
1167
1240
1370
771
1329
1275
1360
1284
1288
878
1400
1346
785
1305
890
1392
1345
REC,
TWh
13.7
23.0
128.4
11.0
141.8
69.6
113.8
79.7
89.1
24.2
139.1
88.6
10.2
152.2
28.7
113.5
43.6
The last column of Table 1 indicates residual electricity

generation quantities of the individual nuclear power
plants in accordance with the Atomic Energy Act. (The
situation on 01-01-2008 is shown).
The Atomic Energy Act was voted in 2002 after
consensus talks (the ,,Atomkonsenz) between operators
and the government in Germany. According to the Act
each nuclear power plant is allowed to generate a fixed
quantity of electricity and these quantities may be
transferred under special conditions from one nuclear
power plant to another. As soon as a nuclear power
plant has reached its volume of electricity production, it
must be withdrawn from the grid.
It seems that the only way for Germany to secure power
supply in the country in the future is to follow an
example of Swedish and Italian governments and to
disclaim nuclear power phased-out politics.
The electricity generation share by fuel type in Germany

is presented in Fig. 11.
70
Renewables
Gas
60
50
Share, %
1975
1977
1986
1977
1988
1982
1985
1984
1985
1979
1988
1984
1976
1989
1980
1985
1979
GC,
MWe
1225
1300
1440
806
1400
1345
1430
1344
1344
912
1475
1402
840
1400
926
1458
1410
GC gross capacity, NC net capacity, REC residual

electricity quantity
6. Nuclear energy in Germany
Nuclear
Coal
Year
40
30
7. References
20
1. The Nuclear relaunch in Europe. European Atomic

Forum, Brussels, 2008
2. Power sector development in a Common Baltic
Electricity Market. Final Report. June 2005.
3. Z. Krishans, A.Mutule, I. Oleinikova, A. Kutjuns.
Application of Risk Assessment Method for Baltic
States Power System Development Optimization. 9th
International Conference. KTH, Stockholm, 2006.
4. BALTSO. Annual Report 2007. Riga. 48 p.
5. The Electricity Year 2007. Annual report. Svensk
Energi. Stockholm. 47 p
6. Analyses of Energy Supply Options and Security of
Energy Supply in the Baltic States. IAEA-TECDOC1541. Vienna, 2007. 323 p
10
0
1990
1994
1998
2002
2006
Fig.11. Electricity generation share by fuel type in

Germany
17 nuclear power plants units with an electric gross
output of 21,457 MW are in operation in Germany
(Table 1). Total generated electric power capacity
was140 billion kWh in 2007.
It is to point out that the operation time of almost all
nuclear power plants in Germany is restricted by the life
design time: most of the nuclear power plants were
commissioned 20-30 years ago. The shutdown of the
last unit is foreseen in 2022.
195

PERSPECTIVES OF NUCLEAR COGENERATION TECHNOLOGIES

Gediminas PETRAUSKAS*, Gediminas ADLYS*, Diana ADLIEN*, Kstutis GEDIMINSKAS**,
Algimantas VALINEVIIUS*
* Kaunas University of Technology, Lithuania; **Ignalina nuclear power plant, Lithuania
nuclear energy as a part of Switzerlands electricity mix.

Both Beznau nuclear power plant and Gsgen nuclear
power plant are supplying the district heating in
addition to power generation. Refuna district heating
system supplied from Beznau nuclear power plant
consists of 35 km of the main network and 85 km of
local distribution pipelines. It has become an integrated
part of the regional infrastructure. The REFUNA district
heating network sells about 130 000 MWh of energy
every year and is thus the largest district heating system
of Switzerland.
Visaginas town district heating system is supplied by
Ignalina nuclear power heat (Fig. 1) too.
Abstract: Energy is one of the most challenging issues

that currently many countries are facing, including
Lithuania and Baltic States. Increasing energy prices,
concerns about dependency on foreign energy sources,
climate change have led to a nuclear energy renaissance.
Modern nuclear technologies are good basis to use
nuclear energy not only for base load electricity
production but also for safe district heating in the
cogeneration nuclear power plants.
Features of cogeneration heat and electricity production
analysis and the review on properties and perspectives
of small size nuclear reactors for cogeneration power
plants is provided in this article.
Production, GWh
Keywords: nuclear energy, cogeneration plants, small

size nuclear reactors, greenhouse gas emissions.
1. Introduction
Generation of the heat within the reactor core is a result
of the nuclear fission process. The heat can be applied
to the generation of electricity or used in providing hot
water or steam for industrial or space heating purposes.
Practically nuclear power reactors were applied to
generate electricity and to produce the heat for industry
or district heating at the same time from the very
beginning of their commissioning. The first nuclear
power reactor at Calder Hall in the United Kingdom
provided electricity to the grid and heat to a neighboring
fuel reprocessing plant. The Agesta reactor in Sweden
provided hot water for district heating to a suburb of
Stockholm.
In 1973-1976 in Bilibino, Chukotka four power units of
the nuclear heat and electricity generating plant were
sequentially commissioned. They were provided with
water-graphite reactors. Each unit's capacity was 12
MWe, each of them was supplying heat for Bilibino
central heating system at a rate up to 25 Gcal/hour.
Romania has two nuclear reactors generating almost
20% of its electricity. The first unit also provides
district heating for Cernavoda town. 148 TJ were
supplied in 2003.
Switzerland operates 5 nuclear reactors generating 40%
of its electricity [1]. A national vote has confirmed
40
35
30
25
20
15
10
5
0
Heat
Hot water
10 11 12
Months
Fig. 1. Heat and hot water production for Visaginas

town by Ignalina NPP in 2008
District heating accounts for about 50 per cent of the
total heating market In Finland. 4/5 of it is being
produced in combined heat and power plants. The City
of Helsinki is going to prepare its own feasibility study
of the nuclear energy involvement to cover the district
heating demands of different districts of the capital in
the Great Helsinki area. A new unit at Loviisa (site, with
two existing 488 MW reactors) will be able to cover the
heating demands of the entire metropolitan area.
2. Features of cogeneration
Cogeneration or combined heat and power (CHP)
means using both: the electricity and heat produced
together. Compared to separate generation of heat in
196
boilers and power in condensing plants, CHP systems

offer considerably higher energy efficiency levels.
Cogeneration leads to wasting less energy in
production, increasing efficient utilization of primary
energy resources and for reducing CO2 emissions.
Surplus heat from an electric power plant can be used
for industrial purposes, or for the heating of space and
water. Otherwise, when steam or hot water is produced
for an industrial plant or a residential area, power can be
produced as a by-product [2].
Significant differences exist between the applications of
electricity and of heat. [2]. Transport of the heat from its
generation places to the end-user is complicated and
expensive. Compared to heat, the transport of electricity
is easy and cheap, even to large distances measured in
hundreds of kilometers. The key to successful CHP is to
design the plant according to the heat demand of the
system. These systems can also be installed close to
users, reducing power transmission losses.
According to the definition of the European
Communities
Commission
of
19.11.2008,
a
cogeneration unit operating with maximum technically
possible heat recovery is said to be operating in full
cogeneration mode. The heat has to be produced at the
site pressure and temperature levels required for the
specific useful heat demand of market. In the case of
full cogeneration mode, all produced electricity is
considered as combined heat and power (CHP)
electricity. If the plant does not operate in full
cogeneration mode under normal conditions of use, it is
necessary to identify separately the amount of the
electricity and heat which was not produced under
cogeneration mode, and to distinguish it from the CHP
production [3].
366 gCeq/kWh are calculated for fossil fuel chains and

2.5-76 gCeq/kWh - for renewable energy.
Table 1. Range of total GHG emissions
from electricity production chains
Energy
Plant
Other
source
emissions
Lignite
217-359
11-14
Coal
181-278
25-79
Oil
121-215
28-31
Natural gas
90-157
16-31
Hydro
0
6.3-64.6
Biomass
0
8.4-16.6
Solar PV
0
8.2-76.4
Wind
0
2.5-13.1
Nuclear
0
2.5-5.7
(gCeq/kWh)
Total
228-366
206-357
149-246
106-188
6.3-64.6
8.4-16.6
8.2-76.4
2.5-13.1
2.5-5.7
The main greenhouse gases are carbon dioxide and

methane. Burning of fossil fuels for electricity and heat
production results to the increased carbon emissions in
the form of carbon dioxide.
The biggest part of the primary energy sources in
Lithuania is used for the energy industry. The energy
industry consists of the public electricity and heat
production, petroleum refining and other energy
production industries (liquefied gas, peat, etc).
Comparison of the utilizations structure of energy
sources in Lithuanian and EU-27 [5] in 2006 is provided
in Fig.2. The indications in the Fig.2 are: EI - for
energy industry, I for industry, T- for transport, H
for household, S for service, O for others.
40
35
Share, %
3. Cogeneration and greenhouse gas emission

Cogeneration is a highly efficient and environmentfriendly way of producing electricity and heat
simultaneously. Practically the entire heat market is
supplied by burning coal, oil, gas, or wood. Greenhouse
gas (GHG) emissions for fossil fuel technologies
comprise the direct (stack) emissions from the
combustion process and the indirect emissions
originating among others from the fuel supply chain.
Fuel use for mining, processing and transport, plant
construction and decommissioning are included into the
full energy production chain. The emissions depend on
generating technology too. Typical results for different
technologies (including CO2, CH4, N2O, etc.) are shown
in Table 1 [4]. The principal emissions are related to the
power plant, and only some 10-20% of the emissions
are related to other steps in fossil fuel chains.
Conversely, for nuclear and renewable energy chains,
there are no GHG emissions from the power plant. But
there are some emissions from the other steps of the
chain.
Hydro power related emissions depend on reservoir
size. Nuclear power is one of the less carbon intensive
generation technologies, amounting to only about 2.55.7 grams of GHG per kWh of electricity produced
(gCeq/kWh) as it is indicated in Table 1. Some 105 to
30
LT
25
EU-27
20
15
10
5
0
EI
Fig.2. Primary energy utilization structure in Lithuania

and EU-27 in 2006
4. Situation in the Baltic States
The possibility of significant climate change, largely
caused by greenhouse gas emissions from fossil fuel
burning, especially carbon dioxide, presents a
fundamental challenge to the goals of sustainable
development. Variations of CO2 emissions during the
years in three Baltic States are shown in Fig.3. It could
be assumed that emissions in Latvia and Lithuania
remained almost constant during the last years and were
lower as compared to the emissions in Estonia, where
the main contribution to the emissions is attached to the
burn up of the oil shale.
197
The share of renewable and indigenous energy

resources (wood, peat, various combustible wastes, etc.)
in the primary energy balance has so far not been
sufficiently used (Fig. 5).
30
EE
LV
LT
20
45
15
40
10
Production, GWh
CO2 Emission, Mt
25
5
0
1990
1994
1998
2002
2006
Fig. 3. CO2 emissions in the Baltic States
EE
35
LV
LT
30
25
20
15
10
5
CO2 emissions in Denmark, Estonia, Latvia, Lithuania,

Finland and Sweden in 2006 are provided for the
comparison (Fig.4)
0
1996
1998
2000
2002
2004
2006
Fig.5. Electricity generation from biomass in the Baltic

states
45
CO2 emission, Mt
40
The use of natural gas as a fuel to generate electricity

and provide district heating, despite its favourable
characteristics, represents a supply security risk and
rising prices.
35
30
25
6
20
15
5. Nuclear energy as a heating source
10
5
Energy production and consumption give rise to

significant health and environmental impacts. Besides
commercial energy sources, large volumes of noncommercial wood and other biomass are burned for
energy supply. Acid gas and particulate emissions from
fossil fuels degrade local and regional air quality. Noncarbon sources such as nuclear energy and renewable
energy sources can make a vital contribution to
reducing emissions.
Use of nuclear energy as a heating source is greatly
challenged by the economic factor since the nuclear
heating reactors have relative small size and often
characterized as having the lower plant load factor.
District heating networks generally have installed
capacities in the range of 10 to 50 MWth in towns and
small communities. 600 to 1200 megawatt-thermal
(MWth) are installed in large cities. Due to the size
effect, nuclear economics are, in general, improved for
larger units. But current design small and medium size
reactors are not scaled down versions of large
commercial reactors. They are intended to be
economically competitive.
Potential radioactive contamination of the district
heating networks is avoided by appropriate measures,
such as intermediate heat exchanger circuits with
pressure gradients which act as effective barriers. No
incident involving radioactive contamination has ever
been reported for any of the reactors used for these
purposes [2]. Heating of Visaginas town from Ignalina
Nuclear Power Plant is an example of a good praxis in
Lithuania.
0
DK
EE
LV
LI
FI
SE
Fig.4. CO2 emissions in the Baltic region countries

About 75% of the primary energy supply in Estonia is
domestic mostly oil shale, wood and peat.
Oil shale is a main fuel for electricity and heat
generation in Estonia and represents the largest fossil
resource of the Baltic region. Russian oil and Russian
gas are imported.
Latvia, contrary, is importing about three quarters of its
primary energy (natural gas, coal and oil products).
Biomass, peat and hydropower are domestic resources.
Two large CHP plants, TPP-1 with an installed electric
capacity of 144 MW and TPP-2 (390 MW), are located
in the city of Riga. CHP is produced mainly in
cogeneration mode, according to the heat-load curve.
During the heating season, when there is a substantial
demand for heating and hot water, Riga CHP plants
produce approximately 80% of the total annual
production volume, while during summer time the
volume of energy production reduces. Operation at full
load during the summer (partly in condensing mode), is
not reasonable from an economic point of view. It is the
feature of cogeneration plant.
About 75% of residential houses in Lithuanias towns
are supplied with heat from district heating systems.
Energy supply to Lithuania is vulnerable, as about 90%
of primary energy is imported from a single supplier
[6]. It will decrease after the closure of the Ignalina
nuclear power plant. The reliable supply of natural gas
can be jeopardised by politically driven interruptions in
the same way as it has already occurred in both
Lithuania and in Latvia concerning oil supplies.
198
metal-filled module sitting in a large pool of secondary

molten metal coolant. The whole reactor sits in a 17
meter deep silo. This reactor is designed for developing
countries and is highly proliferation-resistant.
Some selected characteristics of liquid metal cooled
nuclear battery reactors are provided in Table 2.
Concept of creating nuclear power plant with floating
power source was developed in Russia. It is carried out
on the basis of existing ice-breaker reactor plant and
established shipbuilding technologies. Floating power
unit - KLT-40C reactor plant for heat and power
generation is two-circuit system with pressurized water
reactor.
6. Small-sized nuclear reactors

The IAEA defines small reactors as reactors with power
levels between 0 and 300 MWe while medium size
reactors have power levels between 300 and 700 MWe.
The requirements for worldwide deployment of small
reactors include: nonproliferation assurances; passive
safety; efficient utilization of uranium resources; as well
as simple and reliable plant operation [7]. These new
opportunities are provided by advanced hightemperature and fast neutron reactors.
Fast reactors typically use liquid metal coolants, such as
lead, leadbismuth, or sodium, instead of water.
USA has developed a concept called STAR, the secure
transportable autonomous reactor. The main goal of the
implementation of this concept is to provide reliable and
cost-effective electricity, heat, and freshwater. The
reactor design could also be adapted to produce
hydrogen for use as an alternative fuel for passenger
cars. The STAR proliferation-resistance features include
30 full power years between whole-core refueling [7].
Fuel cycle services and waste management is planned at
centralized facilities operated under international
safeguards oversight.
A set of STAR reactors is under development: SSTAR
(20 MWe, 45 MWth) for small towns; STAR-LM (181
MWe, 400 MWth) for cities and STAR-H. The latest is
a high temperature version of STAR-LM reactor for
production of hydrogen. STAR-LM is modular reactor
with a sequence of clustered modules installed over
time as required for growing electricity demand.
SSTAR is a small, sealed, transportable, autonomous
reactor.
The small fast reactor 4S has been under development
in Japan since 1988. It is thought to contribute to a
multipurpose utilization of nuclear power as an
electrical power supply, heat supply, and desalting of
seawater, especially where the power transmission
infrastructure cannot be maintained. The 4S reactor is
designated as a nuclear battery. The reactor core would
be constructed and sealed at a factory, then shipped to
the site. It can be removed and replaced when the power
is depleted. Core lifetime of 4S reactor is 30 years.
Toshiba proposed to test their new 4S (Super Safe,
Small and Simple) reactor in Galena, Alaska.
Table 3. Small and medium reactors at advanced

development stage
Reactor
Power, MWe Country
CAREM
27
Argentina
MRX
30-100
Japan
KLT-40
35
Russia
IRIS-50
50
USA
Modular
50
USA
SBWR
SMART
100
Korea
NP-300
100-300
France
FUJI
100
Japan, Russia, USA
RSR
100
USA
PBMR
165
South Africa
GT-MHR
285
USA, Russia
BREST
300
Russia
Recently Rosatom/ Russia decided to accelerate the
development of the SVBR-75/100 reactor. SVBR75/100 is a lead-bismuth fast reactor with power level
between 75 and 100 MWe, transportable, having 7- to
10-year refueling interval.
7. References
1. Nuclear power in Switzerland. Country nuclear
power profiles. IAEA, 2003
2. B.J.Csik, J.Kupitz. Nuclear cogeneration: supplying
heat for homes and industries. IAEA. 2002.
3. Commission of the European Communities Brussels,
C (2008) 7294 Commission decision of 19.11.2008
4. Nuclear Energy and Kyoto Protocol. NEA OECD
2002. 53 p.
5. EU Energy in Figures 2009. CO2 Emissions* by
Sector, EU-27. European Commission. DG TREN,
2009, 67 p.
6.National energy strategy. Lithuanian Energy Institute,
2003, 44 p.
7.J.J. Sienicki, D.C. Wade. Nonproliferation features of
the small secure transportable autonomous reactor
(SSTAR) for worldwide sustainable nuclear energy
supply.
Table2. Selected characteristics of liquid metal cooled

nuclear battery reactors
Type
Outlet,
Circulation Power,
MWth
C
4S
485
Forced
30
SVBR-75/100 480
Forced
280
ENHS
<600
Natural
60-400
STAR-LM
590
Natural
400
STAR-H
800
Natural
400
SSTAR
570
Natural
45
The Encapsulated Nuclear Heat Source (ENHS) is
developed at the University of California. ENHS is a
liquid metal-cooled modular reactor concept of 125
MWth. It incorporates a fast neutron spectrum core with
a 15-20 year core life. The core is at the bottom of a
199

ALLOCATING PRODUCTION COST OF CHP PLANT TO HEAT AND POWER

Viktoria NEIMANE*, Antans SAUHATS**, Guntars VEMPERS***, Inga TERESKINA***
*Vattenfall Research & Development, Sweden; **Riga Technical University, Riga, Latvia;
*** Riga Technical University, JSC Siltumelektroprojekt, Riga, Latvia
situation due to low efficiency of the energy

production sources, high transportation losses
and utilization of excessively energy
demanding technologies and buildings.
Furthermore, a number of energy sources
appear to be inefficient from the environmental
and safety viewpoints.
Due to climatic conditions (cold winters) there
is especially high demand towards reliability of
energy supply.
There is considerable number of industrial and
domestic constructions and there is a need for
upgrade of energy production sources, which
creates the possibility for utilization of
progressive technologies for energy production
and consumption.
The systems of district heating are extensively
used in the cities of North-Eastern Europe.
These systems are suited for efficient
utilization of cogeneration in combined heat
and power plants (CHPs).
The intention for fuller utilization of fuel as well as the
obvious advantages of cogeneration resulted in creating
such equipment that provides for constructing CHPs
with small capacities (in the order of tens kWe and
larger). Such power plants can be located near the
energy consumers, which can result in the essential
reduction of energy losses in the heat and electricity
supply networks.
The companies owning CHPs (which can include the
consumers of the energy) have two possibilities to
increase the incomes:
By increasing the heat energy tariffs
By increasing electricity power energy tariffs.
The strategy chosen by the owners of CHP influences
considerably the costs and behavior of the customers.
High heat energy prices stimulate saving measures, for
example better isolation in houses etc. Simultaneously
measures for saving of electrical energy are not
prioritized. It is obvious, that for the rational
development of energy supply systems the objective
relation between the prices on heat and electrical energy
Abstract: This paper is dedicated to the task of cost

determination for heat and power energy production for
small size cogeneration plants. Cogeneration of heat and
power is only possible if there is a demand for both
types of energy. The consumers of both heat and power
are considered as members of the coalition having a
goal to gain additional revenues due to reduced
investments and increased efficiency of power
production. The suggested approach uses the
cooperative game theory for distribution of the
additional revenues between the members of the
coalition. The examples of energy tariffs determination
in the market environment show the rationality and
efficiency of using the considered approach.
Keywords: energy supply, distributed energy resources,
cogeneration, district heating, game theory, decisionmaking.
1. Introduction
Awareness of the global society regarding issues related
to production, distribution and consumption of both
thermal and electrical energy is mainly induced by the
following three factors [1]:
Global reserves of primary energy resources
are limited
Mankind activities increasingly influence the
global climate changes
The demands of modern society towards
reliable energy supply without interruptions
are increasing.
In the countries of North-Eastern Europe the issues of
enhanced efficiency and reliability of energy supply are
especially important due to climatic, historical and
economic reasons. What are the prerequisites there?
Rapid economic development goes along with
increasing energy demand.
The existing energy supply system was built
based on condition that there are inexpensive
primary energy resources available. This
system appears to be inefficient in a new
200
is needed. The methods for determination of costs and

tariffs on different kinds of energy in case of
cogeneration are known [2], [3], [4].
These methods are based on calculation of the summary
revenue R that is sufficient for normal functioning of
CHP. R is gained from selling thermal energy WQ and
electrical energy WE, proceeding from the given rate of
profit. The revenue that is gained from the production
of each type of energy, RE and RQ respectively, is
distributed on the basis of physical laws, for example,
proportionally to the produced energy as follows:
WE
RE
W W
E
Q
WQ
RQ
W W
E
Q
the method is given, and then the task of power supply

is formulated as a game with participation of coalition
with several players. Case study based on real-life data
used in planning of power supply of Riga (Latvia) is
presented. A business plan for construction of a CHP
utilizing wood waste as the main fuel is considered.
2. The theoretical background
2.1. Game- theoretical approach to the planning task
Let us imagine the task of power supply development
planning in the form of a static game with complete
information [5]. The game is presented in normal form
as following:
(1)
I ,
i.e. a list of players I , all situation combinations

S i S i i I and revenues R of each player at all
his strategies and at each combination of the
competitors strategies.
It is assumed that the list and number of the players, i.e.
competing companies, is known; also, that each player
knows the competitors strategies, i.e. the list of their
possible structures and parameters, as well as the
revenues at any combination of structures and
parameters of all the players.
It is known [5] that the solution of the game task,
presented in the form of (3), is a collective choice of
equilibrium strategies. The most often used is the Nash
equilibrium [5], i.e. such a set of strategies s S that
In this case, the unequal value of thermal and electrical

energy is not taken into account.
The method of Exergy is a more common method, by
means of which a different value of energy is accounted
for by coefficient , that changes depending on energy
production technology used.
WE
RE
W W
Q
E
WQ
RQ
W W
Q
E
S i S i I , R R1 , R2 ,...Rn (3)
(2)
for all players i and each alternate strategy si / S i , the

following condition is fulfilled:
For the steam cycle, for example, =0.6 [2] is assumed.

The main disadvantage of this method is the
impossibility of objectively assessing the value of the
coefficient . It is obvious that this coefficient is
different for different power systems. For example, in
case of ample water resources, resulting in production
of cheap electrical energy, the larger value of this
coefficient should be selected.
When selling electrical energy via the power system
grid, its price is determined by market. Hence, using (1)
or (2), we can determine R and RQ. However, in case
of small CHPs, the consumers can receive the electrical
energy from CHP, not using the power system grid. In
this case, the conditions that are dictated by the global
market become non-obligatory for these consumers.
Even more, the considered approaches are not giving
any advantage to the customers whose consumption
profile suits cogeneration profile of heat and power
best. Thus, the contribution of the heat energy
consumers to the operational efficiency from the point
of view of fuel savings and exhaust gases emission
when producing two types of energy is not taken into
consideration in the price setting.
The method based on the cooperative game theory,
which overcomes the limitations described above, is
presented in this paper. First the mathematical base for
i, s i S i Ri ( s i , s i ) Ri ( s i/ , s i )
(4)
i.e. the equilibrium is formed by such a set of strategies,

with which the decision of one of the players to deviate
from such set may only diminish his revenues.
The search for the equilibrium includes the following:
1. Forming the set of all the possible strategies,
excluding the dominated strategies, i.e. strategies si/ of
player i, for which the following condition is fulfilled
[5]:
s i S i Ri ( s i , s / i ) Ri ( s i/ , s / i )
(5)
i.e. the players strategies are excluded, if there is a

strategy that is better, irrespective of the competitors
actions.
2. Search for the equilibrium using (3). Let us suppose
that as a result of such search the only set of strategies
that complies with the Nash equilibrium conditions is
determined. Theoretically, there may be many such
equilibriums; however, some methods are known [5]
that make it possible to diminish their number.
201
3. Considering the rationality and possibility of

organizing a coalition among the players.
4. Choosing the methods for organizing the coalition
and distributing additional revenues among the
participants of the coalition.
If the possibility to form a coalition is taken into
account, the formulation of the optimization task is
modified once more. Due to the need to consider not
only the strategies of individual companies, but also
those of possible coalitions in various combinations, the
dimension of the task increases considerably. Resulting
from the solution of this task, the set of the
sub-optimum plans for each company and their
coalitions at various combinations of possible
competitors plans can be obtained.
possible combinations of the players who form the

coalition are equally probable.
3. Examples of formation of the additional gains
distribution
The district heating plant (DHP) Z is situated in one
of the districts of Riga. At present, the equipment is in
satisfactory condition; however, the growth of heat
loads is expected. Let us assume that there is a search
for the best schematic of power supply to this district
Z. There is a great number of variants for this
schematic (the preventive variant repair work of the
boiler house equipment; individual (consumer) boiler
houses; electrical heating, etc.). Let us suppose that as a
result of comparing these variants, the dominating
strategies have been rejected. In order to increase the
efficiency of power supply, it has been decided to
reconstruct DHP Z into a combined heat and power
plant (CHP) (Fig. 1), which will allow for providing the
consumers of the district not only with heat but also
with electricity. Based on the State support of CHP, it
has been decided to use wood waste as the main type of
fuel. Part of electric energy can be exported or imported
via the distribution grid. Reconstruction of the existing
boiler house that produces only heat energy for the
consumers serves as an alternative for the power supply
schematic presented on Fig. 2.
2.2. Distribution of the gain between the members of

the coalition Shapley value
In case of cooperative behaviour, there is a problem of
revenue distribution between the members of the
coalitions. The simple approach would be to give to
each player his contribution ci :
c i R( S
i ) R( S )
(6)
where R (S ) is the revenue of the coalition S ,

R ( S
i ) is the revenue of the coalition S with
participation of the actor i .
However, such an approach is not anonymous, i.e.
ordering of the players makes difference in the amount
they are rewarded.
In game theory, a Shapley value [6] describes one
approach for the fair allocation of gains avoiding the
mentioned drawback. Fair allocation ensured by
selecting uniformly a random ordering and rewarding
each player his expected marginal cost in ordering.
Since players can form n! possible random orderings,
the probability of set S being ranked exactly before
player i is: S !(n 1 S )! n! . Thus the additional
amount that the player i gets is:
iS N
Fig. 1. The simplified structural schematic of CHP
S ! (n 1 S )!
n!
R( S
i ) R( S ) (7)
In this case, all the electricity is imported by the district

in question (let us call this power supply variant as
Variant A). In Variant A, electricity is imported from
the grid, and its price is fully dictated by market. In this
case, the price for heat is mainly determined by the
price for fuel, capital costs for boiler house
construction, operating costs and credit payment terms.
where n is the total number of players, S is size of the

set S , the sum extends over all subsets S of N not
containing player i .
In the simplest case, when only two players participate
in the game, the expression (7) is simplified and obtains
the following form:
1 2 R( S
i ) R( S ) / 2
(8)
The Shapley value describes the fair (in a sense

determined by the accepted axioms) [6] distribution of
additional gains in case of formation of the coalition. In
particular, the definition is based on the assumption that
202
the coalition of the consumers of both types of energy

gains the additional revenue at the expense of difference
in the prices for imported and exported electricity.
DISTRICT Z
Electricity
Table 1. The principal data used in calculations.
Electricity
DISTRIBUTIVE
GRID
Name
1
2
Investments
Summary costs of fuel (per
hour)
Price of electricity exported
Price of electricity imported
Amount of heat supplied to
the grid
Amount of electricity
supplied to the grid
3
4
5
Fig. 2. The schematic of power supply to district Z

(Variant B1)
In the event of CHP construction, two more variants of

power supply appear:
Variant B1 (Fig. 2.): CHP supplies heat to
district Z, and electricity is exported to the
grid; the district, in its turn, imports the
necessary electrictricity from the grid.
Variant B2 (Fig. 3.): CHP supplies both types
of energy (heat and electricity) to district Z,
i.e. a coalition of consumers has been formed
within the district. Such coalition is possible,
if its expenses Ek are less than the summary
expences E1+E2 of the consumers of heat
(E1)
and
electricity
(E2),
acting
independently, i.e., if:
Ek < E1 + E2
No.
Amount,
Measuring units
6834 thous. EUR
980622 EUR
105 EUR
126 EUR
183,6 thous. MWh
16,6 thous. MWh
It is possible to distribute the additional revenue, using

(7) and (8), which will reduce the prices to an even
greater degree.
DISTRICT Z
Heat
CHP
(9)
Electricity
...
...
Heat
CHP
Fig. 3. The schematic of power supply to district Z

(Variant B2)
The heat consumers provide the conditions for the

functioning of a cogeneration plant. They are interested
in the formation of a coalition with electricity
consumers, provided that the sale price is fixed higher
than the price of energy that is exported to the grid. At
the same time, the power consumers will be interested
in a coalition, if the electricity price is lower than the
price of energy imported from the grid. A question
about fixing reasoned tariffs emerges.
The principal data used in calculations are presented in
Table 1. When calculating investments, it was taken
into account that the State support is 30% of the
expenses necessary for construction of CHP. In this
paper it is assumed that it is possible to calculate the
costs associated with providing energy supply for
district Z. In the considered Variant A, electricity is
imported from the grid, and its price is fully dictated by
market. In this case, the price for heat will be mainly
determined by the price for fuel, capital costs for boiler
house construction, operating costs and credit payment
terms.
In Variant B1, when all the electricity is exported, its
price is also determined by market. On condition that
there is a task for gaining the normalized profit, the
revenue from selling electricity makes it possible to
reduce the price for heat supply. However, in this case,
the district is forced to purchase the relatively expensive
electricity from the grid. In power supply Variant B2,
In price calculation, it was assumed that the district is

fully consuming the electricity produced. In case if
some part of electricity is exported to the grid, the
difference between Variants B1 and B2 will diminish.
Such dependence indicates that the coalition is
interested in consuming the electricity of its own
production. The considered approach regarding the
distribution of the additional revenue can serve as a
basis for creating an appropriate procedure of
calculation with the consumers, depending on their heat
and electric load charts. Notice that the results shown
on Fig. 4 have been obtained, using the real prices for
fuel and equipment. Hence, these results substantiate, in
the first instance, the efficiency of using small-capacity
cogeneration plants, especially, with high prices for
energy carriers.
203
3.
Construction of CHP and formation of a coalition

of heat and electricity consumers makes it possible
to essentially reduce tariffs for both types of
energy.
The formation of a coalition of energy consumers for
the construction of CHP with increased capacity may
result in the additional reduction of energy tariffs.
126,00
100
100
EUR/ MWh
120
EUR/ MWh
120
80
60
40
38,50
Price of heat
80
Price of electricity
60
40
20
115,00
37,60
5. Acknowledgements
20
0
Variant B1
The research activity presented in this paper is

supported by the European Union through the SES6CT-2003-503516 project "EU-DEEP".
Variant B2
Fig. 4. Comparison of heat and electricity tariffs in Variants

B1 and B2 (with the State support)
6. References
Fig. 4 schematically shows the comparison of heat and
electricity tariffs in Variants B1 and B2. The mentioned
prices, for which the consumers of district Z will buy
the thermal and electric energy, differ markedly from
one another. Hence, the application of the method based
on the game approach results in electricity prices that
are sufficiently close to the market prices as well as to
relatively low heat prices. It appears that such price
ratios stimulate the extensive application of
cogeneration processes in energy production.
Fig. 5 represents the prices of heat and electricity,
excluding the State support. In this case, the heat prices
appear to be higher than the average heat prices in Riga.
Hence, the considered Variant turns out to be noncompetitive.
1. Directive 2001/77/EC of the European Parliament

and of the Council of 27 September 2001 on the
promotion of electricity produced from renewable
energy sources in the internal electricity market.
2. N.A.Slavina, E.M.Kosmatov, E.E.Barikin. Methods
of CHP production costs allocation. Power stations
Nr. 11, 2001, p.p.14-17 (in Russian) Thermoflow.
Comprehensive Thermal Engineering software.
Thermoflow Inc. April 2002.
3. Viktoria Neimane, Antans Sauhats, Guntars
Vempers, Inga Tereskina, Galina Bochkarjova.
Approach for Power Supply System Planning
Based on Cooperative Game Theory, The 5th
International Conference on the EUROPEAN
ELECTRICITY MARKET - EEM 08, Lisbon,
Portugal, May 28-30, 2008, ID#206.
4. V. Neimane, A. Sauhats, J. Inde, G. Vempers, G.
Bockarjova. Using Cooperative Game Theory in
energy Supply Planning Tasks, 16th Power Systems
Computation Conference- PSCC2008, Glasgow,
Scotland, July 14-18, 2008, ID#220.
5. Osborne, M.J. and Rubinstein, A. ,A Course in
Game Theory, MIT Press (Chapters 13, 14, 15),
1994.
6. Lloyd S. Shapley. A Value for n-person Games. In
Contributions to the Theory of Games, Volume II,
by H.W. Kuhn and A.W. Tucker, editors. Annals of
Mathematical Studies v. 28, pp. 307-317. Princeton
University Press.
126,00
100
100
80
60
55,40
EUR/ MWh
120
EUR/ MWh
120
115,00
Price of heat
80
60
40
40
20
20
54,50
Price of electricity
0
Variant B1`
Variant B2`
Fig. 5. Comparison of heat and electricity tariffs in Variants

B1 and B2 (without the State support of investments)
4. Conclusions
1.
2.
Methods based on the game theory can contribute

to making the right decision about the
development of energy supply sources. In
particular, the cooperative game taking into
consideration the possibility of building the
coalition should be used. In due course, this
approach will result in a more efficient energy
supply system and acceleration of DER
implementation in power systems.
The known methods of determining tariffs for
energy produced in the cogeneration mode are
poorly adapted for application in the market
environment.
204

CROSS-BORDER CAPACITY AUCTIONS

Rimantas DEKSNYS*, Robertas STANIULIS*, Giedrius RADVILA**
* Kaunas University of Technology, Lithuania,
** AB Lietuvos energija, Lithuania
network development and return on investment.

Investment security and benefit are showing the
parameters determined through effective utilization of
power network, market development level, possibilities
of electricity market expansion and connection.
Therefore one of the most important indicators is
satisfaction to the market situation of cross-border
capacity, controlled by transmission system operators,
and mechanisms of congestion management and
congestion income formation. In continental part of
Europe historically cooperation of big companies, which
were the main industrial consumers, and foreign
generators was developed by buying electricity on the
basis of mutual agreements. Therefore later with the
development of free and organized market mechanisms,
the task of system control for transmission system
operators has emerged. Therefore mechanism of explicit
auctions for cross-border capacity allocation and control
was introduced in continental part of Europe. Nordic
countries, where electricity generation is diversified,
have chosen more advanced implicit auction mechanism
for cross-border capacity allocation.
This paper is actual because of perspective strategy of
Baltic-Nordic and Baltic-Polish electricity systems
connections, which could lead to the integration of
Baltic electricity system and market into united
European transmission system UCTE and European
electricity market. The main task of this paper is to
summarize the cross-border capacity allocation
mechanism applied in European countries and to
evaluate the possibilities of new Baltic interconnections.
Abstract: Paper is actual because of perspective strategy

of Baltic-Nordic and Baltic-Polish electricity systems
connections, which could lead to the integration of
Baltic electricity system and market into united
European transmission system UCTE and European
electricity market. The main task of this paper is to
summarize the cross-border capacity allocation
mechanism applied in European countries and to
evaluate the possibilities of new Baltic interconnections.
Keywords: electricity market, congestion, cross-border
capacity, capacity auction, explicit auction, implicit
auction.
1. Introduction
With the development of international relations in
electricity trade, with the increasing gross domestic
product, import and export, integration of separate
power systems is growing, which enables the
development of common electricity market. In this stage
of development of energy systems common system
infrastructure is being developed, i. e. cross-border
links, which connect separate energy systems with each
other, enable connection of smaller energy systems to
the bigger ones, more effective utilization of energy
resources and appearance of market elements in energy
systems. Together with infrastructure organizational
requirements of cross-border flows control without
discrimination and with understandable and reliable
principles are growing. The most important role falls for
transmission system operator in this process.
Transmission
system
operators
under
strict
governmental regulation have to ensure reliable and
qualitative supply of electricity, ensure free third party
access to the network and ancillary services, which are
related with energy quality and reliability of supply.
Other important task for transmission system operators
is network development, which could meet electricity
quality, control, technical, organizational and market
needs. Determining those needs, objective parameters
are needed, which could show economic benefit of
2. Methods of cross-border capacity allocation

Auctions are well known market mechanism, which are
used for distribution of limited resources, in this case
cross-border capacity. Auctions are transparent and
efficient as they allocate capacities to those that value it
the most.
At present, a variety of different capacity allocation
methods are in use in electricity market. Although each
method is unique, but they can be grouped as follows
[1, 2]:
205
1. First come first served: capacity is allocated

according to the order in which the transmission
requests have been received by the TSO. Starting from
the earliest request, all requested amounts of capacity
are fully granted until the available capacity is used up.
2. Pro rata: all requests are partially accepted in the
way that each applicant is granted a fixed share of his
requested capacity amount, the share being equal to the
amount of available capacity divided by the sum of all
requested capacity amounts.
3. Explicit auction: along with the requested capacity
amount, the applicants have to declare how much they
are willing to pay for this capacity. These bids are
ordered by price and allocated starting from highest one
until the available capacity is used up. In contrast to the
previous methods, network users incur cost for obtaining
transmission capacity (usually the price is set to the bid
price of the lowest allocated bid).
Variations in auction design are possible with regards to
the bidding mechanisms or the time periods, which are
auctioned (days, weeks, months, years). Explicit
auctions separate energy flows from transmission
capacity. Hence, once interconnection capacity has been
secured by a market participant, the participant will need
another transaction for energy. This can be done on the
bilateral market or on a power exchange.
4. Implicit auction: with all previously described
methods, the electricity spot markets are separated from
the transmission capacity allocation procedure and close
after the transmission capacity has been allocated. With
implicit auctioning transmission capacity is managed
implicitly by the spot markets: network users submit
purchase or sale bids for energy in the geographical
zone where they wish to generate or consume, and the
market clearing procedure determines the most efficient
amount and direction of physical power exchange
between the market zones. Hence congestion
management model is integrated with the energy market
and a separate allocation of transmission capacity is not
required.
Since this method does not separate energy flows from
transmission capacity, the process is simpler for market
parties. Such system creates revenue for the market
operator equal to the size of the interconnector capacity
times the price difference. The revenues which the
market operator collects should in theory, be the same as
the revenues from an explicit auction. The main
drawback of this option is that it requires an organized
market to function in the downstream side of each
congested interconnection. This is currently an
important practical limitation in many countries where a
liquid and stable organized electricity markets have not
yet developed.
Especially in the initial stages of market development,
this has included non-market-based methods, such as the
first-come-first-served principle, retention or a pro-rata
allocation, which have been gradually replaced by
auction-based mechanisms over the past few years.
Although explicit auctions can still be regarded as the
prevailing methods for the time being, implicit auctions
are generally regarded as being the most efficient
solution and can be expected to become of increasing

importance as regional market integration proceeds.
3. Economics of cross-border trade
The benefit from the cross-border trade could be
obtained two ways. First, because of cheaper imported
electricity, average electricity price in the importing
country is decreasing and generators of the exporting
country are earning more money. Second, because of
limited cross-border capacity additional incomes are
earned by the owners of interconnection.
Congestion incomes could be calculated evaluating
electricity price difference between interconnected
parties, duration of congestions and capacity of
interconnection:
R MC T ATC ;
where: R congestion revenues; MC difference of
marginal electricity prices between interconnected
parties; T duration of congestions; ATC available
transfer capacity of interconnection.
Economic principles of cross-border trade between
regions A and B are shown in Fig. 1. Without
interconnection between those regions in region A
electricity price would be MC A1 and in B region MC
B1, and MC A1 > MC B1.
If available transfer capacity exists between those
regions, marginal price of A region would decrease till
MC A2 and in B region would increase till MC B2. In
case that there are no congestions between regions,
electricity price would be equal MC AB in both regions.
Price
Net import curve region A (deficit)
MC A1
MC A2
Welfare
Welfaregain
gain
MC AB
Congestion rent
Welfare loss
MC B2
Welfare gain
Net export curve region B (surplus)
MC B1
ATC
ATC for zero

congestion
Volume
Fig. 1. Economics of cross-border trade
In Fig. 1 different areas show distribution of benefit

between regions A and B, when limited interconnection
capacity ATC exists. In A region electricity price
decreases from MC A1 till MC A2 and the benefit of A
region consumers is equal to the area above the line MC
A2 shown in the figure. In B region because of increased
electricity price from MC B1 till MC B2, benefit of B
region generators is shown below MC B2 line in the
figure. Area of the congestion rent is showing the benefit
of interconnection operators because of the price
206
difference between the regions. This benefit is usually

divided between operators in proportion to the amount
of shares or by other methods. Reduction of social
benefit, which represented by the area in the figure, is
appearing because of limited transfer capacity between
the regions. Maximum social benefit is obtained in the
case then there are no cross-border trade limitations
between the regions.
and benefit is obtained only through the deployment of

internal market resources and internal competition.
4.2. Second scenario sufficient interconnection
capacity
Equal electricity price in the common electricity market
is formed simulating the interconnection according to
the second scenario, when there is some certain
available transfer capacity ATC between the markets,
which is available for the common electricity market.
But this price is being formed only until import or
export through the available interconnection is not
restricted by larger supply amounts from the lower to
higher price region.
4. Simulation of new interconnections in the Baltic

region
Success, economic and social benefit of cross-border
interconnection projects between separate electricity
markets depend on the electricity market conditions and
system control possibilities, because these subjects let
objectively evaluate pay-back time of the investments,
its security, social and economic benefit in the long run.
To solve those problems evaluation of technical level
and future interconnection projects, current and future
analyzed and surrounding countries market situation,
evaluation of market isolated situation, perspective
market development tendencies, simulation of market
based interconnection capacities auction mechanisms
are needed.
Simulation of new interconnections between two regions
could be done by three scenarios described below.
Price,
EUR/MWh
Market area B
Import
Quantity, MWh
X MW
Export
Quantity, MWh
Export from B/
Import to A
Fig. 3. Electricity price formation in the common electricity

market with sufficient interconnection capacity
Market area A
The main assumption of this scenario is that available

transfer capacity is larger than actually needed capacity
(X) for electricity import and export.
The influence of interconnection capacity could be
simulated as the additional possibility of lower price
region market players (in this case internal market) to
sell electricity and for higher price region market players
(in this case neighboring external market) to buy
cheaper electricity that is proposed in the isolated
market.
So in this case:
Market area B
Market price P2
0 MWh
0 MWh
Available transfer
capacity[ATC]
Price,
EUR/MWh
Import
Quantity, MWh
Market price P2
Effect of
interconnection
Import to B/
Export from A
4.1. First scenario market isolation
Market price P1
Market price P1
Market area A
Export
Quantity, MWh
Fig. 2. Electricity price formation with the isolated electricity

market
Internal market price MC2 = External market price MC1.

With the condition that available transfer capacity ATC
is higher than actually needed capacity X.
Investigation of market price changes, using marginal
price methodology, having interconnection and
possibilities to import and export electricity shows that
generators in the exporting country could increase their
revenues, but costs for consumers would be higher. And
on the contrary, consumers costs would be lower in the
importing country, because of the lower electricity price.
Revenue of importing country generators would be
lower, because they will have to compete with cheaper
imported electricity. Simulation shows that there is no
congestion rent for interconnection operator, because
there is no price difference between the markets due to
sufficient interconnection capacity.
Marginal electricity price in the isolated electricity

market, without physical connection or with available
transfer capacity equal to zero, would be developing
independently from the price of neighboring countries.
Import and export activities would not be possible and
only supply and demand of internal market players form
electricity price.
So in this case:
Internal market price MC2 External market price MC1.
Analyzing the influence on generation and consumers, it
is necessary to state that in this case there is no crucial
effect on their costs and revenues. Social and economic
benefit of common electricity market is not achieved
207
Lower price market
Higher price market
Price,
EUR/MWh
GENERATION
CONSUMPTION
A area market
price P1
P2
CONSUMPTION
GENERATION
Benefit for
consumers
P1
A area market price =

B area market price
Benefit for
generators
B area market
price P2
Decrease of generation
price = decrease of
customers costs
Increase of generation
price = increase of
customers costs
SELL BIDS
(GENERATION)
Export f rom B
Import to A
Quantity, MWh
Import to A
Export f rom B
Quantity, MWh
Fig. 4. Relation of revenue and costs for generators and consumers
In the situation when electricity is exported from the

market, the greatest benefit would get exporting
generators, because of the possibility to sell their energy
in the high electricity price markets. In the importing
region benefit would be gotten by electricity consumers,
because more expensive local generators would be
replaced by cheaper electricity import.
The tendency of electricity price changes would depend

on the customers elasticity, when customers are
reacting to indicative electricity price and change their
consumption behavior. But the tendency of revenue and
costs distribution is the same as in the second scenario.
In this case:
Internal market price MC2 External market price MC1.
4.3. Third
capacity
scenario
limited
interconnection
With the condition that available transfer capacity ATC
is equal to actually needed capacity X. Available
transfer capacity ATC is lower than the quantity of
proposals to sell electricity using interconnection
capacity, what is corresponding to demand curve in the
higher price area.
Having price difference between external and internal
market, congestion rent is being cumulated by electricity
exchange, which could be distributed among
transmission system operators at the end of the year. The
level of congestion rent is indicating how often
congestions occur and is it necessary to increase
interconnection capacity. On the other side, these
revenues could guarantee the return of investments of
new interconnection, therefore further analysis of
interconnection return of investments is related with
possibility to forecast electricity price in two analyzed
markets and to distribute congestion rent for the
investment.
The third scenario is simulated for the situation when the

quantity of sell bids in the higher price region is higher
than available transfer capacity of interconnection. In
this case different price areas would develop.
Price,
EUR/MWh
Market price P1
Market area A
TSO revenues
Market area B
Import
Quantity, MWh
Market price P2
Effect of
interconnection
0 MWh
X MW
Export
Quantity, MWh
Eksportas i B
Importas A
Import to B/
Export from A
Available transfer
capacity[ATC]
5. Simulation of interconnection return of

investments
Fig. 5. Electricity price formation in the common electricity

market with limited interconnection capacity
Simulating the return of investments of BalticSwedish

interconnection certain assumptions have to be done
which influence calculation results. One of the main
assumptions is the forecast of electricity price
difference in Baltic and Nordic electricity markets for
the pay-back period, which is equal to 7 years. It was
assumed that interconnection is simulated for the period
from 2013 to 2019. This period includes big changes not
only in the Baltic, but also in neighboring markets:
In the third scenario the situation is analyzed when all

available transfer capacity of the interconnection is used
by market players.
The influence of interconnection capacity is maximal,
because all available transfer capacity is used for needs
of higher price area. In this case bigger benefit is for
exporting market, because sell bids are cleared fully
utilizing interconnection capacity.
208
Table 1. Forecast of Nord Pool electricity physical exchange price in 20102019, EUR/MWh
Year
Maximal Nordic electricity
market price
Minimal Nordic electricity
market price
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
53.0
55.1
54.0
56.2
55.0
57.2
56.0
58.2
57.0
59.3
37.9
38.1
41.0
45.0
46.0
47.8
49.8
51.7
53.8
56.0
Table 2. Forecast of Baltic electricity physical exchange price in 20102019, EUR/MWh

Year
Maximal Baltic electricity
market price
Minimal Baltic electricity
market price
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
98.0
101.9
106.0
110.2
114.6
119.2
124.0
129.0
134.1
139.5
72.1
75.0
78.0
81.1
84.3
87.7
91.2
94.9
98.7
102.6
1. Closure of Ignalina nuclear power plant.

2. Start of operation of new Visaginas nuclear power
plant in 2018.
3. Russian electricity market growth, development and
integration into other markets.
4. Installation of new capacities in the Baltic region and
annual growth of demand by 4 %.
5. Buy-out
of
Estlink
interconnection
from
shareholders in 2014 and further exploration as
infrastructural project.
6. Estonian electricity market opening in 2013.
7. Closure of Estonian oil-shale power plants.
8. New period of quotas distribution of greenhouse
gases emission in 2013.
time Estonian and Lithuanian condensing power plants

could be eliminated from the market.
Using electricity price forecast in Baltic and Nordic
countries and modeling the return of investments of
BalticNordic interconnection, it could be stated that
Baltic electricity market will be an importing market.
For the formation of revenues from the interconnection,
two assumptions have to be fulfilled:
interconnection has to be an infrastructural or by
other means free access to use interconnection
capacity for market players has to be ensured;
sufficient electricity price difference and market
liquidity, ensuring sufficient amounts of electricity
sell and buy bids on both sides of interconnection.
Total annual consumption in Nordic countries is 412.8
TWh, while in the Baltic countries it is 25.8 TWh,
therefore possibilities for Baltic countries to use the
advantages of Nordic electricity are much bigger than
the opposite way. Calculating the return of investments
using forecasted electricity prices for the period 2013
2019, it is necessary to determine the capacity of
interconnection. According to the future plan of
interconnection, two scenarios of interconnection
capacity are reasonable 700 MW and 1000 MW.
Therefore further calculations will be for two
interconnection capacity scenarios.
Duration of congestions depends on the price difference
between the markets and technical reliability of the
interconnection. To evaluate technical inaccessibility of
the interconnection due to annual maintenance and fault
conditions could be done by knowing projected
utilization factor of interconnection. Usually this factor
is equal to 97 %, therefore duration of congestions is
decreased by 3 % of annual operation of the
interconnection. Evaluating the tide period, when
situation in the Baltic market is similar to Nordic
market, it is advisable to decrease the period of
congestions by two months, when there will not be buy
bids in Nordic countries, though interconnection
capacity will not be totally used (second scenario). To
ensure system security minimum generation level in the
system will be set, which could ensure hot operational
reserve, guarantee sufficient regulation range for
generation and consumption balance.
With the varying maximum price difference between
Baltic and Nordic electricity markets from 65.2
Taking into account the forecast of long-term physical

market contracts, the most probable that electricity price
in the physical electricity market Nord Pool would reach
53 EUR/MWh in 2010 (Table 1) [4]. Correction factors,
related with economic indicators of different markets,
could more precisely set the margins of electricity price
increase, but to set this tendency more precise economic
evaluation and officially confirmed forecasts of
development scenarios are needed. In the price forecast
annual inflation rate of 4 % is assumed. Assumptions are
evaluated taking into account historical data for 8 years
period, therefore minimal forecasted margin is set taking
into account prices of futures, and the maximal price
forecast of spot market electricity price.
Taking into account generation structure and annual
energy generation in the Baltic countries, after closure
of Ignalina NPP indicative electricity price in Baltic
countries will be formed by Estonian and Lithuanian
condensing power plants. Forecasted electricity price in
the Baltic countries for the period from 2010 to 2019 is
provided in the Table 2. Annual inflation rate of 4 %
was assumed in this forecast. In Baltic and in Nordic
countries seasonal factor has a big influence on
electricity prices. During the tide effectively using
Latvian Daugava hydro power plants, Lithuanian
Kruonis hydro storage power plant and importing the
rest of energy from neighboring countries, influence of
condensing power plant on indicative electricity price
could be eliminated. Taking into account seasonal
factor, energy import possibilities and redistribution of
generation sources, electricity prices in the annual
perspective could decrease by 30 %, because during this
209
EUR/MWh (2013) till 83.5 EUR/MWh (2019)

maximum possible annual congestion revenue with 700
MW interconnection capacity vary from 320 to 410 mill.
EUR, and with 1000 MW interconnection capacity
from 501 to 821 mill. EUR. With the varying minimum
price difference between Baltic and Nordic electricity
markets from 36.1 EUR/MWh (2013) till 46.7
EUR/MWh (2019) minimum possible annual congestion
revenue with 700 MW interconnection capacity vary
from 177 to 229 mill. EUR, and with 1000 MW
interconnection capacity from 253 to 327 mill. EUR.
Not adequate situation in the Baltic market is being
formed after closure of Ignalina NPP. Big price
difference between the markets is causing high level of
congestion revenues, which are equal to big countries
France, Germany and Italy revenues. Physically
isolated Baltic electricity market with high electricity
price has enough potential to initiate new
interconnection projects, which pay-back could be
guaranteed by market connection and integration with
Nordic electricity market.
the end of the year. The level of congestion rent is

indicating how often congestions occur and is it
necessary to increase interconnection capacity. Also,
these revenues could guarantee the return of
investments of new interconnection.
5. Simulating the return of investments of Baltic
Swedish interconnection the forecast of electricity
price difference in Baltic and Nordic electricity
markets for the 7 year pay-back period from 2013 to
2019 is needed. The most probable electricity price
in the physical electricity market Nord Pool would
reach 53 EUR/MWh in 2010 and 59.3 EUR/MWh in
2019. Forecasted maximal electricity price in the
Baltic countries in 2010 could reach 98 EUR/MWh
and in 2019 140 EUR/MWh.
6. Using electricity price forecast in Baltic and Nordic
countries and modeling the return of investments of
BalticSwedish interconnection, it could be stated
that Baltic electricity market will be an importing
market. For the formation of revenues from the
interconnection, interconnection has to be an
infrastructural or by other means free access to use
interconnection capacity for market players has to be
ensured and also sufficient electricity price
difference and market liquidity, ensuring sufficient
amounts of electricity sell and buy bids on both sides
of interconnection, have to be guaranteed.
7. Big price difference between the Baltic and Nordic
markets is causing high level of annual congestion
revenues, which vary from 177 to 410 mill. EUR for
700 MW interconnection and from 253 to 821 mill.
EUR for 1000 MW interconnection and are equal to
big countries revenues. Physically isolated Baltic
electricity market with high electricity price has
enough potential to initiate new interconnection
projects, which pay-back could be guaranteed by
market connection and integration with Nordic
electricity market.
6. Conclusions
1. Success, economic and social benefit of cross-border
interconnection projects depend on the electricity
market conditions and system control possibilities,
because these subjects enable objectively evaluate
pay-back time of the investments, its security, social
and economic benefit in the long run.
2. Marginal electricity price in the isolated electricity
market according to the first analyzed scenario,
without physical connection would be developing
independently from the price of neighboring
countries. Social and economic benefit of common
electricity market is not achieved and benefit is
obtained only through the deployment of internal
market resources and internal competition.
3. Simulation of interconnection according to the
second scenario with sufficient interconnection
capacity shows that generators in the exporting
country could increase their revenues, but costs for
consumers would be higher. And on the contrary,
consumers costs would be lower in the importing
country, because of the lower electricity price. There
is no congestion rent for interconnection operator,
because there is no price difference between the
markets due to sufficient interconnection capacity.
4. Simulation of interconnection according to the
second scenario with limited interconnection
capacity shows that different price areas would
develop. Having price difference between external
and Baltic market, congestion rent is being
cumulated by electricity exchange, which could be
distributed among transmission system operators at
7. References
1. Analysis of Cross-Border Congestion Management
Methods for the EU Internal Electricity Market.
Study commissioned by the European Commission.
CONSENTEC, Final report, June 2004.
2. Market Operator (Regulatory Oversight), Final
Version. Energy Regulators Regional Association
(ERRA), Prepared by KEMA International B.V.,
December 2007.
3. Towards a common coordinated regional congestion
management method in Europe. Study commissioned
by European Commission, CONSENTEC, October
2007.
4. Market information. www.nordpoolspot.com, 2008.
210

THE NEURAL NETWORK AS THE INDUCTIVE ENGINES

STATES ANALYSER
Boguslaw TWAROG, Zbigniew GOMOLKA
University of Rzeszow, Institute of Technology, Poland
Structure of neural network depends as well on

complexity of solving problem. Possibility of
generalization is important advantage of artificial neural
network over traditional diagnostic methods. If neural
network is teach by representative set of data then it is
able to work properly even for cases dont considered
before [2].
Abstract: The idea of object diagnostics with use

artificial neural network and the method of computer
simulation vibrate processes of inductive motor is
proposed. Information about system is contained in
vibration signals or in the current load flow, that allows
to define its state. Changes of air gap and one-sided
forces of magnet attraction are taking into consideration.
The integration of differential equations is realized by
Gear implicit method. Using neural networks as
diagnosis instrument we can determine different degree
of damages.
2. Architecture of diagnostic system

In our case an electric engine is the object of analysis.
The proposed method is non interference and using
artificial neural network. It allows to detect first
symptoms of failure what is important for assuring
lasting work without damages. The measurement signal
is vibration signal of working engine which contain
information about object state.
The proposed system consists of three main parts:
measurements sub-system, sub-system of artificial
intelligence, presentation sub-system.
Realization of such measurement system (working in
real time) was done on laboratory standing with IBM PC
computer and PCI 1711 series measurement module
equipment [5]. That module enables to make
measurement of electric variables or representing by
them physical variables using analog-digital converter.
Advantage of measurement module is tuning possibility
of input and output voltage range. Proper adjustment of
input and output voltage range can prevent over tuning
and increase measurements accuracy. Especially
measurement system contains:
1. microprocessor system: analog interface for system
protection, serial interface for communication with
another computers, microprocessor card with analog/
digital and digital/analog converters,
2. software system: sampling and estimation, taking
decision, error classification.
Described systems have high sensitivity, non
complicated construction and universal applications.
The artificial intelligence systems, which used neural
network, detects failures or their symptoms. That
detection bases on spectrum of object vibrations. The
Keywords: mathematical model, diagnostic, inductive

engine, Fourier transform, spectrum of signal, neural
network, back-propagation.
1. Introduction
The system of defect detection has to shown on-line that
failure occurs. Usually in techniques we have two
categories of detection and error diagnosis: estimation
methods and pattern recognition methods.
Estimation methods require knowledge of mathematical
model, which reflect real process. Model cannot be too
complicated because diagnostics is realized on-line and
solution of model equations shouldnt take too long
time. When error of device functioning or its failure
occur diagnostic system should detect it immediately.
Next kind of error and his location should be properly
classified. If it is possible should isolate influence of
failure in system.
In second category of recognition methods there is no
mathematical model, process of diagnostic and error
classification. There is representation space of process
from measurements to space of decision. Traditional
recognition and classification is divide to three levels:
measurement, analysis, classification. Artificial neural
networks can realize such classification tasks in very
efficient ratio. It connects to choice of network
architecture, which depends on problem which realize.
In that case chosen characteristics of object determine
number of neurons in input, output and hidden layer.
211
xR xm cos ;
Fourier transform is perform on the signal of vibration.

Next the artificial neural network analyses amplitude and
phase Fourier spectrum of that signal.
If defects (symptoms of defect) appearance in system
that causes changes in amplitude and phase of signal.
The neural network respond to that change is generation
of information about actual object state. It is worth
nothing that this process is execute in real time routine.
Depend on case of object different structures of neural
networks and different methods reduce size of input data
(pre-processing) can be applied. Different learning
methods, for example Back Propagation, Counter
Propagation can be applied to different network
architecture. After end of learning process artificial
neural network can work in diagnostic system.
Data presentation system makes possible interpretation
of result obtained from neural network. In that case it is
proper work or type of detected by neural network
failure. Moreover presentation system allows receiving
intermediate results or presentation of statistic
calculation in approachable form (tables, diagrams).
where: R arc tg ( y R / x R ) angular.

On the basis (1) and (2) received:
x R ym cos y R xm sin
x R2 y R2
where:
xm, ym speed of rotor mass centre in coordinates x, y,
angular speed of rotor.
x m xm ;
dS
dt
dR
yR
cR
dt
dR
dt
u p0R S rS iS ;
(4)
p0 ( R ) R rRiR
where:
ik, iik (i=S, R; k= , ) full linkage magnetic fluxes
and currents,
u, u stator voltages,
rS, rR
resistance,
p0
number of pole pairs (indexes S and R
indicate belonging to stator and rotor, and and belong
to separate electric contours).
Stator voltages we find as:
u U m sin(0t p0 R ) (5)
Um, 0 - amplitude and angle frequency of input voltages
0 xR xm
(3)
p0 (R ) R rRiR ;
u U m cos(0t p0 R );
ym
u p0R S rS iS ;
dt
dS
y m ym ;
Differential equations of electromagnetic rotor state in

coordinates and have form:
All classical theory of electric machines is built by

condition that air gap between rotor and stator is
constant, therefore she is useless in practice in this case.
The mathematical model [1] of induction motor describe
sufficient complicated of simultaneous crosswise and
rotation oscillations of rotor and stator, which is attached
to foundation by elastic supports. We simplify the
problem in case of vibration processes of rotor only at
stiff immovable stator, taken in consideration the
susceptible mechanical links of fixation rotor to stator.
The rotor we consider as absolutely hard substance with
mass m and moment of inertia adequately to mass centre
J. The eccentric of rotor construction arise as result
of mass centre cm shift in relation to its axis of
rotation cR, what respond to value . In stationary
coordinates x, y, coordinates of rotor mass centre and its
axis of rotation are marked adequately as xm, ym and xR,
yR. Angle of rotor turn , and its axis of rotation R.
cm
(1)
Lets take orthogonal coordinates and spinning in

way that axis crosses geometrical centre of rotor.
Angular speed of that system R can be obtained from
equation:
R R
(2)
3. The mathematical model of electric engine
y R ym sin
Currents we find as follow Fig. 2.:
imk m mk k , m S , R; k ,
R2
R1
(6)
where S, R are inverse inductance of stator and rotor,

, are main linkage magnetic fluxes
S aS bS R aR bR ;
Fig. 1. The graphical model of the rotor-stator system of the

inductive engine with crosswise engine movements
S mS nS R mR nR
From geometrical dependence we have:
where:
212
(7)
1 TN M TS 2
cos p0
M b
d ;
( )
0
2
N b
0
sin 2 p0
d ;
( )
1 TM N TS 2
The velocities of rotor movement we obtain from

general equation of dynamic
sin p0 cos p0
S b
d ;
( )
0
2
T S R
1 TM 1 TM TS 2 ;
p0
3w2
d xm 1
Fx c x xR x xm sin ;
dt
m
d ym 1
Fy gm c y y R y ym cos ;
dt
m
dym
d
d 1
ME MM m xm sin m
mg cos ,
dt
There b = 3wa0/(p0) is constant coefficient; is

magnetic resistance of main magnetic circuit of electric
machine with the exception of resistance of air gap; w is
number of effective windings of phase stator; 0
magnetic constant; p0 is angle coordinate of stator; ()
is function dependence of value of air gap from angle
dt
dt
(13)
where R1 is inner radius of stator; R2 is external radius of

rotor.
where MM is resistance moment; cx, cy are stiffness of

rotor supports; x, y are dissipation coefficients; g is
gravitations constant.
The system of 11 differential equations (1), (3), (4), (13)
are integrated simultaneous and joint. As result of
integration we find full linkage magnetic fluxes S ,
S , R , R, co-ordinates of centre of rotor mass xm,
ym, angles of rotor turn and movable co-ordinate
system R, linear velocities of movement of centre of
rotor mass, xm, ym and angle rotor velocity .
Fig. 2. The time run of the current changes of the stator of

phase A
Fig. 3. The time run of the current changes of the stator of

phase A of the eccentricity of the rotor of the inductive engine
=5.0e-05 (A) and =2.0e-04 (B)
() xR2 yR2 R12 2R1 xR2 yR2 cos R2
(8)
Electromagnetic moment
ME
3
p0 iS iS
2
Because of high frequency vibrations of mechanical

system the differential equations are very stiff [3]. They are
integrated by implicit numerical Gear method including the
formulas of 6-th order. For this aim we obtained the
analytical formulas of Jacobi matrix, because it is very
complicated we here do not given.
(9)
The one-sided forces of magnet attraction in immovable

co-ordinate axes will be
Fx F cos R F sin R ; Fy F sin R F cos R (10)
where
2
2
;
F
c
sind ,
cos
0
2
)2
0
0 ( )
2
F c0
There
3w
iS iR

w
p0
(11)
cos p0
3w
iS iR

w
p0
sin p0 ,
(12)
Fig. 4. The time run of the changes angular speed
Obtained results confirm that mathematical model can

reach steady state thus can work properly Fig. 4. Further
experiments will realize with using neural network for
where c0=R1l0/2 is constant coefficient; l is length of

stator.
213
gradient method of Levenberga Marquardta has been

applied in connection with the back-propagation
algorithm, implemented in the programming language of
Matlab for Windows [4]. For the final evaluation of the
activity quality of the network there has been the
independent, evaluation sequence used, made up of 80
input vectors Fig.6.
engine with different defects, which can obtain setting

up appropriate conditions in mathematical model.
Proposed diagnostic system allow to elimination of a
human being from participation in measurements. Entire
diagnostic process is realized automatically by kit of
suitably equipped personal computer. In presented
system measurement signal is submitted to Fourier
Transform and results, after appropriate treatment, are
feed as input signal to neural network.
Efficiency [%]
Number of neurons in hidden layer

5
100
4. The system with the artificial neural network
S um -Squared E rror
10
50
1,00E-01
100
200
300
400
E poc h
500
600
700
1,00E-04
1,00E-05
1,00E-06
1,00E-07
5. Conclusions
The main virtues of using the artificial neural networks

in the diagnostic testing of the technical system result
from the attractive network properties, such as the ability
of the parallel information processing, effective
approximation of free nonlinearity, teaching and tutoring
the networks on the basis of the object signals.
Moreover, using the specialized mathematical model and
advanced computational technology led to the proper
network training and allowed for the precise evaluation
of the device state like the induction motor. Below there
are the detailed conclusions that are the result of the task
realization: the special approach has been proposed for
the analysis of the dynamic stresses in the induction
motors considering the mechanical and electromagnetic
vibratory effects, the analytical relations have been
received for the defining of the size of the air-gap in the
induction motor that allow for the mutual interaction of
the mechanical and electromagnetic subsystems in the
motor, the mathematical models have been compared in
terms of the usefulness in computer simulations, the
effectiveness of the artificial neural network application
has been shown for the analysis of the non-linear object
state using among others: their virtues for example the
abilities to generalize, low sensitivity on possible
damages (breaking certain number of the joints between
the neurons), the ability of the parallel processing and
the work in the actual mode.
800
8
Learning Rate
1,00E-03
Fig. 6. Effectiveness of the neural network depending on the

number of neurons in the hidden layer
-5
-3
1,00E-02
Sum - Squared Network Error
x 10
30
60
Training for 819 E poc hs
25
70
frequency value and the component amplitude of
10
20
80
the particular spectrum are the result of the solution of

the mathematical model equations for the steady work
state. The quantitative evaluation of the spectrum will be
done by using the neural networks. This kind of
approach does not need the division of the spectrum of
the stator current into particular component sets,
characteristic of the particular eccentricity kind. At the
neural input there is a vector given, including the
amplitudes and particular phases of the components
related to the value of the grooved component.
Different winding eccentricity values generate different
spectrums (amplitude values and phase angles) of the
winding current components, that is why these changes
can be used in research with the application of the
artificial neural networks.
5
15
90
Because of the lack of the occurrence of nonlinear

relations between the changes of the analyzed quantity
and their causes the non-linear artificial neural networks
have been used in the researches. Such tasks require
solving the problem of the right transformation of the
runs describing the object as well as carrying out
experiments in the range of choosing the right
architecture and data for teaching to obtain the ability of
the network to generalize. The direct pass of the current
flow on the receptors of the neural network is connected
with the necessity of the digitizing of this flow, whereby
the lowest the digitizing step the more precisely the flow
will be represented. There is a necessity of the initial
converting of the current flow stator, that is a modulated
flow. In order to do this there has been the current
distribution into the Fourier series applied. The
10
10
6
4
6. References
2
0
0
100
200
300
400
500
E poc h
600
700
800
1. Tchaban V. Mathematics modeling of electromechanical processes. Lviv, 1997.

2. Kocielny J. M. Diagnostyka zautomatyzowanych
procesw przemysowych. Wyd. EXIT/2001
3. Fortuna Z. i inni. Metody numeryczne. WNT/2005
4. Demuth H, Beale M. Neural Network Toolbox User's
Guide. The Math Works Inc.
900
Fig. 5. The course of teaching error in the function of the

stages number necessary to achieve the established error level
In this case there has been the non-linear neural network

applied for the purpose of the experiment made up of
one hidden layer. For teaching the networks Fig. 5, the
214

INVESTIGATION OF THE INDUCTION MOTOR START UP PROCESS USING

STAR-DELTA START
Marina KONUHOVA, Karlis KETNERS, Elena KETNERE, Svetlana KLUJEVSKA
Riga Technical University, Faculty of Power and Electrical Engineering, Latvia
-big inrush current cause large windings thermal

overload thus accelerating insulation ageing, its
damaging and finally turn-to-turn fault;
-large fluctuation of motor torque in the beginning of
start exceeding 4-5 times of nominal torque create
adverse conditions for mechanical part (kinematics
chain).
Due to provided obstacles in industry it is widely used
star-delta start (reduced-voltage start). This solution
allows limit inrush current [3].
Abstract: This investigation is concerning mathematical

modeling of the induction motor start up mode using
stator windings switching from star to delta connection.
Mathematical model of induction motor realized by
Park-Gorev equations. The analyses of presented start
up mode show, that at the switching time there is rotor
rundown of induction motor. Speed of rotation of the
induction motor due to rotor rundown decreases and in
case of prompt switching in delta connection remaining
rotor flux, which could not drop to zero immediately, it
may be current rush up to 20 times more than rated
current and that is not acceptable for induction motor. In
this investigations there are provided recommendations
for optimal switching time determination for star delta
start up mode of induction motor.
2. Induction motor star-delta start features

Induction motor star-delta start is possible for the
motors intended for operation with delta connection of
stator windings. This mode of start is acceptable only if
the starting torque of the driven mechanism is not big. If
motor has big load, the torque might be too short for
acceleration of the motor up to the speed for switching
of delta connection. For instance, the motors driving
pumps and ventilators in the start moment have small
torque and increase proportionate to rotation speed
squared. When the speed exceeds 80-85% of nominal
speed, the torques of motor from one side and torque of
the load become even and acceleration ends [1].
Inrush current in circuit when start is star switching is
triple smaller then when delta switching used. That is a
large advantage of considered start mode. Taking in
consideration that the on star switching of stator
Keywords: analysis, induction motor, mathematical

model, modeling, Park-Gorev equations, star-delta start.
1. Introduction
Induction squirrel-cage rotor motors (IM) are most
commonly used motor type on the market. They are
relatively cheap and have not expensive service [1].
Generally in asynchronous drive there appear transient
conditions: electromagnetic, resulted by induction motor
windings electromagnetic lag and mechanical, resulted
by the motor mechanical lag [2].
Transient current in the motor rotor and stator during
motor start process vary according to complicated
fluctuating rules and inducted by that electromagnetic
torque is of waving mode and is in complicated time
dependence [2].
Full voltages start up of the induction motor entail two
adverse factors i.e.: inrush current big repetition factor,
exceeding 6-10 and fluctuating damping type of motor
starting torque. Consequences of these factors effects
are:
-big inrush current leads to large drop of the voltage of
the substation main bus-bar (if the power of transformer
and motor are commensurable) thus destroying normal
work of other consumers and motor itself dragging out
its start;
windings the voltage is 3 smaller than on delta

switching, the start torque M decreases 3 times.
Disadvantage of this mode is absence of smooth of start,
having only two starting steps and increased start
process.
Starting with use of start-delta switch presented on the
Fig.1.
215
where TM - time constant of the machine in electrical

radians;
M em = X ad (i2 d i1q i2 q i1d ) - the electromagnetic torque;
u1d , u1d , u 2 d , u 2 q - are components of stator and rotor

voltage;
u 2 d = u 2 q = 0 for induction motor with a squirrel-cage
rotor;
1d , 1q , 2 d , 2 q - are components of flux linkages;
i1d , i1q , i2 d , i2 q - are components of stator and rotor

currents;
- is angular rotation speed of rotor;
0el - is angular rotation speed of coordinate system;
X 1 , X 2 , X ad , R1 , R2 - are parameters of the induction
motor;
M em , M l - are electromagnetic torque and load torque.
For IM rundown process (switching time) modeling we
solve the equation (1) by substitution of equation (3)
concerning derivative currents.
As far the current in stator of IM on rundown
immediately drops to zero, it is necessary to make such
an admission:
Fig. 1. Connection scheme of stator windings starting
At the switching moment, stator windings are switched

off for a while and than switch on again. Switch in
starting position switching stator windings in star and in
working position in delta. For switching usually apply
different automatic schemes (relays, contactors,
actuators).
3. Mathematical model of star-delta start for
induction motor
i1d = i1q = 0.
Developed program allows to model: induction motor

start up mode under undervoltage conditions (switching
stator windings in star), rundown mode of induction
motor (switching time) and induction motor switching
on full voltage (switching stator windings in delta).
IM reduced-voltage start and repeated switching on full
voltage modeling is implemented using IM
mathematical model described by Park-Gorev
equations, which are putted down in coordinates
rotating synchronously with stator field (coordinates d,
q) [4]:
d 1d
U 1d = R1i1d 0 el 1q +
d
d 1q
U 1q = R1i1q + 0 el 1q +
d
U 2d
U 2q
,
d 2 d
= R2 i2 d ( 0 el ) 2 q +
d
d 2 q
= R2 i2 q + ( 0 el ) 2 q +
d
TM
d
= [ M em M l ] ,
d
1d = X 1i1d + X ad i2 d
1q = X 1i1q + X ad i2 q
2d
2q
,
= X 2i2 d + X ad i1d
= X 2i2 q + X ad i1q
(4)
Taking into consideration above mentioned, from the

combined equations, which are modeling IM start up we
receive differential equations, which determine IM
rundown conditions:
di2 d
R
= 2
dt
X2
1
X2
X 1 ad
X2
X 1i2 d
X 1 0el i2 q ,
2
X
X 1 ad
X2
(1)
di2 q
1
=
X 1 0el i2 d
2
dt
X 1 X ad
X2
(2)
R2
X2
1
X1
2
X ad
X2
(6)
X 1i2 q ,
1
d
M l.
=
dt
TM
(3)
(5)
(7)
Voltage inducted by damped rotor field on the stator

terminal is following:
216
R2 X ad
i2 d X ad i2 q ,
X2
R X
= X ad i2 d 2 ad i2 q ,
X2
U 1d res =
U 1q res
(8)
(9)
where U 1d res , U 1q res - are residual voltage on stator

terminal.
Elaboration of this model allows researching
electromagnetic, mechanical and energetic parameters
for induction motor and completing electrical drive as
well.
Fig. 4. Induction motor electromagnetic torque curve starting
4. Modeling and obtained results analysis
At the time of starting between breaking and switching

from star to delta power supply interrupted, thus
rotation speed under the action of resistance of the
driven mechanism decreases. Rundown process of
induction motor in details is shown on the Fig.5.
For induction motor switching from star to delta mode

of start up analysis it was used induction motor
(4A355M4Y3) possessing following parameters [7]:
R1=0.012;
P2N=315kw; P2N=315kw; X=5.7;
XS=5.799;
R2=0.014; X2=0.14;
X1=0.099;
R2p=0.027;
Rk.p.=0.039.
Xk.p.=0.16;
Curves of rotation speed, current and torque of
induction motor depending on time during induction
motor star-delta start and direct start are presented on
Fig.2 Fig.4.
Fig. 5. Rundown mode of induction motor
Every IM possesses own rundown characteristics

because of dependence of load parameters and of driven
mechanism mechanical characteristics, finely of IM
parameters. The bigger moment of inertia and initial
rotation speed and the smaller motor load, the slower
rotation speed decrease during rundown time [5].
IM switching off from power supply, inducing residual
EMF in stator windings on rundown time. Since IM
magnetic flux couldnt instantly break off, the presence
of this flux exert negative influence to transients and
cause maximal transitional torques increase [6]. Relying
on IM rundown process modeling we obtained
hodographs of rotor current and of a voltage on stators
terminals (see Fig.6 & Fig.7), which shows EMF
attenuation process. Traced analysis shows that the
prompt switching from star to delta in case of adverse
switching phase (antiphase) may result with inrush
current amount to multiple nominal current rates.
Observed 25 IN inrush current on switching to delta in
conditions of adverse phase (see Fig.8) another time
closing on favorable phase inrush current was lack (see
Fig.9).
Fig. 2. Induction motor rotation speed curve starting
Fig. 3. Induction motor current curve starting
217
Fig. 9. Curves of current and electromagnetic torque

depending on time. Process successful
So, applying present model there is opportunity to

analyze IM safe start by star-delta start mode, since
residual voltage presence on IM stator at the windings
switching in delta might result with significant windings
thermal overloading, such way processing untimely its
insulation deterioration, injury and finely turn-to-turn
fault.
Fig. 6. Hodograph of a rotor current
5. Conclusions
Provided mathematical model of IM allows to make

complete switching process analysis if chosen motor
start to make in star-delta start mode. Analysis of
proceeded investigations shows that using this mode of
start up it is not useful to apply fast-acting automatic
machinery since each motor has different EMF
attenuation process time interval depending on its own
characteristics and load parameters. If apply fast-acting
automatic machinery for switching, there is risk to close
in antiphase. In such conditions the injury of the IM is
inevitable.
Fig. 7. Hodograph of a voltage on stators terminals
6. References
1. Softstarter handbook, ABB Automation Technology

Products AB, Control, Febryary 2003.
2. .. , M.:
, 2004.
3. Austin, H. (2006) Electric Motors and Drive (3td
ed.). Published by Elsevier Ltd, UK.
4. .. M
. .-3-
.-.: .,2001.
5. ..
- 2- ., . .-.:,
1985.
6. Ketnere E., Ketners K., Kuevska S., Kouhova M.
The research into the self-starting mode of the
induction
motor.
RTU,
Enertika
un
elektrotehnika 4. Srija, 20. Sjums, Rga, 2007.
7. 4: /
..,
..,
..,
...- .:, 1982.
Fig. 8. Curves of current and electromagnetic torque

depending on time. Process unsuccessful
218

RATIONAL GEOMETRY OF A MAGNETIC CIRCUIT OF AXIAL INDUCTOR

GENERATOR
Svetlana ORLOVA*, Vladislav PUGACHOV*, Nikolaj LEVIN*, Edmunds KAMOLINS**
*Institute of Physical Energetics, Latvia; **Riga Technical University, Latvia
Abstract: Axial inductor generator (AIG) is considered

together with tooth windings in the present work. This
generator is being manufactured by plant RER for
power supply of carriages. In order to define the optimal
geometry of tooth zone and magnetic circuit, an
estimation of electromagnetic field in the generator was
performed together with the application of a complex of
programs QuickField. The analysis of the results was
performed with the help of an objective function which
characterizes the relation of nominal power to the
generators mass. The recommendations concerning the
choice of rational geometry of the generator realization,
which will allow improving mass-dimensions indices,
are given on the basis of the analysis of magnetic fields.
2. Owing to the fact, that the magnetic flux is created by

one unmovable coaxial excitation coil situated in the
space between the rotor packets, inductor machines
requires a significantly lower expenditure of copper and
power of excitation, which makes the excitation system
and regulation of the machine simpler.
3. The inductor machine used for alternating current of
heightened frequency has a possibility to perform
significantly smaller tooth points on the rotor in
comparison with clearly polar synchronous machine of
the same diameter.
4. Absence of rotating windings saves the machine from
any contact, and this feature constitutes its most
important advantage [2].
Keywords: axial inductor generator, objective function,

software QuickField.
1.Indroduction
Inductor generators are devices belonging to the type of
brushless electrical machines, which have been widely
used in different industrial fields. Inductor generators
often have been used in the systems of electrical supply
of trains, planes and ships, in wind-driven power plants
systems, as well as in modern technologies of electric
arc welding, induction heating etc. This type of
equipment is characterised by the diversity of its
structural variations, scheme and modifications. In the
present paper the inductor generator with axial
excitation is examined. The main feature of this type of
machines is that by rotating rotor the flux in the teeth of
the rotor doesnt change the direction.
Axial inductor machines (AIM) have a simple
construction (see Fig. 1.) and a high degree of reliability
as well as they have some advantages in comparison
with common constructions of synchronous machines:
1. Absence of rotating windings and the rotor made like
a tooth-type cylinder. In this case, magnetic flux density
in every surface point of the armature bore of an
inductor machine changes only in its value, but the
direction remains the same.
Fig. 1. Two-core axial inductor machine: 1 body, 2 armature cores with windings, 3 rotor cores, 4 rotor shell,
5 excitation winding, 6 shaft, ( is a magnetic flux)
The main disadvantage of axial inductor machine is low

mass-dimensional magnitude, therefore designer shall
take a special attention to the rational selection of
geometry of teeth area of generator.
2. Modeling of magnetic field of an axial inductor
machine
AIM owing simple constructions have an
electromagnetic system that principally differs from
other synchronous machines and magnetic fields in
them and preserves clearly three-dimensional character.
However, with the help of some simplifications the
solution can be reflected in two two-dimensional tasks:
the field in an active zone (a core of the stator, a rotor
219
and a main air gap) is flat parallel, but outside the

boarders of the active zone (in an axial core) can be
received as symmetrical to the axis. Complexity in
modeling of a magnetic field in the active AIM zone is
that it is necessary to model an absent source of the
excitation field, as the coaxial excitation coil is situated
in the space between the cores of the stator and the rotor
[1].
The magnetic field, in the cross-section of the machine
is defined by a vector magnetic potential and is
described by a differential equation (1) in quotient
derivatives:
1 Az 1 Az
jz ,

x y x y x y
In case of optimisation the investigations on numerical

modelling of magnetic fields in AIM presented are not
complete, as it is necessary to take into account the
distribution of magnetic flux if the geometry of tooth
zone is different. Program Quick Field [4] is used in
the present work for estimation of magnetic field in the
machines cross-section, allowing to get a real picture of
magnetic field and on this basis to perform optimisation
of the machines magnetic system on the higher level
and to avoid the disadvantages mentioned before. It is
necessary to take into account the dependence of vector
magnetic potential (VMP) and magnetic flux on
geometric parameters of the tooths zone in the
optimization tasks for the AIM magnetic circuit. Fig. 3
shows an illustration of cross-section of magnetic field
of AIG, where Z1=24; Z2=10; =1,45.
(1)
where components of the tensor of magnetic

permeability x and y , Az is a vector magnetic
potential, jz is density of the current on the co-ordinate,
which coincides with the axis of machine rotation.
When AIM is no-load, the source of the field is an
excitation coil, which is situated outside the calculation
area between the two cores of the stator and the rotor.
In case when the calculation area is half of the machine
(Fig. 2), boundary conditions are set in the following
way:
1. Boundaries and D are axes of symmetry of the
magnetic field and a boundary condition of the 1st line
should be set for them: AAB=const; ACD=const, where
AAB, ACD are the values of a vector potential on the
boundaries of and D.
As the magnetic flux is defined with the help of the
expression (2):
( AAB ACD ) l A l ,
2
Fig. 3. Cross-section of magnetic field of AIG with Z1=24;

Z2=10; =1,45
The present work considers methods of definition of

vector magnetic potential and magnetic flux for
changeable parameters of the tooths zone for the
condition of permanent current in the machines
excitation windings which are outside the investigation
area as it was already stated above. The methods
proposed allow connecting the picture of magnetic field
in cross-section of the machine with magnetomotive
force of excitation winding. Special values of vector
magnetic potential are chosen (that is in our case AAB)
for each geometry of the machine. The relation between
the number of teeth of stator (Z1) and rotor (Z2), ratio of
the width of slot to the width of the tooth of the rotor ()
were chosen as changeable parameters for the solution
of optimization issues. In all cases the magnetomotive
force of excitation winding remains permanent even if
geometrical values are changed. The algorithm of
gaining of vector magnetic potential with the help of
software Quick Field is presented below:
1. Optionally predetermine the value of vector magnetic
potential (AAB).
2. Find magnetomotive force within the tooths zone
(Ft) and total magnetic flux () for the predetermined
value AAB.
3. Calculate magnetomotive force values in axial
magnetic core (Fc).
4. Sum up magnetomotive force values found in the
tooths zone and in the axial magnetic core; gain
magnetomotive force of the excitation winding (Fexc)
under the predetermined vector potential.
5. The value of electromotive force is known and by
using Newtons interpolation we can obtain the
(2)
where lis the length of a machine.
Fig. 2. Calculated geometrical model of magnetic system in

cross-section for AIM with teeth 24/10
The value of a vector magnetic potential can be chosen

freely on one of the boundary, in this particular case:
ACD=0 and AAB=A.
2. Boundary conditions of the 2nd line are set on the
outer (D) and inner () circumferences, as the
boundaries underconsideration have only normal
constituent of flux density (=0) [3].
220
necessary values of vector magnetic potential,

electromotive force of excitation winding and total flux.
The present algorithm of gaining of vector magnetic
potential should be estimated in each separate case, as in
case of changes in the geometry of the tooths zone, the
zones change themselves as well as magnetomotive
force values in the tooths zone and in the magnetic
core.
where
is a number of parallel branches in armature

max(k ) min( k )
windings;
a
is an average
2 Fk
permeance ensured by magnetic flux of tooth, which is
linked with the coil of armature winding dislocated on it
and which is induced by short-circuit peak current
( Fk 2 wk I k ).
Since the frequency (9) is
3. Objective function used for calculation of an axial

inductor generator
In order to obtain optimal geometry of teeth area of twocore inductor generator having axial excitation and
toothed windings it has to be used objective function,
representing correlation between generators nominal
power and its mass (3):
(3)
Ik
2
E0
xS 2
(5)
In that case objective function is represented as follows

(6):
C
mE02
,
2 xS
(6)
where E0 is an electromotive force of no-load stroke; Ik

is a short-circuit current; Xs is a synchronous resistance.
Electromotive force of no-load stroke can be calculated
by equitation (7):
E0 2wk fnk k w1
max min
,
2
a
a
120G
k w2 nN Z1Z 2
max min 2
max(k ) min( k )
(11)
4. Analysis of results obtained from objective

function of inductor generator
(7)
The present paper shows calculation of objective

function for obtaining the most rational geometry of
axial inductor generator with power of 32 kW, produced
by RER which has been used for power supply of
carriages. The rated values of the basic type model are
given below:
nominal power 32 kW;
number of stator teeth 24;
number of rotor teeth 10;
ratio of the width of slot to the width of the
tooth of the () 1,45;
where wk is a number of turns in toothed coil of

armature winding; f is a frequency; nk is a number of
coils in each parallel branch; kw1 winding coefficient;
max, min max and min magnetic flux of excitation,
corresponding to disposition of stator tooth in relation to
rotor tooth and in relation to rotor slot.
The following equitation is true for synchronous
resistance (8):
x S 2fnk wk2
It has been assumed that, by choosing geometry of

teethed area the higher harmonic components of
magnetic flux of excitation are reduced up to 0.
Analysis of said expression shows that by increasing the
number of stator and rotor teeth the nominal power of
inductor generator also should increase. However if Z1
and Z2 values increase, the difference of magnetic flux
(max-min) decreases in quadratic equitation. Therefore
these values cannot increase indefinitely, i.e., optimal
values for Z1 and Z2 do exist, which give maximum
result of objective function. It brings to conclusion that
in order to design optimal geometry of inductor
generator both by size and by the shape of rotor and
stator teeth it is necessary to carefully analyze magnetic
field in the air gap of electrical machine and to
determine conditions under which objective function,
i.e., specific power of electrical machine reaches its
maximum.
and the current (5)
IN
(10)
Fk
(4)
2Z1
,
ma
where Z1 is a number of stator teeth; Z2 is a number of

rotor teeth; nN is a nominal rotational speed.
The final expression of equitation is as follows (11)
where m is a number of phases; UN, IN are the nominal

voltage and the current of a generator; G is a generators
mass.
If a generator runs at its nominal conditions (4), by
reaching maximal power at active load the voltage is
E0
(9)
but the number of coils in the parallel branch of twocore generator (10) is
nk
mU N I N
,
C
G
UN
Z 2 nN
,
60
(8)
221
nominal rotational speed 1000 RPM;

excitation current 3;
number of turns of the excitation winding
1500;
number of cores 2;
length of stator/rotor core 115 mm;
winding three phases tooth winding;
outer stator diameter 458,3 mm;
stator bore diameter 329,3 mm;
height of stator tooth 36 mm;
width of stator tooth 25 mm;
steel grade of the stator and the rotor 2211.
The basic dimensions of the machine are as follows: the
number and form of armature teeth, data about the
winding, dimensions of axial magnetic core and
excitation winding remain without changes. The
parameters changed were: the number of teeth of the
rotor, and their relative width. Modeling of the
electromagnetic field was performed for the value of
the teeth of the inductor Z2=8,10;11;12;13;14,16,17 in
case of ratio of the width of slot to the width of the tooth
of the rotor =1,2;1,4;1,45;1,5;1,55;1,7;1,9. Optimum
had been reached with values Z1=24; Z2=1416,
however when Z2=16, due to the lack of distribution
effect of winding the shape of curve of electromotive
force becomes worse, and when machine is running it
increase noise and vibration. Therefore Z2 = 14 as an
optimal number of inductor teeth was selected. Fig. 3
shows a curve of objection function in relation to
excitation current, where Z2=10, =1,45 for basic
sample, and if Z2=14; =1,7.
The Fig.4 shows, that optimum has been reached when
Z2=14; =1,7. Meanwhile, if teeth are has been changed
from Z2=10, =1,45 to Z2=14, =1,7, it allows to
increase the specific power of generator by 12 %.
5. Conclusion
The present paper describes the results of modelling of

teeth area of the inductor generator with power of
32 kW. There is also illustrated the type of modelling of
magnetic field, taking into account relation between
vector magnetic potential and magnetic flux and
geometrical parameters of teeth area. Proposed
methodology allows defining relation between crosssection of magnetic field and electromotive force of the
excitation winding. In order to obtain optimal geometry
of teeth area of two-core inductor generator with axial
excitation and toothed coils it has to be used objective
function, representing correlation between generators
nominal power and its mass. The calculation of
objective function shows, that if the number of inductor
teeth reaches 14, the specific power of axial inductor
generator can be increased by 12 %.
6. References
1. .., .., ..

. Latvian
Journal of Physics and Technical Science, Riga,
1991, Nr.1, pp.87-94.
2. .. .
- , 1984, .247.
3. Orlova S., Pugachov V., Levin N. Magnetic field in
the tooth zone of an axial inductor machine (AIM).
5th annual Conference of Young Scientists on Energy
Issues 29 May, 2008, Kaunas, Lithuania, CD
proceedings. II-20 II-27
4. QuickField User Guide, version 5.5. WEB
http://www.quickfield.com
Fig. 4. Curves of dependence objective function of excitation

current for Z2=10 =1,45; Z2=14 =1,7
222

THE EDGE EFFECTS INFLUENCE ON THE BRAKING

FORCE OF LINEAR INDUCTION MOTOR
Bronius KARALINAS, Edvardas MATKEVIIUS
In LIM because of their open magnetic circuit there are

manifested the so called longitudinal and transversal
edge effects. Because of that the characteristics and
energy indexes of a LIM are worse than the indexes of
analogous rotary type motors.
LIM practically are confronted with the phenomenon of
braking the secondary element or the other moving
parts. From all the known modes of electric braking the
most extensively investigated is the mode of dynamic
stationary braking, considering that the velocity of the
secondary element at the moment of braking is constant
[3]. However, the influence of edge effects on to the
braking characteristics of a LIM is not sufficiently
analyzed. The main parameter describing the efficiency
of any braking mode is the braking force generated by a
LIM.
The objective of the article is to investigate the
influence of longitudinal and transversal edge effects on
to the braking force of a LIM as well as to present the
results of the computations processed.
Abstract: The main peculiarity when determining the

difference between linear induction motors (LIM) and
rotary motors is considered to be expressed in
longitudinal and transversal edge effects. These effects
in LIM appear due to the finite length and width of
magnetic system of motors. Because of the mutual
correlation of both effects, internal electric and
magnetic asymmetry in linear motors is formed. Rotary
type of motors does not have the characteristics of such
asymmetry.
The work presents the research carried concerning the
the braking force of a LIM. The theoretical study is
based on spectrum analysis method for magnetic fields
and on Maxwell equations and Fuorier integral changes.
The same method is exercised in the assessment of the
the characteristics of a LIM and on transitional
processes.
Keywords: linear induction motor, longitudinal effect,
transversal effect, magnetic field, braking force,
Reynoldss magnetic number.
2. Studies on edge effects of linear induction motors

When solving the problems of braking regarding a LIM,
the main attention is devoted to the analysis of an
electromagnetic field in respect to the influence of
longitudinal and transversal edge effects. The available
literature sources usually present the description of
various reasonable mathematical models.
The processes of a stationary dynamic braking in a
double sided linear induction motor are also analyzed
in work [4]. The influence of edge effects is assessed
there when solving the problem of one dimensional
electromagnetic field by means of corresponding
marginal conditions. There are some researchers who
assess only longitudinal edge effect when analyzing the
dynamic mode of braking, but the influence of
transversal effect is suggested to be assessed by
applying the correction coefficients of A. Voldek,
Russell Norsworthy or Preston Rice. These
coefficients do not depend on the operational mode of a
motor i.e. on the change of velocity of the secondary
element or on its slip. V. Skobelev [5] proposed the
1. Introduction
In new technologies there are widely used both rotary
motors of common structure and linear electric motors.
In cases when special technologies require linear or
sliding motion there are applied LIM with the secondary
element shorter than an inductor. Such an executing
device with the help of a sliding magnetic field not only
expels the secondary element conductive to electricity
from the active zone of the inductor but has the capacity
to perform the functions of an electromagnetic brake, a
damper, a spring, an absorber and executes the other
functions of an energy converter [1, 2]. The motors
mentioned above could be successfully installed in
compressors, in repulsion and shok hammering
mechanisms, in aggressive and hazardous technologies
such as for extraction of conductive metallic parts when
utilizing and decontaminating mercury, luminescence,
daylight lamps and other gas discharge tubes.
223
resultant scheme as well as by the coefficients of

correction.
empirical coefficient which depends only on the

geometric measurements of a LIM:
KF =
1+
0, 63
2c ( 2b 2c )
3. Computation of braking force
(1)
In this work the braking force is calculated following

the theoretical simulation model of a LIM which is
presented in sources [7, 8]. According to the model, for
the analysis of all electric modes of braking there is
applied spectrum method based on the equations of the
electromagnetic
field
and
Fourier
integral
transformations along the longitudinal axis of a motor.
At first, in accordance with the assumed wave of
volume density of braking current there is calculated
complex amplitude H of the strength of the total
magnetic field. There was obtained that the magnetic
field in the air gap of a motor is comprised of two
components:
where 2b is the width of the secondary element; 2c

is the width of the active zone (inductor); is the
pole pitch of an inductor.
H. Bolton [5] suggested a better way. He stated that the
transversal edge effect depends not only on the
parameters of a motor but on the slip s:
KF =
1 Re th(c1 1 is 0 ) sv 0
1 sv 0 + s 2 02 Re th(c1 1 is 0 )
(2)
where Re is a real part of a complex number; th a

hyperbolic tangent; i = 1 imaginary unit; 0
Reynoldss magnetic number; v and s is the velocity
of the secondary element and its slip; 1 =
H = H1 + H 2 ;
where H 1 is the complex amplitude of the primary or

inductor generated magnetic field; H 2 is the complex
amplitude of the magnetic field generated by the
induced currents in the secondary element.
By applying spectrum method together with the
expressions of theory of residuum, it is possible to
calculate by analytic method magnetic fields H 1
is a
special frequency of braking current and primary

magnetic field.
S. Jamamura [6] determined the boundaries, when
trying to reduce hazardous influence of transversal
effect on the characteristics of a motor, the boundaries
up to which it is advisable to widen the secondary
element. Its coefficient is rather simple:
K F = th
(b c)
.
and H 2 concerning any electric braking mode. Then an

electromagnetic braking force generated within the
active zone boundaries of a motor is easy to calculate by
means of the following type of volume integral:
(3)
Felm =
The difference (b c) in the expression (3) indicates

the size of ones sided widening of the secondary
element in respect to the active zone c of a motor. When
changing the argument, the value of hyperbolic tangent
(3) gradually approaches unit one. Starting from the
values 1.3 and more of an argument, K F slightly differs
from unit one, then comes the following:
(b c)
1.3 .
0.41 .
0
2
Re H 2 &j ( x, t )dV ;
(7)
(V )
where 0 = 4 107 , H m is magnetic permeability

of the secondary element; &j ( x, t ) is the joint complex
of volume density of the braking current; V is the

volume of the secondary element where the braking
force is generated and according to which expression (7)
is integrated.
After solving integral (7) for any braking modes there
are derived rather complicated mathematical
expressions indicating that braking force consists of
three components:
(4)
Then the relative value of widening of the secondary

element will be the following:
bc
(6)
Felm = F1 + F2 + F3 ;
(5)
(8)
where F1 is the braking force component, estimating

only transversal edge effect as well as the change of
velocity of the secondary element and validated for a
motor with an infinite long active zone; F2 is a free
component of the braking force, existing only during the
transitional process; F3 is force component evaluating
longitudinal and transversal edge effects and their
mutual correlation.
In case of dynamic braking, when the direct current
The achieved result (5) indicates that single sided

widening of the secondary element behind the
boundaries of an inductor has to be not higher than 41%
from the pole pitch . There is no need to widen the
secondary element, because this measure does not give
a positive effect and doesnt have any possibility for
practical application.
The influence of the edge effects on such technologies
is evaluated by the brought in resistances of the
224
flows through the inductor windings, the following

expression is derived to calculate component F3 :
F3 = F0 Re
1
p 3
(2 K + 1)
0 v0 n sin 2
K =0
( )
( 0v )
+ nK2
4. The results of computations
In order to analyze the edge effects, it is very

comfortable to calculate all the components of the
braking force by relative units. For that purpose braking
force F0 of an ideal motor is taken by the basic quantity.
Following the derived expressions (9) (11) there were
calculated dependencies of relative forces F3 F0 and
coefficient hs on the parameters of a motor and

geometric measurements by means of software Mathcad
2001 Professional.
Dependencies of braking forces F3 F0 on the width
c of an active zone of a motor are presented in Fig. 1.
Fig. 2 presents the curves indicating the variation of
force F3 F0 which depends on number p of the pairs of
poles of a motor. Analogous dependencies on the
relative width of secondary element are presented in
2
1 exp p 0 v + ( 0 v ) + nK
+
2
K 3 ( 0 v0 n K 3 )(1 i K 3 )
1 exp p 0 v + ( 0 v ) + nK2
+
2
K 4 ( 0 v0 n K 4 )(1 i k 4 )
(9)
Fig. 3. Dependencies of coefficient hs on width c of

a motor are presented in Fig. 4.
where F0 is an ideal braking force of a LIM without

the edge effects; p is the number of pairs of an
inductor poles; v0n is the initial velocity of braking of
1.5
F3 / F0
the secondary element; c is relative width of an
1.1
0.9
active zone; = b
is relative width of the secondary

c
element; k 3 and K 4 are the roots of the
characteristic equation:
= 0v
K3
v+
K 4 = 0
1.3
( 0v )
( 0v )
0.7
0 = 50
0.5
0.3
0.1 0.2
0
0.4 0.6 0.8
-0.1
+ nk2
0 = 10
-0.3
(10)
1.2 1.4 1.6 1.8
2.2
0 = 1
-0.5
+ nK2
c /
Fig. 1. Dependencies of relative braking force F3 F0 on the

width c of an active zone, when Reynoldss numbers are
nK =
2K + 1
;
c
different and p=1,0; v0 n = 1,0; = 1,4
K = 0; 1; 2; 3; and etc.
1.5
1.3
In the expressions of force components F1 and F2 the

influence of transversal edge effect is evaluated by
coefficient hs . For a LIM model, where the secondary
element is wider than the active zone of an inductor
( b > c ), there was derived the following expression of
the coefficient of transversal edge:
F3 / F0
0.9
0 = 50
0.7
0.5
0.3
sh ( c 1 i 0 v )
hs = Re 1
c
1 i 0 v
ch c ( 1) 1 i 0 v
ch ( c 1 i 0 v )
0 = 10
1.1
0.1 1
0
-0.1
-0.3
-0.5
(11)
0 = 1
p
Fig. 2. Dependencies of relative braking force F3 F0 on the

number of pairs of poles of a motor, when Reynoldss
numbers are different and v0 n = 1,0; c = 0,6; = 1,4
where v is the velocity of the secondary element at the

moment of braking.
225
5. Conclusions
1.75
1. To calculate braking force of a LIM the authors

applied spectrum analysis of magnetic fields which
is based on Maxwell equations and Fourier integral
transformations along the longitudinal axis of a
motor.
2. There is determined that electromagnetic braking
force is generated by three components in which
there is assessed the influence of transversal and
longitudinal edge effects, their mutual correlation as
well as the change of velocity of the secondary
element at the moment of braking.
3. The results of calculations made indicate that the
influence of edge effects of a LAM with different
Reynoldss numbers is irregular. The transversal
edge effect is the most intensively expressed in
relatively narrow motors. The widening of the
secondary element behind the boundaries of the
active zone of a motor enables to reduce the
influence of transversal effect on the braking force.
By increasing the number of pairs of inductor poles
the influence of longitudinal effect on the braking
force has a tendency to be weakened. If p > 4.0,
then it is possible not to consider the longitudinal
edge effect in braking force calculations.
4. The derived results could be implemented in
designing a new LAM and in compiling the most
optimum systems of their braking .
F3 / F0 1.5
1.25
1
0.75
0.5
0.25
0
1 1.2
1.4
1.6
1.8
2.2
2.4
2.6
2.8
-0.25
-0.5
Fig. 3. Dependencies of relative braking force on relative

width of the secondary element, when Reynoldss numbers
0 are different and p=1,0; v0 n 1,0; c = 0,6
10
hs
4
2
c/
0
-2
-4
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
6. References
-6
1. Laitwaite E. R. A History of Linear Electric

Motors. London, 1986. 178 p.
2. Boldea I. Linear Electric Actuators and Generators.
IEEE Transactions of Energy Conversion, vol. 14,
No. 3, 1999. p. 712-716.
3. . .

. . . . .
, 1978. 252 .
4. Nonaka S. Simplified Fourier Transform Method of
LIM Analysis Based on Space Harmonic Method.
Proceedings of Second International Symposium on
LDIA98, Tokio, Japan, 1998. p. 187-190.
5. Karalinas B. Trinties jg komprnsavimas lktuvo
valdymo sistemoje. Elektronika ir elektrotechnika,
Nr. 1(19), Kaunas, 1999. p. 53-56.
6. .
. , 1983. 180 .
7. Karalinas B. Non Stationary Processes in the
Electric
Machine
Converters
of
Energy.
Proceedings of Electrotechnical Institute, vol.
200/99, Warszawa, Poland, 1999. p. 99-112.
8. Darulien O., Karalinas B. Mathematical Model of
Non Stationary Processes of Electromechanical
Power Converters. Proceedings of 11th
International Conference on Power Electronics and
Motion Control, EPE PEMC2004, vol. 3, Riga,
Latvia, 2004. p. 415-419.
-8
-10
Fig. 4. Dependencies of coefficient hs of the transversal effect

on the width of a relative active zone when Reynoldss
numbers 0 are different and v= 0,8; 1,4
The results of computation indicate that components F1

and F2 of the braking forces depend on the transversal
edge effect, which is characterized by coefficient hs
given in expression (11). Due to that the longitudinal
edge effect for the mentioned above components has no
influence.
From the submitted curves it is obvious that the most
significant influence onto the relative force F3 F0 is
caused by transversal and longitudinal edge effects and
their mutual correlation. However, the influence onto
the motors with different Reynoldss numbers is not
equal. When Reynoldss numbers are low ( 0 3.0),
the edge effects weaken the braking force and therefore,
the braking of these types of motors is not efficient.
When Reynoldss numbers are increasing the
transversal edge effect exhibits itself in the most
significant way in relatively narrow sized motors where
c 0.8. Widening of the secondary element behind
the boundaries of an active zone ( > 1.0) tends to

weaken the transversal edge effect as well.
226

MODELING OF TWO-MASS SYSTEM WITH ELASTICITY

Sigitas JURAITIS, Roma RINKEVIIEN, Jonas KRIAUINAS
Abstract: Dynamics of two-mass electromechanical

system with clearance and elements capable to deform
is considered. Model for investigation of influence of
elasticity and clearance is presented. Induction motor is
used to drive two-mass system. Speed responses of the
system are presented and analyzed. Influence of
elasticity and clearance to speed response is considered.
d 1
M M st1 M12 = J1 dt ;
M + M = J d 2 .
12
2
st 2
dt
(1)
Torque of elasticity is calculated as:

M 12 = c12 (1 2 ),
Keywords: two-mass electromechanical system,

influence of elasticity and clearance, simulation.
(2)
where c12 is stiffness of elastic mechanical part. All

other notations are as in [4].
Rotational stiffness c12 is given by:
1. Introduction
Electromechanical system as object of investigation
comprises
electrical
and
mechanical
parts.
Electromechanical power converter and its control
system depends to electrical part as well as all moving
masses coupled between themselves form mechanical
part. Electromechanical system includes various
mechanical chains, with infinite or finite elasticity and
clearance. Systems with infinite stiffness and without
clearance are one-mass systems and are quite well
analyzed [1, 2, 3]. Systems with capable to deform
chains and clearance are more complex. They are
described by high order nonlinear differential equations,
and without essential simplifying of problem they
cannot be solved in analytical way. In these cases
computer models of solved program must be developed,
using specialized software, and system responses
simulated.
Some problems of two-mass system were considered in
[4, 5, 6, 7, 8, 9, 10].
The paper presents general model of two-mass system,
developed in Simulink. Results simulations at different
stiffness of two-mass system and different clearance are
presented and considered. Dependences of maximal
speed values against elasticity factor and clearance are
formed and presented.
c12 =
(3)
where M is the applied torque, is the rotation angle.

Rotational stiffness is typically measured in newtonmetres per radian. In general, elastic modulus is not the
same as stiffness. Elastic modulus is a property of the
constituent material, stiffness is a property of a solid
body. That is, the modulus is an intensive property of
the material, stiffness, on the other hand, is an extensive
property of the solid body dependent on the material
and the shape and boundary conditions.
Differentiating equation (2) yields:
dM 12
d d
= c12 1 2 = c12 (1 2 ) .
dt
dt
dt
(4)
Equation (4) in frequency domain looks like this:

M M st1 M12 = J1s1 ;
M st 2 + M12 = J 2 s2 ;
sM12 = c12 (1 2 ) .
2. Model of two-mass system with clearance
(5)
Portion of two-mass system with kinematic clearance is

shown in Fig. 1, a. Due to the clearance, dependence
M12 = f ( ) is nonlinear (Fig. 1, b).
Two-mass system is described by set of equation:
227
3. Development of model and simulation of the

system
Computer model is elaborated for a motor, which

parameters are presented in Table 1.
Table 1. Parameters of a motor
Parameter
Motor power, kW
Number of pole pairs
Phase voltage, V
Power factor
Rated torque, Nm
Rated current, A
Inertia, kgm2
M 12
b)
Fig. 1. View of two-mass system with clearance (a) and

dependence M12 = f ( ) (b)
Value
1,1
2
230
0,81
7
3,56
0,00262
The second mass is chosen freely. Let it will by

cylindrical shape body, fastened along mass center (see
Fig. 1). Its inertia is calculated in this way:
Two-mass system movement equations in frequency

domain can be expressed as [4]:
J=
M M st1 M 12 = J1s1 ;
M st 2 + M12 = J 2 s2 ;
M 12 = c12 1 2
, when
2
M 12 = 0, when 1 2
.
2
(7)
where m is mass of cylinder and r is radius of cylinder.

It is assumed in simulation J=0,0025kgm2.
Model of induction motor operating with two-mass
system is presented in [7]. The general simplified form
of model is presented in Fig. 2.
(6)
;
1 2 >
2
m r2
,
2
Equation 6 shows both system masses moving

independently one from other while system operates in
the clearance zone.
At starting of the drive with load three stages of
response can be separated. In the first stage, while
M 12 = 0 , the speed of second mass 2 also is equal to
zero. Due to this effect the speed of the first mass rises
quickly because torque, developed by motor M is
greater than load torque M st1 . Afterwards the second
Fig. 2. Model of two-mass electromechanical system
stage of starting begins. Clearance already is absent, but

speed 2 remains equal to zero, while increasing
torque M12 becomes greater than Mst1. Afterwards speed
begins to rise. This speed influences dynamic load of
the second load.
Response of motor speed at different values of stiffness

is presented in Fig.3.
250
200
300
150
100
60
400
200
30
Rotation speed, rad/s
motor 600
150
10
100
50
0.01
0.02
0.03
0.04
0.05
Time, s
0.06
0.07
Fig 3. Speed transients at different elasticity
228
0.08
0.09
0.1
Simulation results are compared with results of onemass system operating at no-load for exploring
dependences of system parameters influence. Clearance
is determined by time duration after which the first mass
(motor) starts to move the second mass.
Stiffness value was chosen freely for getting
dependences of speed maximum value against stiffness.
Comparison of simulation results indicate motor speed
lag in two-mass system, depending on stiffness, with
speed of motor, operating in one-mass system. That
depends on the load torque, developed by second mass.
After the shaft dislodges the second mass, the motor
speed reaches the greater value than that at no load due
to inertia of the second mass accelerating the motor.
Oscillations of speed in steady-state also are greater
because elasticity of the shaft. The shaft under rotation
of motor is being twisted and deformed. In this way it
accumulates energy and delays the influence of the
X: 0.02182
Y: 194.4
200
X: 0.02638
X: 0.02873
Y: 216.3
Y: 211.8
X: 0.03237
Y: 197.7
X: 0.02458
Y: 214.2
220
X: 0.02262
Y: 204.3
X: 0.01542
Y: 171.6
180
X: 0.01435
Y: 150.4
160
140
X: 0.01382
Y: 122.5
120
second mass, causing oscillations of the speed.

Dependence of the speed against elasticity is presented
in Fig. 4. The greatest values of the first and second
maximum are marked in this figure. The solid thick line
corresponds to motor speed at no-load. With increasing
of stiffness the greatest speed approaches the speed
transient of one-mass system. At first the second
amplitude of oscillations reduces, then increases and
reaches the maximum value, afterwards it begins to
reduce again. According to simulation results, the
system reaches the greatest value at proper elasticity,
which is undesirable due to great speed oscillations. On
the other way, if the elasticity is small, maximum speed
is smaller. This operation mode can be used to avoid
speed impact during starting.
Speed response at constant elasticity and different
clearance values is presented in the Fig. 5.
X: 0.03736
Y: 177.3
X: 0.02087
Y: 183.1
X: 0.01488
Y: 163.5
X: 0.03948
Y: 163
X: 0.04065
Y: 158.6
X: 0.01435
Y: 135.9
X: 0.01329
Y: 111
100
0.005
0.01
0.015
0.02
0.025
0.03
0.035
Time, s
0.04
0.045
0.05
0.055
Fig 4. Speed transients maximum values at different elasticity

300
dz=-5-5
dz=-2-2
dz=-0.5-0.5
dz=0
250
200
dz=-10-10
dz=-20-20
dz=-15-15
dz=-25-25
dz=-30-30
dz=-35-35
dz=-40-40
motor
150
100
50
dz=-45-45
dz=-50-50
0
0.05
0.1
0.15
0.2
Time, s
Fig 5. Speed transients maximum values at different clearance
229
0.25
Curves, presented in Fig. 4 indicate the shorter time

required to reach the speed maximum with increasing
elasticity and greater oscillations frequency as well as
longer set-up time. Fig. 6 presents the peak speed values
at varying elasticity factor, obtained from simulation
results in Fig. 4.
5.
The maximum speed value increases proportionally

to clearance. At the great clearance, when the
movement to second mass is transmitted at steadystate speed of the motor, the maximum speed
during transients is not significant.
5. References
Peak speed value, rad/s
250
1.
Schnfeld R., Hofmann W. Elektrische Antriebe

und Bewegungsteurungen. Vde - Verlag. 2005. s.
454.
2. Vogel J. Elektrische Antriebstechnik. 5. bearb.
Aufl. Httig, Buchverlag, 1991.
3. Masteika R. K. Mechanics of electric drives.
Kaunas: Technologija, 1996. 40 p. In Lithuanian.
4. Smilgeviius A., Rinkeviien R. Simulation of
transients
in
the
Mechanical
part
of
Electromechanical
system
//
Mathematical
Modelling and Analysis Proceedings of the 10th
International Conference MMA2005&CMAM2,
Trakai. 2005. p. 155-162.
5. Rinkeviien R., Batkauskas V. Modelling and
Investigation of Vector Controlled Induction Drive.
Elektronika ir Elektrotechnika. Electronics and
Electrical Engineering. 2008 No.1 (81). p. 53 - 56.
6. Smilgeviius A., Rinkeviien R., Savickien Z.
Operation of electric motor with elastic load.
Elektronika ir elektrotechnika. Electronics and
Electrical Engineering. 2006, Nr. 6(70), p. 15 18.
7. Juraitis S., Rinkeviien R., Smilgeviius A.
Dynamic properties of two-mass electromechanical
system. 6th International Symposium Topical
Problems in the Field of Electrical and Power
Engineering. Kuressaare 2009, p. 18-21.
8. Feiler M., Westermaier C., Schroder D. Adaptive
speed control of a two-mass system.//CCA 2003.
Proceedings of 2003 IEEE Conference on Control
Applications, 2003.Vol. 1, 23-25 June 2003, p.
1112 1117.
9. Kara T., Eker I. Experimental nonlinear
identification of a two mass system. CCA 2003.
Proceedings of IEEE Conference on Control
Applications, 2003.Vol. 1, 23-25 June 2003, p. 66
71.
10. Bolognani S., Venturato A., Zigliotto M.
Theoretical and experimental comparison of speed
controllers for elastic two-mass-systems.//IEEE
31st Annual Power Electronics Specialists
Conference, 2000. PESC 00. Vol. 3, 18-23 June
2000. p. 1087 1092.
200
150
100
synchronous
1 peak
2 peak
50
0
0
10
30
60 100 150 200 300 400 600

Elasticity factor, Nm/rad
Fig 6. Dependence of the speed maximum against elasticity
Dependence of peak speed values, obtained from Fig. 5,

is presented in Fig. 7.
Peak speed value, rad/s
300
250
200
150
100
synchronous
50
peak
0
0 0,5 2
10 15 20 25 30 35 40 45 50
Clearence
Fig 7. Dependence of the speed maximum against clearance
Peak speed value reaches its maximum at clearance

factor, approximately equal to 25.
Dash line in Fig. 7 denotes synchronous speed of motor.
The greatest peak speed exceeds synchronous speed
approximately 45 percent. Motor at this time operates in
regenerative mode.
4. Conclusions
1.
2.
3.
4.
Finite stiffness causes angular speed oscillations of

two-mass system, which amplitude, frequency and
set-up time depends on value of stiffness.
Value of finite stiffness can reduce undesirable
speed overshoot; otherwise this phenomenon can be
used to reach the greatest value of speed.
In the system with clearance beginning of starting
transients corresponds to starting transients of the
drive at no load. During this time motor speed
increases by exponential law.
Clearance
causes
dynamic
torques
in
electromechanical system, whose values increase
with the increase of clearance.
230

THE MEANS OF ELECTROMAGNETIC FLOW METER MAGNETIC CIRCUIT

TESTING
Valdas Jonas PAKNAS, Juozapas Arvydas VIRBALIS
Abstract: The measurement accuracy of electromagnetic

fluid flow meter depends on parameters of magnetic
circuit and distribution of magnetic field inside of
transducer active zone. The stability of magnetic circuit
parameters and magnetic field distribution can be
monitored by control coils or using for monitoring one of
magnetic field exiting coils. The magnetic circuit
parameters can vary because of three main reasons. The
reasons are the following: the magnetic admixtures
inside fluid, the magnetic sediments on channel wall and
the variation of circuit permeability. When two control
coils are used, the unwarrantable variation of magnetic
flux and the reason of this variation can be set.
accounting of fluid and transporting by fluid heat

practically only EMFM and ultrasonic meters are used.
When fluid flows in non-moving magnetic field the
electric field arises. The electric field strength E by
Faradays law of electromagnetic induction in any point
of flow can be expressed as vector product flow velocity
vector v and magnetic flux density vector B in this
point: E F v B . If the electrodes are mounted in the
walls of channel, the difference of potentials Ui arises in
the electrodes. It can be expressed when fluid is directed
by channel axis, i.e, the flow velocity in any flow point
has only one component vezvz, this way:
Ui vzWz d .
Keywords: Electromagnetic fluid flow meter, magnetic

field distribution, control coil.
(1)
There d is element of EMFM sensor active zone

volume , Wz is weight function. Active zone is flow
volume in which electrode signal is forming, i.e,
magnetic field is acting. The weight function is
proportional to component of magnetic flux density By,
perpendicular to flow and to the line which connects the
electrodes: WzKwBy. Usually EMFM are used for
measurement of volume fluid flow Q. By (1) electrode
signal of EMFM must depend on distribution of velocity
in cross-section of channel. But superabundance of
experimental investigations shows [1], that when the
distribution of magnetic field in active zone is close to
uniform, we can suppose that electrode signal is
proportional to fluid flow in all diapason of flow values.
The transfer coefficient of EMFM is not depended on
flow and proportional to total magnetic flux , which
passes via EMFM transducer:
1. Introduction
The quantitative characteristics of fluid flow: local and
mean velocity, the mass and volume flow and quantity
are important for many technological processes and
commercial accounts. Its measurements compose about
40% of all technological measurements which are
performed in industry. The conventional problems are
the accounting of drinking and industrial water and heat
measurement. The actual problems with arising of
ecology requirements are the accounting of sewages, the
batching of different special fluids and other. The
significant part of meters using for this purposes are the
electromagnetic fluid flow meters EMFM. The EMFM
are convenient for use in different informational and
control systems and in the devices, intended for different
combined measurements, par example, in heat meters.
The modern EMFM are intellectual devices, in which the
microcomputers are used for measurement cycle
organization, for subsidiary parameters calculation, for
parasitic signals suppressing, for measurement error
diminishing.
The mean advantages of EMFM are absence of
resistance to fluid flow, practically non-inert
measurement, linear transfer characteristic and small
measurement uncertainty. There is what for commercial
U KQ, K K1 ,
(2)
where K1const. The transfer coefficient K is setting

when EMFM is calibrated and it must be constant along
all time between two metrological assessments.
Coefficient K1 depends on geometrical dimensions of
device and it is practically constant. Therefore the
stability of EMFM transfer coefficient depends on
231
stability of magnetic flow and on its distribution in

EMFM active zone. We can guarantee the EMFM
measurement results metrological reliability when
stability of magnetic flow will be constantly
monitored.
Ie
Ue
VCC
CC1
2. Magnetic circuit of EMFM and means of its

monitoring
Ur1
Ur2
CC2
Typical magnetic circuit of EMFM is presented in Fig.

1. The particularity of this circuit is the large air gap. The
rectangular current pulses are used for magnetic field
excitation in the most of all electromagnetic flow meters.
The frequency of these pulses is not exceeded some Hz.
The invariability of EMFM transfer coefficient will be
warranted when the magnetic flux and its distribution
inside sensor active zone will be invariable.
R0
U0
Fig. 2. The EMFM sensor with control coils
When the switch is turned into position 2, the excitation

current is directed via only one excitation coil EC1. The
voltage Ur which arises in the excitation coil EC2 can be
used for magnetic flux monitoring.
VCC
+
Ue
EC1
R0
2
EC2
Ie
Ur
U0
Fig. 3. The use of excitation coil for magnetic flux monitoring

Fig. 1. Magnetic circuit of EMFM sensor
The some part r of common flux passes via control

coil independently on chosen monitoring technique:
using the special control coils or using one of the
excitation coils. This part is proportional to common
magnetic flux. In turn the common magnetic flux is
proportional to excitation current ie. Therefore we can
write:
The merest way of magnetic flux monitoring is the use

of one or two complementary control coils CC1 and
CC2. They can be mounted between one of magnetic
field excitation coil and channel as it is showed in Fig. 2.
The voltages Ur1 and Ur2, proportional to magnetic flux
of sensor active zone, arise in the coils CC1 and CC2,
correspondingly.
But this way is suited to the case, when sensor is
designing. When the EMFM sensor is produced without
the control coil, the magnetic flux inside the EMFM
sensor can be monitored by one of excitation coils. In
this case we can measure the control coil voltage instead
of the flow measurement.
The possible excitation current generation circuit is
showed in Fig. 3 by bold lines. The switch S is in the 1
position. In this case the magnetic field excitation
current is directed via both excitation coils EC1 and
EC2, when the pulse with amplitude Ue acts on the input
of voltage to current converter VCC. The flow is
measuring in this case.
rKie,
(3)
where Kconst is coefficient of proportionality. The

processor of EMFM forms the series of pulses with
amplitude Ue. The timescale of pulse Tm is equal to the
timescale of pause Tp. The pulse and the pause compose
one measurement cycle. When the pulse with amplitude
Ue begins to act in input of voltage to current converter
(see Fig. 3 and Fig. 4) the excitation current arises
exponentially:
ie I 0 (1 e
);
(4)
where eLe/Rr is time constant of exciting circuit, Le

and Rr are the total inductivity and resistance of exciting
coil (or coils), correspondingly. After time interval
t(45)e the magnetic field excitation current
232
where i is integrator time constant, KIA amplification

coefficient of IA.
The potential equal to zero connects to integrator input
and the integrator output connects to ADC input in this
moment. We obtain in the ADC output the number N0:
practically becomes equal to I0. The excitation current

time diagram of one cycle is showed in Fig 4.
ie
I0
t
T0
Tm
N0KADCUI0KADCKIAI0R0T0/i,
Tp
where KADC is transfer coefficient of converter ADC.

The number N0 is stored into memory M, and integrator
output is setting to zero.
The second measurement performs when other current
cycle begins. The voltage Ur of control coil connects to
IA input. The integration of amplified voltage begins
with current pulse beginning in this case.
The control coil voltage Ur is equal to derivative of total
magnetic flux which passes via this coil. We obtain
evaluating (3) and (4)
Fig. 4. The form of exciting current
The electrodes signal is measured in the time interval

when the excitation current ieI0. By (2) and (3) we can
write:
UK1KI0Q.
(5)
We can suppose, that transform coefficient is constant

when coefficient K value is not varied with respect to
value KK0 obtained in the moment of EMFM
calibration. Therefore the monitoring of coefficient K
must be performed.
Ur
The monitoring technique is the same in the both cases:

when there are used the special control coils or one of
excitation coils for magnetic flux measurement.
For the accurate coefficient K evaluation two
measurements are performed. In the first measurement
cycle the basic voltage U0 is measured and in other
cycle the voltage of control coil is measured.
The monitoring circuit is presented Fig. 5.
U Ir
ADC
Tm
K IS K I 0
K IS K I 0
dt
K IS K I 0
Tm
This voltage analogically to voltage UI0 is converted to

number Nr in ADC:
NrKADCUI0KADCKIAKI0/i.
Ue
(10)
This number is stored into processor memory, too and

the ratio Nr/N0 is computed. The coefficient K can be
expressed of (7) and (10) via this ratio:
K R0T0
Fig 5. The monitoring circuit
There are in this circuit: IA input amplifier, I integrator, ADC converter analog-to-digital, P processor, C commutator, M memory unity.
At first the voltage U0R0I0 is connected via commutator
C to amplifier IA input. This voltage acts in resistor R0
(see Fig. 2 or Fig. 3), when the transient process is finite
and excitation current ie is equal to I0 (interval T0 in the
time diagram, presented in Fig. 4).
Therefore the voltage U0 is connected to IA input during
the interval T0. This voltage amplifies into amplifier IA
and integrates during all interval T0. After the end of
interval T0 the integrator output voltage is equal:
UI0KIAI0R0T0/i,
(8)
(9)
C
IA
K I 0 t
di
d r
e.
K e
e
dt
dt
e
This voltage after amplifying is integrating along all

excitation current pulse Tm. The integrator output
voltage will be equal after integration:
3. The monitoring technique
Ur
U0
(7)
Nr
.
N0
(11)
The numbers Nr and N0 must be measured immediately

after calibration and K0 must be computed. The
monitoring measurements must be repeated periodically
for example one time during day (or during week
dependently on EMFM working conditions) and
obtained value of K must be compared with K0. It is
useful to store the data-bank in memory.
4. The advantage of two control coil use
The coefficient K which relates excitation current with
passing via control coil magnetic flux depends on two
groups of reasons. The first group consists of the
geometrical reasons: the geometrical dimensions of
magnetic circuit, the dimensions of control coil and its
(6)
233
position. The external sensor part is immobile.

Therefore, the probability of the geometrical variation is
very small in normal conditions. The second group of
reasons is related with magnetic resistance of magnetic
circuit variation or magnetic field redistribution. There
are three factors, which can vary magnetic resistance or
magnetic field distribution: variation of relative
permeability of magnetic circuit, magnetic sediments on
channel internal walls and magnetic admixtures in flow
[2, 3]. We can see of Fig. 6, that any of these three
factors can be characterizes by specific magnetic field
distribution in active zone of EMF sensor.
coil, therefore, the magnetic sediments are formed on

the internal walls of channel (see Fig. 6, c).
The accumulation of magnetic sediments and decrease
of permeability are the inertial and monotonic
processes. But the magnetic admixtures can appear
sharply and be episodic or, contrary, the long time the
concentration of magnetic admixtures can be constant.
Examining the dependences KCC1(t) and KCC2(t)
stored in the data-bank we can understand, as the
processes develop and we can surely solve about the
variations about metrological characteristics of EMFM.
5. Conclusions
1.
2.
3.
a
Fig.6. The distribution of magnetic field in active zone of

EMFM : a) when channel is pure, b) flow with magnetic
admixtures, c) channel walls with magnetic sediments
The measurement accuracy of electromagnetic fluid

flow meter depends on parameters of magnetic
circuit and distribution of magnetic field inside of
transducer active zone.
The stability of magnetic circuit parameters and
magnetic field distribution can be monitored by
control coils or using for monitoring one of
magnetic field exiting coils.
When two control coils are used, the unwarrantable
variation of magnetic flux and the reason of this
variation can be set.
6. References
When the two control coils are mounted we can not only
observe the variation of transfer coefficient, but we can
understand the reason of this variation. It is usefully to
mount the coils CC1 and CC2, as it is showed in Fig. 2:
one under the central part of magnetic core, other
under the edge of excitation coil. When the signals of
both coils decrease, therefore, we can suppose, that
permeability of magnetic circuit decreases. When the
signal at least of one coil increases and the signal of
central coil increases more than the signal of peripheral
coil, therefore, the magnetic admixtures appear in the
flow (see Fig. 6, b). When, contrary, the signal of
peripheral coil increases more than the signal of central
1.
2.
3.
234
Virbalis J.A. Elektromagnetiniai impulsiniai

skysio debito ir kiekio matuokliai //
Elektrotechnika: Mokslo darbai. Kaunas,
Technologija, 1996. Nr.20(29). p.121-128.
Padegimas R., Virbalis J.A. Nuosd taka
elektromagnetinio skysio srauto matuoklio
signalui // Elektronika ir elektrotechnika.- Kaunas:
Technologija, 2002.- Nr.5 (40 ).- P.59-64.
D. Baltuninkas, J. A. Virbalis Magnetini daleli
poveikis
elektromagnetinio
skysio
srauto
matuoklio magnetinei grandinei.// Matavimai
2002 Nr.1 (21). - P. 24 28. ISSN 1392 1223.

THE ATTENUATION OF THE ELECTRIC REACTOR MAGNETIC FIELD

STRENGTH
Jevgenij MOROZIONKOV, Juozapas Arvydas VIRBALIS
Kaunas University of Technology, The Department of Electrical Engineering, Lithuania
Abstract: The electromagnetic field in which electrical

devices spread out can be dangerous for human. The
special attention is given to a magnetic component of an
electromagnetic field of industrial frequency. Electric
reactors with the air core concern to the electromagnetic
devices which create the strong electromagnetic field.
For reduction of the magnetic field influence to the
person organism the following measures should be
taken for: shielding of workplaces; use of shielding
waistcoats; exception of long exposure in places with
the raised level of the industrial frequency magnetic
field. Recently many new materials with unique
properties have beem developed. Therefore, the
shielding is one of the most perspective approaches to
the solution of the problem on decrease in inadmissible
levels of electromagnetic fields.
1st, 2002. According to this document the action on a

human body of the industrial frequency magnetic field
strength should not exceed 900 /m in the working day
(8 hours) [1]. Acceptance of the norms specified in the
Instruction of Parliament and the Council of Europe
2004/40/, in Lithuania is postponed till April, 30th,
2012. Thus, a tendency to decrease in maximumpermissible specifications of influence of magnetic
fields it is not observed yet. Now many experts consider
that maximum permissible size of a magnetic flux
density must be equal to 0,2 - 0,3 . These values are
equivalent to magnetic field strength values 0,16 - 0,24
/m.
Electric reactor is one of the most powerful sources of
magnetic fields of industrial frequency in power plants.
Therefore, the adequate attention must be given for
attenuation of the reactor magnetic field.
Keywords: magnetic field, the electric reactor, electric

shielding of magnetic fields, the limiter of a shock
current.
2. The three-phase electric reactor of vertical

installation
1. Introduction
In high-voltage networks short circuit currents can reach

very larges values and it is not obviously possible to
pick up installations which could sustain the
electrodynamic forces arising by these currents. The
main destination of electric reactor is to restrict the
shock current of a short circuit [2].
Usually the air electric reactor is placed in the ground
floor room of the indoor switch-gear, i.e., in the small
area room. In the first floor under the reactor there can
be situated the sectionalized bars, disconnecting
switches and cases with the electronic equipment
(equipment of relay protection and automatics).
The three-phase electric reactor of vertical installation
(see Fig. 1) consists of three single-phase reactors. Axes
of the phases A, B and C windings coincide. Distance
between end faces of phase windings is equal to g [2].
The magnetic field, created by currents of the threephase reactor, has elliptic polarization. The effective
value of the total magnetic field on the big semi axis of
an ellipse is defined by this expression [2]:
The discussions about an influence of the industrial

frequency magnetic field on the person health are sharp
in the last years. There are two opposite sights of
scientists and experts to this question. The opinion
dominates that existent standards of limiting values of
the magnetic fields intensity danger to the person, are
quite sufficient and there is no basis for their revision.
However recently the publications appear which prove
the negative influence of relatively small industrial
frequency magnetic fields on a human.
According to Instruction 2004/40/ of the Parliament
and Council of Europe from 2004.04.29, the strength of
the industrial frequency magnetic field on a workplace
should not exceed 400 /m. In Lithuania the basic
document
regulating
limiting
influences
of
electromagnetic fields on a human body is hygienic
norm HN110:2001 "The electromagnetic field of
industrial frequency (50 Hz) on a workplace.
Admissible numerical values of parameters and the
requirement to measurements" operating since January,
235
H max
2
Z
H R2
2
Z
person body. The average electric conductivity of the

person tissues are equal about 0,04 S/m. The numerical
analysis of this influence can be performed using
Maxvell equations.
There is estimated by such analysis that when the
magnetic field acts to the human body in different
directions, the greatest density of the induced current is
observed around a thorax. The maximum value of the
induced density current is obtained 100,3 /m2, when
the magnetic field strength is equal to 80 /m [4].
The influence of small currents induced by the magnetic
field as on internal organs, as on whole person body is
not known till now. In the absence of such information
the normative documents contain only the limiting
values of magnetic field strength.
At carrying out of the electric reactor repair the person
is in immediate proximity from current carrying parts of
the other reactors which are in the next cells under
loading. The certain position of the person near the
reactor can have the essentially influence to the density
of the currents induced in its body.
The distribution of magnetic field strength was
computed by means of package ANSYS in the reactor
and around it. The magnetic field strength on distance
0,5 m from a reactor winding was obtained equal to
49576 A/m and on distance 1 m was obtained equal to
25083 A/m, when reactor is loaded by rated current (see
Fig. 2).
2
H R2 4 H Z2 H R2 sin 2 Z R , (1)
where Z and R are the phase angles of total

components magnetic field strength on the coordinate
axes.
The approached calculation of magnetic field strength,
, /m, created along an axis of the reactor; can be
performed by expression [3]:
H
In
r2
2
x2
1,5
(2)
where r - reactor radius, m; - distance on an axis of the

reactor from its centre to a measurement point, m; I - the
reactor current, A; n - number of coils.
B, mT
b)
Fig. 1. Distribution of reactor phase coils (a) and the photo of

a phase "C" of the air electric reactor (b)
The approximate calculation of magnetic field strength,

H, /m, created by the reactor in a horizontal plane on
distances more than two diameters of the reactor, can be
performed by the expression [2]:
2
rx cos
d
1.5
2
2
In 0 r x 2r cos
H
,
4 2
r2
d

1.5
2
2
0 r x 2rx cos
75
70
61,97 mT
65
60
55
50
45
40
35
30
25
20
15
10
1,214 mT
0,505 mT
5
00
1600 A
1,25 mT
0,5 mT
31,354 mT
1,214 mT
-X -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0
0,505 mT
9 10 X
Distance, m
Fig. 2. The distribution of magnetic flux density created by the

electric air reactor at loading I=1600 A and distance from the
reactor axis equal to 10 m.
(3)
The obtained results show that for such workplaces

additional actions are required for the magnetic field
strength decrease.
where is the angle between the abscise axis and the

vector directed from the centre of the reactor to the
point in which the field is calculated.
4. Limiters of a current rush

The application of the reactors raises not only the
enlarged level of magnetic field strength, but also the
voltage losses. The self-start problem of powerful
electric motors arises caused by this phenomenon.
These problems can be solved using the current rush
limiters which are established in parallel to electric
reactor. Such limiters are developed and produced in
Germany (ABB-Color Emag Is-limiter), in the USA
(G&W Electric Company CLiP) and in the other
countries [6].
3.Estimation of influence of the electromagnetic field

on the workers at service of the electric reactor
Last years the increasing attention is given to such
mechanism of magnetic field influence on a human
body, as the currents induced in it. The induced currents
values depend on the magnetic field frequency, on the
direction and the value of the magnetic field strength, on
the conductivity of the tissues and on the form of the
236
When the reactor operating conditions are normal the

voltage losses are not excided the 3 4 % in
comparison with the voltage rate value. That is quite
admissible. The maximal value of short circuit current
rush can be calculated by expression [5]:
im 2.54 I N
100 ,
X r 00
(4)
where IN is network rated current; Xr is the reactor

reactance.
Fig. 5. The circuit of Is-limiter and a principle of its work: a)
in a normal operating mode; b) at the moment of giving of an
initiating impulse on a charge; c) at the moment of current
interruption in an explosive breaker; d) the rupture chain; e)
the cross-section of Islimiter in an initial condition and after
operation [6]
Fig.3. Is-limiter and the reactor connected in parallel
In Is-limiter the bus rupture occurs only in one point, in

CLiP - in several points (on 10 V four ruptures). With
more quantity of the rupture intervals the speed of a
current switching from the basic bus to a parallel safety
fuse increases. The electric durability of the
disconnecting device improves after full transition of a
current to a safety fuse [6].
a)
b)
Fig. 6. Is-limiter: a) Truck mounted Is-limiter in a switchgear

panel; b) Is-limiter connected in parallel with a reactor (fixed
mounted)
5. Application of electromagnetic screens

The electromagnetic screens can restrict the magnetic
field strength created by electric reactor, too. This
variant of shielding can be used, when the new electric
reactor is designing, manufacturing or installing. This
method does not suited to reactors already being in
operation because of the design features of the offered
screen. The winding of screen coil K is placed over the
winding of the phase coil N (see Fig. 7, a). The axes of
shielding and phase coils coincide [3].
Fig. 4. Device CLiP: 1 - logic schemes, 2 - input, 3 - the

dividing transformer, 4 - the current transformer, 5 - a basic
insulator, 6 - rupture points on the copper bus, 7 - a safety
element, 8 - replaced disconnecting device
In CLiP the signal on switching-off arises only when the

instant value of current Is achieves the threshold size, in
Is-limiter the system of definition Is by the speed of
increase of a current di/dt is applied.
In the normal conditions the current flows by the copper
bus. When the current rush is beginning the electronic
control device switches the disconnecting device which
tears the bus on some sites. The current is switched to in
parallel connected safety fuse. The safety fuse begins to
melt, providing restriction of the current within the first
half-cycle of current to the first peak [7].
The current limiters are produced in various executions,
both for external installation and in the form of sliding
cells of complete switch-gears.
a)
b)
Fig. 7. The scheme of a shielding winding K arrangement over

the winding N of phase coil (a) and the vector diagram of
magnetic field strength components created by the phase coil
winding and the shielding winding (b)
237
The turns of winding are reeled up in opposite

direction to phase coils N and both windings are
connected in series. The components of magnetic field
strength created by currents in windings N and are
directed in opposite directions (see Fig. 7b). The
magnetic field strength of shielding winding
compensates some part of the magnetic field created by
reactor. The total values of magnetic field strength are
defined by expressions (5) and (6) [3]:
H ZPK
I R0
N 1 2
I R0EK
K 12
j 0 0
HRPK
R0EK R cos
2
2
2
z g K jk R R0EK RR0EK cos
j0 0
1. Despite the overestimated norms the values of the

electric reactor magnetic field strength in normal
conditions are considerable and must be decreased.
2. At parallel connection of the reactor with limiters of a
shock current, the loading current flows bypassing the
reactor. Therefore the reactor magnetic field is
insignificant in normal conditions. The values of the
magnetic field strength exceed maximum permissible
only during short circuit in a network after operation
of the shock current limiter.
3. Installation on the electric reactor of shielding
windings allows to lower considerably the levels of
magnetic field strength created by the air reactor.
(5)
d,
3
4 i0 0
2
2
2
z in R R0 2RR0 cos
. K12
When the V shielding winding is installed the full

inductive resistance of the reactor will decrease to 33 %.
z hp cos
I R0 N12
0EK
(6)
z gK hk cos
3
2
2
2
z gK jk R R0EK 2RR0EK cos
(7)
6. Conclusions
I R
XP = XP + XK + XV - 2MPK - 2MPV + 2MKV .
R R cos
0
d

3
4 i0 0
2
2
2
z in R R0 2RR0 cos
The inductive resistance of the winding N of the reactor

is equal to XP = 0,35 .
The inductive part of the full resistance of the reactor
with shielding windings and V is defined by
expression [3]:
d.
7. References
Using the shielding winding mounted on the top of the

reactor the magnetic field strength over the reactor
decreases almost in 5 times. Sideways from the reactor
the magnetic field strength decreases only in 1,2 times.
Its greatest value is on the area of the reactor winding
middle.
For the magnetic field shielding in the horizontal plane,
the shielding winding V can be mounted outside of the
reactor winding as it is showed in Fig. 8. This winding
is reeled up in opposite direction to coils of the reactor
winding, too. The coils of the reactor and of the winding
V are connected in series [3]. When the shielding
windings K and V are used the magnetic field strength
over the reactor decreases to 30 times, and sideways
from the reactor to 6 times.
1. Lietuvos higienos normos HN 110: 2001 Pramoninio

danio (50 Hz) elektromagnetinis laukas darbo vietose.
Parametr leidiamos skaitins verts ir matavimo
reikalavimai.
2. .., .., ..
. //
,
, 2007. -
2. - . 108 - 113.
3. J. Morozionkov, J.A.Virbalis. Magnetic Field of Power
Plant Air Core Reactor. // Electronics and Electrical
Engineering. Kaunas: Technologija, 2007. No. 7(79).
P. 67 70.
4. .., ..
.//
3-
- 2008, 2008. .
81-90.
5. J. Schaffer. Commutating Current-Limiters an effective
alternative for high current protection.// NETA World, vol.
18, Winter 1996-97. - No. 4. - P. 7-18.
6. .., .., ..,
..
. //
IX-
" 2030.
-.", 2007.
7. ABB AG. Calor Emag Medium Voltage Products. Islimiter.// Leaflet no. DEABB 2243 08 E Leaflet no.
DEABB 2243 08 E. - Germany, 2008. - P. 24.
Fig. 8. An arrangement of shielding windings and V outside

of the reactor winding N
238

DIGITALLY SELECTED SPEEDS FOR

A DIRECT CURRENT (DC) MOTORS APPLICATIONS
Sambwa ADOKO*, Onyekachi ADOKO**
*Asa-Lambda Technology Ltd., Nigeria;
** Benson Idahosa University, Nigeria
Abstract: The authors present a digitally selected speed

for a DC motor using pulse width modulation (PWM).
The 555 timer which generates a PWM output, is
interfaced by a multiplexer channel and a switching
regulator, to digitally select the speeds and position for
DC motors applications. Although, it is an open loop
controller with great level of precision and simple to
configure and design.
Fig. 1. Basic DC motors speed controller topology
Keywords: Pulse generator, switching regulator circuits

design, and applications.
1. Introduction
Ideally, the motor speed controller represents a module
which is used to drive a motor at a pre-determined
speed. In some cases, it measures such speed, hence the
difference between feedback speed controller and open
loop speed controller. The adopted speed controller
circuit in this work does not necessary assume the above
mentioned objectives rather, it is designed to efficiently
control the DC motors speed to a pre-set values as
shown in Fig. 1.
In thru perceptive, the DC motor speed controller is a
variable power supply. This is accomplished by simply
adjusting the voltage sent to the motor. Visibly, such
technique may seem highly inefficient to perform.
Consequently, the process being adopted in this work is
an ON OFF switching technique, which is performed
by a 555 timer (IC) as pulse generator, coupled with a
switching regulator made of power transistor that can
turn very large currents ON and OFF under the control
of a signal width.
Fig. 2. Proposed speed controller topology
The time that it takes a DC motor to vary its speeds

under switching conditions is dependant on the inertia of
the rotor, and how much friction and load torque there is
[1]. The objective of this work has been and remains the
realization of the principle of switch mode speed
control.
Thus, the speed is governed by PWM Pulse width
modulation, following the general principles governing
DC motors in which the speed of a DC motor is directly
proportional to the supply voltage [2]. There exist
several methods and procedures establishing such
proportionalities.
Hence,
the
introduction
of
electronically controlled DC motors speed, as shown in
Fig. 3.
239
Fig. 3. Basic DC motor speed controller using a variable

supply
If the above stated principles are justified, then the main

objective of this work will be to design a regulated
supply and create a sub-circuit which will function as a
variable switching regulator. That is, a regulated supply
having a variable regulated output. This scenario is
necessary as to create a working environment where the
output stage regulators function will be to change the
input supply voltage to another voltage level. The
second stage regulator is the DC motors speed
controller. In other words, the supply voltage will
remain constant while the second regulator will vary its
output to match with the pre-selected motors speeds. It
then creates a voltage of different magnitude from the
voltage that power it, and the process is done digitally
through switching regulator.
Fig. 5. Integrated pulse generator unit
The above circuit diagram as described in Fig. 5 can be

built into a capture/compare/ PWM (CCP) Peripheral.
Configured for a single PWM output, while the step up
regulator is designed to produce the unique output for
the motors control. If cost and availability remain a
critical design factor as it is the case in the sub-Saharan
Africa, used and unserviceable apparatus remain one
of the best alternative options.
3. LM 555 (IC) timing analysis
The basic elements of a 555 timer (IC) are shown in Fig.
6. The switching unit is commutated to the supply, via
the pulse generator. The 555 timer has been re-arranged
to be analysed as a T-Network configuration, using the
existing basic techniques.
Fig. 4. DC to DC Converter as DC motors speed drive

Fig. 6. The projects clock generator circuit
2. Pulse generator (PWM)
4. Switching regulator
The simplicity of this work lies in the selection of the

pulse generator which produces a digital signal. The
width of the modulated signal is critical for the motor
speed control, by adjusting the pulse width, the speed of
the DC motor can efficiently be controlled without
larger linear stages.
In view of maintaining a constant output voltage, an

adoption of a switching regulator became apparent and
appropriate for this work. As it can be shown in Fig. 7,
the primary filter capacitor (c) has been placed on the
output of this regulator. Such arrangement has been
confirmed by several authors as an efficient arrangement
in switching regulators [3].
240
Fig. 7. Switching regulator topology
Keeping the fact of having a constant voltage and

constant current output, a step up switching regulator
remains an ideal consideration [3]. Modulated by an
external pulse generator, the adopted regulator in this
work, functions as shown in Fig. 8-10, that is, the
switching regulator uses a power switch (transistor), an
inductor, and a diode to transfer energy from, input to
the motor. The sequences of such transfer constitute the
main objective of this work. The regulators components
perform special role in their individual characteristics.
Fig. 9. Discharging stage of the switching regulator
5. A charging stage
At this stage, the inductor [L], main assignment is to
limit the current slew rate through the power switch. It
creates a fast energy storage which is expressed as
follows:
E 0.5 LI L2 .
Fig. 10. A graph diagram of the current (i) at a time t =0
di VL
dt L
(1)
(2)
In contrarily, as shown in Fig. 8, when the switch opens,

the inductance voltage is expressed as
VL L
di
.
dt
For a DC motors operating environment, it will be

advisable to maintain the current through the inductor in
a steady state mode. Hence, for a given charge time tON,
and a given input voltage with a lossless energy circuit,
there will be a specific discharge time tOFF, for an output
voltage.
The correlation between the above sequence and the
clock transition period needs to be established. In the
interim, the switching circuit can equally be calculated
as follows:
Fig. 8. Regulators charging stage
VIN
As stated in the engineering theories, the inductors

current IL is 90 degrees out of phase with the voltage
[4]. This will facilitate the recovery of the stored energy
during the discharge phase of the switching cycle. The
Engineering beauty is the fact that this process produces
less heat at an acceptable efficiency.
tOFF
VL
tON
(3)
Working in the steady state mode, we have
t
VOUT VIN 1 ON
tOFF
6. Discharging
(4)
Then, the duty circle is
At the instant when the switch closes, VIN is impressed

across the inductor L. The capacitor C discharges its
energy to the load. The current IL through the inductor
rises linearly with time at a rate proportional to the input
voltage divided by the inductance.
241
t ON
,
t OFF t ON
(5)
VOUT
VIN
.
1 D
can be applied in a wide range of applications. It offers

better efficiency, smaller components and cost effective.
This simplicity has been demonstrated in the work
where it was possible to design and construct a complete
circuit using the components recovered from the
unserviceable apparatus and the final results are
satisfactorily.
While there exists, some buckets of problems in DC
motors speed control, the works done by C. Schaefer
[7], J.P. Karunadasa [8], and N. Matsoi [9] illustrate
useful contributions in this issue of engineering
importance.
(6)
From the above analysis, the adopted controller can be

classified as a voltage mode controller (VMC). The preselected pulses detect the shape and width of the output
voltage, hence controlling the speed of the DC Motor. A
current mode control (CMC) can equally be used.
7. Applications
With the trend of power electronics where the speed and
power handling capabilities of semi conductors have
improved substantially in the last few years in the field
of portable power supply and electric motors speed
control. This development has considerably enhanced
some complex real time algorithms often required in
industries.
In addition, the development of renewable energy has
been made possible through such innovations as DC-DC
converters as an added values from some powerful low
cost semi-conductor devices.
As mentioned by B.K. Bose, that Electrical machine
control was one of the first research fields to take
advantage of this continuing development with both AC
and Dc machines [5]. Although the applications of DC
motors speed controller have spread in several avenues,
among them are, the robots, electrical vehicles, and
renewable energy technologies, to mention but a few.
Equally, there seem to be some unanswered questions in
PWM techniques for DC motors speed control. This
puzzle lies on the load torque and speed relationship.
It has been established that, as the load torque of a DC
motor increases, the speed drops [6]. This is a general
occurrence for an open loop speed control as it is the
case in this work. It will be assumed that PWM under a
constant speed will have its torque equal to the load
torque. But, when the load increases, the motor torque
and load torque have to be balanced out immediately
with the current IL and the equivalent torque value on
the torque current TL. But, obviously speaking, the
speed demand is a DC voltage which is fed to the PWM
generator for motor. This drives motor at the speed
required on the demand voltage, but not on load torque.
Possibilities to incorporate special devices/sensors to the
circuit can be a positive aspect with tremendous
advantages [7].
9. References
1. Marcel P.J., Gaudreau P.E., Casey J.A. , Hawkey
T.J., Kempkes M.A., Mulvaney J.M., Ver Planck P.
Solid State High Voltage DC Power distribution and
control. In proceeding of the Particle Accelerator
Conference, New York, 1999.
2. Adoko S. Effective Paradigm for direct current (DC)
distribution using light emitting diodes (LEDs) in
architectural lighting system. The 3rd Regional
World Renewable Energy Congress (AISEC-10),
Kuwait City, Kuwait, 2008.
3. Adoko S., Irowa O. Integrated power supplys
design and contruction practice for solar lighting
applications. Energy and Environment (EE09)
WSEAS Coference, Cambridge, UK, 2009.
4. Theraja B.L., Theraja A.K Textbook of Electrical
Technology. Ninja Construction & development Co.
(p) Ltd Ram Nagan, New Delhi, 1997.
5. Bose B.K. Adjustable speed AC Drives A
technology status review. IEEE Trans Ind. App.,
1982, vol. 70, no. 2. p. 116-135.
6. Herrit B., Morogavel R. PWM DC Motor Control
Using Timer A of the MSP 430. Application
Report SLAA 120 Texas Instruments December
2000.
7. Schaefer C. Field Weakening of Brushless
Permanent Magnet Motors with Rectangular Current.
Proc. Env. Conf. Power Application (EPE),
Florence, 1991, vol. 3, p. 429-434.
8. Karunadasa J.P., Renfrew A.C. A flexible fast Digital
Controller for a Brushless DC motor. 4th Int. Conf.
Power Electronic and Variable speed Drives, 1990.
p. 429-434.
9. Matsoi N. Sensorless PM Brushless DC Motor
Drives. IEEE Trans Ind. Electronics, April 1996,
vol. 43, no. 2. p 300-308.
8. Conclusion
DC motors speed control technique has been hitherto
very difficult and energy wasteful method to implement.
The modern switching technique as adopted in this work
242

INTENSIFICATION OF HEAT EXCHANGE IN REFRIGERATION

TECHNOLOGIES BY USE OF CORONA FIELD
Stasys EBRAUSKAS*, Irmantas BENIULIS**
* Kaunas University of Technology; Lithuania, **Lithuanian University of Agriculture, Lithuania
boundary layer with a still air, reducing its thermal

resistance and increasing the heat transfer from the wall.
Experiments demonstrate the ability of ionic wind to
decrease the wall temperature substantially in the
presence of a bulk flow over a flat plate, corresponding
to local enhancement of the heat transfer coefficient by
more than twofold [4]. This technology to increase the
heat transfer coefficient, which describes the cooling
rate, by as much as 250 %, is unique, because other
experimental cooling enhancement approaches might
give you a 40-50 % improvement [5]. It is sufficient to
create ionic wind with a velocity of 0.1-0.3 m/s to
demonstrate such an effect [6].
Abstract: Experimental results of the influence of

corona field on the heat exchange in domestic freezer
processes are discussed. Average freezer condenser
surface temperature influenced by negative corona field
at the end of compressor operating cycle decreases by
2.4 K in comparison with the temperature without
corona field. This decrease is 4.0 K at the end of
compressor pause cycle. Decrease of condenser surface
temperature causes the reduction of compressor
operating cycle, and consequently the reduction of
freezer energy consumption by 14 %. Use of negative
corona field for cooling of water from 289 K to 273 K
in the freezer reduces the duration of the process and
freezer energy consumption by 35 %. Heat exchange
processes intensification and energy saving is caused by
small corona device energy consumption.
2. Corona fields in cooling and freezing processes

During the last twenty years daily energy consumption
of domestic coolers and freezers is reduced from 1.5
kWh to 0.5 kWh. This is achieved due to use of more
perfect constructions and new heat insulating materials.
There are several processes in each cooler or freezer
which effectiveness may be enhanced by use of the
corona field: cooling of the product from room
temperature to 0 C, deep-freezing or icing of the
product, storage of the product in deep-frozen state,
defreezing of the product or the ice, cooling of the
condenser of cooler or freezer, heat exchange on the
surface of evaporator, evaporation of refrigerant,
condensing of refrigerant, separation of oil from
refrigerant in the compressor, production of artificial
snow, control of the moisture in refrigerating product,
antiseptic action to refrigerating product. Effectiveness
of using the corona field is caused by small energy
consumption of corona field arrangements. The results
of cooling the surface of condenser and cooling and
freezing of water in domestic freezer by use the corona
field are discussed in this article.
Keywords: rate of temperature changing, corona field.

1. Introduction
Direct current corona field is widely used for heat and
mass transfer enhancement. In this technique, application of high voltage to an electrode with a small
radius of curvature ionizes air molecules, which are
propelled by the electric field, transferring part of their
energy to neutral air molecules, thus creating airflow
and cooling effects. It is shown in [1] that the increase
of convection heat-transfer coefficient is linearly
proportional to the applied voltage for all tested air-gap
separation distances. Experimentally and by means of
modeling is determined that velocities of airflow
initialized by ion movement in the space between
electrodes are of values up to 4 m/s, and maximal value
of convection heat transfer coefficient is 280 W/(m2K)
[2]. Because ionic wind can generate flow with no
moving parts and have low energy consumption, it offer
an attractive method for enhancing convection heat
transfer from a surface. An increase of more than 200 %
in the local heat transfer coefficient is observed due to
the ionic wind for a power input of only 68 mW [3]. In
the presence of a bulk flow ionic wind distorts the
3. Cooling of the surface of condenser

Domestic refrigerator of power, equal to 180 W and
overall length of condenser tube 8.4 m is used for
243
testing in direct current or 50 Hz corona field. The

temperature of ambient air is 296 K. Diameter of
condenser tube is 6.5 mm. Corona wire which diameter
is 0.050 mm is attached at the distance from the tube
surface equal to 15 mm. Refrigerator is connected to
supply network through watt-hour meter. Formulas for
computation of volt-ampere characteristic in electrode
system wire parallel to the cylinder are given in [6]
and [7]. Operating corona voltage is determined from
computed volt-ampere characteristic and is equal to 6
kV. Linear corona current density is 100 A/m.
Condenser tube surface temperature is measured at the
beginning of the tube, at the end and at the other 4
points on the surface of the tube separated by distances
168 cm. Freezing chamber contains a vessel with 5 l of
water. The best effect is detected for negative corona
field. Variation of condenser tube surface temperature at
the 1 point is presented in Fig. 1. Condenser surface
temperature at the end of compressor operating cycle is
309 K, at the end of pause is 298 K for negative corona
field, and correspondingly 311 K, 302 K for control
test. Average duration of operating cycle is 3.5 min for
corona field and 4.0 min at control test. Average energy
consumption
T,K
306
2 min
302
3 min
300
298
- corona
T,K
310
1 min
308
2 min
306
3 min
4 min
304
5 min
302
6 min
8 min
304
300
298
operating
cycle
5 min
Average meaning of condenser surface temperature

influenced by corona field rises from 298 K at the
beginning of operating cycle of the compressor to 309
K at the its end. Corresponding values for control test
are 302 K and 312 K.
306
302
4 min
0
1
2
3
4
5
6 n
Fig. 2. Variation of temperature at 6 equally distant points in
overall length of the condenser tube influenced by negative
corona field for pause of the compressor operation
control test
308
1 min
304
T,K
310
308
pause
t,min
300
0
1
2
3
4
5
6
overall length of the condenser tube during the control test for
pause of the compressor operation
0
2
4
6
8
10
12
Fig. 1. Variation of temperature at the beginning of condenser
tube
is 1.45 kWh/day for corona field, and 1.65 kWh/day for

control test. Energy consumption of corona device is
0.1 kWh/day. Reduction of condenser surface
temperature by 1 K corresponds to a day energy
consumption by 4 %.
Dynamics of temperature variation at all 6 points of
condenser tube surface is given in Fig. 2 Fig.5.
During compressor pause average meaning of
condenser surface temperature influenced by corona
field decreases from 309 K to 298 K (Fig. 2) and from
311 K to 302 K at a control test (Fig. 3),
correspondingly. Temperature decrease wave passes
from the end of condenser tube to the beginning. The
rate of this decrease during the first minute of pause
influenced by corona field exceeds 1.7 time the
corresponding rate at the control test.
Condenser surface temperature increase dynamics at all
6 points for operating cycle of compressor influenced
by corona field and during the control test is shown in
Fig. 4 and Fig. 5.
T,K
308
3,5 min
2,7 min
1,8 min
306
304
302
0,9 min
300
298
296
0
n
0
1
2
3
4
5
6
overall length of the condenser tube influenced by negative
corona field for operating cycle of the compressor
244
T,K
4 min
310
3 min
308
T,K
289
287
2 min
306
304
302
300
285
50Hz corona
283
+ corona
281
1 min
control test
279
- corona
277
275
t,min
273
0
1
2
3
4
5
6
overall length of the condenser tube during the control test for
operating cycle of the compressor
Fig. 6. Variation of freezing water temperature
Temperature rise wave passes from the end of

condenser tube to the beginning. The rate of this
increase during the first minute of operating cycle
influenced by corona field exceeds 1.75 time the
corresponding rate at the control test.
Energy consumption of the process is 3400 kJ (0.944

kWh) for control test conditions, and 2900 kJ (0.805
kWh) for negative corona field conditions (17% reduction). Energy consumption of corona field arrangement
is 20 Wh.
4. Cooling and freezing of water in freezing chamber

of domestic freezer
T,K
275
271
Freezing chamber of domestic freezer which power is

180 W contains of a vessel with a 1000 ml of water.
The area of the open surface of water is 490 cm2. A net
of parallel corona wires of 0.1 mm diameter is arranged
10 mm above the surface of water. The distance
between neighbouring wires is 15 mm. Area of the net
with corona wires is 314 cm2. Formulas for computation
the volt-ampere characteristic are the same as in [8].
Operating value of corona voltage U = 10 kV is selected
from computed characteristic. It corresponds to 400 A
of average corona current value and 4 W corona
arrangement power. Cooling water temperature
decrease process from 289 K to 273 K is shown in Fig.
6.Positive corona field practically has no effect to the
rate of temperature decrease. 50 Hz corona field
demonstrates negative effect. No effect is detected also
for electrostatic and quasistatic 50 Hz electric field.
Only negative corona field demonstrates the cooling
rate enhancement effect. Water cooling from initial
temperature to 273 K is 25 min under control test
conditions, and 18 min influenced by negative corona
field. It corresponds to 240 kJ (67 Wh) energy
consumption without a field, and 177 kJ (49 Wh) with a
corona field (35 % decrease). Energy consumption of
corona field arrangement is only 1.2 Wh. Constant
temperature values for t > 30 min correspond the
freezing point.
Freezing water temperature changing curves in the
interval 275 K 263 K are shown in Fig. 7. Duration of
water freezing to ice in negative corona field (at 263 K
temperature) is 5 h. It corresponds to 20 % reduction in
comparison with control test conditions. Duration of the
process influenced by the positive corona field reduces
by 15 % in comparison with control test. 50 Hz corona
field demonstrates the least effect to freezing process.
273
10
20
30
40
50
60
t,h
50Hz corona
271
control test
269
- corona
267
+ corona
265
263
Fig. 7. Variation of frozen water temperature
Ice melting temperature rise curves in the interval 263

K 283 K are shown in Fig. 8. Contrarily to cooling
and freezing tests the most substantial influence to the
rate of ice melting temperature rise has the positive
corona field. Duration of ice melting process
corresponding to temperature rise from 263 K to 283 K
is 4.5 h under control test conditions. Duration of the
process influenced by positive corona field is 3.7 h (60
% reduction). Flat part of characteristics correspond to
specific solution heat.
5. Conclusions
1. Average freezer condenser surface temperature
influenced by negative corona field at the end of
compressor operating cycle decreases by 2.4 K in
comparison with the temperature without corona field.
This decrease is 4.0 K at the end of compressor pause
cycle.
245
6. References
T,K
283
1. Hsu C. P., Jewell-Larsen N. E., Krichtafovitch I. A.,

Mamishev A. V. Heat-Transfer-Enhancement Measurement for Microfabricated Electrostatic Fluid Accelerators. Journal of Microelectromechanical Systems,
Vol. 18, No. 1, February 2009, p. 111-118.
2. Jewell-Larsen N. E., Krichtafovitch I. A., Montgomery S. W., Dibene J. T., Mamishev A. V. CFD
Analysis of Electrostatic Fluid Accelerators for
Forced Convection Cooling. IEEE Transactions on
Dielectrics and Electrical Insulation, Vol. 15, No. 6,
December 2008, p. 1745-1753.
3. Go D. B., Maturana R. A., Fisher T.S., Garimella
S.V. Enhancement of External Forced Convection by
Ionic Wind. International Journal of Heat and Mass
Transfer, Vol. 51, 2008, p. 6047-6053.
4. Go D. B., Garimella S. V., Fisher T.S. Ionic Winds
for Locally Enhanced Cooling. Journal of Applied
Physics, Vol. 102, 053302, 2007, 8 p.
5. McHale. Purdue Researchers Demonstrate New
Chip-Cooling Technology. Military and Aerospace
Electronics, November 2007, p. 3.
6. ebrauskas S., Mariulionis P. Elektrod sistemos
laidas alia cilindro vienpolio vainikinio ilydio
elektrinis laukas. Electronics and Electrical engineering, 2006, No. 1(65), p. 43-47
7. ebrauskas S., Mariulionis P. Elektrod sistemos
laidas alia cilindro elektrostatinis laukas.
Electronics and Electrical engineering, 2005, No.
6(62), p. 32-36
8. ebrauskas S. Bariauskas R. Using of Corona Fields
in Building Material Industries. Proceedings of
Technologies 2009, Kaunas, 2009 p. 247-250.
+ corona
50Hz corona
278
control test
273
- corona
40
80
120
160
200
t,min
240
268
263
Fig. 8. Variation of melting ice temperature
2. Decrease of condenser surface temperature causes the

reduction of compressor operating cycle, and consequently the reduction of freezer energy consumption by
14 %. Decrease of condenser surface temperature by 1
K corresponds to the reduction of energy consumption
by 4 %.
3. Use of negative corona field for cooling of water from
289 K to 273 K in the freezer reduces the duration of the
process and freezer energy consumption by 35 %.
4. Use of negative corona field for freezing of water to
ice causes the decrease of process duration and freezer
energy consumption by 17 %.
5. Heat exchange processes intensification and energy
saving is caused by small corona device energy
consumption in comparison with freezer energy
consumption.
246

USING OF CORONA FIELDS IN BUILDING MATERIAL INDUSTRIES

Stasys EBRAUSKAS, Riardas BARIAUSKAS
frequencies [6]. Moisture content can be reduced by

using the chilling and heating method [7]. The speed of
water evaporation may be enhanced by using the corona
field. Ion-driven movement of air influences the rate of
mass exchange processes of evaporation. Absolute rate
of water evaporation from porous material increases up
to 15 times by using the corona field [8]. We discuss
there the results of clay panel drying enhancement by
using the direct current corona field.
Abstract: The experimental results of clay panel drying

rate enhancing in direct current corona field are
discussed. Using the direct current corona field for clay
panel drying process increases the rate of 100 %
moisture evaporation 3.2-fold in comparison with
natural drying process in the air. The duration of drying
to 80 % moisture evaporation reduces 10 times, and the
duration of 50 % moisture evaporation reduces 20
times. Energy consumption of 100 % drying is 1.5 kWh
to 1 m2 of clay panel area. It takes 0.6 kWh/m2 for 80
% drying, and 0.3 kWh/m2 for 50 % drying. Using the
direct current corona field to drying process increases
the drying quality. Rate of moisture evaporation in all
points of drying material volume becomes more
uniform. Intensification of drying may be enhanced by
increasing the corona voltage, but it will increase the
energy consumption.
2. Experimental arrangement
We use the direct corona electrode system a set of
parallel corona wires above the plane electrode (Fig.1)
Plane electrode is represented by the surface of drying
clay panel. Test drying clay sample is a plate of 1 cm
thickness and of 10x10 cm2 area. Radius of corona
wires is ro= 0.05 mm, the step of the grid d = 15 mm,
and the height h is variable in the interval 10-14 mm.
Keywords: clay-straw panels, moisture content, corona

field.
1. Introduction
Modern building technologies as a part of a common

industrial technologies ought to meet ecological
requirements. One of new (and ancient) kinds of
buildings is the straw bale one [1]. There are many
advantages inherent to straw bale constructions:
excellent heat insulation quality, reduced heating and
cooling costs, suitability to use local renewable
materials, cheap and fast mounting procedure [2].
Plaster as well as plaster-cartoon panels are used usually
for finishing. There are proposals to use clay-straw
panels in straw bale buildings which are more preferable
in comparison with plaster-cartoon ones for some
advantages [3]. The main disadvantage limiting the
usage of clay-straw panels is comparatively long drying
duration of panels after the formation procedure. One of
successful methods of drying is the freeze-drying [4].
The moisture content of humus decreases to 22 % from
an initial value of 44 % wet basis after 2 h of electroosmosis at 50 V across a 2.9-cm-thick bed in electroosmotically enhanced drying of biomass [5]. The wellknown method of water removal from the paper is
electromagnetic paper drying using the microwave
r0
x
Fig. 1. Electrode system of the experimental device
We use the calculated data for volt-ampere characteristics of corona in the discussed electrode system at
several values of the height h to determine the proper
values of test voltage and current corresponding the
source voltage 10 kV and the current 1200 A. The
used formulas are the same as in [9].
Initial field strength of the discharge depends upon the
radius of the wire r0 and the relative air density :
247
E0 = 3030000 (1 + 0,0298
(ro )),
(1)
the relative air density being determined by the formula:
= pT0 / p0T ,
(2)
where p is the atmospheric pressure, p0 is equal to p for

normal conditions, p0 = 101.325 kPa; T is the thermodynamic temperature, T0 is equal to T for normal
conditions, T0 = 293.15 K. Relative air density equals to
1 if p = p0, T = T0. Onset voltage of the discharge:
r
h
U 0 = r0 E0 2 ln 2 0 .
d
d
(3)
Discharge index depends upon the voltage U:

U
Fig. 2. Computed volt-ampere characteristics for negative

corona
1 2 ln 2 0 .
=
d
U 0 d
(4)
Data of volt-ampere characteristics at the value of

height h = 10 mm are used for choosing the proper high
voltage source.
Linear density of discharge current:

2
r
d
I 0 = 1,6k 0 1 + 1,12 E0 0 1,5 ,

2h
h
(5)
where k is the ion mobility, k = 2.20 104 m2/Vs for
k + = 1.80 104 m2/Vs for

positive discharge, 0 is the permittivity constant, 0
= 8.85 1012 F/m. Initial field strength determined from
(1) for ro= 0.05 mm and = 1 is Eo = 158.0 kV/cm.
The dependence of initial voltage Uo determined from
(3) on the height h is given in Table 1.
negative discharge, and
Table 1. Dependence of initial voltage on the height h

h, mm
Uo, kV
10
6.36
11
6.69
12
7.02
13
7.35
14
7.68
Fig. 3. Computed volt-ampere characteristics for positive
corona
Computed volt-ampere characteristics for negative and

positive discharge are shown in Fig. 2 and Fig. 3. The
values of height h are given in mm. The greater values
of voltage are needed for the same values of current if
the values of height h are increased.
The greater values of the discharge current for negative
corona are determined by the greater negative ion
mobility. The value of height of corona wires above the
plane electrode is selected equal to 10 mm for
experimental arrangement.
Comparison of computed and experimental volt-ampere

characteristics for discussed electrode system is shown
in Fig. 4 and Fig.5. Experimental curves differ from the
computed ones due to the roughness of corona electrode
and the errors of measurement. The discharge voltage
shouldnt exceed the value 2U0 because it is related with
a risk of a spark discharge.
248
Fig. 4. Comparison of computed and experimental voltampere characteristics for negative corona
Fig. 6. Drying process in negative corona field
Fig. 5. Comparison of computed and experimental voltampere characteristics for positive corona
3. Results of an experiment
Drying test is performed at 19 C ambient air temperature, 45 % relative humidity and 101-102 kPa of an
atmospheric pressure. Drying clay sample is arranged
horizontally on the grounded aluminium sheet. Initial
weight of wet clay sample is 0.2410.001 kg.
Periodicity of sample weighing is 1 h. Each drying
session is performed twice: under free convection
conditions and with a sample subjected by direct current
corona field. Total discharge current is adjusted to 200
A at the voltage 7.5 kV (approximately, the value of
the voltage is slightly increased to compensate the
increasing of resistance of drying clay sample). Final
weight of dry clay sample is 0.2060.001 kg. Total
weight of water in wet clay sample is 0.035 kg. Curves
of drying process with negative and positive corona
field in comparison with natural drying conditions in
still air are presented in Fig. 6 and Fig. 7. Letter m
denotes the mass of drying clay sample, and t is the time
measured from the beginning of drying. Drying
experiment is finished at 10 h from the beginning when
the sample drying in corona field has no moisture.
Fig.7. Drying process in positive corona field
Parameters of the drying process with the corona field

and without it are given in Table 2.
Table 2. Parameters of drying process
t, h
1
2
4
6
8
10
mnd, %
0.6
2.8
8.6
14.3
22.8
31.4
m+cd, %
31.0
54.0
88.6
94.3
97.1
100.0
m-cd, %
28.6
51.4
82.8
91.4
97.1
100.0
r+, times
51.7
19.3
10.3
6.6
4.3
3.2
r-, times
47.7
18.3
9.6
6.4
4.3
3.2
Meanings of symbols used in Table 2 are the following:

mnd is the quantity of naturally evaporated water from
drying clay sample, m+cd and m-cd are the quantities of
evaporated water from drying clay sample with positive
and negative corona discharge, r+ and r- is the increase
of drying rate actuated by positive and negative corona
249
evaporation in all points of drying material volume

becomes more uniform.
4. Intensification of drying may be enhanced by
increasing the corona voltage, but it will increase the
energy consumption.
field, respectively, in comparison with natural drying

rate.
Positive corona field influences slightly better to the
rate of moisture evaporation at the beginning of drying
process in comparison with the negative corona field,
but the influence of both corona modes to the overall
drying process is the same. The duration of 100 %
drying under the action of direct current corona field is
10 h, the increase of the drying process rate to evaporate
all the moisture content is 3.2-fold in comparison with
natural drying conditions. The duration of 80 % drying
actuated by corona field is 4 h, the increase of the
drying process rate to evaporate this content of moisture
is more than 10 times. The duration of drying process to
evaporate 50 % of moisture content is 2 h, using the
corona field it reduces by 20 times. Intensification of
the drying process is determined by the ion-driven wind
(electric wind) of the corona discharge and by the
phenomenon of electro-osmosis in the drying material
[5]. Using of corona field improves the quality of drying
because of more uniform moisture evaporation in all the
volume of the material. Under the conditions of
naturally drying process the moisture evaporates more
intensively in the layer near the surface.
Energy consumption of 100 % drying is 1.5 kWh to 1
m2 of clay panel area. If drying process ends at 80 % of
moisture evaporation it takes 0.6 kWh/m2, and the
consumption of energy for 50 % moisture evaporation is
0.3 kWh/m2.
Experiment of drying is performed at the relative corona
voltage U/Uo = 1.67. Intensification of drying may be
enhanced by increasing the relative corona voltage but it
will increase the energy consumption.
5. References
1. Nekroius L. Architektros ir aplinkos darna:

iaudinis namas. http://www.spec.lt
2. Milutien E., Jrmann K., Keller L. Straw Bale
Building Reaching Energy Efficiency and
Sustainability in Northen Lattitudes. International
Conference Northen Sun 2007, 2007-05-30
2007-06-01, Rga, 6 p.
3. King B. Design of Straw Bail Buildings. The State
of Art. New York, Green Building Press. 2006. 218
p.
4. Kaneko T., Hamada K., Chen M. Q, Akashi M.
Freeze-Drying of Soft Nanoparticles with Projection
Coronas Forms Three Dimensional Microstructs.
Advanced Materials, 2005, Vol. 17, p. 1638-1643.
5. Banerjee S., Law S. E. Electroosmotically Enhanced
Drying of Biomass. IEEE Transactions on Industry
Applications, 1998, Vol. 34, No 5, p. 992-999.
6. Monzon J. C. Electromagnetic paper drying. IEEE
Transactions on Microwave Theory and Techniques.
1995, Vol. 4, No 2, p. 299-305.
7. Mishima H., Yasui T., Mizuniwa T., Abe M., Ohmi
T. Particle-Free Wafer Cleaning and Drying
Technology. IEEE Transactions on Semiconductor
Manufacturing, 1989, Vol. 2, No 3, p. 69-75.
8. . ., . ., . .,
. .

.
, 1990, . 5 . 49-53.
9. Remeikis M., ebrauskas S. Heat exchange in the
corona field. Proceedings of the 2nd international
conference on electrical and control technologies,
ECT-2007. Kaunas, Technologija, 2007, p. 209-212.
4. Conclusions
1. Using the direct current corona field for clay panel

drying process increases the rate of 100 % moisture
evaporation 3.2-fold in comparison with natural drying
process in the air. The duration of drying to 80 %
moisture evaporation reduces 10 times, and the duration
of 50 % moisture evaporation reduces 20 times.
2. Energy consumption of 100 % drying is 1.5 kWh to 1
m2 of clay panel area. It takes 0.6 kWh/m2 for 80 %
drying, and 0.3 kWh/m2 for 50 % drying.
3. Using the direct current corona field to drying
process increases the drying quality. Rate of moisture
250

A COMPARATIVE STUDY BETWEEN SMALL TRADITIONAL INDUCTION

MOTORS AND SMALL CONDUCTING SLEEVE INDUCTION MOTORS USING
3D FINITE ELEMENT TRANSIENT ANALYSIS
Seyed Mohsen HOSSEINI*, Javad S. MOGHANI*, Akram HOSSEINI*
* Amirkabir University of Technology, Iran
Abstract: In this paper, transient modeling of a small

induction motor with conducting sleeve is presented.
Conducting sleeve Induction motor has a solid rotor
which is composed of two parts: ferromagnetic core and
conductor hollow cylinder. After two dimensional finite
element transient analysis of the motor, its performance
is compared with traditional induction motor. Then, for
the precision study of the motor, three dimensional
finite element transient simulation is also presented and
compared with two dimensional results. Two and three
dimensional finite element results are relatively similar.
It is shown that in small sizes, the performance of the
conducting sleeve induction motor is better than
traditional squirrel cage induction motor.
number of methods are proposed for analysis of solid

rotor induction motors like finite element and analytical
methods [6, 7], but no work present transient analysis of
conducting sleeve three phase induction motors. In this
paper, with finite element method, transient simulation
of the CSIM is done. First two dimensional transient
simulation of the CSIM is done and its performance is
compared with the performance of traditional induction
motor. Then for precision study of the CSIM, three
dimensional model is analyzed with finite element
method. Results from two and three dimensional finite
element analysis are compared with each other. These
results are agreed together. It is shown that in small
size, the performance of CSIM is relatively better than
traditional induction motor.
Keywords: conductor sleeve, eddy current, finite

element analysis, induction motor, traditional, transient
analysis, three dimensional, two dimensional.
2. Introduction to the Three Phase CSIM Structure

Fig.1 shows three and two dimensional views of a three
phase CSIM structure.
The stator of the CSIM is similar to the traditional
structure.
As shown in Fig.1, the rotor has a ferromagnetic core
that is surrounded by a conducting sleeve.
The material of conducing sleeve may be copper or
aluminum.
1. Introduction
Conducting Sleeve Induction Motor (CSIM) is a special
induction motor with a solid rotor which is made of two
parts: ferromagnetic core and hollow conducting
cylinder. The material of hollow conducting cylinder
could be copper or aluminum. This motor has particular
applications such as helical motion induction machines
[1], [2], [3], [4]. In helical motion induction machines,
induced currents on the rotor must move in helical
direction, so squirrel cage structures can not be used.
Therefore using solid structure for the rotor is
unavoidable. One of important application of solid rotor
induction machine is high speed motors. Due to solid
3. Transient Finite Element Analysis

Governing equations are as following [8]:
B = J
B
.dl = J .ds
dB
J =
dt
B = A
rotor and low cost, solid rotor induction motors are a

good choice for high-speed applications comparative
to other kinds of motors [5]. Conducting sleeve
structure of the rotor is the good counterpart of the
wholly ferromagnetic rotor. The wholly ferromagnetic
rotor structure presents lower torque than the
conducting sleeve structure. In the recent years a
(1)
(2)
(3)
(4)
Where B, J and A are magnetic flux density, current

density and magnetic vector potential respectively.
251
a) Three dimensional view
b) Two dimensional view
Fig. 1. Three and two dimensional views of a three phase CSIM structure
and are electrical conductivity and magnetic
Copper is chosen to be used in conducting sleeve motor.

In every step of transient simulation, induced currents
on conducting sleeve due to stator excitation is
calculated first. Fig.2 shows induced currents on
conducting sleeve at time t=0.05 second.
permeability.
The following equation directly results from the above
equations:
H +
B
=0
t
(5)
The final result is a formulation where vector fields are

represented by first order edge elements and scalar
fields are represented by second order nodal unknowns.
Field equations are coupled with circuit equations for
conductors, because in transient simulations supply
voltages are applied and currents are unknown.
For the nonlinearities, the classical Newton-Raphson
algorithm is used.
The major difficulty for transient simulation is that
induced currents on the rotor conductor must be
calculated in each step, and then the torque must be
calculated. The mentioned procedure is time consuming
and need high memory especially in three dimensions.
Fig. 2. Induced currents on conducting sleeve at t=0.05 second
4. Two Dimensional Finite Element Transient

Simulation
Fig. 3 shows flux lines at t=0.05 second.
Table 1 summarizes the characteristics of the CSIM.

The designed CSIM has small dimensions.
Table 1. Design Data
inner radius of rotor
outer radius of rotor
outer radius of stator
air gap
conducting sleeve thickness
relative magnetic permeability of stator core
relative magnetic permeability of rotor
conductor sleeve conductivity
number of slots
slot wide
current density in stator slot
pole numbers
frequency
number of phases
2 (mm)
4 (mm)
15 (mm)
1 (mm)
0.5 (mm)
2000
250
5.8e+7 (S/m)
6
20
3 (A/mm2)
2
50
3
Fig. 3. Flux lines at t=0.05 second
After reaching to the steady state condition, the transient

torque and speed are obtained and shown in Fig. 4 and
252
Fig. 5 respectively. Note that simulations are done at

free acceleration.
2
1.5
Torque (N.m)
2
1.5
0.5
0
0
0.02
0.04
0.06
0.08
0.1
0.12
0.1
0.12
-0.5
Torque (N.m)
-1
1
-1.5
Time (second)
0.5
Fig. 7. Transient torque in free acceleration
0
0
0.01
0.02
0.03
0.04
0.05
0.06
-0.5
4500
Time (Second)
4000
3500
Speed (rpm)
Fig. 4. Transient torque in free acceleration
3000
2500
2000
1500
1000
3.49E+03
500
2.99E+03
0
0
0.02
0.04
Speed (rpm)
2.49E+03
0.06
0.08
Time (Second)
1.99E+03
Fig. 8. Transient speed in free acceleration
1.49E+03
9.90E+02
1.5
4.90E+02
1.00E-02
2.00E-02
3.00E-02
4.00E-02
5.00E-02
Currents (Ampere)
1
-1.00E+01
0.00E+00
6.00E-02
Time (Second)
Fig. 5. Transient speed in free acceleration
0.5
Phase A
0
Phase B
0
0.02
0.04
0.06
0.08
0.1
0.12
Phase C
-0.5
-1
As shown in Fig. 5, reaching to slips under 5% is

impossible even at no load.
No load currents are shown in Fig. 6.
-1.5
Time (Second)
Fig. 9. Transient currents in free acceleration

2
1.5
Currents (Ampere)
1
0.5
Phase A
0
-0.5
Phase B
0
0.01
0.02
0.03
0.04
0.05
0.06
Phase C
-1
-1.5
-2
Time (Second)
Fig. 6. Transient currents in free acceleration
For comparison with a traditional induction motor, a

traditional induction motor with similar characteristics
to CSIM is designed and simulated.
Fig.7, Fig.8 and Fig.9 show transient torque, speed and
currents in free acceleration respectively.
Fig. 10 shows flux lines in the traditional induction
motor at t=0.05 second.
Fig.4 depicts the pulsating torque in CSIM has lower
amplitude in comparison with traditional induction
motor (Fig.7).
Fig.8 illustrates that the traditional motor speed has
larger amplitude vibration with respect to CSIM speed
(Fig.5)
It is clear from Fig.6 and Fig.9 that the CSIM no load
currents are greater than traditional no load currents.
Fig. 10. Flux lines at t=0.05 second
From transient analysis results one can conclude:

In small scale and few pole numbers CSIM has
relatively better performance in comparison with
traditional induction motor. Moreover in small scale
manufacturing process, a CSIM has lower cost and
easier process than the traditional induction motor.
253
5. Three Dimensional Finite Element Transient

Simulation
3500
3000
Speed (rpm)
2500
In this section for precision study of the CSIM three

dimensional finite element transient analysis is
presented.
As mentioned before, in each step of transient analysis,
eddy currents induced on rotor conducting sleeve should
be calculated first. In three dimensions, this takes longer
time and needs high memory.
Fig.11 illustrates eddy currents induced on rotor
conducting sleeve at t=0.05 second.
2000
1500
1000
500
0
0
0.01
0.02
0.03
0.04
0.05
0.06
Time (Second)
Fig. 13. Transient no load speed curve results from 3D finite

element analysis
1.5
Currents (Ampere)
1
0.5
Phase A
Phase B
0
0
0.01
0.02
0.03
0.04
0.05
0.06
Phase B
-0.5
-1
-1.5
Time (Second)
Fig. 14. Transient no load currents
As shown in Fig.14, no load current is slightly less than

2D result (Fig.6).
Torque versus speed curve in 2D and 3D simulations
are shown together in Fig. 15. As expected, the peak of
transient torque from 3D finite element analysis is lesser
than 2D result.
Fig. 11. Induced currents in three dimensional model of

the rotor conducting sleeve at t=0.05 second
2.2
Fig. 12 and Fig.13 depict transient torque and speed in

free acceleration, respectively.
1.7
Torque (N.m)
1.2
2.00E+00
1.50E+00
2D
0.7
3D
0.2
Torque (N.m)
1.00E+00
0.00E+00
-0.3
5.00E+02
1.00E+03
1.50E+03
2.00E+03
2.50E+03
3.00E+03
5.00E-01
-0.8
Speed (rom)
0.00E+00
0
0.01
0.02
0.03
0.04
0.05
0.06
-5.00E-01
Fig. 15. Torque versus speed curve from 2D and 3D transient

finite element simulations
-1.00E+00
Time (Second)
6. Conclusions
Fig. 12. Transient no load torque curve results from 3D finite

element analysis
In this paper, transient simulation of a small scale

conducting sleeve induction motor is presented.
Simulations are done with finite element method both in
2D and 3D. Results from 2D and 3D simulations are in
good agreement with each other. This ensures the
results validity.
Transient finite element simulation of a traditional
induction motor is presented, too. Results show that in
small scale and low pole number motors, CSIM has
relatively better performance in comparison with
traditional induction motor. Moreover in small scale
manufacturing process, the CSIM has lower cost and
easier to process than the traditional induction motor.
No load currents are shown in Fig. 14.

Three dimensional results are in good agreement with
two dimensional results.
Fig. 12 shows more oscillation in comparison with
Fig.4, because in 3D finite element analysis the real
model of the motor is used while in 2D the length of the
motor is assumed too long.
254
7. References
5. Huppunen J. High Speed Solid Rotor Induction

MachineElectromagnetic Calculation and Design,
Thesis for the degree of Doctor of Science, 2004,
Lappeenranta University of Technology, Finland.
6. Jagiela M., Bumby J. R. and Spooner E. Modeling of
High Speed, Solid Rotor Induction Machines Using
Adaptive Finite Elements Procedures. Proceeding of
International Conference "Electrical Machines
(ICEM)- 2004" , Lodz, Poland, 2004.
7. Davey K.R. Analytic Analysis of Single and Three
Phase Induction Motors. IEEE Transaction on
Magntics, 1998, Vol. 34, No. 5, pp. 3721-3727.
8. Bianchi N. Electrical Machine Analysis Using Finite
Element. CRC Press, 2005, pp. 3-24.
1. Cathey J. J. Helical Motion Induction Motor.

Proceedings of IEE, Mar. 1985, vol. 132, pt. B, no.
2, pp. 112-114.
2. Onuki T., Jeon W. J. and Tanabiki M. Induction
Motor with Helical Motion by Phase Control. IEEE
Transaction on Magnetics, September 1997, Vol. 33,
No. 5, pp.4218-4220.
3. Alwash J. H. H., Mohssen A. D. and Abdi A. S.
Helical Motion Tubular Induction Motor, IEEE
Transaction on Energy Conversion, September
2003, Vol. 18, No. 3, pp. 362-369.
4. Becherini G. and Tellini B. Helicoidal
Electromagnetic Field for Coilgun Armature
Stabilization, IEEE Transaction on Magnetics,
January 2003, Vol. 39, No. 1, pp. 108-111.
255

GRAPHICAL METHOD OF ESTIMATING OPTIMUM VALUES FOR THE MAIN

DIMENSIONS OF A D.C. MACHINE
C. EASWARLAL*, V. PALANISAMY**, M.Y. SANAVULLAH***
*Sona College of Technology, India,
** Principal Info Institute of Engineering, India
*** K.S.R. College of Technology, India
The design is subordinated to the question of economic

manufacture. The major considerations to evolve a
good design are:
Abstract: Design is based on the assumption of values

for the variables. However, the machine is expected to
have minimum cost, weight, volume and losses. The
above factors are conflicting with one another.
However, all the factors are dependent upon the
machine dimensions. In practice, only one of the above
factors is taken as objective and optimum design is
arrived at. But in this paper, the variables are
categorized carefully as free and fixed variables. The
values for the fixed variables are selected from the
range of values available. The value for the free
variables is estimated under specified constraint,
resulting in optimum values for the free variables. The
main dimensions of a DC machine are treated as free
design variables. The objective is suitably stated along
with the constraint. Thus the design problem is changed
to an optimization problem. Graphical method is used
for obtaining optimum values for the main dimensions
besides care is bestowed to obtain rated power as
output. A design problem is taken for illustration and
obtained results both by conventional and graphical
method are compared. The optimality of the graphical
method is verified by popularly available optimizing
method.
I.
II.
III.
Cost
Durability
Compliance with performance criteria as laid
down in specifications.
In most of the situations, it becomes difficult to design a

machine which meets all the performance indices and
also satisfies the cost and durability criteria because
these requirements are usually conflicting.
It is
impossible to design a machine which is cheap and is
also durable at the same time. This is because a
machine which is to have a long life span must use high
quality materials and advanced manufacturing
techniques which obviously make it costly. The
performance indices have to be met for certain.
However, a compromise between cost and durability
can be had. In order to achieve optimality of an
objective, the values that are used for the variables
should be optimum besides assuming which of the
variables are free and fixed variables suitably.
The main dimensions of a d.c.machine can be expressed
in terms of the rated power output, desired magnetic and
electric loadings and speed of the machine in the form
of an equation, which is known as output equation of
the machine. The choice of specific magnetic and
electric loadings is made based on performance factors,
such as:
I.
Maximum flux density in iron parts of the
machine
II.
Magnetizing current
III.
Core losses
for specific magnetic loading and
I.
Permissible temperature rise
II.
Voltage
III.
Size of machine
IV.
Current density
Keywords: power, armature diameter, armature core

length, optimum values, graphical method.
1. Introduction
Design may be defined as a creative physical realization
of theoretical concepts [1]. Engineering design is
application of science, technology and invention to
produce machines to perform specified task with
optimum economy and efficiency. Engineering is the
economical application of scientific principles to
practical design problems. If the items of cost and
durability are omitted from a problem, the results
obtained have no engineering value. The problem of
design and manufacture of electric machinery is to
build, as economically as possible, a machine which
fulfills a certain set of specifications and guarantees.
for specific electric loading.
256
It is clear that these specific loadings are design

variable. Similarly the main dimensions are also design
variables. Speed of the machine may be treated as fixed
variable or constant. Hence the power output for a
constant or fixed speed machine can be expressed in
terms of the above four variables. The general
procedure in design is to assume some variables as
known and fixed and some variables as independent or
free variables and evolve. Since the specific loadings
are dependent on the main dimensions, the specific
loadings are treated as fixed variables and the main
dimensions are treated as independent or free variables.
To find optimum values for the free design variables, an
objective function and constraint, if any is required in
terms of free design variables. The objective must be
such that the machine dimensions are minimum or
optimum.
[2] P.RUpadhyay and K.R.Rajagopal
considered efficiency as the objective function whereas
temperature rise and motor weight are the constraints and
the slot electric loading, magnet-fraction, slot-fraction,
air gap and air gap flux density are the design variables.
[3] Messine, F (et.al) showed the advantage of a
deterministic global-optimization method in the optimal
design of electromechanical actuators. The numerical
methods classically used are founded either on nonlinear
programming techniques (i.e., augmented Lagrangian,
sequential quadratic programming) or on stochastic
approaches which are more satisfactorily adopted to
global optimum research (i.e., genetic algorithm,
simulated annealing). However, the later methods only
guarantee reaching this global optimum with some
probability. In this paper, the non-linear design variables
are limited to two in number. Further the power output is
taken as constraint. The constraint equation is modified
in terms of the same two design variables, which helps
to use [4], [5], [6] a simple graphical method for getting
optimum values for the design variables. The values are
obtained simultaneously, which is a special feature of
this method and satisfy the objective.
3. Description of design problem

[1] To find the optimum values for the main dimensions
of a 5 kW, 250 V, 4 pole, 1500 rpm simplex lap wound
D.C. shunt generator. The loadings are: Average flux
density in the gap = 0.42 Wb/m2 and ampere conductors
per metre = 15,000 A. Assume full load efficiency =
0.87.
4. Solution by conventional method
Design makes use of assumptions, empirical relations
derived out of experience, approximations, etc.,
Neglecting friction, windage and iron losses, the power
developed by the armature in general can be written as
Pa = P/ = 5.75 kW.
Speed, n = 25 r.p.s.
Output equation,
Pa = 2Bav ac D2 Ln x 10-3 .
The output coefficient,
Co = 2Bav ac x 10-3 .
(2)
On substitution, Co 62.1 .
From equation (1),
D2 L = Pa /Co n .
Substituting the values,
D2 L = 0.0037037 m3
(3)
D2 L 0.0037 m3 .
(4)
or
The value as a product in terms of D2 and L is obtained

for D and L. In order to separate D and L, one of the
following techniques depending upon the requirement is
adopted.
2. Symbols
Pa =
D =
L =
n =
Bav=
ac =
P =
=
=
=
Co =
b =
bp =
=
Lf =
X =
p =
(1)
Power developed by armature, kW

Armature diameter, m
Armature core length, m
Speed in rps
Specific magnetic loading, Wb/m2
Specific electric loading, A
Rated power output, kW
Efficiency
Ratio, pole arc to pole pitch
Pole pitch = D/p
Output coefficient
Pole arc
Width of pole body
Lagrange multiplier
Lagrangian function
A constant
Number of poles
i.
ii.
iii.
Square pole section (L = bp)

Length equaling twice the width of
pole body (L=2 bp)
Square pole face (L = b)
The above indicates only pole proportion of the

machine. Let us take the case of square pole face such
that the armature core length equals pole arc.
For a square pole face, core length/pole arc = L/b = 1
or L/ = 1. Taking, = 0.64, L = x D / p =
0.50265D, i.e.,
L 0.503D.
(5)
Substituting eqn. (5) in eqn. (4) and solving,

D 0.195 m
and L 0.1 m .
Thus the values for the main dimensions are found
analytically one after another. Further the values for the
main dimensions would be different for different
257
The armature is viewed in terms of its main dimensions

namely D and L. The objective is to minimize the sum
of linear distances of armature and hence the objective
function equation is written as
requirements. Unique and optimum values can not be

expected.
5. Optimum value by graphical method
[4] Optimization is a process of finding the conditions
that give the maximum or minimum value of a function.
For any optimization problem, an objective is
compulsorily required along with constraint, if any is to
be specified. Here the conditions refer to the values for
the free design variables, namely D and L such that the
objective function is made maximum or minimum. The
function refers only to
D +L = X;
where X is a constant, which must be minimum.
The design problem is now stated as to find the values

for D and L such that total linear distance in the
machine is optimum (i.e. minimum) and at the same
time delivering the rated power. Thus the output
equation which governs the power output is the
constraint equation, and it is expressed in terms of D
and L. The corresponding equations for the design
problem are equation (7) and equation (4) which are
rewritten as
and
D +L = X
(objective function)
D2 L = 0.0037 m3 (constraint equation).
The above equations contain only two variables. The

constraint is an equality constraint type. Since there are
two variables, a simple graphical method is used.
The graph for constraint equation is drawn first.
The
values for D are obtained for different assumed values
of L and a graph between D and L is drawn. The shape
of the graph is obviously non-linear and takes the form
of a parabola as shown in Fig.2. The readings for the
graph are as shown in Table 1.
To draw the objective function graph, a constant value
for X is assumed. Then for different assumed values for
L, the corresponding values for D are obtained. It is a
straight line graph.
This process is repeated for
different X. The points for objective function graph for
different value of X are given in Table 2, Table 3
and Table 4 as examples. The objective function is
thus represented as a series of straight line graphs as
shown in Fig. 2. It may be noted or seen that some
graphs may be above the constraint graph and some
may be below it. But at the same time, one straight line
graph may become tangent to the constraint graph.
Fig. 1. Armature of a D.C. Machine

the objective function. The aim now is to bring in an
objective function. The armature of a d.c.machine
labeling D and L is as shown in Fig.1. Whenever we
refer armature diameter, D it reminds only a circle,
whereas armature core length, L may be thought of a
linear distance. The circle (i.e. armature) diameter is
also considered as a linear distance with a linear
distance of D. Hence the objective function may be
written in terms of D and L, which are variables, as
f(D,L).
(7)
(6)
Table 1. Readings for L and D (constraint equation)
L,
m
0.01
0.02
0.05
0.07
0.09
0.1
0.12
0.14
0.16
0.2
0.25
0.3
D2,
m2
0.37
0.185
0.074
0.053
0.041
0.037
0.0308
0.026
0.023
0.0185
0.0148
0.0123
D,
m
0.608
0.43
0.272
0.23
0.202
0.192
0.1756
0.162
0.152
0.136
0.121
0.111
258
Table 2. D + L =X (= 0.2)
D,m
0.2
L,m
0.2
Table 3. D + L = X (= 0.3)
D,m
0.3
L,m
0.3
D,m
0.4
L,m
0.4
Table 4. D + L = X (= 0.4)
Fig. 2. Graphical Method
The constraint graph and the objective function graphs

are shown in Fig.2. One objective function graph
become tangent to the constraint graph. From the
tangent point, the values for D and L are obtained as
D = 0.192 m ;
L = 0.1 m .
which are optimum values for D and L, irrespective of
the requirement of pole proportions.
Table 5. Comparison of values
Design
variables
D, m
L, m
(D+L), m
The objective function is D + L = X

and constraint equation is D2 L 0.0037 = 0 .
The Lagrangian function Lf can be written as
Lf (D,L,) = D + L + (D2 L 0.0037).
The necessary conditions for optimum are
L f
L
L f
=0
=0
Methods
Graphical
0.192
0.100
0.292
Lagrangian
0.1948
0.0974
0.2922
It is seen that the sum of armature diameter and core

length is less in graphical method than that obtained by
conventional method which indicate that the values for
D and L are optimum. It is verified with the help of
Lagrangian method.
6. Optimality by lagrangian method
L f
Conventional
0.195
0.100
0.295
8. Conclusions
A d.c.machine design problem is solved by
conventional method, which employs assumptions,
approximations, etc., Further it requires an equation in
terms of pole proportions for the separation of D and L,
which results in different values for D and L depending
upon the separating condition equation. None of these
values would be optimum. Hence the conventional
design problem is converted as an optimization
problem.
The procedure for estimating optimum
values for the main dimensions are explained with the
help of graphical method. The values are verified with
the values obtained by popularly available optimizing
method. The values are agreeable. The speciality of
this method is that optimum values for both the free
design variables are obtained simultaneously. Such
method can be used in intermediate steps in the process
of design of an electric machine, provided the constraint
equation and objective function contain only two same
variables. The example may be the estimation of
window dimensions of a transformer. The performance
of the d.c. machine would improve because the
optimum values are estimated at the first step itself
which will be used in further calculations. Hence such
graphical method may be adopted for obtaining
optimum values for the two variables at the time of
design of an electric machine at appropriate places.
(8)
(9)
(10)
= 0
On solving, the optimum values are

D = 0.1948m
and L = 0.0974m .
7. Comparison of results
The values for armature diameter and core length
obtained by all methods are given in Table 5 for
comparison.
259
V. Palanisamy received the B.E.

degree
in
Electronics
and
Communication Engineering, from
PSG College of Technology,
Coimbatore in 1972, the M.E.
degree in Communication systems
in 1974 from University of Madras
and the Ph.D. in Antenna Theory in
1987 from IIT Karagpur, India.
Since 1974 he has been working in various capacities of
Technical Education in Tamilnadu, and he is retired
Principal of Government College of Technology,
Coimbatore, India. Currently he is holding the post of
Principal at Info Institute of Engineering, Coimbatore.
His interest includes Neural Network, Fuzzy Logic and
Optimization.
9. References
1.
2.
3.
4.
5.
6.
Sawhney A. K. A course in Electrical machine

Design. Dhanpat Rai & Co (p) Ltd, Delhi-110006,
2001.
Upadhyay P. R., Rajagopal K. R. Genetic algorithm
based design optimization of a permanent magnet
brushless dc motor. J.Appl.phys.97, 10Q516(2005);
D.O.I : 10.1063/1.1860891.
Messine F., Nogarrede B., Lagowanelle J.L.
Optimal design of electromechanical actuators: a
new method based on global optimization. IEEE
Transactions on Magnetics, Vol.34, Issue 1, Jan
1998, PP.299-308, D.O.I:10.1109/20.650361.
Singirasu S. Rao. Engineering optimization theory
and practice. 3rd Ed., New Age international (p) Lt,
New Delhi-110002, 1999, pp.11-15.
Easwarlal C., Palanisamy V., Sanavullah M.Y.,
Gopila, M. Graphical Estimation of optimum
weights of iron and copper of a transformer.
ISBN:0-7803-9772-X/06/$20.002006 IEEE.
Easwarlal C., Palanisamy V., Sanavullah M.Y.
Optimim full load losses of a Transformer by
graphical method, International Journal of
Electrical and Power Engineering. (3):359-362,
2007, Medwell journals, 2007.
M. Y. Sanavullah was born in

1942. He received the B.E
degree in Electrical & Electronics
Engineering and M.Sc degree in
Power System Engineering in the
years 1965 and 1967 from the
University of Madras, Ph.D
degree
in
High
Voltage
Engineering in 1987 from Anna University, Chennai.
He guided a candidate at Ph.D level in Periyar
University, Salem during 2001 2005. At present he is
guiding four candidates leading to Ph.D in Vinayaka
Missions University, Salem. He has 38 Years of
teaching experience. He is currently a Professor and
Dean with K.S Rangasamy College of Technology,
Tiruchengode 637 209, India. He is interested in finite
Element Method.
10. Biographies
C. Easwarlal was born in 1947.
He studied his B.E (Electrical and
Electronics Engineering) degree
course and
M.Sc (Engg)
(Electrical Machines) degree course
at PSG College of Technology,
Coimbatore - 641 004 and received
the degrees in the years 1971 and
1973 from University of Madras. He has worked as an
Electrical Engineer in a company for two years. Then he
took up the teaching assignment. He has put in 28
Years of experience. Currently he is working towards
his Ph.D degree in Anna University, Chennai, India. He
is a professor of Electrical and Electronics Engineering,
Sona College of Technology, Salem 636 005, India.
His area of interests is design of Electrical Machines,
optimization in design, power system analysis.
260

THE RESEARCH OF CAUSES AND CONSEQUENCES OF PARAMETERS

CHANGES OF GROUNDING ELECTRICAL INSTALLATIONS
Marina KIZHLO, Arvids KANBERGS
Riga Technical University, Institute of Power Engineering, Latvia
Abstract: This paper introduces the resistance of the

ground (soil resistivity) and measurements of the soil
resistivity. It is well known that the resistance of the
ground electrode is heavily influenced by the soil
resistivity in which it is driven and as such,
measurements of the soil resistivity are an important
parameter when designing grounding installations.
Knowledge of the soil resistivity at the intended site,
and how this varies with parameters such as moisture
content, temperature and depth, provides a valuable
insight into how the desired ground resistance value can
be achieved and maintained over the life of the
installation with the minimum cost and effort.
identify the most appropriate ground electrode design.

However, even on the area, selected for the construction
of the grounding system, the ground vary most
frequently in terms of significant heterogeneity in the
vertical and horizontal sectional views; therefore the
soil resistivity, which satisfies the high accuracy of
calculation can be obtained only by direct
measurements. In cases when the high accuracy of
calculation is not required, it is possible to use tabular
values of the soil resistivity of the ground. It is known
that all substances according to their electrically
conducting properties are divided into:
5
conductors 10 m ;
8
insulators 10 m ;
8
5
semiconductors 10 m 10 m .
The rocks, different structures of the ground and natural
aqueous media, as a rule, relate to the semiconductors.
Keywords: resistance of the ground, soil resistivity,

grounding system, measurements of the soil resistivity.
1. Introduction
The electrical characteristics of the ground have an
effect on the resistance of the grounding system, and
therefore to the electrical safety of the personnel, which
operates electrical devices. One of the main objectives
of grounding electrical systems is to establish a common
reference potential for the power supply system,
building structure, plant steelwork, electrical conduits,
cable ladders and trays and the instrumentation system.
To achieve this objective, a suitable low resistance
connection to the ground is desirable. However, this is
often difficult to achieve and depends on a number of
factors:
Soil resistivity;
Stratification;
Size and type of electrode used;
Depth to which the electrode is buried;
Moisture and chemical content of the soil.
The conductivity of surface layer substantially
influences the value of the touch voltage and step
voltage. The resistivity of the ground changes in the
very wide range depending on the geological structure
of the ground. By changing the spacing of the grounding
electrodes, it is possible to develop a profile of the soil
resistivity at various depths, which can be used to
2. Soil resistivity
The property of resistivity can be defined for any
material. As applied to soil, resistivity is an indication
of a given soils ability to carry electric current. The
flow of electricity in the soil is largely electrolytic,
determined by the transport of ions dissolved in
moisture. The soil resistivity is the resistance measured
between two opposing surfaces of a 1m3 cube of
homogeneous soil material, usually measured in m.
The soil resistivity has a direct effect on the resistance
of the grounding system. The typical features of soils
with certain resistivity values are shown in Table 1.
These changes are caused by the influence of
admixtures and by different structure of the mineral
grains, on which the measurements were conducted.
Microscopic cracks and oxidations of surface within the
limits of individual grains produce significant changes
in the values of the measured resistances. The values of
the resistivity of the ground water are given in Table 2.
261
Table 1. Classes of soil and their resistivity

Classification
Low
resistivity
zone
Medium
resistivity
zone
High
resistivity
zone
Resistivity
(m)
100
1001000
1000
interpreted to yield an equivalent model for the

electrical performance of the ground. However, the
results may be incorrect or misleading if adequate
investigation is not made prior to the test or the test is
not correctly undertaken. To overcome these problems,
the following data gathering and testing guidelines are
suggested.
Features
Lowland at the mouths of
rivers or by the sea. Usually
abundant in water.
Midland plains where
ground water is not so
difficult to obtain.
Hilly zones, piedmont
districts and high lands,
where drainage is good.
Table 4. Classes of soil and their resistivity

Class of soil
Paddy of clay and swamps
Farmland of clay
Seaside sandy soil
Paddy or farmland with gravel stratum
Mountains
Gravel, pebble seashore or parched
river bed
Rocky mountains
Sandstone or rocky zone
The resistivity of rocks depends also on temperature.

For the water-containing rocks the temperature effect on
the resistance of rocks is the same as the temperature
effect on the electrical resistance of the located in the
rocks water and the temperature range between the
points of its freezing and boiling.
Table 2. Resistivity of water (approximate values)
Type of water
Pure water
Distilled water
Rain water
Tap water
Well water
Mixture of river and sea water
Sea water (inshore)
Sea water (ocean 3%)
Sea water (ocean 5%)
Resistivity (m)
200000
50000
200
70
20 ~ 70
2
0.3
0.2 ~ 0.25
0.15
20C
10C
0C
0C (ice)
-5C
-15C
2000 ~ 5000
104 ~ 107
4. Method of trial vertical electrode

The method of the trial vertical electrode can be used in
the electrical devices up to 20 kV, if the ground in the
region of electrical device is characterized by a
relatively low specific resistance. As a rule, in such
electrical devices the grounding system is calculated by
the permissible resistance, moreover significant in the
value standardized permissible resistances can be
obtained with a comparatively low accuracy of the
determination of the soil resistivity of the ground. The
method is based on the measurement of the resistance of
the trial electrode, whose geometric dimensions
correspond to those accepted for designing the
grounding system, immersed at the calculated depth at
different points of the selected area. Resistivity is
determined on the basis of the formulas, which
characterize the resistance of the grounding conductor
of the corresponding form. For example, for the pivotal
trial grounding conductor specific resistance is
calculated by the formula (1):
Table 3. Soil temperature vs resistivity

The ground
resistivity (m)
72
99
130
300
790
3300
1000 ~ 5000
Data related to nearby metallic structures, as well as the

geological, geographical and meteorological nature of
the area is very useful. For instance the geological data
regarding strata types and thicknesses will give an
indication of the water retention properties of the upper
layers and also the variation in resistivity to be expected
due to water content. By comparing recent rainfall data,
against the seasonal average, maxima and minima for
the area it may be ascertained whether the results are
realistic or not. For measuring the parameters of the
electrical structure of the ground - the thickness of the
layer and of resistivity of each layer the two methods
are recommended: the method of the trial vertical
electrode and the method of the vertical electrical
sounding. The selection of the method of measurement
depends on the characteristic of the ground and
necessary accuracy of measurement. The following
sections outline the major practical aspects of the
measurement procedure and result interpretation.
Table 3 demonstrates evidently a sharp increase of the

resistivity of the rocks with the freezing of pore
moisture. The resistivity of soil varies with changes in
temperature and the rate of its increase as temperature
declines. The moisture content of the soil in Table 3 is
15.2% of its total weight. The resistivity of the soil at 15C is 45.9 times higher than the resistivity of the
same soil at +20C. The resistivity of soil is, in fact,
influenced by many factors and it fluctuates constantly.
It is lower in summer and higher in winter. Table 4
gives the recommended values of the resistivity of the
upper layer of the ground with a thickness of up to 50 m
for the use in the design calculations.
Temperature
Resistivity (m)
10 ~ 150
10 ~ 200
50 ~ 100
100 ~ 1000
200 ~ 2000
Rate
1.0
1.4
1.8
4.2
11.0
45.9
3. Measurements of the soil resistivity

When designing the grounding system to meet safety
and reliability criteria, an accurate resistivity model of
the soil is required. The purpose of the soil resistivity
testing is to obtain a set of measurements which may be
262
Rl
0,366 lg
2l
d
Table 5. Climatic zones
(1)
Characteristic of
the climatic zone
I
Average longstanding lowest
-20 -15
temperature of air
(January), C
Average longstanding highest
16 18
temperature of air
(July), C
Average annual
amount of
40
precipitation, cm
Duration of the
freezing of water, 190 170
days
where R resistance of the grounding conductor in the

form of
electrode, obtained by the
measurement, [];
l depth of the laying of trial electrode, [m];
d the diameter of trial electrode, [m].
Obtained result of measurement of the soil resistivity
represents a certain averaged value (2), which calculates
the heterogeneity of the ground on the basis of vertical
line. In order to consider the uniformity of the ground
along horizontal, one should conduct several
measurements at different points of the selected area
and for the calculations use arithmetic mean of the
results of measurements:
1 2 ... n
-14 -10
-10 0 0 +5
18 22
22 24 24 - 26
50
50
30 - 50
150
100
Table 6. Seasonal coefficients for the design conditions of the

soil resistivity for the calculation of the resistance of trial
vertical grounding conductors
(2)
Value of coefficient k with the submersion depth of

electrode
23m
45m
For the moisture of soil or amount of precipitation in the
period before the measurements
Increased Average Lowered Increased Average Lowered
I
1,85
1,65
1,50
1,45
1,35
1,30
II
1,60
1,45
1,30
1,30
1,25
1,20
III
1,45
1,30
1,20
1,25
1,15
1,10
IV
1,30
1,10
1,0
1,20
1,10
1,0
where R soil resistivity, obtained as a result of

measurement at different points of the area,
[m];
n number of measurements.
A change of the soil resistivity along the horizontal is
not considerable with the small sizes of areas and it is
possible to limit to conducting one measurement.
Climatic conditions in the region of the construction of
the grounding system cause, as a rule, the essential
fluctuations of the soil resistivity of the ground in the
course of the year as a result of freezing, moistening and
drying soil; therefore it is necessary to carry out the
measured specific resistance (m) to the design
conditions by means of the seasonal coefficient (3):
Table 7. Seasonal coefficients for the design conditions of the

soil resistivity for the calculation of the resistance of trial
horizontal grounding conductors
Value of coefficient k with the length of grounding
conductor
23m
45m
For the moisture of soil or amount of precipitation in the
period before the measurements
Increased Average Lowered Increased Average Lowered
I
5,4
3,2
2,3
4,3
2,6
2,0
II
3,8
2,2
1,7
3,2
2,0
1,5
III
3,1
1,7
1,4
2,6
1,5
1,3
IV
2,0
1,3
1,1
1,8
1,2
1,0
Climatic zones
eq avg k ,
IV
Climatic zones
avg
Climatic zones
II
III
(3)
where eq calculated equivalent soil resistivity, (it

corresponds to the most unfavourable climatic
condition of the ground), [m];
k seasonal coefficient, which considers an
increase the resistance of the ground.
Seasonal changes in the soil resistivity depend on
climatic conditions in the region of the grounding
system and depth of the ground layer for accounting for
seasonal changes, the entire territory of Latvia is
divided into four climatic zones (Table 5). Seasonal
coefficients are selected depending on climatic
conditions in the period, which precedes measurements.
In this case accept different coefficients for vertical
(Table 6) and horizontal grounding conductors (Table
7). Different influence of climatic conditions on the
surface and deep layers of the ground thus is considered.
The method of the trial vertical electrode is
characterized by simplicity of measurements, and it also
requires the use of a minimum quantity of minimum
number of measuring equipment and calculations.
Its basic deficiency is complexity of sinking the

measuring electrodes at the calculated depth, especially
under the conditions of the solid ground. However, the
method is not practically acceptable for the
measurements on areas with the high specific resistance
of the surface layers of the ground (rock, permafrost and
so forth). The need for immersing grounding conductors
at the significant depth under such conditions can arise
and, to determine the calculated specific resistance it is
necessary to use the method of vertical electrical
sounding.
5. Method of vertical electrical sounding (VES)
For the measurements four electrodes (Fig. 1),
immersed at the small depth 40-50 mm are used.
Measuring the strength of the current I , that flows from
the source through the current electrodes 1, 4, and the
263
voltage between the potential electrodes 2, 3, determine

the resistance (4)
R U /I ,
(4)
which is equal to the resistance of a certain volume of

the ground, which adjoins the measuring unit.
Conditionally this volume is taken as the equal to the
hemisphere with the diameter, determined by the
distance between the current electrodes. The measured
soil resistivity carries to the point of the ground on the
vertical axis of the hemisphere at the depth, equal to
fourth of its diameter. Consequently, for measuring the
soil resistivity at the depth h it is necessary that the
distance between the current electrodes would compose
l1 4 4 h . In this case if the potential electrodes divide
the distance between the current electrodes into equal
parts l 23 1 / 3l1 4 , the soil resistivity (m) can be
Fig. 1. Method of vertical electrical sounding
However, the obtained equivalent soil resistivity of twolayered structure of the ground cannot be used directly
for calculating the resistance of the grounding system.
In fact, the grounding system consists of different type
of the grounding conductors, which occupy different
position relative to the equivalent layers of the ground
of the calculated two-layered structure of the ground.
Therefore for the basic types of grounding conductors
and grounding system tables and drawings have been
developed. That makes it possible to calculate the
equivalent soil resistivity respecting the soil resistivity
of two-layered structure of the ground and relative
placement of the grounding conductors in it.
calculated according to the formula (5)
VES 2Ra ,
where
a l1 2
l 2 3
(5)
l3 4 distance between the
6. Conclusions
adjacent electrodes, [m];

R measured resistance, [].
Changing the distance between the current electrodes,
the conditional soil resistivity at the different depth
The soil resistivity data is the key factor in designing a

grounding system for a specific performance objective.
All soil conducts electrical current, with some soils having
good electrical conductivity while the majority has poor
electrical conductivity. The soil resistivity varies widely
throughout the whole territory of Latvia and changes
dramatically within small areas. Factors that affect the soil
resistivity may be summarised as:
Type of the ground.
Stratification; layers of different types of soil.
Moisture content; resistivity may fall rapidly as
the moisture content is increased.
Temperature; the ground resistivity is lower in
summer and higher in winter.
Chemical composition.
Topography; rugged topography has a similar
effect on the resistivity measurement as the local
surface resistivity variation caused by weathering
and moisture.
Factors such as maximum probe depths, lengths of cables
required, efficiency of the measuring technique, cost
(determined by the time and the size of the survey crew)
and ease of interpretation of the data need to be
considered, when selecting the test type.
h for the different layers of the ground is measured. If

the distance between the potential electrodes composes
l 23 1 / 3l1 4 , it is possible to transpose the current
electrodes only, leaving the potential electrodes
constant. In this case the value of the soil resistivity is
calculated from the formula (6)
VES
R(
l12 4 l 223
2l 223
).
(6)
For obtaining the minimum error it is not allowed to

take a too small distance between the potential
electrodes.
It
should
be
assumed
l 23 (0, 41 0, 43)l1 4 . The measured soil resistivity
by the method of vertical electrical sounding is a
conditional value. For bringing it into the calculations,
use the special graphs - theoretical templates for the
multilayer ground, used in geophysics. In this case it is
necessary to attempt to reduce the results of vertical
electrical sounding to a smaller number of layers. If
bringing the result to the design conditions, the structure
of the ground has several layers, accept it as the twolayered ground. The tentative depth of the laying of the
horizontal and vertical conductors of the grounding
system is selected for this; moreover, it is necessary to
select the length of vertical electrodes proportional to
the layer depth of the ground with the high
conductivity. As the first equivalent layer, accept the
layer, which penetrates vertical electrodes. All the
remaining layers of the ground are taken as the second
equivalent layer.
7. References
1.
2.
3.
264
Kilo M., Kanbergs A., The special features of the

grounding equipment, master paper, Riga, RTU,
2008. 120 p.
.., . .
.:
2005.
IEEE Std 80-200 (Revision of IEEE Std-1986),
LVS valid from 12.07.2004.
4.
5.
6.
7.
8.
Sverko R. E., Ground measuring techniques:

electrode resistance to remote earth and soil
resistivity, ERICO, Inc. Facility Electrical
Protection, U.S.A.
Armstrong, H. R., Grounding electrode potential
gradients from model tests, IEEE Transactions on
Power Apparatus and Systems, pp.618-623, 1960.
Switzer, W. K., Practical guide to electrical
grounding, ERICO
Soil resistivity testing. Four point wenner method,
LEP-1001, Lyncole XIT Grounding
Internet sources: www.building.lv, www.likumi.lv,
www.vdi.gov.lv.
265

A SIMPLE TRANSIENT-BASED WIDE-AREA SELECTIVITY TECHNIQUE

FOR EARTH-FAULTS IN ISOLATED AND COMPENSATED NEUTRAL
MEDIUM VOLTAGE NETWORKS
Mohamed F. ABDEL-FATTAH and Matti LEHTONEN
Department of Electrical Engineering, Helsinki University of Technology (TKK), Espoo, Finland
Abstract: This paper presents a simple selectivity

technique for earth-faults in isolated and compensated
neutral medium voltage networks. It is based on
investigations on the flow of the residual currents in the
network during the earth fault. The fault weight is
calculated as the transient RMS residual current ratio
between the feeder/section and the background network.
The fault indicator will be activated for higher fault
weights; around 100%. The faulted feeder can be
detected easily and the faulted section can be selected
accurately. The proposed technique is less dependent on
the fault resistance and the faulted feeder parameters; it
mainly depends on the background network. The MV
network is simulated by ATP/EMTP program.
Extensive simulations that cover different fault
conditions are performed to validate and realize the
technique.
the possible disturbance in the system. It can detect

high-impedance and intermittent arcing faults; leading
to improvement in the reliability and power quality.
Many research works have been done in this area in the
last few years, but the goal is still not fulfilled [2-9].
In this paper, a simple selectivity technique for earth
fault detection is proposed. It is transient current-based
technique. The fault weight factors will be calculated as
the ratio between the transient residual currents. It will
be used for detection of the faulted feeder (first stage)
and selection of the faulted section (second stage). A
wide-area communication link is required for the
coordination between the two stages. A 20 kV, 251 km
medium voltage distribution network is simulated by
ATP/EMTP program. Extensive simulations are
performed to validate and realize the technique, for
different fault conditions; different fault inception
angles, fault locations and high fault resistances.
Keywords: earth faults, transient based protection,

residual current, isolated and compensated neutral
medium voltage distribution networks.
2. Transient RMS Frequency Response

The transient period is the first few milliseconds
directly after fault incidence. The fault detection
window is proposed to be the most sensitive part of the
transient period, in which the polarity comparison
technique is verified. The window width is not constant;
it varies according to the fault characteristics, mainly
depends on the fault resistance and incidence angle.
Ref. [2] proposes a sensitive varying fault detection
window, in which the average values of the transient
voltage and current are calculated. In practical
application, the constant sliding window is preferred,
which makes some limitation to the average calculations
that are highly affected by opposite polarities contained
in the fault detection window. In this paper, it is
proposed to use the transient RMS values which are not
affected by polarity changing, and can be used with
constant sliding fault detection window technique. The
transient RMS value will be calculated in a constant
sliding window. The width of the window will be very
1. Introduction
As a result of climate change, windiness and thunder
storms in particular are considerably increasing. This in
turn leads to a higher fault probability. Together with
the climate change, the modern societys growing
dependence on uninterrupted power supply will greatly
influence the development of distribution networks and
organizations. Therefore, the requirement of reliable
and efficient protection system is increased. Earth faults
detection in isolated and compensated neutral medium
voltage distribution networks presents a challenge to
protection system due to small levels of fault currents
that can not be detected by conventional relays [1].
Although several techniques have been proposed, and
much progress has been made, a reliable and flexible
solution has still not been found. Transient-based
protective schemes utilize the transient component in
the fault signals, which provides fast information about
266
small; about few milliseconds to be able to extract the

transient features from the considered waveform.
The residual value is equal to the summation of the
instantaneous phase values (a, b and c) as given by (1).
The residual currents are very small in normal operation
and become meaningful in fault condition. They are
very sensitive for earth faults and from practical point
of view, it can be measured easily by one sensor for
each feeder; hence it is suggested to be used for fault
detection.
120
Transient RMS Frequency Response (%)
ir (t ) ia (t ) ib (t ) ic (t ) .
Therefore, the response will be investigated only for

frequencies up to 5000 Hz.
(1)
The transient RMS value of the residual current (Ir) will

be calculated using the following equation:
Ir
1
N
100
80
60
40
0.5 ms
1.0 ms
1.5 ms
2.0 ms
2.5 ms
20
500
1000
1500
(i
k 1
r ,k
)2 .
(2)
2000 2500 3000

Frequency (Hz)
3500
4000
4500
5000
Fig. 1. The transient RMS frequency response for different

values of the sliding window width (0.5, 1.0, 1.5, 2.0& 2.5
ms), up to 5 kHz, at sampling frequency of 100 kHz
Where: ir,k is the instantaneous value of the residual

current at sample k, calculated from (1) and N is the
total number of samples in the window. Equations (2) is
calculated in the constant sliding fault detection window
which starts at sample (k=1) and terminates at sample
(k=N). If TS is the sampling time, then the width of the
fault detection window will equal to NTS. The sliding
(moving) window will be updated after collecting the
new samples, after constant time step TS. The smaller
value of the sampling time enables to cover high
frequencies contained in the waveform which increases
the sensitivity of the detection. A sampling frequency of
25 kHz is used.
Ground fault signals consist of different frequency
components, which result from charging or discharging
of the network capacitances. The range of these
frequencies is varying from few hundreds to few kilo
Hertz. According to practical experience, the
frequencies of discharge and charge components vary in
the range 500-2500 Hz and 100-800 Hz respectively.
The amplitude of the discharge component is relatively
small, typically only about 5-10 % of the amplitude of
the charge component [10].
The evaluation of the frequency response of the
transient RMS values should be investigated for
different frequency components contained in the
transient period. This can be achieved by calculating the
transient RMS value for a sinusoidal signal with
different frequencies. After that, investigate if the RMS
value of each frequency can be estimated by the
transient RMS method or not. The transient RMS value
will be estimated in a constant width sliding window
and the effect of the sliding window width will be
determined to propose the suitable value of it.
Fig. 1 and 2, present the transient RMS frequency
response for different values of the sliding window
width (0.5, 1.0, 1.5, 2.0& 2.5 ms), at sampling
frequency of 25 kHz. From preliminary analysis, it is
investigated that the response of the transient RMS is
highly affected by the lower frequencies (less than 2000
Hz) but the error is small for higher frequencies.
120
Transient RMS Frequency Response (%)
F=100 Hz
RMS=100%
100
80
60
40
0.5 ms
1.0 ms
1.5 ms
2.0 ms
2.5 ms
20
50
100
150
200
250
300
Frequency (Hz)
350
400
450
500
Fig. 2. Enlarged view, up to 500 Hz, for the transient RMS

frequency response shown in Fig. 1
The response is evaluated as a percentage ratio between

the estimated RMS value and the actual expected RMS
value. From Fig. 1 and 2, it can be seen that, the lower
frequency components can be estimated by the transient
RMS calculations using a suitable sliding window
width. The performance is highly improved by
increasing the widow width but more calculations are
required considering the same sampling frequency.
A suitable minimum window width of 2.5 ms is
adequate and proposed to cover the minimum transient
frequency of 100 Hz by 100% (as shown in Fig. 2). The
sampling frequency is mainly related to frequency
components contained in the transient period. To cover
different frequency components up to 10 kHz, the
minimum required sampling frequency is 20 kHz.
3. The Simulated Network
A 20 kV, 251 km medium voltage distribution network
is simulated by ATP/EMTP program. A single line
diagram of the simulated network is presented in Fig. 3.
267
The network consists of a 66 kV supply connected to a

66/20 kV substation transformer.
substation transformer. The connected high-voltage and

low-voltage networks do not affect the fault currents.
Healthy Feeders
F5
Background Network
4 Feeders (198 km)
F3
Background
Network
198 km
Ir,back
F4
66/20 kV
/Y
F2
66/20 kV
/Y
Fault
Indicator
CB F1
Communication Link
7 km
7 km
Fault
Point
Ir
4 km 3
5 km
5 km
Wireless Communication
Remote Terminal Unit (RTU)
5 km
5 km
5 km
Ce
Faulted Feeder, 53 km
66 kV
Supply
B 10 km
A
66 kV
Supply
Fault
Point
Fig. 4. The flow of the earth fault residual currents, during a

single line-to-ground fault
Fig. 3. The simulated, 20 kV, 251 km, medium voltage

distribution network
In the first stage, the faulted feeder will be detected. For

each feeder, the transient RMS residual current (Ir) will
be calculated using (2), and the transient RMS
background residual current (Ir,back) which is equal to
the summation of the residual currents in the other
background feeders will be calculated using (3).
The transformer feeds five 20 kV-overhead line feeders.

Each feeder is terminated by a 0.4 kV load through
20/0.4 kV transformer. The network is implemented
using ATP (Alternative Transients Program), version of
EMTP program where the circuit was realized using
ATPDraw [11]. The transmission line frequency
dependent model of EMTP program is intentionally
selected to account for the unsymmetrical faults. The
feeder lines are represented using the frequency
dependent JMarti model. The two cases of isolated and
compensated neutral of the main transformer are
simulated. A sampling frequency of 25 kHz is used. The
ATPDraw circuit and the configuration of the feeders
are given in Fig. 10 and 11 in Appendix.
I r ,back
r
back feeders
(3)
It can be concluded that, for faulty feeder the residual

current should equal to the background residual current
but for healthy feeder they are not equal. The fault
weight factors are proposed to be used as fault
indicators. It will be calculated as the percentage ratio
between the residual current of the feeder and the
residual current of the background network as in (4),
considering availability of all residual currents at the
substation.
4. The Proposed Technique

The proposed technique is based on the investigations
on the flow of the residual currents in the medium
voltage distribution network during the earth fault. The
transient residual currents for each feeder and section
are calculated and used to detect the faulted feeder (first
stage) and select the faulted section (second stage)
without voltage measurements. Considering the
simulated network in Fig. 3, the flow of the earth fault
residual currents, during a single line to ground (SLGF)
fault is presented in Fig. 4.
The earth fault residual current, for the faulted feeder, is
composed of the residual currents flowing through the
earth capacitances of background network, and hence
should be equal to the summation of the residual
currents in all healthy feeders as shown in Fig. 4. The
other impedances of the network components are small
compared to those of the earth capacitance and hence
can be neglected. The earth capacitance of the network
depends on the types and lengths of the lines connected
in the same part of the galvanic connected network. In
practice, for radial medium voltage distribution
networks, it is the area supplied by one HV/MV
Ir
I r ,back
100 .
(4)
For the faulted feeder, the fault weight should be equal

to 100%. For the healthy feeder, the fault weight should
be small. It will depend on the number of feeders in the
considered network and its lengths. Assuming network
of n number of equal feeders, then the fault weight for
healthy feeder will be given by (5).
Whf
1
100 .
2n 3
(5)
Where n is the number of feeders in the assumed equal

feeder networks. Considering the simulated network
shown in Fig. 3, and assuming equal five feeders, then
the fault weight for the healthy feeder should be equal
to 1/7 (i.e. 14.29%). For higher number of feeders in the
networks, the fault weight will be very small which
improves the performance of the proposed technique
and enhances its selectivity. Once the faulted feeder is
268
healthy feeders, the fault weight are less than 20%,

which presents a high selectivity performance of the
proposed technique.
diagnosed, the proposed technique will continue to

select the faulted section in the faulted feeder.
In the second stage, the fault weight factors for all
sections in the faulted feeder will be calculated using
the residual current of the background network for the
faulted feeder as a reference, considering the
availability of suitable wide-area communication link
between the fault indicators for all sections and the
substation, as shown in Fig. 3. The fault weight for the
faulted section should be higher than 100% due to the
extra circulated residual currents through the branched
sections. The faulted section can be selected easily by
following the radial flow of the active fault indicators.
For the healthy section, the fault weight should be very
small. It will depend on the number of sections in the in
the faulted feeder, its lengths and its configuration.
Assuming feeder of m number of radial equal sections,
then the fault weight for healthy section will be given
by (6).
Whs
1
100 .
m (n 1)
Instantaneous Feeder Residual Currents (A)
100
Minimum detection tim e

(after one com plete w indow )
60
40
20
0
-20
-40
-60
(6)
ir(t) (F1)
ir(t) (F2)
ir(t) (F3)
ir(t) (F4)
ir(t) (F5)
80
10
15
Time (ms)
20
25
30
Fig. 5. The waveforms of the instantaneous feeder residual

currents for a single line to ground fault on Feeder 1 (section
D1) after 3 ms incidence time and 10 fault resistance
Where m is the number of radial equal sections and n is

the number of radial equal feeders in the networks.
Considering the simulated network shown in Fig. 1, and
assuming equal nine sections in the faulted feeder (F1),
then the fault weight for the healthy section should be
equal to 1/36 (i.e. 2.78%). For higher number of
sections in the feeder, the fault weight will be very
small which improves the performance and selectivity
of the proposed technique.
50
Transient RMS Feeder Residual Currents (A)
45
5. Results and Setting
For the simulated network shown in Fig. 3, the

instantaneous feeder residual currents are directly
measured or calculated from the phase values (using
(1)) at the substation measuring points; F1, F2, F3, F4
and F5. Fig. 5 shows the waveforms of the
instantaneous feeder residual currents for a single line
to ground fault on feeder 1 (section D1). The
waveforms for all healthy feeders (F2, F3, F4 and F5)
are in phase and approximately the same but the
situation is different for faulted feeder (F1); higher
magnitude with opposite polarity, which confirms the
suitability of using transient RMS method. The
waveforms are equal to zero before the fault incidence
at 3 ms. The transient RMS values of the residual feeder
currents are calculated in the proposed sliding window
(2.5 ms) using (2). The waveforms of the transient RMS
values, in time domain, are shown in Fig. 6. The
transient RMS background residual currents for each
feeder are calculated, in the sliding window, using (3).
Then, the fault weights for each feeder are calculated, as
the percentage ratio between the residual current of the
feeder and the residual current of the background
network, using (4). The fault weights results in the
sliding window for each feeder and section are shown
in Fig. 7 and 8. It can be seen that; for the faulted
feeder, the fault weight is about 100% but for the
Minimum detection tim e

Ir (F1)
Ir (F2)
Ir (F3)
Ir (F4)
Ir (F5)
40
35
30
25
20
15
10
5
0
10
15
Time (ms)
20
25
30
Fig. 6. The waveforms of the transient RMS feeder residual

currents for a single line to ground fault on Feeder 1 (section
D1) after 3 ms incidence time and 10 fault resistance
160
Transient Fault Weights for Feeders (%)
140
W
W
W
W
W
Minim um detection tim e

120
(F1)
(F2)
(F3)
(F4)
(F5)
100
80
60
40
20
0
269
10
15
Time (ms)
20
25
30
Fig. 7. The percentage transient fault weights for all feeders

for a single line to ground fault on Feeder 1 (section D1) after
3 ms incidence time and 10 fault resistance
Fig. 9. The threshold fault weight limits for isolated and

compensated neutral networks for a SLG fault with higher
fault resistance up to 100 k
160
Transient Fault Weights for Sections of F1(%)
140
6. Conclusions
Minim um detection tim e

A simple transient-based technique for earth faults in

isolated and compensated neutral medium voltage
networks is proposed. The proposed technique is used
to detect the faulted feeder and select the faulted section
exactly without any voltage measurements. Transient
RMS values of the residual currents are calculated in a
moving sliding window of 2.5 ms and then the fault
weights are determined for all feeders and sections. The
higher fault weights indicate the faulted feeder and
section. Extensive simulations using ATP/EMTP are
performed to validate the proposed technique for
isolated and compensated networks. The proposed
algorithm is simple and applicable
120
100
W
W
W
W
W
W
W
W
W
80
60
40
20
0
10
15
Time (ms)
20
25
(AB)
(B1)
(BC)
(BD)
(C1)
(C2)
(C3)
(D1)
(D2)
30
Fig. 8. The percentage transient fault weights for all sections

of the faulted feeder (F1) for a single line to ground fault on
section D1 after 3 ms incidence time and 10 fault resistance
7. Appendix
As shown in Fig. 8, the fault weights for all faulted

sections (AB, BD and D1) are greater than or equal to
100%. The faulted section can be selected easily by
following the radial flow of the higher fault weights
(active fault indicators). Actually, the highest fault
weight is for the faulted section. For all healthy section,
the fault weights are less than 20 %.
For compensated networks (Peterson coil/resonance
grounded/earthing networks), the impedance of the
compensation coil is relatively high at transient
frequencies. Consequently, the transients are about
similar in both of isolated and compensated neutral
networks. This is clearly investigated from the results of
the simulation data. The effects of the fault resistance
(up to 100 k) for both of isolated and compensated
networks are investigated.
The threshold fault weight limits are shown in Fig. 9.
There is an adequate isolating gap between the healthy
and faulty condition, and no overlapping between the
two conditions. The setting value of the critical fault
weight is proposed to be at the middle of the isolating
gap, which is equal to 60%.
Threshold fault weight limits
Background Network;
4 Feeders (2,3,4 & 5)
Total Length of 198 km
Overhead Line Feeders

Feeder 5 (48 km)
20 kV
Feeder 4 (50 km)
66/20 kV
Feeder 3 (50 km)
Feeder 2 (50 km)
Feeder 1 (Total Length of 53 km)

A
B1=10 km
BC=5 km
C1=5 km
C2=7 km
C3=4 km
AB=5 km
BD=7 km
D1=5 km
D2=5 km
D
140
SLG Fault
(Phase A)
120
Fault Resistance
251 km MV Distribution Network
100
Fig. 10. The ATPDraw circuit of the simulated, 20 kV, 251

km, medium voltage distribution network
80
60
Min.
Min.
Max.
Max.
W-faulty (Iso.)
W-faulty (Comp.)
W-healthy (Iso.)
W-healthy (Comp.)
Setting Weight = 60 %
(at the middle of the gap)
1.1 m
40
20
0
0
10
8.1 m
1
10
10
10
Fault resistance (ohm)
10
10
270
1.1 m
1.1 m Raven
d= 10.11 mm
r= 0.536 /km
Soil resistivity: 250 m
compensated MV network, IEEE Power Tech 2003

Conference Proceedings, Bologna, vol. 2, 23-26
June 2003, pp. 1-6.
10. M. Lehtonen and T. Hakola, Neutral Earthing and
Power System Protection - Earthing Solutions and
Protective Relaying in Medium Voltage Distribution
Networks, ABB Transmit Oy, Vassa, Finland, FIN65101, 1996.
11. L. Prikler and H. Hidalen, ATPDraw Users
Manual, SINTEF Energy Research, TR A4790,
Nov. 1998.
Fig. 11. The configuration of the feeders in the simulated

medium voltage distribution network
8. References
1. L. Kumpulainen et al., Distribution network 2030,

Vision of the future power system, Technical
Research Centre of Finland (VTT), Research Report
VTT- R-03835-07, Espoo, Finland, May 2007.
2. M. F. Abdel-Fattah and M. Lehtonen, A Transient
Fault Detection Technique with Varying Fault
Detection Window of Earth Modes in Unearthed
MV Systems, Power Quality and Supply
Reliability, 6th international conference (PQ2008),
Prnu, Estonia, 27-29 August 2008, pp. 181-186.
3. N. I. Elkalashy, M. Lehtonen, H. A. Darwish, A.-M.
I. Taalab and M. A. Izzularab, DWT-Based
Detection and Transient Power Direction-Based
Location of High-Impedance Faults Due to Leaning
Trees in Unearthed MV Networks, IEEE
Transactions on Power Delivery, vol. 23, no. 1, Jan.
2008, pp. 94-101.
4. W-Y. Huang and R. Kaczmarek, SLG Fault
Detection in Presence of Strong Capacitive Currents
in Compensated Networks, IEEE Transactions on
Power Delivery, vol. 22, no. 4, Oct. 2007, pp. 21322135.
5. X. H. Zhang, H. X. Ha, Z. C. Pan and B. Y. Xu,
Grounding faulty line selection in non-solidly
grounded systems using transient energy,
International Power Engineering Conference (IPEC
2007), 3-6 Dec. 2007, pp. 1147-1150.
6. Q. Yilong, T. Weipu, C. Shuan and Y. Yihan,
Study on Fault Line Selection based on Transient
and Mathematical Morphology in Resonant
Grounded System, International Conference on
Power System Technology (PowerCon 2006). 22-26
Oct. 2006, pp. 1-5.
7. W-Y. Huang and R. Kaczmarek, Line-to-ground
fault location in compensated distribution networks
by 3 parameters least-squares curve fitting without
voltage measurements, IEEE Power Tech, Russia,
27-30 June 2005, pp. 1-4.
8. H. Belka and M. Michalik, Application of the
continuous wavelet transform to intermittent high
impedance ground fault detection in MV networks,
Eighth
IEE
International
Conference
on
Developments in Power System Protection, vol. 2, 58 April 2004, pp. 473-476.
9. J. Lorenc, K. Musierowicz and A. Kwapisz,
Detection of the intermittent earth faults in
9. Biographies
Mohamed F. Abdel-Fattah was born in Qualiobia, Egypt on
June 11, 1972. He is a lecturer and researcher at Zagazig
University, Egypt. He received his
B.Sc. with distinction and first class
honors, M.Sc. and Ph.D. degrees from
Zagazig University (Faculty of
Engineering, Department of Electrical
Power and Machines Engineering) in
1995, 2000 and 2006, respectively.
He is a member of IEEE. Currently, he
is a Post-Doctoral Fellow at Helsinki
University of Technology (TKK),
Department of Electrical Engineering, Espoo, Finland.
His main research interest is protection of electrical power
systems including fault diagnosis and location of high-voltage
transmission lines, transient/ultra-high-speed and wide-are
protective schemes and compensation of the high resistance
effect in earth faults. (e-mail: mohamed.abdel-fattah@tkk.fi ,
mohamed.f.abdelfattah@ieee.org ).
Matti Lehtonen (1959) was with VTT Energy, Espoo,
Finland from 1987 to 2003, and since 1999 has been a
professor at the Helsinki University of
Technology (TKK), where he is now
head of Power Systems and High
Voltage Engineering.
Matti Lehtonen received both his
Masters and Licentiate degrees in
Electrical Engineering from Helsinki
University of Technology in 1984 and
1989, respectively, and the Doctor of
Technology degree from Tampere
University of Technology in 1992.
His main activities include power system planning and
asset management, power system protection including earth
fault problems, harmonic related issues and applications of
information technology in distribution systems (e-mail:
matti.lehtonen@tkk.fi ).
271

SENSORS FOR ON-LINE PARTIAL DISCHARGE DETECTION IN COVEREDCONDUCTOR OVERHEAD DISTRIBUTION NETWORKS
G. Murtaza HASHMI*, Matti LEHTONEN*, Mikael NORDMAN**, Muzamir ESA*
*Power Systems and High Voltage Engineering, Helsinki University of Technology (TKK), Espoo, Finland
**Ensto Utility Networks, Porvoo, Finland
wave propagation and attenuation, sensor spatial

sensitivity, frequency response, calibration, noise, and
data interpretation.
For electric power distribution industries, continuous
monitoring of installed and operating HV apparatus is
of particular importance from the point of view of
safety and reliability. The relatively new and
challenging application is conducting on-line high
frequency PD measurements for the monitoring of
falling trees on the CC overhead distribution lines [2].
The advantage of on-line PD monitoring allows for
conductor insulation diagnostics during normal
operation, and specifically, when the trees are leaning
on the conductors. The falling trees produce PDs,
whose magnitude increases when trees cause the
surface of the conductors to deteriorate. Automatic
detection of the falling trees will reduce visual
inspection work after storms and it will improve
reliability and safety of the distribution system.
Recently, a great interest has been shown by electric
utilities using CC system in distribution networks in
Finland to develop an on-line automatic system that
should be capable of detecting falling trees on the lines.
The system can be planned to be integrated into the
distribution automation system to reduce the overall
costs of the CC lines [3].
For the development of the on-line PD measuring
system using a Rogowski coil to detect falling trees on
the CC lines, the following few experiments were
conducted to investigate the PD wave propagation over
the CC line as well as to analyze the behavior of the
Rogowski coil for high frequency measurement
Abstract: The covered-conductor (CC) overhead lines

have been frequently used in medium voltage (MV)
networks in Finland for a long time. The falling trees
produce PDs, which may cause the insulation of the
conductors to deteriorate with the passage of time. In
this paper, the Rogowski coil is proposed as a partial
discharge (PD) sensor for this application. This sensor
is non-intrusive and superior to the conventional PD
detecting methods. The PD measurement methodology
is introduced. The concept of wireless sensors and
challenges to implement it for on-line condition
monitoring are also described.
Keywords: covered-conductor, medium voltage, partial
discharge, Rogowski coil, and wireless sensors.
1. Introduction
The CC lines have been used in MV networks
throughout the world since 1970 [1]. One compelling
reason to use CC lines is that they are more compact
and environment-friendly than bare conductors. The
CC line also withstands clashing and fallen trees for a
considerable time without interruption to the power
supply. In this way, the line can continue to function
despite the tree contact and the removal of the trees can
be scheduled appropriately. A drawback of CC lines in
distribution networks is that falling trees on the line
can neither be detected with normal protection relays
nor be localized by advanced high impedance relays
because the fault current is approximately nothing due
to the CC insulation and tree resistances. However,
these leaning trees produce PDs in the insulation of the
CC lines as mentioned in the previous chapter, which
may rupture after the passage of a certain time,
resulting in different kind of faults being introduced
into the network. By monitoring these PDs on-line,
progressive deterioration of the insulation can be
indicated. Early detection of developing faults leads to
better power quality and increased customer
satisfaction. PD monitoring involves an analysis of
materials, electric fields, arcing characteristics, pulse
2. Conventional PD detectors and their limitations

When a PD occurs, a current pulse is produced and this
current pulse interacts with the insulation capacitance
as well as the external elements in the test circuit.
Consequently, a voltage pulse is superimposed on to
the high voltage (HV) supply. The conventional
detection methods generally employ matching
impedance consisting of resistors, inductors and
capacitors in the PD current path. The measuring
272
window processing, zooming and auto-advance

features. Such techniques promise to be superior to the
currently used conventional PD detector methods.
Additional benefits may be gained if monitoring can be
performed as a continuous on-line measurement with
automatic PD data analysis [7]. Costs and need for
personnel expertise are reduced and the reliability of
condition assessment is improved [8]. Recently, the PD
detection frequency range has been extended up to the
radio frequency band with the development of new
sensors e.g., Rogowski coils.
resonant circuit expands the discharge current pulses in

the time domain for easier detection. In the event of PD
taking place, a quantitative parameter is required to
decide whether the apparatus needs to be repaired or
replaced. This means that the detected signal needs to
be accurately calibrated. In most cases, calibration is
done by injecting a known amount of charge and
measuring the voltage amplitude from the detector.
There are following two different methods of
conventional PD investigation [4].
i)
Straight detection including direct and
indirect methods
ii)
Balance detection method (bridge circuit)
The conventional PD detection technique can identify
discharges in short isolated cable lengths only.
Unfortunately, it has insufficient sensitivity for a long
circuit because of the large capacitance involved. It is
also required to isolate the cable from the circuit. PD
testing requires, besides the PD detector, additional HV
components such as a test-voltage supply and a
coupling capacitor, which are heavy and expensive and
not suitable for on-site tests [5]. The capacitor is a very
high impedance to the high ac voltage, while being a
very low impedance to the high frequency PD pulse
currents. The PDs are detected via a high-pass filter
established by this capacitor and the resistive
measuring impedance [6]. This method has two main
disadvantages for on-line application:
a) Since the HV capacitors have to be connected
to the phase conductors, the cable has to be
switched off, and power delivery is
interrupted.
b) The HV capacitor has to be connected for a
long measuring time, or even permanently.
HV capacitors are not always reliable in the
long term and can, therefore, become a cause
of faults themselves [6].
Instead of installing a lumped component, capacitive
coupling can also be realized through a metal
electrode, e.g., a plate can be installed at a certain
distance from a phase conductor. The obtained
capacitance, however, is now highly dependent on the
geometry of the substation, cable termination, and
positioning of this electrode. Furthermore, the
capacitance is relatively low. In modern substations, a
capacitor is sometimes integrated in the switchgear to
detect the power frequency voltage on each phase
conductor. The use of this capacitor for measuring
small signals like PDs is in practice hard, due to its
small value (usually <100pF). Moreover, this method
would not be very universal, since it depends on the
presence of this particular type of switchgear. In short,
these conventional techniques experience severe
limitations when it comes to on-line monitoring due to
the influence of background noise, absence of nonintrusive sensors and processing facilities. They are,
therefore, restricted to testing in a laboratory
environment.
Recent studies have shown that radiation from PDs is
impulsive in nature and consists of individual highenergy, wideband impulses of a few s in length. The
digital storage oscilloscopes and advanced digitizers
enable study of the PD signals more closely using
3. Rogowski coil as a PD sensor

In 1912, the Rogowski coil was introduced to measure
magnetic fields. Since the coil output voltage and
power were not sufficient enough to drive measuring
equipments, it has been widely used for measuring fast,
high-level pulsed currents in the range of a few megaamperes rather than in power system [9]. With the
improvement of todays microprocessor-based
protection relays and measurement devices, the
Rogowski coils are more suitable for such applications.
They have generally been used where other methods
are unsuitable [10]. They have become an increasingly
popular method of measuring current within power
electronics equipment due to their advantages of low
insertion loss and reduced size as compared to an
equivalent current transformer (CT) [11]. They are the
preferred method of current measurements having
more suitable features than CTs and other iron-cored
devices.
The coils have also been used in conjunction with
protection systems particularly high accuracy systems
or where there are direct current (DC) offsets which
would degrade the performance of CTs. This is useful
when measuring the ripple current, e.g., on a DC line.
The features of the Rogowski coils which make them
particularly useful for transient measurements stem
from their inherent linearity and wide dynamic range.
3.1. Construction of Rogowski coil
A Rogowski coil is basically a low-noise, toroidal
winding on a non-magnetic core (generally air-cored),
placed around the conductor to be measured, a fact that
makes them lighter and smaller than iron-core devices.
To prevent the influence of nearby conductors carrying
high currents, the Rogowski coil is designed with two
wire loops connected in electrically opposite directions
[12]. This will cancel all electromagnetic fields coming
from outside the coil loop. The first loop is made up of
turns of the coil, and the other loop can be formed by
returning the wire through the centre of the winding as
shown in Fig. 1, where d1, d2, and drc are the internal,
external, and net diameters of the Rogowski coil,
respectively. The coil is effectively a mutual inductor
coupled to the conductor being measured and the
output from the winding is an electromotive force
proportional to the rate of change of current in the
conductor. This voltage is proportional to the current
even when measuring complex waveforms, so these
transducers are good for measuring transients and for
273
applications where they can accurately measure

asymmetrical current flows.
The self inductance of the coil is fixed, and its mutual
inductance with the HV test circuit varies to some
extent (or slightly) depending on the position of the
coil in relation to the conductor. But once the coil is
clamped and held stationary, the mutual inductance
remains constant. Usually the coil is not loaded and the
voltage appearing across it is used as PD measurement
signal. Under this condition, the coil voltage is directly
proportional to the derivative of the current in the
conductor. Owing to the small value of the mutual
inductance it acts as a high pass filter; the sensor,
therefore, attenuates the power frequency component
of the current in the conductor and minimizes EMI at
lower frequencies. However, the PD signals are in the
range of MHz generally and induce sufficiently large
voltage in the range of mV.
The coils are designed to give a high degree of
rejection to external magnetic fields, e.g., from nearby
conductors. The coils are wound either on a flexible
former, which can then be conveniently wrapped
around the conductor to be measured or are wound on a
rigid former, which is less convenient but more
accurate. Both of these transducer types exhibit a wide
dynamic range, so the same Rogowski coil can often be
used to measure currents from mA to kA. They also
exhibit wideband characteristics, working well at
frequencies as low as 0.1 Hz, but are typically useful
up to hundreds of kHz, too. All this with low phase
error and without the danger of open-circuited
secondary (as could happen in case of CT). A typical
mutual inductance of a standard flexible Rogowski coil
(used in this research work) is up to 200-300 nH and its
resonant frequency lies in the high frequency spectrum
[13], [14].
v rc (t ) = M
di (t )
dt
(1)
where vrc(t) is the voltage induced in the coil by the

current i(t) flowing in the conductor due to the mutual
inductance M between the main current and the coil,
which is practically independent of the conductor
location inside the coil loop.
For simplified analysis, the behavior of the Rogowski
coil with terminating impedance Zout can be
represented by its equivalent circuit of the lumped
parameters as shown in Fig. 2 [15], where Rl, Ll, and Cl
are the lumped resistance, inductance, and capacitance
of the coil, respectively. The high frequency behavior
of the coil, in particular its bandwidth and
susceptibility to high frequency oscillations, is
significantly influenced by the terminating impedance
Zout. There is a trade-off between the bandwidth and the
sensitivity of the coil. The transfer function (Vout/Vrc) of
the Rogowski coil lumped parameters model (see
Fig.2) can be calculated as:
Rl
Ll
+
vrc(t)=Mdi(t)/dt
Cl
Zout vout(t)
Fig. 2. The Rogowski coil equivalent circuit (lumped

parameters model)
Vout
Z out
(2)
= 2
V rc
s Ll Z out C l + s ( Ll + Rl Z out C l ) + ( Rl + Z out )
d2
d1
drc
i(t)
Conductor
Usually the air-cored coil is used in combination with

large load impedance, i.e., Ll<< Zout. is the angular
velocity (rad/s), where =2f, and f is the frequency
(Hz) of the propagated signal. By assuming that the
Rogowski coil has negligible resistance, the
approximate measured voltage at terminals becomes:
Rogowski coil
turns
Internal return
wire loop
v out (t ) v rc (t ) M
di (t )
dt
(3)
So the output voltage at the terminals of the winding

wounded around the toroidal coil is proportional to the
time derivative of the current flowing in a conductor
passing through the coil. An integrator is incorporated
with the coil, which integrates the output voltage vout(t)
according to the following equation to convert it into
the current following through the conductor,
Zout
vout(t)
Signal to the measuring instrument
Fig. 1. Geometry and construction of the Rogowski coil
3.2. Working principle of Rogowski coil
i (t ) =
The Rogowski coil operates on the basic principle of

Faradays law. The air-cored coil is placed around the
conductor, where current pulses produced by PDs are
to be measured. This variable current produces a
magnetic field and the rate of change in current induces
a voltage in the coil given as:
274
1
M
out (t ) dt
(4)
3.3. Advantages of using Rogowski coil
model can be used to measure the PD magnitude by the

Rogowski coil (see Fig. 3).
In the conventional PD sensor, the test circuit

capacitance determines the frequency bandwidth and it
is usually not very wide compared with this sensor.
The frequency bandwidth of the Rogowski coil is not
influenced by the capacitance of the test circuit. It is
determined largely by the self inductance and the
capacitance of the coil and signal cables. It has the
following advantages:
The frequency response of the Rogowski
i)
coil sensor is very wide.
There is no conductive coupling between
ii)
the coil sensors and the HV test circuits.
Furthermore, the coil installation does not
necessitate
disconnection
of
the
grounding leads of the test objects and
therefore becomes a non-intrusive sensor
which is a very important aspect for onsite, on-line monitoring.
It has the advantage of possessing high
iii)
signal to noise ratio (SNR) with wide
frequency bandwidth.
There is no saturation due to air-cored
iv)
coil; therefore, it is not damaged by over
current.
It has very good linearity due to the
v)
absence of magnetic materials.
The
Rogowski
coil
based
PD
vi)
measurement system is a low cost
solution and can be easily implemented
on-site due to its light weight.
These advantages are essential for on-line PD
measurements; therefore, the Rogowski coils are
preferred over conventional PD sensors to take
measurements for detecting falling trees on CC
overhead distribution lines.
FDATool
vout (t)
Voltage signal obtained
from Rogowski coil
1
M
out (t ) dt
|u|
1
s
Abs
Integrator-2
Division
PD measured in
coulombs
q(t)
Incident noisy pulse
Modified filtered pulse
Mutual inductance of Rogowski coilM
Fig. 3. Simulink model for PD measurements
5. Wireless sensors concept for on-line PD

measurements
5.1. Motivation for wireless sensors
The rapid development in electronics, microcontroller

performance, digital signal processing, and wireless
communication has opened up possibilities to
implement new industrial sensor solutions on the
process level. One such promising concept is the
wireless sensor [16]. Wireless sensors can be used in
several applications like environmental monitoring,
medical diagnosis, and different industrial condition
monitoring applications. The wireless sensors may
become a key technology, especially in on-line
condition monitoring, as they are cheap and can easily
be embedded in the processes (also as retrofit). They
do not need wiring, which is a source of noise and
unreliability [17].
The motivation for using wireless sensors in power
system instrumentation is generally twofold: economy
and safety. From a system operation perspective,
wireless sensors give an opportunity to safely and cost
efficiently increase measurement coverage of the
network, including locations where wiring is
impossible. Hence, more extensive and accurate realtime information regarding the state of the system
becomes available to the operator. This means that the
components and the network can safely be run closer to
their technical limits and that vital information for
condition based maintenance of the network assets can
be elicited.
Today, one of the biggest shortcomings in distribution
automation is the lack of simple and cheap
instrumentation solutions that are easy to implement
and are applicable in system refurbishment. The cost of
instrumentation in a distribution automation system
with a high degree of automation is approximately 25%
[18]. The overall cost of installing and wiring a sensor
exceeds the cost of the sensor by more than ten times
[19]. Using wireless communication, installation costs
are significantly reduced; no problems appear with
damaged signaling cables that would need
maintenance, and instrumentation is possible in
applications where wiring is unfeasible [17]. Typically
these relate to rotating machines as well as medium
and HV environments. In these environments, isolation
becomes a problem and maintenance activities can be
dangerous [19]. In addition, refurbishment and
installation without de-energizing the power network is
possible. This favors wireless over power line
communications (PLC), although PLC has similar
advantages to wireless technologies for sensor
communication in industrial environments [20].
The PD data captured by the Rogowski coil is

essentially an oscillatory voltage pulse, which needs to
be processed to obtain PD characteristics such as peak
value, apparent charge, phase position, repetition rate,
and PD energy. The apparent charge q(t) entering into
the system due to PDs is given as:
1
s
Integrator-1
200e-9
4. PD measurement methodology
q (t ) = i (t )dt =
IIR band-pass filter
(5)
where vout(t) is the oscillating voltage appearing at the

output terminals of the sensor, M is the mutual
inductance of the Rogowski coil (which is taken as 200
nH in this case), and i(t) is the current flowing in the
conductor due to PDs.
The fast Fourier transforms (FFTs) of the acquired
pulses show the spectrum of frequency contents
present in the signals. An infinite impulse response
(IIR) band-pass filter (Butterworth type) of order 16
having frequency band (1-6 MHz), depending on the
frequency contents of the interferences and
disturbances in the signal, is applied for noise
elimination. Using the above steps and mathematical
expression of equation (5), the following Simulink
275
reason to use CC lines is that they are more compact

and environment-friendly than bare conductors. The
additional investment cost is often fully compensated
by savings in line spacing, reduced maintenance, and a
better quality of network. A drawback of CC lines is
that falling trees on the line cause a very high
impedance fault which cannot be detected with normal
or advanced protection relays. However, these leaning
trees produce PDs in the insulation of the CC lines,
which may rupture after the passage of a certain time,
thus introducing different kinds of faults in the
network.
A relatively new and challenging application is
conducting on-line high frequency PD measurements
for the monitoring of falling trees on CC lines. The
advantage of on-line PD monitoring allows for
conductor insulation diagnostics during normal
operation, and specifically, when the trees are leaning
on the conductors. By monitoring these PDs on-line,
progressive deterioration of the insulation can be
indicated. Early detection of developing faults leads to
better power quality and increased customer
satisfaction. PD monitoring involves the analysis of
materials, electric fields, arcing characteristics, pulse
wave propagation and attenuation, sensor spatial
sensitivity, frequency response, calibration, noise, and
data interpretation.
The measurements are taken using a wired Rogowski
coil, which can be converted into a wireless sensor in
the future. Due to limited computation capacity and
energy constraints in wireless sensors, signal
processing techniques can not be implemented as is
possible in the case of the wired Rogowski coil. We
must collect PD energy over a long period of time at a
specific bandwidth. This PD energy will be sensed by
the sensor to detect the intensity of PDs. Wireless
technology is fairly inexpensive and it can be
integrated into modern protection relays and to the
distribution automation systems for detecting falling
trees on CC lines. The proposed system will improve
the safety of CC lines and make them more attractive
to utilities due to reduced maintenance costs and visual
inspection work.
5.2. Challenges in on-line condition monitoring

using wireless sensors
Wireless sensor networks in general pose considerable

technical problems in data processing, communication
and system management. These problems are typically
caused by a harsh and dynamic environment; combined
with bandwidth and energy constraints that affect the
communication, data, and signal processing that can be
done. One of the biggest limitations is the fact that the
wireless sensors are self-powered (the power is drawn
from a battery or from energy harvested from the
environment). This means that the energy resources are
scarce (especially as the sensors should be small, easy
to embed in components and economical); only very
low frequency sampling and limited data processing
can be done in the sensor [17], [21], [22].
For on-line monitoring, especially of fast phenomena
like PDs, this means that the sensor has many design
challenges. For example, the design of the signal
processing filters must be very energy efficient and the
designer must know which frequency component are
the important ones for the application (to reduce the
impact of noise). Also the analogue-to-digital (A/D)
conversion is difficult to implement because the A/D
converters working at high frequencies have rather
high energy dissipation. The samples received from the
conversion must be processed by the microcontroller,
which means that the requirements on processing and
memory capabilities are high.
Minimizing the power consumption is the most
important design aspect for wireless sensors. In
addition to optimized components and functionality,
shut-down and power management strategies are
used. In wireless sensors for power systems, minimized
energy dissipation is essential for the feasibility of
integrating the sensor within equipment and
components. A light, small sensor is easy to integrate
and thus minimizes the impact the sensors have on the
power system reliability and operability. To achieve
this goal, wireless sensors in power systems should not
operate continuously. They instead need properly
designed shut-down strategies and the means to
minimize the pre-processing done by the sensor.
However, they must preserve functionality, i.e., the
capability to participate in, for example, state
estimation and fault management [16].
Hence, it is clear that the design of a wireless sensor
for the measurements of PDs in CC overhead lines
needs a totally new design philosophy. To be able to
find the right solutions for the design of the critical
sensor components, and to be able to find the trade-off
that must be made between energy dissipation and
functionality, a generic model of the sensor and the PD
behavior should be created.
7. References
1. P. Heine, J. Pitknen, and M. Lehtonen, Voltage

Sag Characteristics of Covered Conductor
Feeders, 38th International Universities Power
Engineering
Conference
(UPEC
2003),
Thessaloniki, Greece, September 1-3, 2003.
2. G. M. Hashmi, M. Nordman, and M. Lehtonen, A
Partial Discharge Detection Concept for Wireless
Sensors in Covered-Conductors Distribution
System, Europe's Premier Conference on
Electrical
Insulation
(INSUCON2006),
Birmingham, UK, May 24-26, 2006.
3. P. Pakonen, V. Latva-Pukkila, M. Bjrkqvist, and
T. Hakola, Condition Monitoring and Fault
Detection of Medium Voltage Covered Conductor
Lines, 4th Nordic Distribution Automation
Conference (NORDAC), Trondheim, Norway, May
22-23, 2000.
6. Conclusions
The use of CC lines has been expanding in MV

networks throughout the world over the last 30 years.
CC systems are developed to reduce failure rates
compared to bare wire MV networks and hence, to
improve the security of the supplies. One compelling
276
4. IEC60270, High-Voltage Test Techniques Partial

Discharge Measurements, CEI/IEC 60270, third
edition, 2000-12.
5. N. H. Ahmed and N. N. Srinivas, On-line Partial
Discharge Detection in Cables, IEEE Transaction
on Dielectrics and Electrical Insulation, Vol. 5,
N0.2, April 1998, pp.181-188.
6. P. C. J. M. van der Wielen, J. Veen, P. A. A. F.
Wouters, and E. F. Steennis, Sensors for On-line
PD Detection in MV Power Cables and their
International
Locations in Substations, 7th
Conference on Properties and Applications of
Dielectric Materials, Nagoya, June 1-5, 2003.
7. P. Pakonen and V. Latva-Pukkila, On-line Partial
Discharge Measurements on Covered Conductor
Lines, Nordic and Baltic Workshop on Power
Systems, Tampere, Finland, February 4-5, 2002.
8. B. A. Lloyd, S. R. Campbell, and G. C. Stone,
Continuous On-line Partial Discharge Monitoring
of Generator Stator Windings, IEEE Transaction
on Energy Conservation, Vol. 14, No. 4, December
1999.
9. G. Xiaohua, L. Jingsheng, Z. Mingjun, and Y.
Miaoyuan, Improved Performance Rogowski
Coils for Power System Transmission and
Distribution Conference and Exposition, 2003
IEEE PES, Wuhan, China, 7-12 September 2003.
10. D. A. Ward and J. L. T. Exon, Experience with
using
Rogowski
Coils
for
Transient
Measurements, Pulsed Power Technology, IEE
Colloquium, February 20, 1992, pp.61-64.
11. Z. Mingjuan, D. J. Perreault, and V. Caliskan,
Design and Evaluation of an Active Ripple Filter
with Rogowski Coil Current Sensing, Power
Electronics Specialists Conference, PESC99, 30th
Annual IEEE, Vol. 2, 1999, pp.874 -880.
12. L. A. Kojovic, Rogowski Coil Transient
Performance and ATP Simulations for Applications
in Protective Relaying, International Conference
on Power Systems Transients (IPST'05), Montreal,
Canada, June 19-23, 2005.
13. D. A. Ward and J. La T. Exon, Using Rogowski
Coils for Transient Current Measurements,
Engineering Science and Education Journal, June
1993.
14. L. A. Kojovic, Rogowski Coil Suit Relay
Protection and Measurement of Power Systems,
IEEE Computer Applications in Power, Vol. 10,
No. 3, July 1988, pp.47-52.
15. M. Argueso, G. Robles, and J. Sanz,
Implementation of Rogowski coil for the
Measurement of Partial Discharges, American
Institute of Physics, Review of Scientific
Instruments, 76, 065107, June 2005.
16. M. Nordman, An Architecture for Wireless
Sensors in Distributed Management of Electrical
Distribution Systems, Doctoral dissertation,
Helsinki University of Technology (TKK), Finland,
November 2004.
17. T. Brooks, Wireless Technology for Industrial
Sensor and Control Networks, Sensors for
Industry Conference, Sicon01, Rosemount,
Illinois, USA, 5-7 November 2001.
18. E. Antila, Developing Distribution Automation in

Medium Voltage Networks, Masters Thesis,
Helsinki University of Technology (TKK), Finland,
February 2003.
19. G. Shamble, J. Schutz, and C. Apneseth, Novel
Wireless Power Supply System for Wireless
Communication Devices in Industrial Automation
Systems, IEEE Annual Conference of the
Industrial Electronics Society (IECON02), Sevilla,
Spain, 5-8 November, 2002.
20. J.
Ahola,
Applicability
of
Power-line
Communications to Data Transfer of On-line
Condition Monitoring of Electrical Drives,
Doctoral dissertation, Lappeenranta University of
Technology (LUT), Finland, 2003, ISBN: 951-764783-2.
21. A. Sinha and A. Chandrakasan, Dynamic Power
Management in Wireless Sensor Networks, IEEE
Design & Test of Computers, Vol. 18, March-April
2001.
22. D. Wentzloff, B. Calhoun, R. Min, A. Wang, N.
Ickes,
and
A.
Chandrakasan,
Design
Considerations for Next Generation Wireless
Power-aware Microsensor Nodes, International
Conference on VLSI Design, Mumbai, India,
January 2004.
277

FAILURES AND DIAGNOSTICS OF POWER TRANSFORMERS

Sandra VITOLINA, Janis DIRBA
however small and initially insignificant defects are

frequently determined. This confirms that a well timed
detection of the faults in their early stage is extremely
important as it provides high coefficient of readiness,
reduces outages and costs for repair, as well as prolongs
the working life of the power transformer.
Abstract: The situation in the field of power

transformers diagnostics is analyzed, focusing mainly
on the most frequent causes for transformer outages in
Latvian distribution system. Classification of diagnostic
tests according to a fault type as well as some aspects of
control parameters selection is given. A version of
condition appraisal system is also given this article.
Keywords: control of technical condition, diagnostics,
power transformers, reliability.
1. Introduction
Power transformer is one of the main elements of
energy system determining largely the reliability of
electric supply. Effective control of technical condition
is one of the ways how to reach and to maintain high
reliability of the unit.
It has been proved that in many cases prolonging power
transformers lifetime for a 20-30 years is economically
more profitable than replacement of the old unit with
the new one. The use of rather expensive control and
diagnostics systems is considered as appropriate. The
expenses of diagnostic systems amount to 10% of costs
of a power transformer [1]. Thus a conversion from
previously scheduled repair system to a more operative
repair system based on units technical condition
prognosticating also units work with acceptable risk
level is prerequisite.
Sudden failure of a power transformer can cause high
expenses. For example, in year 2002 transformer fire
caused by a failure in one of the 345kV bushings was
experienced in Labadie power plant in the USA (Fig.1).
This transformer was put into operation in 1968 and
served one of the four generators in the plant with 600
MW rated generating power. Estimated loss at this point
is $ 5 million, including $ 1.5 million business
interruption. 15 days were needed for a primary repair
to set up a spare unit and to proceed generation [2].
Experts clarified the reason of the fire after this
accident. Investigation showed that the fire was caused
by small fault in one of the high voltage bushings that
caused subsequent failures. Severe failures of power
transformers are very rare in Latvian power system
Fig. 1. Transformer fire caused by a failure in high voltage

bushing and aftereffects of the transformer fire
2. Control parameters, measuring methods and

diagnostic tests for power transformers
Generally power transformers are affected by outer
factors or abnormal conditions in power system. It is
considered that material deterioration, construction
imperfections, manufacture faults, as well as inadequate
maintenance strategy and not sufficiently performed
278
repair works are the factors that mostly reduce

reliability of transformers.
Determination of the units technical condition can be
divided into 3 stages:
1. Selection of control parameters that
characterise ongoing processes in the unit;
2. Measurement of chosen control parameters;
3. Condition assessment of the unit based on
gained data analysis.
The main control parameters for oil-filled power
transformers are following:
Concentration of dissolved gases in oil, interfacial
tension, acid number, dielectric strength, moisture,
density of mechanical admixture, concentration of
soluble acids and many other oil parameters;
Cellulose insulation parameters cellulose
polymerization degree, moisture and mechanical
strength;
Oil-cellulose dielectric parameters including
dielectric
loss,
conductivity,
capacitance,
resistance, absorbability etc.;
Units heating parameters such as hottest point
temperature, temperature apportionment in the
winding, temperature of local overheating in the
winding;
Mechanical qualities of the winding condition of
pressing, short circuit impedance of the winding,
frequency characteristics etc.;
Parameters of partial discharges intensity,
incidence, continuance, primary voltage of partial
discharge, medium current value, discharge power
etc.
Thus there are very many diagnostic control parameters.
The larger half of mentioned control parameters is
measured accomplishing the complex diagnostic of the
unit. However complex diagnostic of the unit is timeconsuming and very expensive, thus optimization of
choosing control parameters is done in two stages. In
the first stage minimum obligatory characterizing
parameters are measured, allowing determining
origination of faults in transformers. After determining
the existence of the fault the second diagnostic stage is
started in which extra control parameters are measured,
fault type and its stage of development are designated.
This approach is cost-effective, but it isnt possible yet
to nominate optimal minimum obligatory control
parameters whose measurement results could show
origination of all type faults.
One of the basic requirements for measuring control
parameters is high metering precision. The accuracy of
the gauge has to be higher for at least an order than
acceptable value of measuring parameter prescribed in
standards. It is very important to ensure reiterative
measurements which means to provide conditions as
equal as possible when measurements are carried out.
Thus it is possible to compare the results to the
accumulated diagnostic data from previous tests.
Another aspect is necessity to gain data in the most
suitable form for units technical condition assessment.
For example, its hard to make a decision if only total
mass of mechanical alloy in oil is given, because there
can be a lot of small and harmless particles as well as
paucity of large particles that are hazardous for the

transformer. In this case for determining mechanical
alloy in oil lasers are used to screen off alloy groups in
order of their proportions.
The second stage of diagnostics is the measurement of
chosen control parameters. For the needs of diagnostics
it is necessary to choose such metering methods that can
be applied for a loaded transformer. This requirement
reposes on several aspects. Firstly in many cases it is
substantially to intensify supervision of units technical
condition by increasing frequency of periodical
inspections or by using automatic on-line monitoring
systems, and that means that measurements have to be
done for loaded transformers. Secondly several
parameters as for example dielectric loss of paper oil
insulation measured for loaded transformers give larger
and more reliable diagnostic information.
The third stage of diagnostics is condition assessment
based on gained data analysis. Determination of
transformers technical condition that is based on
gained data analysis is a very complicated diagnostic
stage that requires comprehensive knowledge about
ongoing processes in the unit and diagnosing
experience. In the last 15-20 years tendency to apply
expert systems for diagnosing the transformer can be
observed [3].
The most characteristic faults of power transformers,
their caused problems as well as diagnostic test
techniques are briefly summarized in Table 1[4].
Table 1. Selection of diagnostic test according to a fault type
Fault type
Dielectric
fault
of
insulation
Problems caused by a
fault
Winding inter-turn short
circuit
Winding shorted to core
Leads shorted to tank
OLTC failure due to
short circuit
Gas generation
Electrical
connection
fault
Winding
failure
Tank
leakage,
mechanical
fault
Bushing
fault
OLTC
failure
Cooling
system
failure
Overheating
between
defective contacts at
terminal bushings
Winding open circuit
Winding deformation or
displacement
Oil leaks
High
impurities
or
moisture content
Cracked bushing
OLTC not functioning
Overheating of winding
and insulation
Paper degradation
279
Diagnostic test
Turns-ratio
measurement
Insulation
resistance
measurement
Insulation and
winding resistance
Dissolved
gas
analysis (DGA)
Thermography
Winding
resistance
measurement
Frequency
response analysis
Visual inspection
Oil quality
analysis, Tan delta
test
Tan delta,
inspection
Tap changer
functional tests
Thermography,
fans functional
test
Furfural analysis
Degree of
polymerization
distribution of in-service power transformers (more than

40% of all power transformers have served more than
30 years)
Outages caused by inadequate maintenance as well as
by outer and environmental effects occur comparatively
rare (respectively 1.45%, 0.72% and 0.72% of whole
proportion).
Construction
imperfections
and
manufacture failures (6.97%) are external factors.
However high proportion of outages is caused by not
sufficiently performed repair works (9.01%), which
suggests that higher reliability can be achieved by
increasing the quality of repair works.
Reliability level of power transformers maintenance
generally can be evaluated by outage flow parameter
which characterizes the number of outages in definite
time period
3. Analysis of the most frequent causes for

transformer outages
The specificity of power transformers diagnostics is the
use of paper-oil insulation. High voltages and rated
powers prescribe high operation ratio of active
materials. These materials must resist heightened heat,
considerable strength of electric and magnetic fields, as
well as mechanical strokes during short circuits. Since
condition of insulation is the main factor determining
working life of the transformer, it is substantially to
control ageing of insulation during exploitation.
Important tasks are also periodical fixation of windings,
avoiding additional moisture and air from environment
and other preventative activities.
The combination of chemical and electrical diagnostic
testing can provide a better assessment on the condition
of the transformer. Any fault in the transformer core,
winding, tap changer or any other part can be diagnosed
by various parameters as described previous.
With so much data provided by several diagnostic
parameters and diagnostic techniques engineers have
difficulties to evaluate the risks and proposing
corrective solutions on problematic transformers.
Many energy companies and institutes such as US
Bureau of reclamation have developed the algorithm for
testing and evaluating transformers using condition
index [4, 5, 6]. However equipment and situations is
very different so it is impossible to use their experience
directly, development of an algorithm appropriate for a
particular power system is required.
The most frequent causes of transformer outages are:
1. Material deterioration (aging);
2. Construction imperfections and manufacture
failures;
3. Inadequate maintenance;
4. Outer and unpredictable stress (such as
atmospheric overvoltages);
5. Not sufficiently performed repair works;
6. Environmental effects.
The proportion of the above mentioned outages for the
power transformers of Latvian distribution network for
the time period from 1998 to 2008 is given in Fig.2.
( t )
n(t )
,
N t
(1)
Where n(t) the number of outages;

N total amount of power transformers;
t time period (generally t =1 year).
Reliability level of power transformers maintenance can
be also evaluated by such parameter as probability of
no-failure operation
p 1 ( t ) .
(2)
Table 2 shows values of these parameters for Latvian

distribution system power transformers in the time
period from 1998 to 2007.
Summarizing power transformer outages in the last
decade (see Fig. 3) it is obvious that the highest
intensity of outages are semi-season stages (March-May
and August-October), which characterizes with changes
of electrical load and accordingly with changes of
operational and environmental temperature.
Table 2. Estimated outage flow and probability of no-
failure operation values

Year
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
Number of
outage
n(t)
3
3
3
5
2
4
2
3
6
3
(t),
per 1
year
0.03
0.03
0.03
0.05
0.02
0.04
0.02
0.03
0.06
0.03
p
0.97
0.97
0.97
0.95
0.98
0.96
0.98
0.97
0.94
0.97
4. Selection of power transformers diagnostic

techniques
Fig. 2. The most frequent causes of transformer outages in
Latvian power system in a time period from 1998 to 2008
Fig. 2 shows that the most severe problem is aging and

deteriorating of insulation materials (81.13%). This
aspect is not anything extraordinary considering age
280
A complete condition evaluation of a power transformer

should include several subsequent steps (proposed
version of this article is shown in Table 3). Each one of
these steps has several elements which promote
diagnostic data evaluation process, as well as many
times identify defects, some of which may be avertable

thus leading to transformer life extension.
followed by off-line methods included in the Level 2

and internal inspections. Level 3 consists of additional
diagnostic tests the results of which allow drawing to
conclusions about reducing the load, preparing for
major repairs or even considering replacement. Since
most severe problem is aging and deteriorating of
insulation materials this appraisal level includes several
techniques devised for cellulose insulation technical
condition assessment.
5. Conclusions
The stated material allows concluding that:
It is hard to nominate optimal minimum control
parameters whose measurement results could
show origination of all type faults in power
transformers. Research in this field to improve
diagnostic parameters continues.
The most severe problem of Latvian power
system transformers is aging and deteriorating
of insulation materials (81.13% from all
outages).
Proposed condition appraisal system could be
used to mark out transformers with higher
probability of a defect which need an
additional supervision. System could render
assistance for such decision making as
reducing the load, preparing for repairs or
replacement of a certain unit.
Fig. 3. Time chart of detection of defective power transformers

Table 3. Proposed transformer condition appraisal system
Level
Basic level
Level 1
Level 2
Level 3
Diagnostic method
Design analysis:
manufacturer;
winding
configuration;
materials etc
Operating envirnoment:
protection scheme;
ambient temperatures;
lightning level ect.
Maintenance:
loading and
overloads;
fault frequency;
previous repairs etc.
Dissolved gas analysis
Oil quality analysis
Infrared thermography
External inspections:
cooling system;
bushings;
tank;
lightning arresters ect.
Turns-ratio measurement
Winding resistance
measurement
Tan delta measurement
Insulation resistance
measurement
Internal inspection:
coils and clamp;
LTC switch;
oil level ect.
Furfural analysis
Degree of polymerization
Frequency response analysis
Partial discharge analysis
Use
Databas
e
Databas
e
Databas
e
6. References
1.
On-line
On-line
On-line
On-line
2.
3.
Off-line
Off-line
Off-line
Off-line
4.
Off-line
On-line
Off-line
Off-line
On-line
5.
6.
Basic condition appraisal level provides a starting point

condition which is very important and increase the
effectiveness of technical condition assessment and
predictive maintenance.
The other three levels subdivide diagnostic techniques
used in Latvian power system, starting with on-line
methods and external inspection of the unit (Level 1)
281
..
()
. // .: - ,
2002.
Transformer
fire.
The
internet:
http://www.electrical-contractor.net/forums/
ubbthreads.php/ topics/114034/1
M.Kosaki,
M.Ieda,
K.Itoh,
Y.Tsutsumi,
Application experience of expert system to
insulation diagnosis of power apparatuses in Japan
// CIGRE Session paper, 1990, Ref.No: 15/33-04.
Young Zaidey bin Yang Ghazali, Mohd Aizim bin
Talib, Hannah binti Ahmad Rosli. Condition
Assessment of Power Transformers in TNB
Distribution System and Determination of
Transformer Condition Index // 17th Conference on
the Electric Power Supply Industry, October 27-31,
2008.
Facilities Instructions, Standards, and Techniques
Vol. 3-31 // Hydroelectric Research and Technical
Services Group, US Dept. of the Interior, Bureau
of Reclamation, June 2003.
Franchek M.A., Woodcock D.J. Life cycle
considerations of loading transformers above
nameplate rating // Proceedings of the Sixty-Fifth
Annual International Conference of Doble Clients,
1998, Sec 8-10.1

INFLUENCE OF THE DISTRIBUTION OF THE AXIAL, RADIAL AND

CIRCUMFERENTIAL COMPONENT OF THE MAGNETIC INDUCTION
VECTOR WITHIN THE SPACE BETWEEN SWITCHES ON RUPTURING
CAPACITY OF VACUUM SWITCHES
Dr in. Bogdan KWIATKOWSKI *, Dr Robert PKALA **
* Rzeszow University , Technology and IT Institute; **Rzeszow University, Technology and IT Institute, Poland
surfaces presented in Figure 1 consists in applying the

measurement probe. There has been the way of placing
the probe towards the surface, where the field
distribution is being measured, presented in Figure 1
below.
Abstract: The paper presents experimental tests results

of magnetic field distribution for highcurrent switch.
Provided tests show influence of switch shape on field
distribution to obtain sustained vitality of vacuum
switch. The method shown in the paper let obtain all
components of magnetic induction vector. There is
aspired to gain the highest value of axial component
designing switches shape. It warrants favorable
influence on fixity and turn off ability of switch system.
There are lower values of peripheral and radial
components making bipolar field in space between
switches.
cewk
i
Pytka stykowa
rdo napicia
Cewka pomiarowa
Cewka
pomiarowa
rejestrator
Keywords: vacuum chamber, bipolar switch, magnetic

field.
g
1. Introduction
Fig. 1. Measurement of the field distribution with the use of

the probe [1]
There have been the results of experimental research of

the magnetic field distribution on chosen vacuum
switches construction (bipolar) presented in this article.
The construction of switches, that generate the axial
field is technologically difficult to perform in
comparison with the switches with the radial field. Such
a switch must consist of several parts that must provide
the proper mechanical strength of the whole switch
construction. The main element of the switch is the
copper axle, where the crown is mounted, copper too
that causes division of current flowing in the axle into
the equal parts, where the switch cover is attached, made
of the special copper and chromium alloy. Preparing
such a connection is the most difficult one because its
quality can not worsen the connection durability of the
whole switch chamber.
The switch cover is placed in the magnetic field in this

method, which is generated by the coil shown in figure
1. The measurement has been done on the switch
surface, and the measured tension has been compared
with the coil tension of the frame of reference placed
outside the measurement system
3. Description of the research system
There has been the HIOKI type 8841/42 MEMORY
HiCORDERrecorder used for the laboratory
measurements. In order to do measurements of the
electromagnetic induction in the space between the
switches there has been a measurement probe
constructed, consisting of three coils. These coils are
placed in a way that one can measure all components of
the electromagnetic induction. The figure below shows
the measurement probe. (fig. 2).
2. Measure method of the magnetic field distribution

There are several measure methods of the magnetic field
distribution known between the vacuum switches. The
way of magnetic field distribution between the switches
282
The electromagnetic induction was counted with the

conversion of the following formula (1).
Measurement coils
Vn S n d
potentiometer
There have been 50 coils rolled up on each measurement

coil, the coil diameter is 8,7 mm. The measurement
probe together with the base was mounted to the device
supporting switches (fig 3). This device enabled to lift
and lower switches on a given height. Moreover, the
table where the measurement probe was mounted, was
equipped with the plate allowing the setting of the
measurement probe every 150 on the circuit.
supplying
(1)
where: Vn- induced voltage on n - coil , Sn- switch field

included by the n-coil, Bn-density of the
electromagnetic field stream. So that is the final formula
calculating
the electromagnetic induction in the
following form:
Fig. 2. measurement probe
Bars
t
Bn
dt
200
abs(u offset ) * f
i 1
the
t
offset
Table with the scale
4* z *S
2 * 50 *103
max(abs ( I ))
(2)
min (u ) max(u )
1 200
1 200
(3)
where: u voltage measured on measurement coils,

z=50 number of coils of the measurement coil, S=59
mm2 measurement coil surface, fp frequency of the
sample recorder, I current flowing through the switch
system.
Fig. 3. Table with the plate with cuts
5. Measurement results of the chosen high-voltage

switches
4. Measurement and results analysis
The construction of switches under test
In the measurement method applied, the directly

measured quantity is the voltage induced on measured
coils. These coils are in the space between the switches
while current is flowing through the switch system and
are put from the outside to the inside of the switch
plate. The measurement of the voltage induced on coils
was done by the HIOKI recorder with the sample
frequency 100 s. It allows for receiving about 25
thousand measurement points for each measurement
probe location. The measurement probe in the initial
phase of measurements of each switch system, is placed
in so called zero position (00) and is successively moved
every 150 on a switch plate circuit. The measurement
method described here has been realized according to
the following diagram (fig.4).
Fig. 5. Bipolar switch
The main aim of the measurements was to show that

thee is a possibility of the evenness improvement of the
magnetic field distribution through the choice of the
construction and setting the switches towards
themselves, in the vacuum chamber of the switch in
order to achieve the best conditions of the magnetic field
distribution and at the same time the conditions of the
arc-quenching.
There were three field components measured: axial,
circumferential and radial, measured with three probes
at the same time from the outer edge of the switch into
inside.
The choice of the arial component have been
approached in these measurements with even field
distribution on the switch surface, what gives the even
diffusion arc distribution.
Fig. 4. Electric diagram of the measurement system.
While doing the measurements, the frequency of the

sample recorder was 100 s what gives at the network
power supply the amount of 200 samples per period.
283
The circumferential component should be small because

it focuses the diffusion arc and can cause excessive
burning of the switch in its eg. Central part and forming
the big amount of metal pairs.
The radial component influences the movement of the
arc on the circuit and its values are big in the central part
of the switches.
During measurements the current of 50 kA was flowing
in. In order to standardize the nomenclature there is a
switch setting shown in figures 6 and 7 towards
themselves, for which the measurements were done.
c)
Styk bipolarny (skadowa osiowa)
600
500
400
B [mT]
r=12 mm
r=20 mm
300
r=30 mm
200
100
0
0
20
40
60
80
100
120
140
160
180
kt [stopnie]
d)
450
400
350
B [mT]
300
r=12 mm
250
r=20 mm
200
r=30 mm
150
100
50
0
Fig. 6. Bipolar switch (setting 1, setting 2)
20
40
60
80
100
120
140
160
180
kt [stopnie]
Fig.8. Magnetic field induction axial components distribution

a)setting 1, b)setting 2,c)setting 3,d)setting 4
It appears from the following routes that the axial

component has the biggest values for setting 1. In case
of this setting this component has the most even
distribution on the switch surface. The following routes
show the distribution of the axial, radial and
circumferential components of the magnetic induction
for all settings of the measuring angle 450
a)
Fig. 7. Bipolar switch (setting 3, setting 4)
Styk bipolarny (ustawienie 1)
For each setting there have been the magnetic field

induction components measurements done. The
following routes have been presented in order to depict
the influence of the switches location on the distribution
and value of the magnetic induction components.
a)
1000
B [mT]
800
600
osiowa
400
promieniowa
200
obwodowa
0
12 14 16 18 20 22 24 26 28 30
r [mm]
b)
600
500
B [mT]
400
r=12 mm
r=20 mm
300
r=30 mm
200
B [mT]
100
0
0
20
40
60
80
100
120
140
160
180
kt [stopnie]
800
700
600
500
400
300
200
100
0
b)
osiowa
promieniowa
obwodowa
12 14 16 18 20
22 24 26 28 30
r [mm]
c)
400
350
300
r=12 mm
r=20 mm
200
r=30 mm
150
B [mT]
B [mT]
250
100
50
0
0
20
40
60
80
100
120
140
160
180
kt [stopnie]
800
700
600
500
400
300
200
100
0
osiowa
promieniowa
obwodowa
12 14 16 18 20
22 24 26 28 30
r [mm]
284
d)
6. Conclusions
The presented construction of the vacuum switches
construction does not exhaust all possibilities within this
range. The research has been conducted in the world all
the time on new interesting constructions that allow for
the increase of the rupturing capacity and rated voltage
often with the simultaneous decrease of the vacuum
chamber size. There are works on new constructions and
research methods of the switch systems done because
vacuum switches are ecological, have huge mechanical
and breaking capacity and allow for the switching off
the huge short-circuit currents.
The measurements results done on a chosen switch
system in the vacuum chambers of the switches shown
that there is a huge influence of the switches setting
towards themselves on magnetic field components
distribution. The measurement results of the magnetic
field distribution between switches used in the vacuum
chambers of the switches determine the original
approach in switch construction research. They also
allow for the evaluation of the magnetic field strength
value between switches and for the choice of the optimal
field distribution on the switch surface for achieving the
most profitable solution.

1000
B [mT]
800
600
osiowa
400
promieniowa
200
obwodowa
0
12 14 16 18 20 22 24 26 28 30
r [mm]
Fig.9. Magnetic field induction components distribution for

the measuring angle 450.
The axial component takes the biggest values for the

setting 1and setting 3 with small values of the
circumferential component. This kind of proportion
between the components is the most profitable between
the diffusion arc switches in space. The next four routes
show the magnetic induction components distributions
for the following settings for the measuring angle 900.
0
Skadowa osiowa dla kta 90

600
B [mT]
500
ustawienie 1
400
ustawienie 2
300
ustawienie 3
200
ustawienie 4
7. References
100
0
12
14
16
18
20
22
24
26
28
30
1. Yoshihiko Matsui, Hidemitsu takebuchi, Masayuki

Sakaki, Akira Nishijima, Masami Takakura, Yasushi
Noda, Analysys and Measurement of Axial Magnetic
Field in Vaccum Interrupter, 19 Int. Symp. On
Discharges and Electrical Insulation in Vacuum,
Xian 2000, China
2. Homma M., Somei H, Niva Y., Yokokura, K.Ohsima,
Physical and Theoretical Aspects of a New Vacuum
Arc Control Technology - Self arc diffusion by
electrode; SEDE, IEEE 18 International Symposium
on Discharges and Electrical Insulation in Vacuum
Eindhoven 1998
3. Dobrogowski J., Moscicka-Grzesiak H., Seidel S.
Zaucki Z., Podstawy wyadowa i wytrzymao
elektryczna w prni, PWN Warszawa-Pozna 1983
4. Jimei Wang: Discussion on future development of
vacuum switschgears, IEEE 18 International
Symposium on Discharges and Electrical Insulation
in Vacuum, Eindhoven 1998
r [mm]
Skadowa obwodowa dla kta 90

300
B [mT]
250
ustawienie 1
200
ustawienie 2
150
ustawienie 3
100
ustawienie 4
50
0
12
14
16
18
20
22
24
26
28
30
r [mm]
0
Skadowa promieniowa dla kta 90

1200
B [mT]
1000
ustawienie 1
800
ustawienie 2
600
ustawienie 3
400
ustawienie 4
200
0
12
14
16
18
20
22
24
26
28
30
r [mm]
Fig.10. Magnetic field induction components distribution for

the measuring angle 900.
5. M. Honma, K. Watanabe, M. Nishihara, I. Oshima,

New Vacuum Arc Technology; SADE - Application
to Vacuum Interrupters, IEEE 18 International
Symposium on Discharges and Electrical Insulation
in Vacuum, Eindhoven 1998.
6. H. Fine, M. Heimbach, W. Shang: A New Contact
Design Based on Quadropolar Axial Magnetic Field
and its Characteristics, European Trans, on Electrical
Power Engineering, Vol.20. 2000, pp.75
7. Siemens Co.: Low Voltage Vacuum Circuit Breakers
with Intelligential Functions, Siemens Product
Specification, 1994.
The axial component of the magnetic field reaches the

high values for the setting 1 and setting 4. In setting
4 there is the possibility of receiving, in the central part
of the switch, small values of the circumferential
component of the magnetic field strength and huge on
the outer edges of the switch.
The value of the axial component of the field for the
setting 4 is big the biggest in the central part of the
switch .
285

THE RESEARCH OF THE INFLUENCE OF THE BOUNDARY CONDITIONS OF

THE ATMOSPHERIC OVERVOLTAGES ON THE INSULATION OF HIGH
VOLTAGE EQUIPMENT
Saulius GUDIUS, Linas Andronis MARKEVIIUS, Alfonsas MORKVNAS
Abstract: The situation when difference characteristics

protection devices against overvoltages are mounted in the
same partly reconstructed substation frequently occur in the
high voltage electric power network. The typical example of
such situations is when dischargers near power transformers
are replaced by arresters and the left part of the substation is
protected using discharges. For the evaluation of such typical
situation the model of power plant substation has been
analyzed. The dangerous conditions for the insulation of the
equipments mounted near incoming lines, when wave of the
overvoltages rose by the lightning discharge arrive to the
substation, has been analyzed during research and break-down
data investigation.
and the boundary characteristics of electromagnetic

waves because of the overvoltages caused by a lightning
discharge, when overvoltages arresters and dischargers
are used for the protection of the substation equipment
insulation.
The main attention in the work is assigned to the
evaluation of the working conditions of the insulation of
the equipment of the 110 kV substations and the
investigation of the arresters efficiency and reliability.
2. The characteristic of the overvoltages calculation
model
Keywords: overvoltages, arresters, cable insertion.
The calculation scheme of the substation consists of the

parameters of the distributed buses segments and
lumped parameters, which model the current and
voltage instrumental transformers, power transformers
and coupling capacitors.
The modelling scheme with five lines has been created
for the electromagnetic transient processes analyser (Fig
1), meant for the investigation of the maximum
overvoltages in the 110 kV electrical substation.
1. Introduction
The strategy of the protection of the high voltage
equipment insulation against overvoltages has been
changed when overvoltages arresters started being used
instead of dischargers. The valve dischargers can protect
the insulation of the equipment mounted in the
substation against violation in the zone of 100 120 m.
for its feature during its switching on, when the
discharge gap is broken down, to generate the returning
electromagnetic wave of the opposite direction which
dampers the overvoltages wave that comes from the
line.
During the process of the replacement of valve
discharges by overvoltages arresters in the electric
energy system, there might be a situation when a valve
discharger is left for the protection of the equipment
insulation, and the overvoltages arrester is connected to
the transformer. In such a combination of the protection
devices, a dangerous level of overvoltages can occur in
the case when the overvoltages arrester lowers the level
of overvoltages and the valve discharger cannot switch
on. In such a situation the insulation of the equipment,
mounted in the substation near the incoming lines, can
be damaged.
The aim of the work is to analyse the conditions of the
possible violation of the insulation, to evaluate the
change of the characteristics of the network elements
L5
42
6
Ci2
Ci1
43 L4
2
7 C
i3
Cr1
17
10
Ci6
45 L2
40
4
Cr2
12 Ci8
11
Ci5
Ci4
16
L3
44
40
40
40
1
24
30
86
Cui1
19
31
27
21
36
CT1
37
CT3
26
32
Cui2
86
38
CT2
Cui3
Cr4
21
Ci10
20
28
22
21
29
33
39
15
14
Ci9
25
23
40
Cr3 5
13
Ci7
18
46 L1
Cui4
34
40
Cu1
35
41
Cu2
CT4
Fig. 1. Modelling scheme of the 110 kV substation: 40 the

lenght of the segment betwen the nodes ( unshown lenghts has
been calculated according to the step of the cell 9 m) ); 10
the number of the node
286
ri ir ni the radius and number of the conductor in I

wire.
The modeling of the electromagnetic processes has been
carried out using special coordinates , q and 0 for
wave channel separation.
Voltage coordinates , q, 0 are calculated as follows:
Because of their minor influence on the overvoltages,

other apparatuses of the substation have been
equivalented.
When analysing the transient processes of overvoltages,
it is important to correctly calculate the scheme
parameters, especially those which depend on the
frequency because of the skin effect.
Partial line parameters capacities of aerial lines scheme
are calculated as follows:
C Cik
1
N,
2 0
u 3 x1 TU1 u 3 x1 ,
(1)
, q, 0;
= , b, c.
Current coordinates are calculated as follows:
0 dielectric air constant;

0 = 1/36109, F/m;
N = (Nik);
2hi
;
N ii ln
rei
D
N ik ln ik ;
d ik
hi, rei, dikn, Dik average conductor hanging,
geometric mean radius of the conductor, the distance
between i and k conductor, the distance between the
mirror reflection on the earth surface of the i and k
conductor.
Partial line parameters inductivities of aerial lines
scheme are calculated as follows:
L Lik 0 N F M ,
2
i 3 x1 TI1 i 3 x1 ,
here: TU1 ir TI1 inverted modal and TI . matrix values.
Phase coefficients and decrement of dissipation are
calculated as follows:
diag G TU1 LC TU ,
Surge resistance:
diag Z diag G TU1 LTI

1
(2)
F = (Fik) ;
Fik
H1 1 1 1 12 H12 1 2 1 ;
2
21
1 22
1,2 = rik exp (j(ik));

k z2 k 02 ;
arg k z2 k 02 ;
ik arctg
bi bk
hi hk
k0 0 0 ;
k z2 k 02
j 0
H1 Y1 Struve function of the first type Struve and the

second type Besel function of the first type;
bi abscise of the i conductor;
z specific ground resistance (z = 100 m);
angular frequency rad/s
= (ii);
k 1 1 J 0 k 1 ri
M ii
,
j 0 ri J 1 k 1 ri n i
k1 j
(4)
Specific scheme models parameters when current

frequency is 200 Hz are equal:
0 air magnetic permeability ;

0 = 410-7, H/m;
rik D ik
(3)
0.1188 - 0.0126 - 0.0027 - 0.0158 - 0.0029
- 0.0126 0.1194 - 0.0126 - 0.0092 - 0.0092

- 0.0027 - 0.0126 0.1188 - 0.0029 - 0.0158 1010, [F/m];
- 0.0158 - 0.0092 - 0.0029 0.0660 - 0.0034

- 0.0029 - 0.0092 - 0.0158 - 0.0034 0.0660
0.1526 0.0633 0.0495 0.0733 0.0514
0.0633 0.1526 0.0633 0.0647 0.0647

0.0495 0.0633 0.1526 0.0514 0.0733 105, [H/m]
0.0733 0.0647 0.0514 0.2332 0.0557
0.0514 0.0647 0.0733 0.0557 0.2332
0.1639 0.1450 0.1440 0.1419 0.1410
0.1450 0.1639 0.1450 0.1417 0.1417

0.1440 0.1450 0.1639 0.1410 0.1419 103 , [/m].
0.1419 0.1417 0.1410 0.4301 0.1381

0.1410 0.1417 0.1419 0.1381 0.4301
3. The effect of the overvoltages on the insulation of

the electrical equipments
The level of the overvoltages depends on the length of
the electrical lines, their configuration, construction and
load, the characteristics of the protection devices against
overvoltages and voltage level in the network.
The situation in the power station 110 kV closed
substation is being analysed, when two coupling
capacitors exploded, when overvoltages occurred near
1
;
1
1 ir 1 conductor magnetic permeability and specific

resistance
287
the substation after the lightning discharge in the aerial

power line.
Up to 700 kV amplitude voltage wave (10 x 70 kA)
might occur in the spot of the lightning discharge in the
aerial line when the insulator garland is electrically recovered. When there are no protection devices against
overvoltages, the voltage can exceed the tested voltage
of the equipment insulation. Because of the overvoltage
caused by the lightning discharge, a wave coming to the
substation can evoke overvoltages which are dangerous
for the equipment insulation. Fig. 2 shows the processes
in in different spots of the power station 110 kV
substation when there are no protection devices against
overvoltages.
wave that comes from the line. As it is seen from the

results of overvoltages test (Fig. 3), the overvoltages
arrester next to the transformer lowered the
overvoltages level at the discharger connection spot, and
the discharger did not connect.
Fig. 3. Lightning overvoltages in the power plant 110 kV

substation when the 700 kV 1,2/50 wave comes through the
line L1; Overvoltages protection means are intalled; 1- voltage
in the coupling capacitor; 2 - voltage in the connection spot of
the dischargesr; 3 - voltage in the connection spot of the
arrester
If there were only a discharger in the substation, the

lengths of the impulses influencing the insulation would
be obviously shorter (Fig. 4)

line L1; Overvoltages protection means are not intalled; 1voltage in the coupling capacitor; 2, 3- voltage in the
connection spot of the arrester and dischargesr
The level of the overvoltages exceeds the specimen

voltages guarnteed by the producer, that is why
overvoltages lowering means must be installed in the
substation. In the 110 kV voltage substation, discharges
are erected next to the buses voltage transformers,
andovervoltages arresters are erected next to the power
transformers. In the scheme investigated, auxiliary
power transformers are installed 21 meters and block
power transformers are installed 86 meters from the
main buses. Overvoltages arresters protect the insulation
of the equipment which are beyond the connection spot.
Because of their characteristics to evenly lower the
voltage, overvoltages arresters practically do not protect
the equipment which is in front of the arresters. The
protection zone of the overvoltages arresters is rather
narrow.
The arrester, which is the closest to the substation
transformer, lowers the voltage next to the discharger,
and the latter does not connect when the overvoltages
wave comes from the line. The length and amplitude of
the impuls are dangerous for the insulation of the
equipment; in the case investigated, it is dangerous for
the insulation of the coupling capacitors (Fig. 3); the
longer the length of the impuls, the lower electric
resistance of the inner insulation.
The valve dischargers can protect the insulation of the
equipment mounted in the substation against violation
in the zone of 100 120 m. for its feature during its
switching on, when the discharge gap is broken down,
to generate the returning electromagnetic wave of the
opposite direction which dampers the overvoltages

line L1; Only dischargers are installed; 1- voltage in the
coupling capacitor; 2 - voltage in the connection spot of the
dischargesr; 3 - voltage in the connection spot of the arrester
As dischargers (near the buses) and overvoltages

arresters (right next to the power transformers) were
erected in the closed substation of the 110 kV voltage
electric power station, the coupling capacitors were
affected by the overvoltage higher than its impulsive
resistance. It could have been the main reason why the
coupling capacitor exploded. Besides, when the line L1
is connected from the other side, the highest
overvoltages are formed at the end of the line, i.e. on the
coupling capacitor. Because of these overvoltages and
as it was damaged by the initial overvoltage caused by
the lightning discharge, the insulation of the coupling
capacitor was punched.
288
When overvoltages transient processes in the 110 kV

voltage substation of the electric power station without
arresters were modeled by a digital analyzer, a condition
was accepted that overvoltages do not span the
insulation of the substation equipment even if the
impulsive resistance of the insulation is exceeded. In
such a way the most dangerous spots in the substation
are determined, in which the overvoltages of the highest
amplitude are most likely to occur, when the wave
comes from line L1.
When there are overvoltages arresters instead of the
discharges, the levels of overvoltages in the 110 kV
voltage substation of the electric power station are
estimated for the same regimes electromagnetic wave
forms as there are in the scheme without arresters.
When estimating and choosing the devices for the
protection of the insulation of 110 kV voltage
equipment in electric energy system the most attention
must be paid to the overvoltages caused by a lightning
discharge, and if the voltage is more than 110 kV, the
attention must be paid to all the possible overvoltages.
When the initial working voltage of the protection
equipment against overvoltages is being lowered, it is
necessary to take into account the energy which is
absorbed by them, if switching overvoltages are limited
. Valve discharges, which are used in the electric net up
to 110 kV, do not react to switching overvoltages
because their switching-on voltage is higher than the
level of switching overvoltages.
The results of the research show that overvoltages
arresters erected before the power transformers and in
the place of discharges can not protect the insulation of
all the substation equipment. Dangerous overvoltages of
higher amplitude than the testing impulse voltage can
influence the insulation of part of the current
transformers and coupling capacitors.
2.
3.
4.
5. References
1.
2.
4. Conclusions
1.
which was affected by initial lightning overvoltage

discharge.
The overvoltage levels in 110 kV substation of the
power plant can have higher amplitude than pulse
test voltage of equipment insulation if surge
arresters are mounted instead of dischargers. The
simulation results show that the surge arresters
mounted before power transformers at the location
of dischargers can not protect all the equipment of
the substation. The insulation of part of the current
transformers and the coupling capacitors can be
affected by dangerous overvoltages higher than
pulse test voltage, which secures reliable operation
of equipment insulation.
The analysis shows that surge arresters must be
mounted at the first supports of the overhead lines
in order to protect the insulation of the 110 kV
closed substation equipment of the power plant.
Surge arresters are not necessary near power
transformers in the closed substation.
The energy characteristics of surge arresters show
that the dissipated amount of energy by surge
arresters mounted near the first supports must be
equal or larger to 5.8 kJ/kV. According to energetic
capacity, additional surge arresters must be of the
III class.
The analysis of the accident shows that the failure

in coupling capacitors could be caused by
improperly tuned overvoltage protections. In 110
kV closed substation, the dischargers were mounted
near the buses and surge arresters were mounted
near the power transformers and the coupling
capacitors were affected by overvoltage higher than
impulse resistance. The main reason why the
capacitor exploded was that the discharger didnt
work. Furthermore, the highest overvoltages occur
in the end of the line, i.e. at coupling capacitor
during the line L1 switching. These overvoltages
damaged the insulation of the coupling capacitor
3.
4.
289
Gudius S., Markeviius L., Morkvnas A. Recursive

method to examine electromagnetic transient processes in
cable network // IV Sympozjum, Metody matematyczne
W elektroenergetyce MMwEE98, Zakopane, 1998. P.
235-244.
Gudius S., Markeviius L., Morkvnas A. The research
of overvoltages arresters efficiency in high voltage
substation // XI International Conference on
Electromagnetic Disturbances: EMD2001, Overvoltage
in Power Electronic and Computer Engineering /
Bialystok Technical University, Vilnius Gediminas
Technical University, Kaunas University of Technology.
Bialystok, 2001. P. 36.
Gudius S., Markeviius L. A., Morkvnas A., Navickas
A. S., Stanionien R. 110 kV kabelini intarp apsaugos
priemoni nuo virtampi darbo slyg tyrimas //
Elektros ir valdymo technologijos 2007. 2 tarptautins
konferencijos straipsni lietuvi kalba rinkinys Kaunas:
"Technologija", 2007. P. 20-23.
Gudius S., Markeviius L., Morkvnas A., Navickas A.,
Stanionien R., Roanskas A. V. The influence of the
overvoltages on the reliability of the 110 kV power line
with cable insertion // XVII International Conference on
Electromagnetic Disturbances: EMD2007, 19-21
September 2007,
Bialystok, Poland. Kaunas:
"Technologija", 2007. P 187-190.

INSULATION LIMIT RESISTANCE PERFORMANCE ASSESSMENT MODEL

Alfonsas MORKVNAS, Renata STANIONIEN, Povilas VALATKA
Kaunas University of Technology, Electrical Systems Department, Lithuania
Abstract: In order to increase power grid reliability, it is

necessary to use the cost-based evaluation tool which
determines the impact of various characteristics in order
to reduce the duration of the emergency interruptions,
by using a properly chosen insulation protection from
overvoltages and by monitoring characteristics which
impact the equipment. In this paper, the analysis of
factors which influence the electrical equipment and
processes which take place in isolation during
overvoltages is given. Also, there are defined the
characteristics of factors which influence the nature of
the parameters change. The digital model of the
processes which take place in the equipment insulation
is designed.
cases it is the chromatographic analysis of the gasses

which are dissolved in insulation [2, 9].
High-voltage equipment monitoring system monitors
the characteristics which impact to the power system
(overvoltage parameters and their duration, asymmetry,
harmonics, parameters of the emergency regimes).
The aim of this work is to determine the limit of
insulation resource according to two factors- voltage
and temperature. To identify the impact factors
influence to the consequences of exposure parameters.
2. Factors which effect the electrical insulation
During the maintenance of electrical equipment
insulation is exposed to operating voltage, short-term
overvoltages, temperature. The parameters of impact
factors, which effect the equipments isolation, depend
on the equipment location in the network, load
variation, the frequency of overloads, short-circuiting
faults.
The electrical power system maintained equipments
insulation condition evaluation and prognosis of defects
is relevant and quite a complex task involving the
processes occurring in the network, and its insulation.
This task determines the electrical equipment efficiency
and reliable power system operation. The electrical
equipment reliability, quite often is determined by
insulation [1, 2]. In order to evaluate accurately the
condition of electrical equipment and clarify the defect
it should be investigated processes in the power system
and in electrical equipment insulation, also to determine
the impact factors influence to the consequences of
exposure parameters and to identify their variation
regularity. Analysis of the overvoltages caused changes
in isolation can give the predicted insulation aging
speed and the evolution of defects and also could allow
to remove defects when necessary and prevent accidents
and power system losses.
The reliable operation of electrical equipment and its
insulation condition is determined by various factors
(operation current and voltage, overvoltages,
overcurrents, vibration, humidity, etc.) which appear in
power system. The electrical equipment isolation
(internal and external) must be resistant to long-term
maximum permissible voltage, switching and
Keywords: grounding device, concentrated grounding

device, specific resistance, coefficient of discharge.
1. Introduction
Reliable operation of the power system electrical
equipment is a key factor in guaranteeing a reliable and
economical operation of whole power system. Reliable
operation of the electrical equipment is based on the
appropriate equipment condition monitoring, scheduled
repair tasks, and the nature of impact factors.
High-voltage equipment insulation condition and the
reliable duration of working time is determined by
various factors of the power system: overvoltages,
partial discharges, overcurrents, vibrations and so on.
Insulation aging rate and the appearance of defects
mainly are caused by overvoltages, heating, and partial
discharges. Duration of the Overvoltages and its
amplitude can be various in the power system.
Overvoltages and equipment operating temperature are
the main factors of insulation aging speed [1, 2, 7].
During the maintenance of the high voltage substations
equipment, which is effected by the high electric field,
overvoltages, discharges, thermal effects, moisture,
oxidation, vibration, aging and other factors, the various
defects appear. The in operation equipment monitoring
systems are applied for the high-voltage equipment,
which monitors such characteristics, which analysis can
give the answer of the inner insulation, and other units,
technical condition and assess the accident risk. In most
290
atmospheric overvoltages [1, 10]. Overvoltages

transient process, if the insulation of electrical
equipment is properly maintained and protected from
overvoltages, can reliably serve all set time.
Because of the overvoltages impact in solid and liquid
isolation begins or increases the partial discharge
processes [3]. During the partial discharge processes
deteriorates the insulating properties of electrical
installations, increases quantities of gas in oil, increase
the surfaces of micro cavities in solid insulation an also
accelerates the aging process [2, 5, 9]. Overvoltages
impact to the insulation may be considered according
arranged insulation model, which evaluates the partial
discharge process. The partial discharge process in the
electrical equipment insulation increases during the
overvoltages impact. Partial discharge level increases 58% per year [3].
Limit insulation resistance conditions are affected by
the operating voltage, its influence time and
temperature.
Overvoltages level and impact duration characteristics
for insulation are studied by various authors [1, 7]. To
evaluate insulation resources the experiments are done,
where the insulation resistance was explored according
to dependence on time and temperature [7]. Such
studies are carried out in order to measure the insulation
resistance to stress exposure duration (volt-second
characteristic), and thus determine the limit of resource.
In the power system equipment is installed in different
locations and is exposed to the different factors with
different parameters (amplitude of overvoltage, the
duration of exposure, occur rate, overloads, etc.).
Therefore, electrical equipment insulation aging
processes take place differently. In addition, in various
locations overvoltage parameters, because of network
configuration, technical characteristics, ongoing
complex physical phenomena, are dispersed uneven.
Secondary electrical equipment insulation control in
terms of qualitative characteristics, is performed
periodically or permanently, and can evaluate the
condition of the device, but the nature of the factors and
the influence still remains unclear [3, 4, 5, 6, 8]. By
registering the operating characteristics of the factors
(overvoltages amplitudes and exposure duration), and
simultaneously the consequences of impact (partial
discharge characteristics: the apparent charge, discharge
currents, the frequency of recurrence, etc.) it is possible
to evaluate arising defects of equipment in the initial
stage.
Energy density consists of thermal component and the

part of electromagnetic field energy which had been
converted into energy of molecular oscillations.
The energy density of Thermo electromagnetic field
HTEM consist of the following components [7]:
thermal HT=cT;
electrical H E E 2 ;
2
and magnetic H M M 2 .
2
(1)
(2)
(3)
The thermo-electromagnetic field density can be written

in the following expression:
1
H TEM c T ( E 2 M 2 );
2
(4)
here density of insulating material;

c isochoric heat capacity;
and electric and magnetic permeability;
E and M strengths of electric and magnetic fields.
The actual exploitation time for an insulating material
starts at time ts and ends at the time tD. This time
interval is considered as a resource:
R tD tS .
(5)
This value is directly proportional to its dimensionless

modification:
1
t D t S ( D S ) am
z 1 .
On the other hand dimensionless

temperature at constant load is:
(6)
conversion
~
r
C R (T
E 2 ) w.
c
(7)
Expression in measured values can be assumed as:
1
( D S ) am
~
r 2
E )
w R (T
c
.
e
(8)
It can be noted that in (8) equation there are included

parameters which depends on the properties of material
and the activation energy for the reaction. One does not
necessary to determine each parameter separately. To
make a prognosis of the insulating service life it suffices
to find experimentally such values:
Insulation scale time
3. Limit insulation resistance characteristics

Maximum economically based equipment utilization
effects on higher equipments working temperature.
Device temperature increase is limited by the insulating
properties of the materials- at a higher temperatures
increases materials thermal decomposition and aging. If
the temperature is increasing the speed of chemical
reactions accelerates. For organic type of insulation
chemical processes often increases of two times when
the temperature increases 10 C.
The insulation break down is considered as a phase
transition, which has been depended on energy density.
R ( D S ) 1am
z 1 P1 , [h]
(9)
Activation temperature
~
Ta w / R P2 ; [K]
here w activation energy [J/mol];

~
R universal gas constant 8,134 [J/molK] .
291
(10)
The basic electrothermal constant is described as

B
r
P3 ; [Km3J-1].
c
Second m aterial
(11)
800
700
Based on these values the insulation resource can be

presented as
T BE P e P T P E .
2
Resurce R, h
R R eTa
600
(12)
It can be noticed that at infinitely high conversion

temperature the scale time lim R R~ .
500
R given
400
R calculated
300
200
To determine the rates of three values R, Ta, B, one

needs at least three experiments to be performed at three
differing conversion temperatures and according to the
test results it is necessary to solve system of equations
according (11) equation. The solution of the system
equations gives parameters R, Ta, B. It is noted that
there can be done large number of experiments, so the
solutions of equations may also be a very large number.
A.N. Stepanov conducted a series of experiments to
verify the conversion theory of insulation ageing [7].
There was investigated the breakdown time for two
materials for temperatures +20 C and +150 C, and
electric field strengths from 4 MV/m up to 12 MV/m
with step of 2 MV/m. Under such conditions duration of
tests made up from 2,7 to 728,5 hours [7].
According A.N. Stepanov tests results, and using of the
above explained methods, the solution of equations
according (11) gives us the parameters of digital
resource model. For the correct parameters evaluation
and selection it is necessary to perform a statistical
evaluation of results.
According A.N. Stepanov experimental results and
choosing the estimated model parameters according to
the lowest average error of approximation, there was
determined such parameters of the model:
First material R1=2,310-3 h, T1a=4633,4 K, B1=2,126
[Km3J-1];
Second material R1=7010-3 h, T1a=3393,7 K, B1=2,006
[Km3J-1].
In figures 1 and 2 are shown insulation resource
according A.N. Stepanovs measurements results and
the calculated ones according to the numerical model
where two parameters, rated voltage and temperature,
are involved.
100
0
1
Test No.
Fig. 2. Second materials test results compared to the

calculated according digital model
In figure 3 and 4 are given materials model marginal

insulation resource surfaces.
Fig. 3. First materials insulation resource marginal surface

according to electric field strength and temperature
First material
700
600
Resource R, h
500
400
R given
R calculated
300
200
Fig. 4. Second materials insulation resource marginal surface

according to electric field strength and temperature
100
0
1
The prognosis of electrical equipment insulation

resource utilization can be determined by the recorded
integral characteristics of the overvoltages impact [10].
Test No.
Fig. 1. First materials test results compared to the calculated

according digital model
292
More detailed resource usage model could be arranged

if the two parameters of numerical model of resource
were applied, where the integral characteristics of the
overvoltages impact and the thermal component are
implied.
5.
Gibeault, J. P. On-line monitoring of key fault

gases in transformer oil. 10th International
Symposium on High Voltage Engineering,
Montreal, 1997. P. 62-71.
6. Hamed, M.; and El Desouky, A.A. Computerized
inspection for the high voltage insulating surfaces.
Electric Power Systems Research, 2000, Nr. 53. P.
91-95.
7. Gerhards, J.; and Temkins, A. Prognostics of
electrical insulation resources: Nordic Insulation
Symposium Stockholm, 11-13 June 2001. P. 125132.
8. Cavallini, C.M.; and Montanari, G.C. On-field
partial discharge measurements: A feasible
approach to noise rejection and defect assessment.
The 9th INSUCON International Electrical
Insulation Conference Messe Berlin, Germany, 1820 June 2002, Berlin, 2002. P. 183-188.
9. Morais D. R.; and Rolim J. G. A hybrid tool for
detection of incipient faults in transformers based
on the dissolved gas analysis of insulating oil. IEEE
Transactions on Power Delivery, April 2006,
Vol.21, No.2. P.673-680.
10. Gudius, Saulius; Markeviius, Linas Andronis;
Morkvnas, Alfonsas; Navickas, Algimantas
Stanislovas; Stanionien, Renata. The Analysis of
the Integral Characteristics of the Overvoltages
Effect Duration // EMD 2006: Proceedings of XVI
International Conference on Electromagnetic
Disturbances, September 27-29, 2006, Kaunas,
Lithuania, Kaunas: Technologija. 2006. ISSN
1822-3249. P. 72-77.
4. Conclusions
1.
2.
It was arranged digital insulation resource model

where the operating voltage and temperature are
included.
According A.N. Stepanov experimental results it
was calculated parameters of mathematical model
and assuming the lowest average error of
approximation the marginal insulation resource
digital model parameters were set.
5. References
1.
2.
3.
4.
, ..; , ..
. : , 1979.
304 .
, . .
.
:
, 1988. 128 c.
Kreuger, F. H. Partial discharges detection high
voltage equipment. London: Butterworth & Co.
Ltd. 1989, 192 p.
Aubin, J.; and Gervais, P. On-line monitoring of
key fault gases in power transformers and shunt
reactors. Sixty-second Annual International
Conference of Doble Clients, 1995. P. 122 135.
293

THE MODELS OF SOURCES OF ELECTROMAGNETIC INTERFERENCES

OVER UHV LINE FOR TRANSIENT PROCESSES INVESTIGATION
Linas A. MARKEVIIUS, Vytautas IOINYS
Abstract: the conditions of surge development due to

lightning strike are described in the article. The insulator
breakdown model error is assessed comparing
calculation results with electrode system rod rod
experimental voltage time characteristics. The fault
source model is described for transient electromagnetic
processes analysis.
Keywords: electromagnetic transient, insulation,
lightning, electromagnetic interference
1.
ts
(1)
there u t - the gap voltage; g - the gap length in m; Bconstant 0.42 for positive voltage and 0.5 for negative.
The leader propagation time tl could be calculated from
[4]:
u t
dl
k l u t
E str ,
g
l
dt
Introduction
The assumption, that the fault source is the step voltage

with rise time equal to calculation time step is made for
transient processes calculation in electric power
network. However, that simplicity is narrowing
understanding about processes. It is worth to mention,
that lightning current flow overcome difficult charge
generation and neutralization processes via grounding
impedance. Additionally, lightning current induced
surge overlap insulator, which breakdown mechanism
overtake streamer and leader propagation time.
Probably, easiest way is to describe the lightning current
magnitude, rise time, impulse time as the stochastic
processes with exponential distribution. However, fault
place circuit and the surge impedance could be modeled.
So, the factors, which make influence on lightning surge
and its parameters, should be known for the
electromagnetic transient processes investigation.
2.
0.5
,
u (t )
B
g
(2)
there l - the leader length in m; E str - initial electric

field strength in the gap, before leader formation; kl leader coefficient or ion mobility [3], for the positive ion
kl 1,2 m 2 kV 2 s 1 ,
for
the
negative
ion
kl 1,3 m 2 kV 2 s 1 .
The gap flashover time, according to [24] and [25]:

t tc t s tl tn ,
(3)
there t c - corona inception time in s; t n - the ion

neutralization time in s.
Corona inception and ion neutralization time are
neglected because of the shortest in all gap flashover
mechanism.
Electric field strength inside the streamer according to
[1, 4, and 5] varies in range for the positive
5 kV / cm , and for the negative

discharge E str
Insulators flashover mechanism
Ionization process in the gap of electrode system rod

rod do not start if the applied voltage is less than corona
inception voltage. The higher intensity and non uniform
electric field strength area is created by space charge
due to increased voltage which is higher than corona
inception voltage. The later on, voltage increase raises
higher intensity electric field strength area till leader has
formed. This area could be described as streamer
propagation time t s [6]:
6 10 kV / cm . However the voltage gradient

one E str
10 kV / cm .
for the pulse voltage could seek E str
The assumption is made for leader propagation time
calculation. The idea is that the leader stops develop,
than the voltage gradient inside the unabridged part of
gap drops below value E str . If the applied pulse voltages
rise time f is short, than gap flashover time become
equal to voltage pulse time. The voltage magnitude
294
The calculation results, dependences of leader length

upon time, after impulse (1.5/40 s ) voltage was
applied, depicted in figure 1.
However another way to assess insulators flashover
dynamics is to use the fifty percent impulse flashover
voltage upon the time characteristic. The electrode
system rod rod voltage - time characteristics are
equivalent to UHV line insulator voltage time
characteristics, as it could be seen from [5, 6]. The
experimental electrode system rod rod voltage time
characteristics after pulse voltage (+1.5/40 s ) was
applied are depicted in figure 2.
The leader propagation dynamics equations (1-3)
simulation results satisfy electrode system rod rod
voltage time characteristics for full wave voltage
(+1.5/40 s ).
The leader propagation calculation (2) starts after
overall breakdown time t t s , there t s is the streamer
propagation time from (1).
should be raised to seek gap flashover. After that, than

the pulse voltage rises till maximum value, the space
charge move from rod and the voltage gradient
significantly decreases.
At the same time leader
propagation is unfavorable due to decrease of generated
amount of charge. The flashover propagation starts near
the impulse voltage maximum at the voltage u U 50%
and its rise time equal to critical f kr . Decrease of
voltage rise time leads to gap flashover over the voltage
maximum. The gap average electric field strength
increase due to increase of impulse voltage magnitude
and the leader propagation time decrease. In our point of
view, the gap flashover starts at the maximum of
impulse voltage and at the rise time for the higher
voltage magnitudes as it is depicted in figure 1.
6
2 x 10
a)
u(t)
u(t)[50%], s=100cm
3. The lightning as the source of

electromagnetic wave in transmission line
0 -6
10
-5
t,s
-5
t,s
10
From 60 to 100% of all lighting starts from negative

charged clouds as it is described in [2, 5]. Step by step
streamers in space and time forms leader which is heated
up to a few thousands degree of Celsius. The same sign
as cloud charge, space charge is generated by floating
ions in conductive medium. After leader seek the ground
the space charge surrounding channel neutralization
process starts, due to formed new streamer with speed
v 2 10 7 1,5 108 m / s . So the resistance of the lightning
channel at the stage of formation and neutralization
could
be
described
by
Tepler
equation
RK 0,08 / q [ / m] , there q charge, formed in
lightning channel.
1.5
b)
l(t)
0.5
0 -6
10
10
Fig. 1. Leader propagation upon time (b), after applied

different magnitude impulse voltage (+1.5/40 s ) (a)
2.4 x 10
S=100 cm
S=150 cm
S=200 cm
2.2
2
x 10
1.8
Um, V
I max
1.6
1.4
1.2
Teup
Te
0.8
0.6
0
0.2
0.4
0.6
0.8
1.2
1.4
-5
t, s x 10
Ted
1
Fig. 2. The fifty percent full wave impulse (+1.5/40 s )

flashover voltage - time characteristic for the gap rod rod
Tf
0
0
0.5
1.5
2.5
3.5
4
x 10
-5
Fig. 3. Typical lightning discharge current
The leader propagation speed and duration yield

expression (2) after numerical integration for example
with 4th order Runge Kutte method. Numerical
integration of expression (2) starts, than the condition
u t U 0 is met. There U 0 is the initial corona voltage.
However another assumption was made voltage drop
u t across gap through propagated leader are neglected.
The minimal lightning current rise time could be

calculated T f H l / c 600 / 3 108 2 s assume, that the
minimum cloud height from ground surface is
H l 600 m . The typical lightning discharge current is
depicted in figure 3. There [2] Te up - electron spreading
time in streamers, formed in upper lightning channel
295
layers on the moment of channel neutralization; Te electron spreading time in streamers, formed in the
middle of channel; Te d - electron spreading time in
there - time constant, proportional to discharge

duration, the time then discharge starts to discharge
current seek 37% of magnitude;
- correction coefficient.
The time constant T f could be evaluated from [4]:
streamers, formed in the lower layers of channel.

The lightning discharge current could stop flow on time
T f , if the lightning channel could seat all neutralizing
charge. However it is not possible and the opposite sign
than lightning channel streamers are formed from
ground surface. So the electromagnetic wave spreads
from ground surface to the centre of charged cloud due
formation of streamers.
The lightning current depends on sign of discharge,
clouds height from ground surface [2]:
(4)
(5)
I max
2 0 ve E str
E str
H l 10,8 H l ,
I max
4 0 ve E str
E str
H l 21,7 H l .
0,828
0,0834 I max
, I max 20 kA;
,
Tf
0,624
0,154 I max
, I max 20 kA.
Discharge current magnitude is the main criterion for

impulse rise time calculation us it could be seen from
(12). Commonly statistical approach is used for
assessment of discharge current magnitude. However,
equation (11) mostly is used for statistical analysis of
lightning process. The influence of full discharge circuit
should be evaluated and equations (4-10) involved to
electromagnetic transient on power network analysis.
The current steepness [22]:
4.
a max
a max
I max
Tf
I max
4 0 ve v E str
E str
6,5 109 A / s .(7)
Tf
The voltage gradient in streamers, electron movement

speed ve 1,5 105 m / s and wave propagation speed in
lightning channel make influence on lightning current
steepness as it could be seen from (6) and (7).
The discharge process change its view then lightning
strikes to the high resistance earthed object. The
potential difference between leader space charge and
lightning channel decreases due to increase of potential
on earth resistance. The decreased potential difference
create slighter gradient. As the result charge
neutralization process become slower and decreases the
lightning current magnitude. The lightning current pulse
duration increases, because the amount of total charge
stays constant. Influence of equivalent earthing
resistance z to the lightning current magnitude,
steepness, impulse rise time is described as [2]:
z
a Req a Req 0 / 1
z
lightning
z
,
T f Req T f Req 0 1
z lightning
there z lightning 300

lightning
channel
u u0 z I ,
there
(9)
I max
t / T f 2 e t / ,
1 t / T f 2
z lightning z line / 2
- wave
z zlightning / 2 .
(14)
(10)
Then the wave propagates approximately 100 m
distance, insulation breakdown after t 3,3 ns as it was
described in part 2 starts. After insulator breakdown,
tower and ground impedances should be involved in to
the equivalent lightning channel surge impedance:
surge
impedance.
The Heidler function [4] describes negative polarity
lightning current:
i t
2 z line / 2
(13)
refraction index; z line 250 400 - line surge

impedance.
Equivalent lightning channel surge impedance before
insulation breakdown:
(8)
UHV transmission line with rope
The earthing resistance makes influence on lightning

discharge current shape, as it was described above.
Thats why processes at the fault place should be well
known for fast transient processes over high voltage line
investigation. The typical UHV line consists of phase
conductor, steel or ferroconcrete tower and rope. Three
calculation schemes are used for lightning generated
transient processes calculations:
1. The lightning overlap phase wire, after that
lightning overvoltage breakdown insulator at
the tower;
2. The lightning overlap the tower, after that
lightning overvoltage breakdown insulator;
3. The lightning overlaps the middle of the rope,
after that lightning overvoltage breakdown the
gap between rope and phase wires.
Than the lightning overlap phase wire, the voltage at the
fault place:
2 0 ve v E str
E str
3,25 109 A / s (6)
z
I Req I max / 1
zlightning
(12)
z 1 z 2 Rearthing / Rearthing z line ,
(11)
296
(15)
there Rearthing 10 - tower grounding impedance.
5.
Respectively, after the marginal value of impedance

involved to equations (14) and (15) we acquire
z 70 120 , z 1 5 7 .
After lightning overlap the tower, the equivalent surge
impedance at the fault place:
The research was supported by Lithuanian State Science

and Studies Foundation.
6.
1.
z z lightning Rearthing / Rearthing z lightning . (16)
After an insulator breakdown, equivalent fault place

impedance:
z 1 z lightning z x / z x z lightning ,
2.
(17)
there z x z line Rearthing / Rearthing z line .
3.
After the marginal value of impedance involved to

equations (16) and (17) we acquire z 9,7 ,
z 1 9,4 .
For the third transient calculation scheme, equivalent
fault place surge impedance before an insulator
breakdown:
z z lightning z rope / z rope 2 zlightning ,
1.
After an insulator breakdown:
(19)
2.
The marginal values of impedances for equations (18)

and (19) yields z 120 , z 1 120 150 .
It is worth to mentioned, that the lightning current surge
shape is influenced by equivalent lightning channel
surge impedance.
The surge wave is half wave, then the lightning strikes to
phase wire. The lightning current steepness could be
evaluated according (9) and (14). The impulse time
would be restrained by the time from lightning current
flow starts till insulator breakdown as depicted in figure
1. The lightning current magnitude is calculated
according (9), (11) and (14).
The surge voltage in the tower rises slowly due to many
second wave reflections from grounding, then the
lightning strikes to tower. The insulation breakdown
after voltage magnitude seeks the breakdown voltage.
The surge in the line rise time is equal to insulator
breakdown time.
Equation (8) and (18) describes the lightning current
magnitude, then lightning strikes to the middle of the
rope. The surge rise time is proportional to insulator
breakdown time leader propagation time, as depicted
in figure 1b.
3.
there z x1 2 z rope zline / z line z rope .
4.
5.
6.
297
Conclusions
The experimental rod rod voltage time

characteristics with in high accuracy could be
modeled using equations which describes
leader propagation dynamics.
The general commutation law states, that
charge can not change on bounce. The
equivalent surge impedance before and after
insulation breakdown should be evaluated for
precise surge rise time evaluation.
The factors which make influence on lightning
current rise time, impulse time and magnitude
are described above. Because of that, these
factors should be appended to classical
lightning current versus time equation for the
electromagnetic
transient
processes
identification task.
7.
(18)
there z rope 400 rope surge impedance.
z 1 z lightning 2 / 1 z x1 zlightning ,
Acknowledgment
References
, . .; , . .; , .
.
.
, 75, . 9, 2005, p. 4451.
, . .
: .
, 76, . 12,
2006, p. 101-105.
, . . .
, , 1975.
Dudurych, I. M.; Gallagher, T. J. EMTP analysis of
the lightning performance of a HV transmission
line. IEE Proc. Generation Transmmission
Distribution. Vol. 150, No. 4, July 2003, p. 501506.
, . . .
, , 1973.
Shindo, T. Suzuki, T. A New Calculation Method
of Brekdown Voltage Time Characteristics of
Long Air Gaps. IEEE Trans. on Power Aparatus
and Systems. Vol. PAS-104, No. 6, June 1985,
p.1556-1563.

APPLICATION OF ACTUAL POWER DEFICIENCY VALUE ON

LOAD-SHEDDING
Vladimir CHUVYCHIN, Nikolay GUROV, Santa KIENE
Riga Technical University, Faculty of Power and Electrical Engineering, Latvia
activate spinning reserve quickly enough to restore the

system to its normal operating frequency, frequency
actuated automatic load shedding (UFLS) serves as a lastresort tool to prevent the system from collapse [2,3].
Existing UFLS automation has drawbacks, which limit
adaptability of emergency automation to a change of
under-frequency situation in a power system. UFLS
tripping frequency settings are selected for some specific
emergency situation, which is considered more probable
for specific power system. It is not possible to foresee all
situations that can occur in the power system. Problem of
value of a load to be shed is also very topical [4].
This papers deals with the determination of necessary
amount of load to be shed in order to improve the operation
of under-frequency load shedding system during real
emergency time.
Abstract: Paper considers the emergency load shedding

process in power system using the calculated active
power deficiency value as control impact.
Power deficiency value is continuously determined on
the busbars of power plants during real time.
Disturbance in power system related with deficiency
appearance and subsequent frequency decline is
eliminated with the operation of load shedding system.
Calculated active power deficiency is implicated in the
under-frequency load shedding system.
Keywords: active power deficiency, under-frequency
load shedding.
1. Introduction
Frequency is one of the most important parameters
describing power system, and frequency control is one
of the main power system control tasks [1].
Successful operation of power system requires that
frequency is constant and close to nominal value.
Frequency deviation limits and duration is rationed and
different for separate power systems. Parameters for
normal and emergency operational conditions in
different power systems are quite distinctive. Distinctive
is assessment of quality of electrical energy, as well as
the allocation and characterization of severity of
frequency
emergency
operational
condition.
Requirements for electrical energy quality are more
rigorous in European (UCTE, NORDEL) power
systems in comparison with Russian IPS/UPS.
Frequency is determined with the balance of generated
and consumed power. Any change in power balance
causes frequency deviation [1].
At normal operational condition frequency within
acceptable limits is kept by providing primary and
secondary frequency control, which accomplishes
generation control at power plants.
At emergency conditions, when fast frequency changes
take place, in order to eliminate frequency drop and to
prevent frequency avalanche the involvement of
spinning reserves or under-frequency load shedding
automation are activated. If governor action cannot
2. Power deficiency determination method

This section is devoted to the description of power
deficiency determination method.
Power deficiency determination is provided for single
machine power system considering the equation of
motion, which is expressed as follows:
TJ (d/dt) = Pmeh - Pel,
(1)
where TJ generator inertia time constant in seconds,

rotor speed deviation in per units, Pmeh and Pel
mechanical and electrical power at current frequency in
per units accordingly.
When there is a load change, it is reflected
instantaneously as a change of electrical power of the
generator. This causes a mismatch between the
mechanical and the electrical power which in turn
results in speed variations as determined by the equation
of motion (1). System frequency varies depending on an
active power [5].
However, use of expression (1) is not rational for
determination of power deficiency, since the power
balance between mechanical and electrical power can
reinstate at frequency different from nominal value.
Here, power system will be balanced, but at emergency
deficient operational condition, reason of which is the
298
dependence of values Pmeh and Pel on frequency. In

order to restore the power balance at nominal frequency,
current parameters of equation (1) are to be reduced to
nominal frequency, and then the power deficiency value
could be determined at any level of frequency.
As it is known, the governor speed droop R
characterizes the ratio of speed deviation () or
frequency deviation (f) to change in power output (P)
[5]. Mechanical power change due to frequency
deviation can be expressed as Pmeh = (-/R). Also
the power system load has the frequency sensitivity,
which is characterized with the load damping constant
D. Load change after frequency deviation can be
expressed as Pel = D [5].
Equation of rotor motion in an extended way taking into
account the effects of governor speed droop R and the
frequency sensitivity of load D on the net frequency
change can be expressed as:
As it is seen from Fig. 1, most generation is

concentrated in eastern part of power system, though
power consumption is located mostly in western part of
power system causing power flows at great distance.
In order to analyze the operation of UFLS in
mathematical power system model the emergency
situation is examined, when at time moment t = 50s the
generated power of amount 120MW, 200MW, and
300MW is tripped at power plant Nr.3, which causes
active power deficiency of 11%, 19%, and 28% of total
power system load.
UFLS operation is examined both for existing operating
principle and new the active power deficiency
application principle.
3.1. Operation of UFLS under existing principle
To simulate the operation of existing UFLS the load
shedding system used in Baltic countries was modeled.
UFLS in Baltic countries consists of different types:
UFLS-I, UFLS-II and special setting of UFLS.
The aim of the special setting of UFLS is to increase the
spinning reserves of power system at first moment of
frequency emergency by shedding less important
consumers. Practice shows that amount of load
connected to the special setting of UFLS is in range of
3-10% from maximal power system load and actuation
time is in range of 0.2-0.5s.
UFLS-I is fast-acting automation, the aims of which are
fast preventing of deep frequency drop by shedding
consumers with small portions, as well as keeping
frequency within allowable frequency-time limits.
Practice shows that amount of load connected to the
UFLS-I automation is in range of 30-50% from
maximal power system load. Actuation time is in range
of 0.1-2.0s, usually the settings of 0.4s or 0.5s are used.
Actuation frequency is in range of 49.1-47.0Hz.
UFLS-II is slow-acting automation, the main purposes
of which are the automatically restoration of frequency
to rated values by shedding consumers with small
portions and large actuation time, as well as
automatically restoration of frequency to rated value in
order to synchronize parts of power system. Practice
shows that amount of load connected to UFLS-II
automation is in range of 12-40% of maximal power
system load, and actuation time settings are in range of
10s up to 80-120s.
It should be noted that UFLS-II automation is used only
in the power systems of former USSR, in other power
systems frequency restoration is provided manually by
dispatcher instructions. In most of European power
systems only fast-acting UFLS system for preventing
deep frequency decline is used.
For simulation of frequency emergency processes it is
sufficient to consider only fast activation of spinning
reserves (primary frequency control), which is involved
at emergency initial period. Spinning reserves are in
average about 3-5% of actual power system load and
they are evenly distributed along the power system.
Further, frequency response processes in power system
considering operation of UFLS are simulated using data
of UFLS automation of Baltic countries in 2003. In
addition, in this paper is simulated only prevention of
TJ (/dt) = (Pmeh /R) (Pel + D). (2)

This means that an increase of system load results in a
total generation increase due to governor action and a
total system load reduction due to its frequency
sensitive characteristic.
Consequently, from expression (2) the power deficiency
value taking into account frequency change can be
derived as
P = TJ(d/dt) + /R + D.
(3)
Power deficiency value from expression (3) allows

determining the exact power impact value to the power
system in order to restore the nominal frequency.
3. Mathematical simulations of load shedding
This section deals with the analysis of under-frequency
load shedding (UFLS) operation in power system
consisting of three parallel operated thermal power
plants connected via long 330 kV power transmission
lines with remote load centers. Operational condition of
considered power system with respective parameters is
represented in Fig. 1.
L7
UFLS
1 km
L1
UFLS
1 73 k m
L8
Nr.1
77MW
Nr.2
200MW
L2
P2-6 =1 5 0MW
144MW
380MW
45MW
P6 -5 =29 7MW
P5-1 =30 7MW
P2-6 =18 8M W
1 51 k m
62km
L6
22 k m
L5
UFLS
UFLS
70 km
P4 -5 =154 MW
144MW
144MW
L3
P3-6 =26 9M W
1 50 k m
P3-2 =1 8 7M W
1 km
Nr.3
787MW
P3 -4 =3 06MW
16 7 k m
L4
UFLS
25MW
144MW
Fig. 1. Examined power system model
Detailed model of power system is provided in PSCAD

software including speed governors and excitation
systems for all power plants.
299
deep frequency decline, which is realized with operation

of special setting of UFLS and UFLS-I. Operation of
UFLS-II here is not analyzed.
Amount of load connected to UFLS automation in
power systems of Baltic countries in year 2003 is given
in Table 1.
some overcorrection takes place and after about 7

seconds frequency stabilizes at value 50.07Hz.
Emergency trip of 300MW causes frequency drop to
47.32Hz and actuation of 7 settings of UFLS
automation, which perform load shedding in amount of
428MW. Redundant tripped load creates 0.5Hz overfrequency situation in power system.
Change of active power deficiency taking into account
operation of UFLS under current principle is
represented in Fig. 3.
Table 1. Amount of load connected to fast-acting UFLS

automation in power systems of Baltic countries in 2003
Units
Amount of load at UFLS

Estonia Latvia Lithuania
Load maximum
MW
1200
1200
1580
51.6
Special setting of
UFLS, fact =
49.2Hz, tact = 0.5s
UFLS-1, fact =
48.8 47.2Hz, tact
= 0.5s
MW
36
36
47
50.8
MW
504
504
664
42
42
42
Frequency (Hz)
Parameters of
UFLS
50
49.2
48.4
47.6
46.8
39
In power systems of Baltic countries special setting of

UFLS consists of one setting with connected load of 3%
from load maximum. Actuation frequency fact = 49.2Hz
and actuation time tact = 0.5s.
Current UFLS-I has 9 settings with actuation frequency
fact = 48.8 - 47.2Hz and setting step 0.2Hz. Actuation
time of each setting is tact = 0.5s. Total amount of load
connected to automation in percents of forecasted
maximum is 42%, in addition, average load of one
setting is 4.67%.
In experimental model realized in PSCAD software the
data of Latvian UFLS automation are used, and control
actions are performed in 5 load centers (substations).
Load connected to special setting of UFLS at each
substation is following:
PUFLS(i) = PUFLS_sp /k = 36/5 =7.2 MW,
40
41
42
43
44
45
46
47
48
49
50
Time (s)
200MW trip
130MW trip
300MW trip
Power deficiency (MW)
Fig. 2. Frequency change during emergency processes

considering the operation of UFLS under existing principle
360
300
240
180
120
60
0
-60
-120
39.5
40
40.5
(4)
41
41.5
42
42.5
43
Time (s)
200MW trip
where PUFLS_sp total amount of load connected to

special setting of UFLS (from Table 1), k = 5 number
of substations prescribed for operation of UFLS.
One setting of UFLS-I at each substation disconnects
load in amount of
130MW trip
300MW trip
Fig. 3. Active power deficiency change after emergency

processes considering the operation of UFLS under existing
principle
After emergency impact deficiency sharply increases

close to the tripped capacity value. Deficiency change
process characterizes the operation of existing UFLS
automation.
Frequency restoration and deficiency elimination
processes using current UFLS automation of Baltic
countries showed that UFLS automation is not adaptive
during different emergency situations. The main
problem is the proper selection of load to be shed.
PUFLS-I(i) = PUFLS-I /(nk) = 504/(95) =11.2 MW, (5)

where PUFLS 1 total amount of load connected to
UFLS-I automation (from Table 1), n = 9 amount of
UFLS-I settings.
From (5) it can be determined that one setting of UFLSI realizes disconnection of load in amount of 56MW.
Frequency response during emergencies as trip of
120MW, 200MW, and 300MW generation capacity at
power plant Nr.3 taking into account operation of UFLS
under existing principle is showed in Fig. 2.
Emergency trip of 120MW causes frequency drop to
48.94Hz, which actuates special setting of UFLS
automation, which in turn sheds 36MW of load.
Remainder deficiency of about 84MW keeps frequency
at 49.6Hz level. Emergency trip of 200MW causes
frequency drop to 48.34Hz and actuation of 4 UFLS
settings. UFLS automation performs load shedding in
amount of 204MW. Frequency restoration process with
3.2. Operation of UFLS under new principle

New operational principle of UFLS automation is based
on the application of calculated deficiency (3) to the
UFLS automation as actuating parameter.
To avoid frequency drop in considered power system
below 49.7Hz, deficiency value for actuation UFLS
automation is determined in accordance with following
expression [1]:
300
PUFLS_act = (f P)/(f0 s) = 60 MW,
(6)
n =P /PUFLS_act,
Power deficiency (MW)
where P total power system generation (1060MW),

f frequency deviation (0.3Hz), f nominal
frequency, s average droop of all rotation speed
governors in power system (10%).
Number of actuated settings is determined as follows:
(7)
360
300
240
180
120
60
0
-60
39.5
where P actual calculated deficiency value (3) in

power system.
One setting of UFLS is ordered to shed the amount of
load equal to actuated deficiency value, which is 60MW
according to (6). One setting at each substation is
ordered to shed the load in amount of
PUFLS(k) = PUFLS_act/k = 60/5 =12 MW.
40.5
41
41.5
42
42.5
43
Time (s)
200MW trip
130MW trip
300MW trip
Fig. 5. Active power deficiency change after emergency

considering the operation of UFLS under existing principle
Efficiency of operation of new UFLS automation can be

characterized with the depth of frequency drop, duration
of power deficiency elimination and frequency
restoration process, and restored frequency value.
Above mentioned parameters of the efficiency criteria
were improved with operation of new UFLS.
(8)
Therefore, appearance of 120MW of active power

deficiency according to (7) actuates 2 settings of UFLS,
which performs load shedding of about 120MW.
Appearance of 200MW power deficiency in power
system causes actuation of 3 UFLS settings. Load to be
shed is 180MW or 90% of deficiency. Appearance of
300MW power deficiency in power system causes
actuation of 5 UFLS settings. Load to be shed is
300MW or 100% of deficiency Frequency response
during emergency process taking into account operation
of UFLS under new principle is showed in Fig. 4.
Actuation time of settings tact is assumed 0.5s.
4. Conclusions
Under-frequency load shedding operation during
emergency situation was analyzed using current
operating principle of UFLS automation in Baltic power
systems, as well as new suggested by authors active
power deficiency application principle.
Results showed that UFLS operating under new
principle more efficiently prevents frequency drop,
faster and more accurately restores frequency to rated
values, as well as prevents from frequency
overcorrection.
Proper selection of load to be shed can improve the
operation of UFLS automation preventing deep
frequency drop and avoiding over-frequency situation
after control.
51
50.5
Frequency (Hz)
40
50
49.5
49
48.5
48
47.5
39
40
41
42
43
44
45
5. References
46
Time (s)
200MW trip
300MW trip
1. Sterninson L.D., Transient Processes During Power

System Frequency and Power Control, Moscow,
Energija 1975, 216 p (in Russian).
2. Rabinovich R.S., Automatic Frequency Actuated
Load Shedding System, Moscow, Energija, 1980,
344 p (in Russian).
3. IEEE Status Report on Methods Used for System
Preservation During Underfrequency Conditions.
IEEE Transactions on Power Apparatus and
System, Vol. PAS-94, 1975.
4. Chuvychin V., Gurov N., Rubcov S., Adaptive
underfrequency Load Shedding and Underfrequency
Load Restoration System, International Conference
Power Tech05, Saint-Petersburg, 2005.
5. Kundur P., Power System Stability and Control,
Tata McGraw-Hill, New Delhi, 2008.
6. Chuvychin V., Gurov N., Kiene S., "Frequency
Control and Determination of Controlling Actions to
Enhance Stability of Power System", Proceedings of
4th Scientific Symposium "Elektroenergetika-2007'',
Slovakia, pp 709 712.
130MW trip
Fig. 4. Frequency response during emergency process

considering the operation of UFLS under new principle
Operation of UFLS automation limits frequency drop to

49.2Hz, 48.8Hz, and 48.1Hz after emergency trip of
120MW, 200MW, and 300MW generated capacities
accordingly. Fast load shedding and frequency
restoration process is followed. After about 4 seconds
frequency is restored to acceptable limits for all
considered emergency situations.
Change of active power deficiency considering
operation of UFLS under new principle is showed in
Fig. 5.
In order to decrease possible inaccuracy of deficiency
calculation, initial value of calculated deficiency
provided to the UFLS automation could be decreased
for 20-40% and selected with portions using the law of
geometrical progression [6].
301

TECHNICAL ADVANCE IN THE AUTOMATION OF DISTRIBUTED NETWORK

CONTROL
Virginijus RADZIUKYNAS, Arturas KLEMENTAVIIUS, Sigitas KADIA, Algimantas LEONAVIIUS, Neringa
RUTKAUSKAIT, Antanas NEMURA, Feliksas BIELINSKIS, Viktorija NEZVANOVA
automatically. Operators really want to monitor the grids

and to respond to troubles, less to control.
Despite of the market forces pressing on DNOs to
become more operationally efficient, there are several
barriers for long-term extension of DA and practical
steps onto smart grids. As it seems, the psychological
and awareness barriers are principal ones, associated
with missing of the visionary views. DNOs often too
much adhere to risk management strategies that
overshadow the innovation strategies.
To pursue the objectives of intelligent grids, a lot of new
devices and ICT technologies should be integrated. This
is a challenge for DNOs how to realize such
fundamental changes. One of their responses the
architecture of intelligent grids shall be developed on
plug-and-play basis. The concept of plug-and-play
is probably best defined by Microsoft Corporation: it
denotes A combination of hardware and software
support that enables a computer system to recognize
and adapt to hardware configuration changes with little
or no intervention by the user (www.microsoft.com)..
Nevertheless, DNOs use connotation of this concept
similar to effortless for a user and sometimes expect
the ICT and equipment industries should do all job in
developing and implementing major solutions..
It is observed that sometimes names of concepts might
give a pretext to DNO personnel to underestimate the
concept and visions. For instance, the name of
automated metering reading (AMR) system is often
understood as AMR only, with scepticism for other
capabilities to be get out of the system. Anyway, the
clue to resolve the hesitations as mentioned above is to
take a leap of faith and develop consistent visions and
plans for future intelligent grids. As for other barriers,
regulatory views also play significant role. Whereas
distribution business is regulated, the regulators, not
DNOs, set energy tariffs for customers. A DNO cannot
invest to DA projects, unless the tariffs ensure
recovering of capital costs. Another characteristic
barrier in DA field is technical and economical
feasibility of automation projects in rural areas.
Abstract: The paper presents short review of state-of-art

and recent technical advance in automation of
distribution network control. It covers issues of
intelligent grids, substation automation, ICT, automated
meter reading, feeder automation, SCADA. A number
of new automation technologies are referred to. With
permanent technology changes in process, the
distribution networks move towards digital distribution
networks,
with
widespread
intelligence
(microprocessors, controllers) across the networks and
high-speed communications between intelligent devices.
Both expectations and reservations from distribution
network operators related to further automation are
discussed.
Keywords: distribution networks, automation, technical
advance, intelligent networks, control, communications.
1. Introduction
Network operators and power utilities are transitioning
from labor-intensive electromechanical power system
designed and built in the mid-20th century to the
advanced technology of 21st century; nowhere is this
more evident as in the distribution system [1].
The advanced distribution technology relies on distribution automation (DA) and is aimed at higher reliability of power distribution. The improvements of DA are
steadily bringing it closer to its ideal functionality: to
monitor, protect and control all processes and equipment
in distribution substations and on feeders remotely, with
minimum intervention of supervising dispatchers.
2. Contemporary problems of automation and
innovations in distribution networks
First and foremost, distribution network operators
(DNOs) quite positively envisage the further automation
of networks and their progress to intelligent (smart)
grids. Sometimes operators prefer monitoring the
network versus operating the network. It means that
more activities are expected to be performed
302
The moving to more intelligent grids promises for

DNOs to get more out of the grid to mine additional
distribution capacities from the same grid infrastructure.
Meanwhile customers expect to benefit another value
to get a higher reliability of delivery, particularly those
in low-density areas. However, the operation of
intelligent grid will call for new skills and new
knowledge from operating staff. There are real
challenges for DNOs to integrate information from
variety of sources and to cope with existing information
systems. Actually, some standard branches of DNOs are
rather slow for innovations. They face data overload
and, with the growth of information flows, their ability
to interpret the data has decreased. On the other hand,
there is a technical expansion need, mainly for standard
protocols, more communication bandwidths, and more
communication tools for remote sites [10].
3) solid-state (semiconductor-type) relays;

4) digital protection relays and microcontrollers;
5) programmable logical controllers (PLC) and IEDs,
integrated into local area networks (LAN), with
advanced logic, monitoring (including video
surveillance), multiple control functions, sensing and
extended communications between substations and
between RTU and SCADA masters and central stations;
6) digital distribution substations of the next future
featuring smart sensing and telemetering, and data
traffic in various high-speed communication media, high
intelligence, smart decisions, automated or not,
extended monitoring with two-way communication
between distribution substations, local substations
(MV/LV), feeder switches, distributed generators,
electricity storage units and customer loads.
4.1. Substation intelligence
3. Intelligent networks
If intelligence in substations is referred to as digital

intelligence (computer intelligence), in the period of
electromechanical and solid state relays there was no
substations intelligence, only logic of protection and
control. This logic was composed from a number of
relays, each having one or two protective functions,
reaching sometimes a rather complex logic.
In a digital relay, many functions are implemented by
the microprocessor programming. A digital relay can
analyze whether the relay should trip or restrain from
tripping based on current or voltage magnitude, complex
parameters set by the user, relay contact inputs, the
timing and order of event sequences.
The modern versions of digital relay will contain
advanced metering and communication protocol ports,
allowing the relay to be a focal point in SCADA system.
Digital relays are clasiffied to IEDs microprocessorbased controllers of power system equipment. Other
types of IEDs include load tap changer controllers,
circuit breaker controllers, capacitor bank switches,
recloser controllers, voltage regulators etc.
Typically,
IED contains around 5-12 protection
functions, 5-8 control functions controlling separate
devices, an autoreclose function, self monitoring
function, communication functions etc.
Several IEDs can jointly control a distribution
substation. Another control alternative is PLC. In
contrast to IEDs, PLC relies on a processor, not on
microprocessor. It contains more extended software and
much more functions as an IED does. Unlike generalpurpose computers, the PLC is designed for multiple
inputs and output arrangements, extended temperature
ranges, immunity to electrical noise.
Like IEDs, PLC receives inputs from sensors (analog or
digital signals) and power equipment. Where substation
design includes IEDs, inputs come from IEDs as well.
PLC issues outputs to actuators such as relays, electric
motors, pneumatic or hydraulic cylinders etc. Some
input/output arrangements provide external communication with SCADA network (via RTU) and other computer networks as LAN or WAN (wide area network).
Embedding the intelligence into substations needs more
sensors, i.e. devices that measure physical quantities and
The general trend in modern development of DA is its

combination with distributed intelligence. It lays
foundations for upgrading the distribution networks up
to intelligent (smart, digital) grids. Therein electricity is
linked with communications and computer control.
Major features thereof will be densely deployed
sensors, monitors, controllers, high-speed communications and data-trading software. The intelligent grids
will offer such values as greater reliability, increased
capacity, choice, safety and flexibility. Intelligent
networks are expected to be self-healing, with selfmonitoring feeders problems anteriorly they happen.
Since the concept of intelligent network is rather vague
for a lot of customers and even DNO employees, it can
be defined, in simplistic mode, as an electric network,
that will monitor, think, act, repair and prepare itself to
respond quickly to customer needs.
The intelligent network features combination of highspeed computers, intelligent electronic devices (IEDs)
and intelligent software.
The state-of-art of intelligent networks today might be
seen as lying between visions and emerging networks.
Majority of intelligent devices and technologies are
available on the market. The current types of
information systems will continue to overlie networks in
their progress to intelligent ones. The principal systems
seem to be SCADA/EMS (Supervisory Control and
Data Acquisition/Energy Management System),
SCADA/DMS (distribution management system), OMS
(outage management system), AMR systems and WMS
(work management system).
4. Substation automation
Historically, automation of distribution substation
(HV/MV) might be loosely classified to following levels
of technical advance:
1) electromechanical protection and control relays,
initially without a remote terminal unit (RTU) of
SCADA system in a substations;
2) electromechanical relays as data transducers to a
microprocessor-based RTU in a substation;
303
convert them into signals that can be read by intelligent

devices. Here noteworthy seems to be recently
developed so-called Esubstation model (Cooper Power
Systems) a new approach to distribution substation
real-time monitoring of equipment and feeders. Based
on Cannon Yukon software platform, it provides the
dispatchers with a complete real-time picture of
substation using integrated browser-based application
with Cannons equipment sensors.
The sensors located on substations transformers
measure (or monitor, as it is said) loadings, critical
temperatures, cooling equipment status, combustible gas
etc. Those located on substations power circuit breakers
monitor vital functions such as gas pressure, density and
temperature along with the tripping status. Other
Esubstation sensors can be installed in control house to
monitor feeder loading, relay targets, voltage regulation
and batteries [2]. In addition to sensor data, Esubstation
offers a video surveillance opportunity that could
replace on-site surveillance by a field crew.
The IEDs in Esubstation poll the sensors via fiber-optic
(transformer data) and wireless (circuit breaker data).
IEDs exchange data (aggregated and not) with RTU.
IEDs and RTU are integrated into Substation Advisor
Gateway of Esubstation. The Gateway communicates
with dispatchers via SCADA and internet.
There is a lot of innovative, low-cost voltage and current
sensors today in the market. Furthermore, new smart
sensing technologies are introduced, maybe marking a
threshold of a new sensor generation. For instance,
classical current transformers might be replaced by one
new IED model that acts as noncontact device and
simultaneously measures 3 phases. Here classical
sensors would need a contact with the bus works [3].
Similarly, another new current-monitoring device is
used in top-pole applications and also has no contact
with a conductor [2].
As for other smart sensing technologies, at least two of
them rely on insulators. First one is defined as currentand voltage- monitoring insulators for use on switches
and lines to monitor in remote locations. The second one
refers to high-accuracy line-post current sensors that
feature a line-post insulator with an embedded coil to
measure currents through an inductive coupling. As for
contact devices, a faulted-circuit indicator has been
designed that clamps on the phase wire to indicate the
exact location of a fault [2].
It can be credibly envisaged that substation equipment
manufacturers will continue to develop proprietary
technologies. Therefore, from manufacturers and
DNOs perspective, the major problem related to
intelligent substations will be to ensure the interoperability of substation devices [3] (see section 5).
National Grids company (Massachusetts, USA) has

designed an integrated design model for breaker-and-ahalf substation, which is most common in high-density
load areas serviced by the company. The design pursued
objectives to reduce the total-life cost of the secondary
systems
within
substation
and
to
replace
electromechanical devices by IEDs. The design themes
were adopted [4] as follows:
1) modularize the control and protection subsystems
(two formerly vis--vis located protection and control
bays A and B should be integrated into one modular
cabinet, say, bay A located over bay B).
2) maximize the use of protective relay data (data should
be collected once and used many times to provide
information for various sources);
3) minimize the amount of hardware and wiring;
4) improve the functionality of substation user interface;
5) enhance remote diagnostics, data collection and test
capabilities.
As a result, with the fewer cabinets in the modular
design, the footprint of substation was diminished from
17 m2 to 7 m2. The control system design brought cost
saving of 42 % over the previous design.
The new design retained an operator desk with humanmachine interface (HMI), so substation is refered to as a
control house. The computer of operator desk communicates with RTU and PLC within the substations LAN.
Considering the outside of distribution substations, it
seems there is too much wiring and too many devices
there. According to some experts (Mark Dudzinski, GE
Energy) more work is required on defining
communication standards and architecture outside the
substations to support the advanced distribution
infrastructures and the vision of intelligent grids [5].
5. Information and communication technologies
In a vision of intelligent network, a vast amount of the
automated control actions will be co-ordinated by
intelligent devices in real-time. The major condition for
successful co-ordination will be existence of extended
robust communication infrastructure and its high-speed
functioning. Switching decisions, for instance, require
sensor data sampling rates faster than 1 msec.
Bandwidth is a key technical factor for both the amount
and speed of communicated data analog and digital
signals. Any set of data other than simple text message
is very large against a message. Radio medium does not
suit for transferring video (surveillance) data.
State-of-art of communications in distribution
companies covers variety of physical media. Wired
options include twisted pairs (copper cable), telephone
line, coaxial cable, fiber-optic cable, power line carrier
(PLCr) and broadband over power line (BPL). Wireless
options include paging, radio, cellular, wireless fidelity
(Wi-Fi), satellite and microwave. Most of the wired
media have distance limitations, unless the amplifiers
are added. Wireless media often have signal strength
limitations [5].
In a substation environment twisted pair and fiber-optics
are major physical media to carry sensor data to local
meters, controllers and RTUs. First medium is
4.2. Substation integration

Significant benefit of substation automation (SA) is
reduction of substation
space. DNOs (with other
stakeholders in the sector) have undertook customized
substation integration projects aimed at development of
optimal integrated design.
304
inexpensive, but rather sensitive to intensive

electromagnetic interference, contrary to more costly
second medium. Somewhere Wi-Fi and PLCr have been
installed for communications inside a substation.
Computer-to-computer
communications
usually
function on fiber-optic, coaxial or copper cable, because
compu-ters are interconnected typically as a LAN or
WAN.
The RTUs access the SCADA central stations over
dedicated telephone and all other wired media. In rural
areas, pole-top mounted RTUs on rural feeders communicate with central sites via radio, cellular or satellite.
If not overestimated, the BPL has enormous potential in
next coming years. One recent study envisages the
potential of broadband network to reach app. 2 billion
rural inhabitants (number of electricity customers will
be several times less) around the globe who are still
unreachable for modern voice and data services and who
otherwise would remain without access to modern, highspeed telecommunication services [6].
Since more than 3 decades, the interoperability of
various protection devices and controllers in both
substations and power grids was largely restricted by
different communication standards embedded into them
by different manufacturers.
The existence of different communication protocols
required from an operator to implement large amounts
of commissioning efforts with subsequent adverse
effects on matching the multi-protocol equipment to
newly added devices in future. Therefore there have
been only few substations where devices and controllers
of different origin were good interfaced.
By now most installations in substations, LANs and
WANs today communicate via DNP3 serial protocol,
followed by older protocol Modbus. Often the
proprietary protocols (not serial, i.e. standardized by
IEC) still are used. SCADA central stations and RTUs
communicate on DNP3 as well. The emerging option to
DNP3 is nonserial Internet protocol/Ethernet. New
IEDs, RTUs as well as video cameras are Ethernetbased devices. Ethernet-core substation networks are
considered as modern technical structures. The
emerging services in ICT market just include assistance
in gradual modification of substations with lowbandwidth media and serial protocols to higher
bandwidth and Ethernet.
Nevertheless, the Internet protocol/Ethernet is not the
long-term benchmark. In the first half of 2004 the
unique communication standard IEC 61850 for
substation automation was released. It ensures smooth
exchange of data with one another because all the
connected devices communicate a common language.
The first substation in distribution networks (and first
substation in world as such) with complete IEC 61850based control was installed in a 16-kV substation in
Switzerland, 6 months after the issue of the standard. It
was done by Siemens Power Transmission and
Distribution, the same entity that was protagonist in the
development of a new standard. The old protection and
control system was replaced by new one of SICAM type
(it is Siemens equipment family consisting of products
for digital power automation). The new communication
standard ensured the required interoperability in the

substation.
6. SCADA/EMS and SCADA/DMS
The SCADA history begins with the advent of digital
computers in the mid 1960-ties. Those computers were
capable for real-time control. Prior to that time, the
remote control existed but it was incapable to cover a
power system (network), just only single objects, e.g. a
larger power plant. Since then SCADA had a long
history of efficient use on the transmission grid, while
the predominantly radial distribution network has not
generally benefited to the same extent.
Today SCADA is categorized to SCADA/EMS and
SCADA/DMS. SCADA/EMS supervises, controls,
optimizes and manages generation and transmission
systems (so-called Generations and Transmission
SCADAs). SCADA/DMS performs the same functions
for distribution networks (so-called Distribution
SCADA). Both systems enable DNO to collect, store
and analyze data from hundreds of thousands of data
points in national networks, perform network modeling,
simulate power operation, detect faults, prevent outages,
and participate in energy trading markets [7].
Nowaday SCADA/DMS is just getting into wide-spread
usage. It covers MV&LV distribution networks and
improves customer service by providing accurate and
prompt information about maintenance and outages.
SCADA/DMS masters communicate with substation
and feeder RTUs, smart relays and SA systems to
monitor the real-time status of the network and provide
remote control of devices such as switches, capacitor
banks and voltage regulators.
It can be concluded from the SCADA definitions
referred to that SCADA/DMS is conceived as covering
the MV and LV network points, but excluding
distribution substations (HV/MV). Thus the abovementioned statement SCADA had a long history of
efficient use on the transmission grid implies namely
the HV part of distribution substation! Therefore,
conceptually, distribution network is shared by
Transmission SCADA and Distribution SCADA.
Smart Energy Alliance (UK) refers to 9 primary benefits
to installing a Distribution SCADA system: 1) monitoring&control; 2) advanced power flow analysis; 3) network switching; 4) load management; 5) fault detection,
isolation, and restoration; 6) volt/VAR control; 7) integrated outage management systems; 8) integrated
trouble call system; 9) mobile workforce management.
Since SCADA is, first of all, a system consisting of
many objects (network points), the minimal threshold of
system might be an issue. The technology developers
are providing to the market microSCADA for minor
applications for selected number transmission and
distribution substations.
The SCADA systems had developed from conventional
architecture as RTUs of distribution substation
SCADA masters in central stations with HMI to
contemporary innovative architecture asconventional
architecture plus distribution substation automation
305
systems plus feeders RTUs plus selected IEDs and

digital relays in MV/ LV network equipment.
Substation RTUs collect sensor data on substation
equipment and transmit it to SCADA masters and
central stations in batch mode on periodic basis, app.
each second (it is data acquisition). SCADA masters
(computers, servers) analyze data and automatically
monitor the process (it is supervisory control) and, when
necessary, issue switching orders to RTUs in response
to the process (it is active or direct control). Similarly, a
dispatcher via HMI monitors the process (supervisory
control) and intervenes (active control).
Some RTUs also perform control of substation
equipment, but it is considered that PLCs do it better,
while RTUs more suit for communication with masters.
Distribution SA systems do not have RTUs. They
communicate to SCADA masters from PCLs or from
IEDs. It is called SA/SCADA systems integration. It
adds new or, at least, unconventional functions to
SCADA system (additional SCADA dispatch centers,
adaptive relay settings, access to substation metering
data, feeder automation support, automatic load
restoration, power quality monitoring etc. [8]).
As yet SCADA coverage on distribution substation level
is practically completed: each larger substation in highdensity areas typically has a RTU. Small substations and
those in low-density areas do without it. Presumably,
only limited potential for SCADA coverage extension
can be envisaged therein. However SCADA systems on
this level will extend in data flow dimension, because of
new SA schemes, SCADA/DSM extension and
cooperation/integration with other grid control systems.
Considering compatibility issues, new SCADA
architecture need adaptation to old substations and old
RTUs, and vice versa new SA schemes require
adaptation to old RTUs.
Today geographical data are embedded in SCADA
systems, after their alignment with GIS.
It should be noted that SCADA is not the one and
exclusive system of distribution network control. There
are some other dispatching, telemetering and monitoring
systems that do work without SCADA masters, e.g.
AMI (see section 7). They can be integrated with
SCADA.
Considering the future of remote distribution control, it
is still vague whether intelligent networks will be a
driver or a barrier for SCADA, despite of the growth of
remote data traffic. Some uncertainties arise from recent
reporting on BPL breakthrough and perspectives. Many
observers see BPL as eventually able to replace other
forms of remote asset management that include
supervisory control and data acquisition, teleprotection
and the use of narrow-bandwidth forms of PLCr [6].
customer meters with central processing units of DNO.

They perform the functions of gathering, transport and
dissemination of metering data. Started with larger
customers, AMR technology is currently coming to
residential customers. In result, some DNOs declare that
AMR have decreased the meter-related billing complaints, inaccurate billing and work order errors by half.
The wire-based networks have proved their value but
recently the wireless networks seem be a winning
technology. Particularly, then wireless technology is
combined with new generation of meters.
The previous wireless network architecture was based
on the huge amount of data collectors, for instance,
hundreds or even thousands in urban area of 500 km2.
Modern network architecture, referred to as a new
advanced AMR, is implemented in various power
utilities in USA. It allows endpoint meters to
communicate directly with AMR base stations. For the
area of the mentioned above, the coverage is ensured by
3 tower-mounted base stations, with total elimination of
complex, intermediate tier pole-top or roof-top-mounted
data collectors. These new AMR systems provide
synchronized hourly read interval data, including
demand data. The integration of meter reads to billing
systems is entirely automated [9].
Albeit the current AMR practice proved its technical
and economical feasibility in urban areas, the analogous
statement still hardly can be fitted for rural applications:
as in general, DA of rural networks is complicated.
Nonetheless, the function of automated reading might be
seen as only incipient application of AMR systems. The
monitoring of customers energy usage enables
dispatchers to have a more comprehensive real-time
monitoring of network operation. A next application
might be recording of outages and customer
interruptions. Further the monitoring of power quality, if
advanced meters send power quality data to network
dispatch centers, can be envisaged.
The next function would be based on two-way
communication between a meter and a control center.
Introduction of such communication will transfer an
ARM system to higher level of advanced metering
infrastructure (AMI). It will provide the remote connectdisconnect service of a customer. Finally, the AMI
concept might be extended to include home portals for
customers. Such a portal will be able to deliver price
signals and to motivate a customer to engage to demand
response. Moreover, a portal could provide information
on outages and power quality. [10].
Eventually, one unexpected driver for AMI deployment
might appear. It is BPL communication technology,
still being in its infancy. It 2006 it covered only 1.3
million electricity customers worldwide, with forecasts
o 20 and 70 mllion in 2008 and 2010, respectively [6].
7. Automated meter reading

8. Feeder automation
The conventional practice of manual meter (electromechanical) reading is superseded by AMR (solid-state)
systems. AMR enable DNOs to track customer energy
usage, and to improve billing. In the last decade various
AMR systems were implemented and tested. Such
systems work as information networks that connect the
Feeder automation deals with automatic load transfer

from failed feeder to a backup feeder. This transfer is
carried out by circuit breaker or other change-over
switch located either in a substation or in remote feeder
306
section. According to some estimation, the automation

at feeder level in North America is less than 20% [1].
As for restoration of interrupted feeders, majority of
utilities still rely on first call from a customer reporting
on interruption, even with AMI systems installed (with
capability of two-way communication). AMI or other
monitoring technologies should be used to notify
dispatchers on customers out of service. Having entire
picture of outage, dispatcher can adequately choose the
actions of remote control and dispatch field crews [1].
As yet customers served via a fuse-protected feeder
usually experienced a sustained interruption after fuse
had blown due to the momentary disturbance, e.g. the
tree limb touching a power line. In 2007, advanced
protection technology was successfully applied socalled TripSAver Dropout Recloser. This device senses
the severity of the problem and restores the delivery if
the problem no longer exists. It proved to decrease
outage time significantly.
The value of high-scale feeder automation in terms of
outage duration can be quite impressive. Naperville
(Illinois, USA) DA project (providing 70 % of
distribution network with automation) demonstrated
good response to out-of-power accident of one 138-kV
distribution substation in Naperville city (normally, the
city is supplied from 4 such substations). In an instant,
25% of Naperville customers, i.e. app. 13,600, lost
power. In less than 2 minutes, 9100 customers (67%)
regained electricity service it was restored automatically. 5845 thereof were re-energized by switches in remote feeder sections, while the rest 3255 customers by
automatic load transfer process at the substation level.
Later on another 4380 customers (32%) regained service
in the next 14 minutes by means of dispatcher control
actions to operate substation bus ties and distribution
assets (via SCADA). The remaining 1% of customers
(143) was re-energized by repair crews within 81
minutes after the accident. Hence the automated feeders
drastically reduced the number of customers exposed to
a typical outage. That accident has completely validated
the significant investment into citys DA [11].
As for urban environment, the optimum distribution
circuit design is considered to consist of 2 feeders (MV),
with 10 km total length (mainline), each feeder (5 km)
fed from different substation. There are 3 circuit
breakers installed on this circuit: at the open point
between feeders and at the mid-point of each feeder.
Thus the circuit is sectionalized into 4 sections. The
total circuit can be fed from either substation in
emergency case. The embedded auto-sectionalizing and
reconfiguring intelligence allows the switches to
reconfigure for faults and loss of power. A faulted
feeder section is isolated automatically and the half the
faulted circuit is restored within 60 sec. On a monthly
basis, maintenance crews shall attend each circuit
breaker to ensure that it is intact and all automated
functions are in ready-to-transfer status [11].
If feeder automation is aimed at reduction of SAIDI
minutes, the design of optimal distribution circuit might
rely on sectionalizing feeders to maximum number of
segments [2].
9. Conclusions
1. Automation, as it is in process, will transform
distribution networks to digital (intelligent) distribution
networks where computer (microprocessor) network
will overlie power network.
2. Intelligent network vision is gradually turning to
emerging network technologies that will help convert
data to intelligence, intelligence to knowledge and
knowledge to intelligent business decisions.
3. The major difficulty related to future intelligent
substations and networks is referred to as interoperability of proprietary devices (different manufacturers
keep to different communication protocols).
4. Sometimes views of distribution system operators on
future intelligent networks are overcome by
apprehensions of complexity in introduction and
operation of intelligent technologies. Operators desire to
implement plug-and-play type technologies and are
apprehensive of superabundant information flows.
5. Distribution automation, particularly on feeders, has
proved its high effectiveness in urban areas in terms of
reliability performance (SAIDI minutes).
6. Automated meter reading systems are progressing to
advanced metering infrastructures (AMI) featuring twoway communications for switching control applications.
7. With the communication technology advance leading
to low-cost products, the automation of rural distribution
networks, including Distribution SCADA or microSCADA, is entering into cost-effective applications.
10. References
1.
Wolf G. The road to assets management.

Trans.&Distr. World, April 2007. p. 8-10 (in
supplement Digital distribution).
2. Wolf G. Embedded intelligence. Trans.&Distr.
World, April 2007. p. 3-7 (in supplement Digital
distribution).
3. Wolf G. Substation intelligence. Trans.&Distr.
World, November 2007. p. 12-16 (in supplement
Digital transmission).
4. Gwyn B. and Panas W. IEDs: Enhanced monitoring
and control in a tiny package. Trans.&Distr. World,
February 2005. p. 41-48.
5. Tani M. Critical connections. Trans.&Distr. World,
April 2007. p.12-14 (in supplement Digital
distribution).
6. Research firm releases study on broadband over
power lines and power-line-carrier communications
technologies. Trans.&Distr.World, Aug. 2007. p. 10.
7. http//www.abb.com.
8. Carpenter D., Foster V. and McDonald J. Kentucky
utility fires up its first SCADA system.
Trans.&Distr. World, February 2005. p. 33-39.
9. Alabama Power begins trial of 500 AMDS Meters.
Trans.&Distr. World, February 2005. p. 14.
10. Bush R. and Musser P. Focus on efficiency.
Trans.&Distr. World, February 2005. p. 44-49.
11. Geynisman O. and Schaub J. Automated switching
benefits customers. Trans.&Distr. World, September
2007. p. 46-50.
307

DISTRIBUTED GENERATORS: DEFINITION IDENTIFICATION

Dainius SLUNYS
Abstract: Electricity supply structures and energetic

institutions associated with installation, exploitation and
development of distributed energy resources have
different definitions for distributed generation. Different
types of generators are installed in electric power
systems. Those units can be either high or low power,
installed at customers or far from them, traditional or
renewable. Large amount of generating units and
different definitions for distributed generators leads to
problematic definition for real and exact distributed
generators. Sometimes it can be a reason of
misunderstanding.
Keywords:
distributed
generation, definition.
generators,
2. Review of definitions for distributed generators

The largest electricity supply structures and energetic
institutions give different definitions for distributed
generators. Let us have a look at some of the existing
definitions [1, 5]:
1. By CIGRE (International Conference on High
Voltage Electric Systems) distributed generators are
called:
Power less than 100 MW;
Connected to the distribution network;
Decentralized;
Amount of electric energy is not planned.
2. By IEA (International Energy Agency) distributed
generators are called:
Connected to the distribution network;
Connected straight to the load;
Small scale (internal-combustion generators, small
power gas turbines generators, micro turbines, fuel cells
and solar units).
3. By DPCA (Distributed Power Coalition of America)
distributed generators are called:
Connected to the consumer;
Small scale generators;
Connected to the consumer network, distribution
network or to the transmission network.
4. By US Department of Energy distributed
Connected straight to the load;
Small scale generators.
5. By Swedish Electric Power Utilities distributed
Power is not important;
Connected directly to the distribution network or at
the customer site.
Other electricity supply structures and energetic
institutions suggest that distributed generators have the
following features [5]:
1. Standardized and modular generation source using
renewable energy source in a range of up to one MW
(Austria);
2. Source connected to the Distribution network
(Norway);
distributed
1. Introduction
Power of generating units in electricity supply system
varies from few kilowatts to hundreds megawatts.
Those units can be a part of small, large and
interconnected electric power systems. Generating unit
connection to the grid can be different. Unit can be
connected straight to the load, both to the load and to
the distribution network, only to the distribution
network and to the transmission network. If unit is
connected to the electricity supply system, consumer
can be near, not far and far away. This variety is the
problem for exact definition identification for
distributed generators. Electricity supply structures and
energetic institutions give very different definitions for
distributed generators. The reason of different
distributed generators definitions is a diverse
understanding of traditional big power units. Mostly, if
unit or units are not recognized as a traditional, they are
called distributed generators. Sustaining definitions of
electricity supply structures and energetic institutions, it
is not possible to detect exact limits for the definition of
distributed generators. Definitions are selected by
subjective understanding. Objective definition must be
chosen by arguments and proofs.
308
3. Source less than 10 MW, not centrally planned and

connected to the distribution network (Bulgaria);
4. Source not operated by utility (Czech Republic);
5. Source less than 50 MW for local consumption
and/or for selling to the utility (Estonia);
not centrally dispatched, and connected to the
distribution network (Finland);
7. Integrated or stand-alone modular source close to the
point of consumption (Germany).
8. Generation not active in system balancing
(Netherlands).
9. Electricity or heat source connected to the user
(Poland).
10. Decentralized source less than 50 MW rating
(Romania).
dispatched, and connected to the distribution network
(Slovakia).
12. Source connected to the distribution network or to
the customer site (Sweden).
13. Source not connected to the transmission network
(Great Britain).
All those definitions are a proof of different
understanding of distributed generators and distributed
generation. It means that distributed generators
definition is not identified exactly. Large number of
definitions already exists and there is no clear
consistency between them.
Lithuanian terminology vocabularies have no comments
about distributed generation definition. Some
international terminology vocabularies have comments
about distributed generators but without exact limits and
descriptions.
3. Generating unit status in power system;

4. Loads location.
Mostly distributed generators are identified as a result
of correspondence to one of the criterions.
4. The aims of distributed generators establishment
The main goals for the establishment of distributed
generators are [2, 3, and 4]:
1. To reduce electricity transmission losses.
2. To increase electricity supply reliability.
3. To make a convenience of renewable resources.
4. Electricity market liberalization.
5. Distributed generators definition and proof
The first step is to identify affinities of the distributed
generators. For the objectiveness, certain criterions must
be chosen: range limits, critical values and correlation.
The first of criterions is generating unit power and
amount of electric energy. In already existing
definitions, value of power varies from some kilowatts
to one hundred megawatt. The amount of electric
energy is taken into account only for renewable
resources. Here we can see that the existing definitions
of generating power unit greatly differ. It means that
electricity supply structures and energetic institutions
associated with distributed energy have different view
point to the distributed generation. The explanation of it
may be that various countries, electricity supply
structures and energetic institutions have different
installation and novelty levels of distributed generators.
Projection, planning, designing, application, and
management of distributed generators are in quite
highly different levels too. The highest defined power
of distributed generators is known as one hundred
megawatt (CIGRE). The highest power is defined by
load range limit possibility, but the demand, production
and loads of electric energy is always increasing and it
is the reason to increase defined power of distributed
generators. For example, distributed network load is
150 MW (such high loads are possible in industry
plants), generating unit 130 MW, last 20 MW part is
covered by electric system. Question is, is it possible to
call 130 MW generating unit as a distributed generator
even if the power is over limited? There is an important
relation between load and generating unit. 130 MW
generating unit uses all his power to cover local load. It
is possible to satisfy three of four aims of distributed
generators establishment. That is the reason to call 130
MW generating unit as a distributed generator even if
the power is over limited. The development of
distributed generators and simple examples provide a
conclusion of power definition: power and electric
energy amount can not be a direct criterion for the
identification of distributed generators definition.
The power of generation unit alone does not correspond
to the goals of distributed generators establishment.
There are many types of generating units. The type of
generator completely depends on demand and primary
energy resources. In most cases, generators are divided
into two groups: traditional and renewable energy.
3. Distributed generators indeterminations and

comments
Definitions prove that there are no exact limits between
traditional generators and distributed generators. In
definitions above distributed generators are identified
by certain characteristics, which are as follows:
1. Power.
2. Generation unit type.
3. Connection location.
4. Generating unit distance to the load;
5. Load connection to the generating unit;
6. Generating unit status in the power system
(centralized or decentralized);
7. Planning and forecast of power (planned or not
planned energy amount);
8. Dependence of generating unit on the power system
(generating unit operating in the power system,
autonomous operation of generating unit, but with
possibility to be connected to the power system and
only autonomous operation of generating unit).
The following four criteria can be distinguished that
have a prominent influence on the definition of
distributed generators origin:
1. Generating unit power and amount of electric
energy.
2. Generating unit type and connection location.
309
Traditional generators can be built almost everywhere

considering the demand. Renewable energy generators
can be built only at renewable energy resources. The
renewable energy unit power depends on amount of
primary energy resources. The traditional generators
can be connected to the electricity network almost
everywhere. Meanwhile, the renewable energy
generators can be connected to the network only at
renewable resources places or close to them. The
connection of generating unit to the net depends on
power, stability of energy production, load location and
load power. If to remember the example above, 130
MW can be either separate generating units, groups of
generating units or different types. A conclusion can be
made regarding the type of generator: generator type
can not be a direct criterion for the identification of
distributed generators definition, but it can be a
reason of connection location. Generating unit can be
connected to the electricity network in three ways:
1. Directly to the load.
2. To the distribution network.
3. To the transmission network.
A generator or a group of generators connected directly
to the load can satisfy all four goals of distributed
generators establishment or at least three of them.
If a generator or a group of generators are connected to
the distribution network, all four, three, two, one or
none of the goals of the distributed generators
establishment can be satisfied.
A generator or a group of generators connected to the
transmission network can satisfy only one or none of
the goals of the distributed generators establishment.
The goals of distributed generators establishment are
directly influenced by a connection location of a
generator or a group of generators. Here a conclusion
can made regarding the connection location of the
generator or the group of generators: the location of
connection of a generator or a group of generators is
one of the main factors for the identification of
distributed generators definition.
Independent generators can be connected to the
electricity grid, if electricity market is liberalized. The
type and power of the generator or the group of
generators is not important and can be neglected in this
case. Those generators can be centralized and
decentralized. The amount of electric energy can be
centrally planned or not planned. A Centralized,
decentralized, centrally planned or not centrally planned
generator or a group of generators can be connected to
the power network in all three mentioned ways. Here a
conclusion can be made regarding the status of the
generator or the group of generators in power system:
the status of the generator or the group of
generators in the power system can not be a
criterion for the identification of distributed
generators definition.
Load locations are available everywhere in the
distribution grid. A single load power can vary from
few watts to the ten megawatts and more, while groups
of loads can reach hundreds of megawatts. Electric
energy demand is changing all the time. Technological
processes, various industrial requirements and anti-
pollution standards can be a reason for load location in

the area. Load location and generator location can be
the same. In most cases, loads are connected to the
distributed network. A load supply voltage can vary
from 0.4 kV to 10 kV (in exceptional cases load supply
voltage can be higher). If a generator and load location
are the same or close, it may be a distributed generation
case. Here a conclusion regarding loads location can be
made: loads location is one of the main factors for
the identification of distributed generators
definition.
Various definitions of distributed generators, proofs,
criterion associations and the aims of distributed
generators establishment suggest an objective
conclusion that: there are two main direct criteria for
the identification of distributed generators definition
a connection location of a generator or a group of
generators and a location of loads. Other criterions
are only secondary factors for the identification of
definition.
Main criterions it suggest that there is an important
distance between generating unit and load. Both
criterions are sufficient for the identification of
definition, but there are two indeterminations:
1. Definite greatest distance between generator and load
in the distribution network.
2. Definite greatest power of generating units or definite
greatest power proportion ratio in comparison with
loads.
Both indeterminations must be researched and the
conjunction between them defined.
The distances between loads and distribution points are
selected to satisfy range of work parameters at loads.
Distribution network can be divided into one, two or
three steps of voltage. The generator can be connected
to each of them in the distribution network. At
generator and voltage step, a load or loads can be
connected to the following voltage steps:
1. Connected to the same voltage step as a generating
unit.
2. Connected one voltage step down.
3. Connected one voltage step up.
4. Connected two voltage steps down.
The first goal of the distributed generators
establishment says that electric power losses must be
reduced by using distributed generators. It is the main
object to research. If a load is in the same voltage step
as generator or generators, power losses are at the
minimum and a generator or generators can be called
distributed. If load is at one voltage step up or down as
a generator or generators, power losses increase but less
than in transmission. It means that generator or
generators can be called distributed in this case. If load
is at two voltage steps down like a generator or
generators, power losses are almost the same as in the
transmission. It means that generator or generators can
not be called distributed in this case.
In the first three cases a generator or generators can be
called distributed, if they correspond to other conditions
mentioned above. Here a conclusion regarding the
greatest distance between a generator and a load in the
distribution network can be made: the distance
310
between a generator or generators and load in the

distribution net is not important, but a generator or
generators and a load must be connected in the same
voltage step or not further than one voltage step up
or down.
Usually power supply in the distributed network is
obtained from the transmission network. All electric
power transmission and distribution components are
selected to supply a load power with required reserve.
The power of a generator or generators can be selected
to supply loads and to transfer the rest of the power to
the transmission network. It means that the selected
power of connected generator or generators can be
double as it is in the load. Here a conclusion regarding
the greatest power of a generator or generators can be
made: a generator or generators can be called
distributed if the selected power is not greater than
double as the one in loads. If the power of a generator
or generators is higher than double as the one in the
loads, and if the distribution network is modified to
transfer more power to the transmission network, then
the generation can not be called distributed.
On the grounds of logic objectives, distributed
generators definition can be exactly identified and
represented as follows: a distributed generator or
generators are electric energy sources connected to
the distributed network to the same voltage step or
not further than one voltage step as loads, operating
in normal conditions, power is not higher than
double as the one of load and have a possibility to be
connected to the system and partially or completely
cover load demand. Other factors are not important for
the identification of distributed generators definition.
A distributed generator or generators are units with the
following features:
1. Generator or generators connected to the distributed
network.
2. Generator or generators connected to the same
voltage step or not further than one voltage step as
loads.
3. The power of a generator or generators must be not
higher than double as the one in load and have a
possibility to be connected to the system; and the same
as load without a possibility to be connected to the
system.
4. Generator or generators partially or completely cover
load demand.
5. Normal operation conditions.
A generator or generators satisfying the conditions
mentioned can be called distributed generators and
electric energy produced by them can be called
distributed generation.
If a generator or a group of generators are similar to
both distributed and the traditional generators, and the
definitions of distributed or traditional generators can
not be applied, such generator or generators can be
defined as dispersed generators. Electric energy
produced by dispersed generators can be called
dispersed generation. A definition of dispersed
generators can be applied for generator or generators
with the following features:
1. Small scale generator or generators connected to the

transmission system (for example: wind farms, hydro
power plants, etc.)
2. Generator or generators remote from the load more
than one voltage step.
3. Generator or generators connected to the distribution
network with power greater than double in comparison
with a load (it can be all types of renewable energy
power plants).
4. Large amount of small scale generators in the quite
large area.
6. Conclusions
1. Generator or generators type, status in electric
power system, centrally planned or not centrally
planned, can not be a reason for distributed generators
definition identification.
2. Generator or generators total power can be only
secondary reason for distributed generators definition
identification.
3. Location of connection and load of a generator or a
group of generators are the main criteria in
identification of distributed generators definition.
4. Distributed generator or generators are electric
energy sources connected to the distributed network to
the same voltage step or not further than one voltage
step as loads, operating in normal conditions, power is
not higher than double as the on of load and have a
possibility to be connected to the system and partially or
completely cover load demand. Other factors are not
important for the identification of distributed generators
definition.
5. If a generator or a group of generators are similar to
both distributed and the traditional generators, and the
definitions of distributed opr traditional generators can
not be applied, such generator or generators can be
defined as dispersed generators. Electric energy
produced by dispersed generators can be called
dispersed generation.
7. References
1. CIGRE, Working Group 37.23. Impact of Increasing
Contribution of Dispersed Generation on the Power
System, 1999.
2. Vaclovas Mikinis, Egidijus Norvaia. Paskirstytoji
elektros energijos gamyba Europoje ir Lietuvoje.
ENERGETIKA. 2007. T. 53. Nr. 4. P. 18.
3. Distributed generation in liberalized electricity
market, IEA Publications, 2002.
4. ENIRDGnet, D2, Concepts and Opportunities of DG,
WP1 - concepts and opportunities of DG: the driving
european forces and trends.
5. A.A. Bayod Rujula, J. Mur Amada, J.L. BernalAgustin, J.M. Yusta Loyo, J.A. Dominguez Navarro.
Definitions for Distributed Generation: A revision.
International Conference on Renewable Energies and
Power Quality. 2005.
6. Ackermann T., Andersson G., Soder L. Distributed
generation: a definition // Electric Power Systems
Research. 2001. Vol. 57(3). P. 195204.
311

NEW ACCURATE FAULT LOCATION METHOD

FOR MULTI-TERMINAL POWER LINES
Przemyslaw BALCEREK*, Marek FULCZYK*, Jan IZYKOWSKI**, Eugeniusz ROSOLOWSKI**,
Murari Mohan SAHA***
* ABB Corporate Research Center, Krakow, Poland; ** Wroclaw University of Technology, Wroclaw, Poland;
*** ABB Power Technology Systems, Vasteras, Sweden
specific interval of expected values and which indicates

the actual place of the fault, is selected from the
hypothetical locations calculated in this way.
Use of yet another set of the fault locator input signals,
which differs from those applied in the cited approaches
[1][2], is proposed in this paper. Three-phase current
acquired synchronously [5] at all line terminals and
additionally three-phase voltage from the terminal at
which the fault locator (FL Fig. 1) is installed, are
taken as the input signals. Such specific availability of
measurements for locating faults has been assumed with
the aim of simple adding the fault location function to
current differential relays for multi-terminal lines.
After derivation of the fault location algorithm, the
results of its evaluation with use of the ATP-EMTP [6]
simulation data are delivered and discussed.
Abstract: This paper presents a new method for locating

faults on multi-terminal power lines, based on threephase current from all (n) terminals and additionally
three-phase voltage from the terminal at which a fault
locator is installed utilization. The fault location
algorithm consists of (2n-3) subroutines designated for
locating faults within particular line sections and a
procedure for indicating the faulted line section. Testing
and evaluation of the algorithm has been performed
with fault data obtained from comprehensive ATPEMTP simulations. Sample results of the evaluation are
reported and discussed.
Keywords: ATP-EMTP, current differential relay, fault
location, simulation, multi-terminal line.
1. Introduction
2. Fault Location Techniques
Accurate location of faults on overhead power lines for

the inspection-repair purpose [1][4] is of vital
importance for operators and utility staff for expediting
service restoration, and thus to reduce outage time,
operating costs and customer complaints.
A system and a method for fault location on a multipleterminal parallel transmission line is known from [1]. In
that method a multi-terminal transmission system is
equivalented to a three-terminal transmission system,
for which differential current amplitudes are calculated
in each station. Then, the distance to the fault point is
calculated from their relations. A method for fault
location using voltage and current phasor measurement
in all stations at the ends of a multi-terminal line has
been presented in [2]. That method relies on reducing a
multi-section transmission line into systems of twoterminal lines assuming that the fault is located in one
of these sections and then hypothetical fault locations
are calculated for this assumption. Next, calculations of
successive hypothetical fault locations are made
assuming that the fault is located in further successive
sections of the line. One value, which is contained in a
Modern microprocessor-based current differential relays

exchange the locally measured current phasors over
long distances. For this purpose different forms of
communication means are utilized.
T4
l(2n-6)
l6
l4
I2
I3
3
T(n-2)
I4
4
T(n-1)
In
n-2
l(2n-4)
T3
I1
l2
FL
T2
l(2n-3)
l(2n-5)
l5
I(n-1)
V1
l3
I(n-2)
l1
n-1
Fig. 1. Fault location on multi-terminal line with the assumed

input signals
The current differential protection principle requires

synchronization of digital measurements performed at
different line terminals. This is accomplished using the
well-known Global Positioning System (GPS) or the
other techniques [5]. Phasors of three-phase current
312
coefficients (Table I), as well as in all further

symmetrical components calculations.
from all line terminals: I1, I2, , In, together with the
locally measured three-phase voltage phasor (V1) are
taken as the fault locator input signals (Fig. 1). In
natural way, these measurements are considered further
as synchronized.
The developed fault location algorithm consists of
(2n-3) subroutines designated for locating faults on
particular line sections (as depicted in Fig.1) where n
is the number of terminals. Since the position of a fault
is a random factor, the faulted line section is not known
in advance. Therefore, the valid subroutine (for the
faulted section) will be chosen using a special selection
procedure.
d1
I 1i
n-1
Fault type
a1
a2
a-g
a0
b-g
0.5 j0.5 3
0.5 + j0.5 3
c-g
0.5 + j0.5 3
0.5 j0.5 3
a-b, a-b-g
a-b-c, a-b-cg
1.5 + j0.5 3
1.5 j0.5 3
b-c, b-c-g
c-a, c-a-g
j 3
1.5 + j0.5 3
j 3
1.5 j0.5 3
0
0
It is proposed to calculate the total fault current from (1)

by using the following generalized fault model:
I F = a F1 I F1 + a F2 I F2 + a F0 I F0
(4)
V 1p , I 1p fault loop voltage and current;

positive sequence impedance of the section L1;
I F total fault current (fault path current).

Fault loop voltage and current are composed
accordingly to the fault type, as the following weighted
sums of the respective symmetrical components of the
measured signals:
Z 0L1
I 10
Z 1L1
Z iL(2n4)
Table 1. Weighting Coefficients in Fault Loop Signals (2)(3)
unknown distance to fault distance to fault

from
the beginning of the line to the fault point
(p.u.);
unknown fault resistance;
I Ap = a1 I 11 + a 2 I 12 + a 0
(1 d1 ) Z iL1
n
Z iL(2n3)
Fig. 2. Equivalent circuit diagram of the network for the i-th

symmetrical component assuming that a fault is on the first
line section
where
d1
V 1p = a1V 11 + a 2 V 12 + a 0 V 10
T(n-1)
The subroutine SUB_A, designed for locating faults

within the line section L1 (between nodes: 1, T2) (Fig.
2), is based on the following generalized fault loop
model [4]:
(1)
V 1p d1 Z 1L1 I 1p R1F I F = 0
Z 1L1
T2
Z iL2
2.1. Fault location algorithm subroutine SUB_A
R1F
d1 Z iL1
where
a F1 , a F2 , a F0 share coefficients (Table2).
The i-th sequence component of the total fault current is
determined as a sum of the i-th sequence components of
currents from all line terminals (1, 2, , n-1, n):
(5)
I F1 = I 1i + I 2i + I 3i + ... + I (n -1)i + I ni
(2)
where subscript i denotes the component type:

i=1positive, i=2negative, i=0zero sequence.
From the analysis of the boundary conditions of faults
[4] results that it is possible to apply different
alternative sets of the coefficients, which are used in (4).
In order to assure high accuracy of fault location, use of
the particular set has been recommended. The following
priority for usage of particular sequence components
(the respective coefficient in (4) is not equal to zero) of
measured currents is proposed (Table II):
- for phase-to-ground and phase-to-phase faults: use of
negative sequence components;
- for phase-to-phase-to-ground faults: use of negative
and zero sequence components;
for three phase symmetrical faults: use of
superimposed positive sequence components.
Excluding of the positive sequence components
( a F1 = 0 ) from the total fault current (4) for all faults,
except three-phase balanced faults, is characteristic for
the share coefficients from Table II.
In case of three-phase balanced faults, only positive
sequence components are contained in the signals
(3)
where
a1 , a 2 , a 0 - weighting coefficients (Table I);
V 11 , V 12 , V 10 - symmetrical components (the second

lower index) of voltages measured in station 1
(the first lower index);
I A1 , I A2 , I A0 - symmetrical components (the second
lower index) of currents measured in station 1(the first
lower index);
Z 0L1 , Z 1L1 - zero, positive sequence impedance of the
line section L1.
Fault loop signals (2) and (3), and also those used in the
remaining subroutines, are expressed in terms of the
respective symmetrical components. Use of such
notation is convenient for introducing the compensation
for line shunt capacitances, however, it is fully
equivalent to the description traditionally used for
distance protection [4]. Natural sequence of phases: a, b,
c was assumed for determining the weighting
313
the line for the i-th symmetrical component, assuming

that the fault is located in the final section of a multiterminal power line. Again, the subscript i is used for
denoting the respective: i=1positive, i=2negative and
i=0zero sequence, components.
Transferring of voltage from the bus 1 to the tap point
T2:
transf.
V T2i = V 1i Z iL1 I 1i
(9)
Table 2. Share Coefficients used in Fault Model (4)

Fault type
a F1
a F2
a F0
a-g
b-g
1.5 + j1.5 3
c-g
1.5 j1.5 3
a-b
1.5 j0.5 3
b-c
j 3
c-a
1.5 j0.5 3
a-b-g
3 j 3
j 3
Z iL1 impedance of section L1 respectively for the

positive, negative and the zero sequence
component.
b-c-g
j2 3
j 3
Next, the voltages for symmetrical components: V T(n -1)1 ,
c-a-g
3 j 3
j 3
V T(n -1)2 , V T(n -1)0 in the end tap point T(n-1) are
where
transf.
transf.
a-b-c
1.5 + j0.5 3
1.5 j0.5 3 *)
0
a-b-c-g
*) a F2 0 , however, negative sequence component is not
transf.
calculated:
1
Z iL1 T2
(1 d(2n 3) ) ZiL(2n3)
T(n-1)
present under three-phase balanced faults.
Z iL2
real( Z 1L1 I 1p )imag ( I F ) imag ( Z 1L1 I 1p ) real( I F )
d(2n3)
Z iL(2n4)
n-1
Fig. 3. Equivalent circuit diagram of a network for

symmetrical components, under a fault on the end section of a
multi-terminal line
V T(n -1)i = V T(n - 2)i Z iL(2n -5)

transf.
n2
1 I
transf.
j=
i =1, 2 , 3
ji
(10)
where:
Z iL(2n -3) impedance of line section L(2n-3) for the
sought fault distance ( d1 ) is determined as:

real(V 1p )imag ( I F ) imag (V 1p ) real( I F )
d(2n 3) Z iL(2n3)
It is proposed to calculate positive sequence of the total

fault current as the following sum of the superimposed
(superscript: superimp.) positive sequence currents
from the line ends 1, 2, , (n-1), n, respectively:
superimp.
superimp.
superimp.
(6)
I F1 = I 11
+ I 21
+ ... + I n1
Finally, negative-, zero- and superimposed positivesequence components of the measured currents are used
in calculation of the total fault current (4). This assures
accurate calculation of the fault current flowing through
the fault path resistance. This is so, since the positive
sequence components, for which the shunt capacitance
effect is the most distinct, are excluded.
After resolving (1) into the real and imaginary parts,
and eliminating the unknown fault resistance ( R1F ), the
d1 =
I ni
positive, negative, zero sequence components,

respectively.
transf.
transf.
transf.
In turn, voltages at the kth tap point V Tk1 , V Tk2 , V Tk0
(7)
Having the fault distance calculated (7), the fault

resistance RF1 can be also determined, as for example
from the real part of (1) as:
real(V 1p ) d1real( Z 1L1 I 1p )
(8)
R1F =
real( I F )
are determined from the following formula:

transf.
V Tki
= V T(k -1)i Z iL(2k -3)

transf.
k 1
1 I
j=
i =1, 2 , 3
ji
(11)
where:
transf.
V T(k -1)i - calculated voltage at point (k-1),
2.2. Fault location algorithm subroutine for

locating faults on the end section of multi-terminal
line
Z iL(2k-3) - impedance of line section L(2k-3) for the
symmetrical components.
transf.
transf.
transf.
The values of current I T(n -1)n1 , I T(n -1)n2 , I T(n -1)n0 , flowing
An analytic transfer of three-phase measurements: V 1 ,
from tap point T(n-1) to station n in line section L(2n3) are calculated, as follows:
I 1 , I 2 ,, I (n -1) , I n to the beginning of the section L2

is performed. The superscript transf. is used to
distinguish the analytically transferred signals from the
measured signals. Certainly, such transfer has to be
performed separately for each of the ith type of
symmetrical component of three-phase voltage and
current. Fig. 3 presents an equivalent circuit diagram of
I T(n -1)ni =
transf.
n 1
1, 2 , 0
i=
ji
(12)
j =1
The fault loop equation for faults on last line section has
the following form:
V T(n -1)np (1 d (2n -3) ) Z 1L(2n -3) I T(n -1)np R(2n -3)F I F = 0 (13)
314
where
V T(n -1)np = a1V
transf.
T(n -1)1
+ a 2V
transf.
T(n -1)2
Z 0L(2n -3)
I T(n -1)np = a1 I T(2n -1)n1 + a 2 I T(n -1)n2 + a 0

transf.
+ a 0V
transf.
T(n -1)0
transf.
Z 1L(2n -3)
After substituting (18)(19) into (17) and the respective

mathematical manipulations one obtains the searched
distance to fault d (2k 2) and fault resistance R(2k - 2)F .
(14)
transf.
I T(n -1)n0 (15)

2.4. Fault location algorithm for locating faults
between two tap points
where
Z 1L(2n -3) impedance of section L(2n-3) for the positive
Z 0 L(2n -3)
Symmetrical
seq.,
impedance of section L(2n-3) for the zero seq.
transf.
components
of
transf.
I TkT(k +1)1 ,
currents
transf.
I TkT(k +1)2 , I TkT(k +1)0 flowing from the tap point Tk
Writing (13) separately for the real / imaginary parts

and
after
performing
further
mathematical
transformations, the solutions for the searched fault
distance d (2n 3) and fault resistance R(2n-3)F are
towards the tap point T(k+1) in the respective line

section (Fig.5) are calculated:
I TkT(k +1)i =
k 1
transf.
1, 2 , 0
i=
ji
(20)
j =1
obtained.
d (2k1)
1
2.3. Fault location algorithm subroutine for

locating faults on tapped line section of multiterminal line
transf.
transf. d (2k 1) Z iL(2k1)

V Tki
Z iL1
transf.
I TkT(k+
1)i
(1 d (2k k) ) Z iL(2k1)
Z iL(2k2)
Z iL2
transf.
Z iL(2n3) n
transf.
V T(k+
1)i
Z iL(2n4)
Z iL(2k)
Voltages for symmetrical components V Tk1 , V Tk2 ,

transf.
V Tk0
at the kth tap point Tk (Fig.4) are calculated from
the formula (11), by assuming for k the number of the

considered station, which is equal to the number of the
tap point Tk, from which the faulted line goes to the
station k.
transf.
V Tki
Z iL1
transf.
I Tkki
Z iL(2k)
transf.
V T(k+
1)i
V TkT(k +1)p (1 d (2k 1) ) Z 1L(2k 1) I TkT(k +1)p R(2k 1)F I F = 0
Z iL(2n3) n
(21)
where
V TkT(k +1)p = a1V Tk1 + a 2V Tk2 + a 0V Tk0
transf.
Z iL(2k)
Z iL(2n4)
transf.
k+1
n-1
transf.
transf.
i=
j =1, j k
(16)
ji
I Tkkp = a 1 I Tkk1 + a 2 I Tkk2 + a 0

transf.
transf.
Z 0L(2k-2)
Z 1L(2k-2)
The presented fault location algorithm consists of (2n-3)

subroutines and only one of them the valid subroutine,
yields the results, which corresponds to the actual fault.
The remaining subroutines give false results.
In the first step of the selection, the subroutine, which
yields the distance to fault indicating the considered
fault as occurring outside the section range (outside the
range: 0 to 1.0 p.u.), or/and the calculated fault
resistance of negative value, is surely rejected.
If the first step is not sufficient, then the remote source
impedances (behind the terminals 2, 3,, (n-1), n in
(18)
transf.
I Tkk0
(19)
Z 1L(2k -2) impedance of line section L(2k-2) for the

Z 0 L(2k -2)
transf.
I TkT(k +1)0 (23)
3. Selection of valid results
The fault loop equation takes the form:

V Tkkp (1 d (2k - 2) ) Z 1L(2k - 2) I Tkkp R(2k - 2)F I F = 0 (17)
where:
transf.
transf.
transf.
V Tkkp = a1V Tkk1 + a 2V Tkk2 + a 0V Tkk0
Z 1L(k +1)
sequence component,
k number of the tap point.
After substituting (22)(23) into (21) and the respective
mathematical manipulations one obtains the searched
distance to fault d (2k 1) and fault resistance R(2k 1)F .
transf.
1, 2 , 3
Z 0L(k +1)
(22)
positive sequence component,

Z 0 L(2k 1) impedance of line section L(2k-1) for the zero
Then, the values of current I Tkk1 , I Tkk2 , I Tkk0 flowing

from tap point Tk to kth station in the tapped line
section, labeled as L(2k-2), are calculated:
transf
transf.
transf.
Z 1L(2k 1) impedance of line section L(2k-1) for the
Fig. 4. Equivalent circuit diagram of a network for

symmetrical components, assuming that the fault is located on
a section of a tapped line
I Tkki =
transf.
I TkT(k +1)p = a1 I TkT(k +1)1 + a 2 I TkT(k +1)2 + a 0
d ( 2 k 2) Z iL(2k2)
d (2k2)
n-1
k+1
Fault loop equation has the following form:
(1 d (2k 2) ) Z iL(2k 2)
Z iL2
Fig. 5. Equivalent circuit diagram of network for symmetrical

components, assuming that the fault is located between two
tap points
positive sequence component,

- impedance of line section L(2k-2) for the
zero sequence component.
315
Fig. 6 presents the wave-shapes of the fault locator

input signals, while Fig. 7 the results for the subroutines
SUB_2 and SUB_4, which indicate the fault as within
their sections. The results of the remaining three
subroutines are not visualized since yield the distance to
fault exceeding the section range, and thus have to be
rejected.
There is for selecting between the subroutine: SUB_2
(the results: d2=0.7035 p.u., RF2=9.83 ) and SUB_4
(the results: d4=0.7006 p.u., RF4=10.41 ). Both,
distance to fault and fault resistance according to these
two subroutines are not to be rejected. Calculating the
source impedances behind buses: 2, 4 one obtains:
SUB_2
actual
SUB_2: Z1S2 =(198.2+j0.26) , while Z1S2 =200 ,
Fig. 1), are calculated for the respective subroutines. For

example, in case of considering the subroutine for
locating faults on the end section (Fig. 3, Section IIB),
the source impedance behind the bus n for the positive
(negative) sequence is calculated as:
V
(24)
( Z 2Sn ) = n2
I n2
where:
transf.
transf.
V n2 = V T(n
-1)2 (1 d (2n 3) ) Z 2L(2n - 3) I T(n -1)n2
transf.
d (2n 3) Z 2L(2n - 3) ( I T(n
-1)n2 I F 2 )
(25)
Note, in case of three-phase balanced faults the negative

sequence quantities involved in (24)(25) have to be
changed into the incremental positive sequence
quantities.
If the source impedance, calculated according to the
considered subroutine, lays outside the Ist quadrant of
the impedance plane, then this subroutine is false and
has to be rejected. Otherwise, if there are still at least
two subroutines remain, then the selection has to be
continued. The particular subroutine can be rejected
also if the calculated value of the remote source differs
from the actual impedance. For this purpose certain
knowledge about impedances of the actual equivalent
sources behind the line terminals has to be possessed. In
particular, such impedance can be calculated using prefault measurements, if there is no generation source at
the line terminal.
SUB_4: Z1S4 =(576.6j3.41) , while Z1S4 =400 .

Therefore, the subroutine SUB_2 yielding the
impedance more close to the actual one is selected
correctly as the valid one, which gives the fault location,
with acceptable 0.35% error. For the majority of the
other considered fault cases the fault location error was
kept below 1%. However, in case of faults on the
shortest line section (T33) the maximum error slightly
exceeded 2%, which for such short segment could be
also treated as acceptable. For this selection knowledge
of the actual effective load impedances is required. In
field applications it will be done based on the
experience of fault locator user, which is familiar with
the level of effective load impedances at particular
buses. In this case they were calculated by performing
an analysis of pre-fault load flow.
SUB_4
4. ATP-EMTP evaluation
ATP-EMTP simulation program [6] was applied to

evaluate performance of the developed fault location
algorithm. Different multi-terminal networks were
modeled for generation of fault data used in evaluation
of the presented fault location algorithm.
In particular, the 130 kV four terminal network (Fig.1:
n=4) including the line sections: 1T2 (15.5km), 2T2
(22.5km), 3T3 (2.77km), 4T3 (12.9km), T2T3
(4.81km). The positive and zero sequence impedances
Z 1' L = (0.0276 + j0.3151),
for
all
sections:
actual
5. Conclusion
In this work, the new algorithm designed for locating

faults on a multi-terminal line has been presented. The
specific set of the fault locator input signals: three-phase
current from all line terminals and additionally the
locally measured three-phase voltage, has been
assumed. This fault location algorithm is intended for
application with current differential relays protecting a
three-terminal line.
The developed fault location algorithm consists of the
subroutines, designated for locating faults within the
respective line sections. The subroutines are formulated
with use of generalized fault loop and fault models.
Multi-criteria selection procedure is applied for
selecting the valid subroutine, i.e. for indicating the
faulted line section.
The simulation results show that the accuracy of fault
location is acceptable under various fault types, fault
resistances, fault locations, pre-fault loading conditions
and source impedances. The evaluation also proves that
the developed multi-criteria selection procedure allows
reliable indication of the valid subroutine, i.e. indication
of the faulted line section.
For the presented sample fault on the four-terminal line
there are two subroutines (out of five subroutines),
which indicate this fault as within their line sections.
Indication of the faulted section is performed by
Z '0L = (0.275+ j1.0265) /km.

Equivalent sources behind the terminals 1, 3 were
modeled with EMFs and the impedances:
Z 1S1 = Z 1S3 = (0.7 + j10.5) ,
Z 0S1 = Z 0S3 = (1.0 + j15) .
The EMF at the terminal 3 was delayed by: 30o with
respect to the terminal 1. The loads at the terminal 2:
actual
actual
Z1S2 =200 and at the terminal 4: Z1S4 =400 , were
also included. Current and voltage transformers were
modeled as well.
Analogue anti-aliasing filters with the cut-off frequency
of 350 Hz were included, and 1 kHz sampling frequency
was used. Full-cycle Fourier orthogonal filters were
applied for determining phasors of the processed signals
Fig. 6 and Fig. 7 present the example of fault location
on the four-terminal line: ag fault at the section T22,
fault distance: d2=0.7 p.u., fault resistance: RF2=10 .
316
comparing the estimated impedances behind the

respective line terminals with the impedances
determined from the pre-fault measurement.
4000
Bus 1 - Phase Currents (A)
Bus 1 - Phase Voltages (105V)
1.5
0.5
0.5
2000
2000
4000
1.5
0
20
40
60
80
100
120
20
40
Time (ms)
600
b
200
100
100
200
300
20
40
60
Time (ms)
80
100
120
200
200
400
20
80
100
120
3000
400
600
0
300
60
Time (ms)
40
60
80
100
120
2000
a
1000
1000
2000
3000
20
40
Time (ms)
60
Time (ms)
80
100
120
Fig. 6. Fault location on four-terminal network the example: input signals of the fault locator.
11
0.74
SUB_4 (10.41 )
0.72
Fault resistance ()
Distance to Fault (p.u.)
10.5
SUB_2 (0.7035 p.u.)

0.7
SUB_4 (0.7006 p.u.)

0.68
10
SUB_2 (9.83 )
9.5
8.5
0.66
0
10
20
30
40
50
60
Fault Time (ms)
10
20
30
40
50
60
Fault Time (ms)
Fig. 7. Fault location on four-terminal network the example: results of the subroutines SUB_2 (valid), SUB_4 (invalid).
6. References
1.
2.
3.
4.
Kenji Murata, Kazuo Sonohara, Susumo Ito, Kyoji

Ishizu and Tokuo Emura, Fault location method
for a parallel two-circuit transmission line with n
terminals, US patent 5, 485,394, Jan. 16, 1996.
Kai-Ping Lien, Chih-Wen Liu, Joe-Air, Ching-Shan
Chen and Chi-Shan, Novel Fault Location
Algorithm for Multi-Terminal Lines Using Phasor
Measurement Units in Proc. Thirty-Seventh
Annual North American Power Symposium in
Ames, Iowa, USA, October 23-25, 2005.
IEEE Std C37.114: IEEE Guide for Determining
Fault Location on AC Transmission and
Distribution Lines, IEEE Power Engineering
Society Publ., pp. 1-42, 8 June 2005.
5.
6.
317
J. Izykowski, E. Rosolowski and M.M. Saha,

Locating faults in parallel transmission lines under
availability of complete measurements at one end,
IEE Gener., Transm. and Distrib., Vol. 151, No. 2,
pp. 268-273, March 2004.
IEEE Std. C37.118: IEEE Standard for
Synchrophasors for Power Systems, IEEE Power
Engineering Society Publ., pp. 1-65, 22 March
2006.
H. W. Dommel, Electro-Magnetic Transients
Program, BPA, Portland, Oregon, 1986.

TRANSIENT ANALYSIS OF VARIABLE SPEED WIND TURBINE

*Djilali KAIROUS, *Bachir BELMADANI, **Mostapha BENGHANEM
*Faculty of Science and Sciences Engineering, UHB, Chlef, Algeria, **Dpt. of Electrotechnics, UST, Oran, Algeria
- Reactive power control by the excitation system to

keep constant voltage.
- Reduce the mechanical torque by pitch angle control to
avoid rotor acceleration.
Now doubly fed induction generator is well known. The
advantage of the DFIG is that the rotor converter needs
thus only to be rated for a fraction 25% (slip power) of
the total output power. By controlling the bidirectional
power flow in the rotor, the DFIG can operate as a
motor or a generator, at sub and super synchronous
speeds. The reactive power compensation can be
supplied from both the supply side and the rotor side,
and thus has potential to provide voltage support.
Different control strategies for such kind of generator
have been proposed for active and reactive power
control or voltage and frequency control. Direct field
oriented control for active and reactive powers is
presented in [2], [3].
In the proposed paper we present the control of the
Active and reactive Powers by sliding mode. The
system has the ability for simultaneous maximum wind
power generation for large speed range of operation and
achieves soft and fast synchronization to the grid. Power
converters and associated control strategies are
simulated using Simulink.
Some simulation results are presented to validate the
theoretical analysis and to show the behaviour and
performances of the proposed structure for power
control.
Abstract: Large wind turbines are often equipped with

Doubly Fed Induction Generator - DFIG.
Today the penetration level of wind power on the
network is high. The power system stability defined as
the ability of the system, for given initial operation
condition, to regain a normal state of equilibrium after
being subjected to a disturbance is the major problem.
In numerous papers the stability problems are pointed
out and there are also many papers describing control
strategies for wind turbine equipped with DFIG.
In this paper we study dynamic behavior of the DFIG
during voltage sag. Also we study the ability of the
control to improve the stability of the system.
Keywords: DFIG, stability, voltage sag, sliding mode
control.
1. Introduction
Wind energy is one of the most important and
promising sources of renewable energy due to its clean
character and free availability. Many variable speed
winds turbines have been used with induction generator,
permanent magnet and eventually switched reluctance
generators. Nowadays the market is oriented on the
design of high power wind generation systems based on
multi-pole synchronous machine or doubly fed
induction generators (DFIG). The wind energy market
(production) at present represents only a fraction of its
potential 47.616.4 installed megawatts at the end of
2004. 80% of this is just located in four countries-Spain,
Germany, United States and Denmark.
Power system stability has been recognized as an
important problem for secure system operation [1]. The
pining of the market of electricity obligates us to make
production of the generators near their physical limits,
for economic reason. It is thus necessary to evaluate
these limits, in particular the risk of instability of
network voltage.
This instability is often started by:
- Voltage recovery as fast as possible after clearance the
grid faults and without wind turbine disconnection.
- Grid synchronisation for reconnection as fast as
possible in order to feed maximum possible active
power to the grid.
2. Description of the studied system

The basic configuration of the whole system is
presented in Fig. 1. The rotor of DFIG is connected to
the grid through two back to back bridge converters.
The grid side converter (GSC) is used to control the
DC-link voltage and to keep it constant regardless to the
magnitude and direction of the rotor power.
The DFIG is controlled by the rotor side in order to
generate the optimal active power depending on the
wind speed and turbine characteristics.
318
Id
Im
Vdc
inet
abc
dq
Vq
Ws L Id +
Ws L Id -
PI
PI
Iqms
Idms -
Fig.1. Representation of DFIG-based wind turbine
PI
The wind turbine gives a mechanical torque, the

rotational speed increases, so we can begin the process
of coupling. The conditions of synchronization are the
same as these of synchronous machine, voltage RMS,
frequency and phase's angle on the stator side must be
the same as those of the grid. The DC bus voltage is
controlled through the front end converter and is kept to
constant value.
For this project the impact from the blade, gear and
rotor system is neglected for simplifying the simulations
results
Vd
+
0
Vdc
Fig.2. Blocks diagram of the Grid side converter control
d id
v dr = R id + L d t s L iq + Vdn
v = R i + L d iq L i + V
q
s
d
qn
qr
dt
3. Modelling of the network, transformer and

transmission line
(2)
The direct axis current is used to regulate the reactive

power and the quadratic axis current is used to regulate
the DC-link voltage. The reference frame is considered
oriented along the stator voltage vector
This method gives possibility to realise independent
control of the active and reactive power between the
GSC and the supply side.
The voltage components in park frame are given at the
output of the PI regulator
The voltage source components are: v d = 0 and
The network represents the simple model with a threephase voltage source. The transmission line between the
network and the transformer is modelled with its
resistance, inductance and capacitance by equivalent
circuit p.
The transformer is considered to be linear (the magnetic
saturation has not been taken into account).
4. Grid side converter control
v q = V and the powers can be written as:
The role of the grid side converter is to keep the DClink voltage constant regardless to the magnitude and
direction of the rotor power. A vector control approach
is used, allowing independent control of the active and
reactive powers flowing between the supply and the
GSC.
Fig.2 shows a simplified representation of the grid
connected converter. AC-side series inductances
accounts for transformer leakage reactance, and series
resistances represents inverter and transformer
conduction losses
The model of the system is given by the following
equation:
P = vq iq and Q = v q id
(3)
By neglecting converter losses, we have
v dc idc = v q iq and C
vdcC
dv dc
= idc im
dt
dvdc
= P Pm
dt
(4)
(5)
Based on the above relations, the GSC control diagram

can be easily deduced as presented in Fig.2
Va Van
1 0 0 ia
ia
d
R
1
ib = 0 1 0 ib + Vb Vbn (1)
dt
L
L
Vc Vcn
0 0 1 ic
ic
5. Sliding mode control

We will use the sliding Mode Control to control the
rotor currents of the DFIG. The goal is to have a more
stability in a disturbance cases then the PI regulators.
we write the Park model of the DFIG [4] :
With the Park transformation, the equation system of the

GSC is given by:
319
d dqs
m s . qds
Vdqs =R s Idqs +
dt
V =R I + d dqr m .
r
qdr
dqr s dqr
dt
(6)
dqs =Ls Idqs +M.Idqr
dqr =Ls Idqr +M.Idqs
(7)
1
0
g(x,t)= 0
0
1
0
0
1
0
Lr
0
0
0 (III.31) (13)
1
Lr
f(x,t)=
M
Rs.ads +Rs.cLrIdr + Vqs +sqs
Lss
Rs.aqs +Rs.cLrIqr + Vds sds
Lss
1
M
M
Rr.bLrIdr + Vqs +Rr.cds +r LrIqr + Vds
Lss
Lss
Lr

1 R .bLI + M V +R.c LI + M V
Lr r r qr Lss ds r ds r r dr Lss qs
By neglecting the stator resistance of the stator we can

write:
Vds = 0
Vqs s ds
(8)
dr =L r Idr + L . Vqs
s
s
=L I + M V
qr
ds
qr r
Ls .s
(9)
(14)
The general model of the machine is given by the
following equations:
The slip surfaces in the Park reference are defined to

control the rotor currents.
They are given by the following equations [5]:
Vds
V
qs =
Vdr

Vqr
d
M
Rs.ads Rs.cLrIdr + Vqs sqs + ds
Lss
dt
dqs
Rs.aqs Rs.cLrIqr + Vds +sds +
Lss
dt
dIdr
M
M

Rr.bLrIdr +L Vqs Rr.cds r LrIqr +L Vds +Lr dt
s s
s s

dIqr
M
M
Rb
.r LrIqr + Vds Rr.cqs +r LrIdr + Vqs +Lr
Lss
Lss
dt
d = ( I drref I dr )
q = ( I qrref I qr )
Where
The vector of Udqeq control is obtained by imposing
dq = 0
f ( x, t ) + g ( x, t ) Vdq = 0
(17)
The state model is put in the following form:
The Fig. 3 presents the global diagram for the DFIG

with sliding mode control
U dq = Vds Vqs Vdr Vqr
(16)
M
M
Rr.bLrIdr + Vqs +Rr.cds +r LrIqr + Vds
Lss
Lss

Ueqdq =
R.bLI + M V +R.c LI + M V
r r qr L ds r qs r r dr L qs
s s
s s
M
c=
LsLr
X =f(x,t)+g(x,t)U dq
Vdr and Vqr are the two control vectors of, they
force the trajectory of the system to converge towards

surfaces dq = 0 .
(10)
1
1
, b=
,
With: a=
Ls
Lr
(15)
(11)
(12)
320
cFig.3. Sliding mode control of the DFIG

6. Simulation results
In this section we present the simulation results of the
system presented in Fig.1. (The DFIG with PWM
voltage sources converters).
The rotor voltage, the stator voltage, the speed and the
powers are keeping at their nominal values. The DFIG
used has a nominal power of 20kw.
The model is simulated between t=0s and t=1.6s. We
make a voltage sag of 50% during 200 ms (Between
t=0.45s and t=0.65s)
The results in the Fig.4 (figures of the currents, powers
and the torque) prove the validity of proposed model,
and the good stability of the system.
d-
eFig. 4. Dynamic behaviour of the DFIG during Voltage

sag of 50 %: a- Stator Current, b- Rotor current,
c- Active Power. d- Reactive Power. e- Torque
7.
Abbreviations
J , C fr : Inertia and viscous friction
-a-
: Mechanical speed
turbine : is the turbine angular frequency
Vds , Vqs , Vdr , Vdr Tow-phase stator and rotor currents
voltage
ds , qs , dr , qr : Tow-phase stator and rotor flux
ids , iqs , idr , iqr Tow-phase stator and rotor currents

I dms , I qms Tow-phase measured rotor currents
Cr , C e Prime mover and electromagnetic torque

Rs , Rr : Per phase stator and rotor resistance
M : Magnetising inductance
-b-
321
Ls , Lr T: Total cyclic stator and rotor inductances

s , r : Pulsation
g : Generator slip
id , iq Tow-phase grid side converter currents
9. References
1.
2.
3.
4.
v d , vq Tow-phase network voltages

vdc , idc : Dc-bus voltage and current delivered by the
grid side converter
P , Q : Active power delivered by the grid side
converter and
im , Pm ,: Current and Active power delivered to the
5.
rotor side converter

C : Capacitance of the DCink
R , L Resistance and Inductance of the grid
Ps , Pr : Stator active power and rotor active power
6.
7.
Qs Qr : Stator reactive power and rotor reactive power

v ds , v qs , v dr , vqr :Tow-phase stator and rot
8. Conclusion
In this paper a doubly fed induction generator
connected to the grid was presented.
The aim of the paper was to develop the decoupled d-q
vector control technique of DFIG supply by a back-toback PWM converter in the rotor side. The control
strategy contains two control levels: DFIG control level
(control of active and reactive power using sliding mode
approach) and the control of the grid side converter.
The mathematical model of the system (DFIGConverters-Grid) has been implemented in MATLAB &
Simulink.
The main goal of the grid side converter control is to
keep the dc-link voltage constant by balancing the real
power on the machine side and on the grid side
converter, and to compensate reactive power of the
DFIG to get the unity power factor.
The control system is applied to the rotating reference
frame fixed on the gap flux of the generator. It has been
proved that this control system can control the active
and reactive power independently and stably.
The simulation results show that we can make
decoupling between active and reactive power and in
the same time have a good stability in a case of voltage
sag.
Also, the DC-link voltage was kept constant to the
reference value. The torque has the same dynamic like
the active power.
322
P. Kundur, J. Paserba, Definition and

Classification of Power System Stability,
IEEE Transaction on
Power Systems, Vol. 19, N2, May 2004.
R. Pena, R.C. Clark, G.M. Asher, Doubly Fed
Induction Generator using Back-to-Back PWM
Converters and its Application to VariableSpeed Wind-Energy generation, IEE ProcElect. Power Appl, Vol. 143, N5. 380-387.
1996.
L. Mihet-popa, I. BOLDEA, Control
Strategies
for
Large
Wind
Turbine
Application, Journal of Electrical Engineering,
Vol. 7, Edition 3rd, ISSN 1582-4594. 2006.
P.C. Krause, Analysis of Electrical
Machinery, New york, McGraw-Hill. 1994.
B. Toufik, Commande Structure Variable et
Etude de lIntegration dEolienne base de
Machine asynchrone Double Alimentation sur
le Rseau Electrique, PhD thesis (in French),
University of NANTES, 2007.

SELECTED HIGH-FREQUENCY POWER SUPPLY SYSTEMS

Elbieta SZYCHTA, Kamil KIRAGA
Technical University of Radom, Faculty of Transport, Poland
Table 1. Sources of power supply to induction heaters

[6]
Abstract: Principle of operation of basic power supply

systems of induction heating equipment is presented.
Principal inverter systems employed in induction
heating, i.e. transistor voltage and current inverters, are
discussed and design of E-class resonant inverter is
presented.
Keywords: power supply systems, frequency, series
inverter, parallel inverter, E-class inverter, MOSFET
transistor.
1.Introduction
Source
Frequen
cy range
Power range
efficiency
Advantages
Disadvantages
Comments
Cycloconvertors
16Hz
25Hz
Several
dozen kW
hundreds
of kW
=70. .80%
Simple desing of
the power part and
control system
Many valves
Rarely used only

to heat billets of
greatest diameters,
good power
source for mixers
Power mains
50(60)
Hz
Several
dozen kW
hundreds
of kW
MW
Easy access to the

power source
(mains), high
power, easy
choice of typical
elements
Low frequency,
low power
density, nonsymmetrical
mains loading
Static magnetic
frequency
multipliers
150Hz
450Hz
Several
dozen kW
hundreds
of kW
=60...80%
Nature of the
operation, simple
design, low
investment, high
reliability, simple
operation
Constant
frequency, large
dimensions, heavy
weight
Rarely used, good

power source for
pool furnaces
Electric
mechanical
generators
960 Hz
2000 Hz
8000 Hz
10000 Hz
Hundreds
of kW
2MW
= 60...80%
High reliability
High investment,
troublesome
operation, large
dimensions, heavy
weight, necessary
to build
foundations, long
start-up, water
cooling, noise
The only mediumfrequency power

source until mid1970s, superseded
by semiconductor
static inverters
Semiconductors
statistical
converters
300 Hz
500 Hz
Severel kW
Hundreds
kW for MW
= 90...95%
Completely static
power and
frequency, high
reliability,
small overall
dimensions and
weight,
immediately ready
to work
Requirements
qualified persons
for operation,
Gentle
Basic sources of
power supply to
induction heaters
Lamp
generators
300 kHz
MHz
Tens of kW
hundreds kW
= 50... 70%
Completely static
operation, high
reliability,
immediately ready
to work
Low power,
constant
frequency, lamp
life 4000...6000h
The only highfrequency power

source, now
superseded by
semiconductor
static connverters
The frequency range of power supply to induction

heating coils is very broad, starting from 16
Hz to as
many as several mega Hertz. Therefore, designs and

operating principles of power supply systems to such
equipment are varied. The equipment supplied with the
mains voltage frequency, i.e. 50 Hz, is the primary
group of induction equipment. It will not be addressed
in this paper, however.
The induction heating equipment of frequency increased
to approximately 10 Hz deserves consideration.
Originally, this equipment was supplied from such
electromagnetic sources as electric mechanical
generators or magnetic frequency multipliers. These
solutions became obsolete, however, as inverters came
into use [3, 7].
High-frequency equipment is supplied by means of
transistor inverters including DC intermediary circuit.
Employing IGBT transistors, power supplies of up to 50
kHz can be designed, while MOSFETs provide for
frequencies of even 500 kHz. When E-class inverters
(special transistor circuits) are utilised at low power,
induction heating systems can operate at frequencies of
up to 1 MHz [7].
Reduced frequency, e.g. 16
A power supply system consists of a rectifier assembly

with a smoothing filter, transistor direct voltage
regulator, and a system of overvoltage or shorting safety
switches. These subassemblies, depending on the
inverter type, provide a current or voltage source of
regulated capacity.
Both in the case of series and parallel inverters, noncontrol rectifiers are most commonly utilised,
constituting a voltage source together with an
appropriate filter. An LC filter in the DC circuit or a
filter consisting of alternating voltage chokes and DC
circuit capacitor may be employed.
With regard to high-power (over 25 kW) parallel
inverters, thyristor control rectifiers including a
smoothing choke acting as filter are utilised as well (the
control rectifier also acts as a regulator) [7].
Hz, servmalles to supply
power to induction equipment heating charges of very

great dimensions, where highly uniform temperature
distribution across the charge is required. Cycloinverters
are used to supply power to this equipment.
Table 1 summarises the most common sources of power
supply to induction heaters.
323
Series voltage inverter power supply circuits contain the

classic buck voltage regulator. It has a voltage output in
the case of series inverter and a current output in the
case of parallel inverter.
IGBT transistor, mentioned before, is the most
commonly used valve. Due to design considerations,
chopper type insulated electric modules are best.
A properly designed transistor snubber enables to
increase the switching frequency to several dozen kHz.
This provides for reduced dimensions of the capacitor
and choke regulator.
Resonant DC/DC buck inverter may be employed as
regulator where a inverter must meet the additional
condition of low electromagnetic interference (EMI).
with the aid of a receiver

. The operating principle
of this inverter is shown in Figures 3 a and b.
a)
b)
Fig.3. Waveforms of: current across the receiver
2. Transistor voltage inverters
capacitor voltage
Transistor voltage inverters comprise a series resonant

circuit and are powered from voltage sources, where the
supplied direct voltage can be regulated. There are two
designs of transistor voltage inverter most commonly
used to supply power to induction equipment: one
including a bridge circuit (Fig.1) and one including a
half-bridge circuit and capacitance voltage divider,
shown in Fig.2. Both the circuits can be modified by
parallel connection of commutation capacitors.
T1 -
, and voltage across the transistor
a) operating range II 0,5

b) operating range III
< <
>
The inverter may operate in three frequency ranges:

- range one 0< <
- range two
< <
- range three >

The operating conditions are best at = between the
2nd and 3rd working range. The load current is close to
sinusoidal, the transistors are on and off at zero current.
The backward diodes do not conduct the current. In
effect, power losses across the semiconductor valves are
the lowest [5].
The output power may be controlled using one of the
three methods:
a) the amplitude method involves varying the average
Fig.1. Design of voltage inverter supplying power to induction

equipment [4]
direct voltage
. This method is the most common, yet
it requires application of a regular-value source of direct
voltage.
b) the frequency method involves varying the frequency
f of transistor switching in a range comprising the
operation ranges, usually range 2 and 3, where
, for
instance, remains constant. It is not employed in
inverters working at high frequencies since the
switching losses rise during resonant tuning.
c) the method of voltage pulse band width regulation at
It involves an appropriate phase shift of
the signals controlling the transistors T1 and T4 in
relation to the signals for T2 and T3 at a constant
frequency of their turn-on.
Fig.2. Design of voltage inverter supplying power to induction

equipment half-bridge circuit including a capacitance
voltage divider [4]
3. Transistor current inverters
The inverter in Fig.1 consists of four transistors (T1, T2,

T3, T4) connected to form a bridge circuit. In addition,
each transistor is shunted by means of a backward diode
(in MOSFET inverters, backward diodes are included
within the transistor's design). The series resonant
circuit comprises: capacitor
Transistor current inverters contain a parallel resonant

circuit and are supplied from direct voltage sources of a
current nature, therefore, the shape of transistor current
approximates a rectangle, and the form of the receiver's
voltage is approximately sinusoidal.
and an inductor shown
324
Figures 4 a and b illustrate principal designs of the

parallel current inverter.
a)
b)
Fig.6. Basic designs of E-class inverter [1]:

a) main circuit of the inverter
b) possible solutions of the systems matching load parameters
to the inverter
a)
Single-transistor resonant inverters may implement

three types of operations:
a) optimum operation, where transistor switching is
most beneficial: turn-on at zero current and voltage,
turn-off at zero voltage and non-zero current[6].
b) sub-optimum operation, where load resistance
decreases, for example. The transistor is turned on and
off at zero voltage and non-zero current. Switching
power losses are somewhat greater than in the optimum
operation.
c) non-optimum operation, where the transistor is turned
on at non-zero voltage and non-zero current and turned
off at zero voltage and non-zero current. This operation
may occur when parameters of the receiver change.
Power losses are greater than in the sub-optimum
operation, therefore it can be accepted at a relatively
low load current [7].
b)
Fig.4. Designs of the parallel current inverter [7]:

a) basic design a bridge circuit and, below, a bridge circuit
including a shunt choke
b) half-bridge circuit including a double current source; below,
a half-bridge circuit including a double current source and
commutation chokes
a)
b)
Fig.5. Waveforms of selected magnitudes in the parallel

current bridge inverter:
a) operating range
<
b) operating range >
Similarly to voltage inverters, transistor switching

losses are lowest at the resonant frequency =
It is impossible to decide clearly which of the inverters,
voltage or current, is better suited to supply power to
induction equipment. A final choice of a inverter
depends on the following factors: power rating of the
equipment, operating frequency, economics, types of
transistors used, and manufacturer's research
experience.
Fig.7. Voltage and current waveforms in the basic inverter

circuit (without a matching circuit) during optimum operation
: transistor's voltage and current; 1 the transistor is
4. Single-transistor resonant (E-class) inverters
on, 2 the transistor is off, T =
- operating interval of the
inverter
In E-class inverters, transistor switching losses are

reduced to a minimum. This enables high-efficiency
operation at very great frequencies of the input voltage
(up to 1Mhz). It can be achieved by means of
appropriate control techniques, so that the transistor
should be switched when its voltages and/or currents are
equal or close to zero [5].
5. Conclusion
This paper was intended to present methods of
supplying power to induction heating systems in the
range from medium to very high (MHz) frequencies. In
practice, their designs vary. The differences are
primarily determined by the intended use and operating
conditions of the inverters, and by the type of power
325
electronic elements (their limitations and capabilities)

used to design the inverters. Economics, or savings, are
of major importance as well. With regard to inverters,
this is related to reducing power losses in the process of
transistor turn-on and switching. It is therefore essential
to design the inverters in such a way that the power
losses could be minimum with increased operating
frequency of the transistor switching.
6. References
1. Junak J. Analysing Possible Application of E-Class
Inverters Including MOSFET Transistors to HighFrequency Induction Heating.
2. Kempski A., Smoleski R., Strzelecki R. Common
Mode Current Paths and their Modeling in PWM
Inverter- Fed Drives PESC 02, p. 1551- 1556.
3. Luft M., Szychta E., Szychta L. Method of designing
ZVS boost inverter, Proceedings of the 13th
International Power Electronics and Motion Control
Conference, 1-3 September 2008, Pozna Poland,
p. 478-482.
4. Rodacki T., Kandyba A. Electrothermal Equipment,
Gliwice 2002.
5. Skoczkowski T. Induction Heating, Przegld
Elektrotechniczny LXXII 10/1996.
6. Sajdek C., Samek E. Induction Heating,
Wydawnictwo LSK Katowice 1985.
7. Szychta E. Resonant Inverter Powered by DC
Voltage Source, Automation and Control
Technologies 2004, Proceedings of the V
International Conference, Kaunas 2004, p. 26-30.
326

IDENTIFICATION OF PARAMETERS OF SYNCHRONOUS GENERATORS

DYNAMIC MODELS
Mindaugas AUBALIS*, Vaclovas AUBALIS*, Audrius JONAITIS*, Ramunas PONELIS*,**
*Kaunas University of Technology, Lithuania; **Lietuvos energija AB, Lithuania
and short circuit and rotor inertia. The detailed list of

parameters used in dynamic model is presented in
table 1.
Abstract: The identification technique for determination

of parameters of synchronous generators dynamic
models for large scale power system stability studies is
presented in the paper. The described methods allow to
identify parameters according to field test data as well as
to the data presented by manufacturers.
Table 1. Parameters of synchronous generator
Inductive reactances and resistances, in p.u., at rated

power and voltage SN and UN
d-axis synchronous inductive
1.
xd
reactance
q-axis synchronous inductive
2.
xq
reactance
3.
d-axis transient inductive reactance
x'd
4.
q-axis transient inductive reactance *
x'q
d-axis subtransient inductive
5.
x"d
reactance
q-axis subtransient inductive
6.
x"q
reactance
7.
Leakage reactance
x
8.
Stator resistance
ra
9.
Field circuit resistance **
rf
Field circuit time constants
d-axis open circuit transient time
10.
T'd0
constant **
q-axis open circuit transient time
11.
T'q0
constant *
d-axis open circuit subtransient time
12.
T"d0
constant
q-axis open circuit subtransient time
13.
T"q0
constant
Total inertia of generator, turbine and exciter
14. Inertia constant, s
TJ
15. Moment of inertia, kgm2 (GD2/4) or
J
16. GD2, kgm2
GD2
Open circuit saturation
17. Saturation at rated voltage UN
S(1.0)
18. Saturation at voltage 1.2UN
S(1.2)
Note: * the parameter is not used for salient rotor
(hydro) generators;
** T'd0 and rf are given at certain temperature of field
winding during measurement.
Keywords: synchronous generator, dynamic model,

identification, power system stability.
1. Introduction
Accuracy of power system stability investigation and
reliability of the study results depends on the accuracy
of the used dynamic models of generating units. More
accurate models ensure more reliable evaluation of
maximum permissible capacities of power flows,
loadings of separate generating units and possibilities of
transient processes control.
Identification of dynamic models for power system
stability and transient studies requires large amount of
information about the equipment and its characteristics.
The lack of data may make uncertainties in evaluation of
dynamic models of generating units as well as
simulation of operating conditions of power system.
The identification techniques for determination of
parameters considering the field test data and data
presented by manufacturers or operators are described in
this paper.
2. Parameters of synchronous generator model
The producing of synchronous generator dynamic
models and the calculation of their parameters need to
know the type of generator, and the nominal total power
SN, active power PN, nominal power factor of cos, the
nominal generator voltages and currents UN and IN, the
field voltage and current UfN and IfN and the power
diagram QG=f(PG).
The main generator dynamic parameters are the
inductive reactances of direct-axis and quadrature-axis,
time constants of field winding at open circuit (no load)
327
Generator dynamic model parameters can be identified

from the regime test or the frequency response test data.
U0
U"0
3. Evaluation of dynamic model parameters

according to field test data
U t U U 0 U e
or
U 0 U 0 e
U t U 0 I 0 xd U 0 I 0 xd U e
U 0 I 0 xd U 0 I 0 xd e
Td//0
I 0 xd xd e
At the subsequent voltage curve point (when the shortterm voltage component extinct) derived tangent to the
line that corresponds settled voltage U a direct axis
transient time constant T'd0 can be determined.
Extrapolating the curve of the exponential transient
voltage U' to the stoppage time the value of U'0 is
determined. According to it, the direct axis transient
inductive resistance x'd value is determined as follows:
xd
(1)
xd
(2)
(3)
where U0, I0, U are initial voltage and current and the
steady state voltage of the disconnected generator; U'0,
U0 initial values of transient and subtransient
voltages, T'd0, Td0 direct axis open circuit transient
and subtransient time constants, xd, x'd, xd direct axis
synchronous,
transient and subtransient inductive
reactances.
The variation in voltage when the generator that is
loaded with capacitive reactive load was switched off is
shown in fig. 1.
The initial value of the subtransient voltage U0 can be
expressed from the first voltage jump and a subtransient
inductive resistance value can be found:
U 0 U 0 I 0 xd ;
xd
U 0 U 0
.
I0
(6)
U0 U
.
I0
(7)
The value of the d-axis subtransient time constant Td0 is

defined by extrapolating the tangent at certain point of
the U0 curve to the derived U' curve.
The leakage resistance of the stator inductive x value is
not normally determined with the tests. It can be
assessed in accordance with the manufacturers data or
approximately - according to longitudinal resistance xd:
x 0.08xd.
Appropriate parameters of the q axis can be extrapolated
in accordance to typical d and q axis ratios that is
presented in table 2.
t
Td//0
U 0 U 0
.
I0
The d-axis synchronous inductive resistance value is

determined similarly:
t
Td/0
Fig. 1. The variation in voltage when the generator loaded

with capacitive reactive load was switched off
or
U t U 0 I 0 xd I 0 xd xd e
T'd0
T"d0
t
Td/0
I 0 xd
The parameters of the synchronous generator can be

identified according to two types of field tests:
regime test the disconnection of the generator
loaded by only reactive load;
frequency response test of the stopped generator.
It is possible to identify parameters of the generator daxis using regime test data with sufficient accuracy.
During the test of unloaded generator which consumes
the reactive power from the network, the terminal
voltage and current are registered.
Processing of test data according to the voltage
variation, the d-axis parameters xd, x'd, xd, Td0, T'd0 are
determined.
Terminal voltage of the disconnected generator can be
expressed as follows:
t
//
Td 0
I0 x'd
U'0
3.1. Inductive reactances, resistances and time

constants
t
/
Td 0
I0 x"d
Table 2. Typical ratios for the d and q axis generators

parameters
Parameters
xq
x'q
xq
T'q0
Tq0
(4)
Non-salient pole
generator
0.9 xd
1.5 x'd
xd
0.3 T'd0
Td0
Salient pole
generator
(0.6-0.7) xd
xd
0
Td0
In both cases of parameters identification according to

regime test and frequency response test of the stopped
generator, the test temperature B and the identified time
constant T'd0B must be taken into account and the value
of excitation windings resistance rfB need to be adjusted
to the winding temperature of the nominal regime N.
(5)
328
234.5 B
Td0 Td0 B
;
234.5 N
rf rfB
Lq ( s ) Lq
(8)
234.5 N
.
234.5 B
The data of the stopped generator frequency response

test allows identify main parameters of the d and q axis:
xd, x'd, xd, Td0, T'd0, xq, x'q, xq, Tq0, T'q0 and
resistances of the stator and rotor ra and rf.
The analyzed synchronous generator may be presented
by the operator expressions for the direct and quadrature
axes:
(10)
q ( s ) Lq ( s ) iq ( s ) ,
(11)
Ld ( s)
Lq ( s)
Equivalent
Scheme of
d axis
id
b)
1 sT1d
Ld ( s ) Ld
(1 sTd ) (1 sTd )
(1 sTd0 ) (1 sTd0 )
(15)
Td Td
Td0 Td0
Tq Tq
Tq0 Tq0
Ld Ld ( j) Ld
From the expressions of the flux linkage generator

transfer function and inductances expressions were
made:
(1 sTd0 ) (1 sTd0 )
(14)
(16)
(17)
(18)
Dynamic expressions of the inductances without the

damping windings (the second rotor contour, contour
with large time constants) will be less complicated and
expressed as follows:
Fig. 2. Generalized schemes of the synchronous generators

direct axis (a) and quadrature axis (b)
G ( s) G0
Z q ( s) ra
Lq Lq ( j) Lq
Equivalent
Scheme of
q axis
iq
Z d ( s) ra
,
s
Ld Ld ( j) Ld
efd
a)
where ra is the active resistance of stator windings,

measured during the test at the temperature of windings.
With operator transfer functions and inductance values
and expanded with polynomial ratio the dynamic
parameters of Ld , T'd0, Td0, T'd, Td and Lq, T'q0, Tq0,
T'q, Tq were determined. Parameters normally used to
determine by frequency identification methods [1].
During the rapid changes of measured value, when
s=jj, marginal values of the inductances Ld(s),
Lq(s) will be equal to transient inductance values Ld and
Lq . According to (2.11)-(2.12) expressions, Ld and Lq
expressed following:
Where efd(s), d(s), q(s), id(s), iq(s) are exciter

voltage, the variation of flux linkage and current
components, G(s) transfer function between rotor and
stator, Ld(s), Lq(s) inductances in operator form of d
and q axis.
The one-line diagrams corresponding to the (10) and
(11) and representing the synchronous machine are
shown in fig. 2. The variation of input and output signals
are measured or simulated at the corresponding
terminals of the blocks.
(1 sTq0 ) (1 sTq0 )
During the test, stator voltage and current and rotor

current instant values were recorded.
During the test data processing complex input reactance
of the d and q axis Zd(s) and Zq(s); transfer function
rotor-stator sG(s) for the different frequencies
(s=j=j2f) were determined. Operator values of the
inductances Ld(s) and Lq(s) were determined estimating
stator resistance
(9)
d ( s ) G ( s ) efd ( s ) Ld ( s ) id ( s ) ,
(1 sTq ) (1 sTq )
Lq Lq ( j) Lq
Td0
Tq
Tq0
(19)
(20)
Rotor poles of the hydro units that are made of the steel
sheets and the free currents closes through damping
windings of rotor transverse axis. Hydro generators that
is usually modelled with 2.1 model, now are designed
with one contour in the transverse axis. In case, when
damping windings time constants are smaller than the
excitation windings time constants, it is considered that
(12)
Td
(13)
329
there is no transient inductance or transient time

constants, just transient inductance Lq and the open
circuit and short circuit time constants Tq0, Tq.
Expression of the transient inductance Lq is similar to
the expression of the transient inductance of the
turbogenerator:
Lq Lq ( j) Lq
Tq
Tq0
I fi Eqi D e
(21)
The inertia time constant TJ of the generating unit is

determined from the generator tripping test, where the
d
initial speed
t 0 is measured and the generator is
dt
loaded with low active load P* :
ln
C
(22)
3.3. Saturation characteristic
I fM EqM
EqK EqM
I fK EqK
e
C EqK Eqt
(25)
(26)
Almost all data necessary for identification of dynamic

model can be listed in technical specifications of the
modern generating units. The missing parameters may
be identified according to startup tests. However, the list
of required parameters in old generators specifications
usually is not sufficient. If there is no possibility to
perform field tests, the parameters can be evaluated
approximately. The main factors for determination of
parameters are type of the synchronous generator, rated
power, voltage and rotation speed.
The direct axis synchronous inductive reactance xd can
be evaluated according to ratio of field currents rated
values IfN and open circuit value If0 and rated power
factor cosN.
If the open circuit characteristic is known as well as
rated values of field current and power factor, the xd can
be calculated considering the rated value of field current
IfN* in per unit on base of field current value
corresponding the source voltage at rated voltage of
linearized characteristic (Ib= If,te1):
Eq
1.2
1.0
Eq,t
I
I
If1.0(0)f1.0 f1.1
I fK EqK
4. Evaluation of dynamic models according to data

presented by manufacturers and operators
Generator saturation values of S(1.0), S(1.2)

characteristics are determined by the open circuit (no
load) characteristic (fig. 3).
If,,t
(24)
where EqK*, EqM*, IfK*, IfM**, Eqt* are source voltage

values of the points of K and of M and the
corresponding excitation currents and the values of the
nonlinearity start in source voltage expressed in per
units.
The initial rotor speed should be recorded with

sufficiently high sampling frequency, and the time
interval should be within 0.01-0.1 s range.
D
C E E
e qi qt ;
Eqi
Idle characteristic is usually measured to 1.1UN and it

continues to be extrapolated to 1.2UN.
According to the two nonlinear idling characteristic
points K and M and the coordinates of the start of
nonlinear approximation source voltage to find the
coefficients C and D:
3.2. Inertia constant
P
.
d
t 0
dt
(23)
where Eqi* and Ifi* source voltage and field current at

ith point of open circuit characteristic, in p.u.; Eqt*
source voltage value at the end of linear characteristic,
in p.u.; C and D coefficients of the approximated
characteristic.
The parameters of the transverse and direct axis

equivalent schemes could be determined according to
basic dynamic parameters values.
The test of the frequency response for the stopped
generator is recommended only when the routine
maintenance is completed and the generating unit is off
for long time.
TJ
and the values of the saturation curve:

S Eqi
C ( E qi E qt )
If1.2 If
Fig. 3. Open circuit saturation characteristic
Any source voltage Eqi* corresponding to the excitation

current of Ifi* can be expressed as follows:
330
EqN I fN
I fN
.
I f, te1
(27)
xd
The rated source voltage for non-salient pole generator

equals
E qN U N I N* ra* jxd* ,
and the absolute value is
xd
(30)
(31)
EQN EqN I dN* xd* xq*
EQN
(32)
Marking the ratio of xq and xd as

k qd
xd
a quadratic equation can be written:
U N* cos N 2 U N* sin N I N* k qd xd 2
I fN* 1 k qd I dN* xd
0.
1 .1
.
K scr
(39)
Ratio
x"d / xd
Turbogenerator
0.10
Hidrogenerator
0.2
x'd / xd
0.16
0.3
TJ
(35)
J 2N 0.00274 GD2 n 2
,s
SN
SN
(40)
where n rotating speed in min-1, GD2=4J , kgm2 and

SN rated total value in VA.
In the literature of the post-Soviet countries, the value of
inertia constant is given on base of the rated active
power (TJPN). The inertia constant on base of total power
is calculated by
For rated conditions, the (35) can be rewritten as
(38)
Inertia constant of generating unit equals 5-10 s [4]. The

data about units inertia can be given in three ways:
value of moment of inertia J=GD2/4, kgm2; value of
GD2, kgm2 or inertia constant TJ, s.
Inertia constant can be determined according to the total
moment of inertia of generator, turbine and exciter:
(34)
Table 3. Approximate ratios of inductive reactances of

synchronous generators
U N* cosN 2 U N* sinN I N* xq* 2 . (33)
xq
If the power factor is unknown, its value can be taken

approximate: the power factor of turbogenerators under
120 MW and of hydrogenerators under 50 MW equals
cosN=0.8, of turbogenerators 120-500 MW and
hydrogenerators beyond 50 MW cosN=0.85, and of
turbogenerators beyond 800 MW equals cosN=0.9.
According to xd value, other parameters of synchronous
generator can be determined. Approximate ratios of
inductive reactances can be used (table 3) [3].
The approximate values of time constants are: T'd0 = 6.09.0 s, T"d0 = 0,12 s.
When the stator resistance is not taken into account, the

source
voltage
of
salient
pole
generator
(hydrogenerator) EQN* can be expressed via open circuit
source voltage EqN* and rated voltage and current of the
generator:
Also, the d-axis synchronous inductive reactance can be

determined according to short circuit ratio Kscr:
Usually, the stator resistance is less than 0.5% of xd and

can be neglected. In this case xd in p.u. can be written as
I fN I dN* xd* xq* ,
2
xd 5 I fN
3I fN 2 I fN 2 .
where UN* and IN* - rated voltage and current of

synchronous machine, in p.u.; ra* and xd* - resistance and
inductive reactance of the stator, in p.u.
At rated conditions, the voltage and current equals to
one, then
The majority of hydrogenerators has the ratio of

reactances kqd0.6 and IdN*=0.75-0.8, therefore, xd can
be approximately expressed as
EqN UN* cos I N* ra* 2 UN* sin I N* xd* 2 ; (29)
2
2
xd I fN
* cos N sin N .
2
kqd
2
kqd sinN IfN* 1 kqd IdN*
IfN
* 1
.
2 2
2
2
2
kqd 1 kqd 2 IdN

kqd 1 kqd IdN*
*
(37)
(28)
2
2
xd I fN
* cos N ra * sin N .

2
1 kqd I dN
*
kqd sin N I fN* 1 kqd I dN*
2 I dN2 *
2
xd 2kqd sinN IfN* 1 kqd IdN* IfN
* 1 0 . (36)
2
xd2 k qd
1 k qd
The value of d-axis synchronous inductive reactance of

salient pole generator is the solution of (36):
331
TJ TJPN cos N , s.
(41)
6. References
The inertia can be described as a relative kinetic energy

H of rotor of rated operating conditions, in s:
H TJ / 2 , s.
1.
(42)
5. Conclusions
2.
The identification techniques of synchronous generators

dynamic models parameters are presented in the paper.
The most accurate model can be identified if the special
field tests of generating units are performed to obtain the
data need for parameters evaluation. If such tests are
not available, the parameters can be determined
approximately according to some data given by
manufacturers.
3.
4.
332
Azubalis V., Virbickas D., Jonaitis A. and

Azubalis M. Synchronous Generator and Excitation
System Dynamic Model Parameters Investigation.
Proceedings of International Conference Electrical
and Control Technologies 2006, Kaunas, 2006.
p. 110-113.
Kundur P. Power System Stability and Control.
New York, 1993. 1176 p.
Neklepaev B.N., Krjuchkov I.P. Electrical part of
power stations and substations. Moscow:
Energoatomizdat, 1989. 608 p. (in Russian).
Sterninson L.D. Transient processes of frequency
and power control in power systems. Moscow,
1975. 216 p. (in Russian).

POTENTIAL OF NON-TRADITIONAL ENERGY SOURCES

FOR COLD RECOVERY IN ABSORPTION REFRIGERATING MACHINES
Michail G. BERENGARTEN*, Semyon I. VAINSHTEIN*, Aleksandr S. PUSHNOV*, Vytautas ADOMAVIIUS**
*Moscow State University of Ecological Engineering, Russian Federation
** The Centre for Renewables Energy Technologies at Kaunas University of Technology, Lithuania
Abstract: Two types of refrigerating units for air

conditioning systems vapor compression and
absorption types are examined in this paper. The
examined equipment is applied for creation of special
cooling systems in various large-scale objects (for
example: the large area storages). Several advantages of
the absorption refrigerating machines are shown there.
There is emphasized that efficiency of such units may
be increased by the substitution of individual boiler
plant as a hot water source for the non-traditional
energy source solar collectors for water heating. It
would result in savings of the fossil fuels and decrease
in pollution of environment.
cooling cycle. The energy is necessary only for

transportation of liquid and air to make the pump and
fan drives work.
Another serious feature of absorption refrigeration units
are low operating costs. They are approximately two
times lower in comparison with vapor compression
units [1].
In line with the positions mentioned above the
absorption refrigerating unit needs a source of hot
water. For example, single-stage absorption cooling unit
needs a hot water with temperature in a range
TW=9098 0.
Autonomous boiler plant traditionally is used as a hot
water source for needs of absorption refrigerating unit.
As an alternative there may be applied flue gas boiler,
gas turbine or gas-piston cogeneration unit as a source
of heating medium. The mentioned boiler units
permanently are under further development and they are
widely used around the world. They have a great
capacity and minimal nitrogen oxide emissions into the
atmosphere. Some technical features of modern back
boilers are shown in Table 1.
Nevertheless, effective boiler plants WRTSIL that
provide high capacity (up to 9094%) permanently need
the fuel consumption gas or liquid fuel that have a
huge price at world market and is ranged. As for the
efficient boiler plants BioEnergy produced by
WRTSIL (Finland) that use the woodworking
wastes, the peculiarity is following: this type of fuel
isnt available everywhere in a sufficient amounts. The
last WRTSIL design project is three-generation
system that presumes production of the power and heat
(hot water) in winter time and the cooling generation in
summer as well.
But these units may be considered as the threegeneration ones only taking into consideration the
whole annual period, because during the season periods
(summer or winter) they work as co-generation units, or
in other words, they produce the power and heat in
winter months and the power and cooling in
summertime.
Keywords: air conditioning systems, absorption

refrigerating machines, cogeneration, cooling, solar
collectors.
1. Introduction
In many countries and regions the climate conditions
require creating special refrigerating systems for various
large-scale objects including storage and production
workspaces with area up to 100.000 m2. In several
cases the mentioned above goal can be achieved by
usage of air conditioning systems equipped with
autonomous gas boiler plants in combination with
absorption refrigeration units or vapor compression
units.
2. Comparison of various types of refrigerating units
As indicated by calculations performed in reference [1],
the outlay for usage of absorption refrigerating units and
vapor compression units are three times better for
absorption refrigeration units. The mentioned above
comparative calculations [1] were made by taking the
air conditioning systems capacity Q=1000 kW at the
taken cooling coefficient =3,0.
Absorption refrigerating units have an important
advantage in comparison with vapor compression units
they never consume the energy for the providing of
333
where T is the number of days when refrigeration is

required, N=200;
k l loading factor of the refrigerating unit,
(0,5);
k md coefficient of the dimensions matching,
(3600. 103).
Boiler oil 100

kg/h
Gas, nm3/h
Efficiency, %
Diesel fuel,
kg/hour
Fuel consumption
width
21
MW
22
MW
23
MW
35
MW
415
MW
Overall
dimensions, m
length
Boiler plant
Table 1. Technical features of back boiler plants S3V

produced by WRTSIL. Source: [2], [3]
5,3
9,8
198
200
220
922
6,9
9,8
396
400
440
922
6,9
9,8
594
600
660
922
10,0
13,8
1485
1500
1650
922
24,3
14,3
5940
6000
6600
922
After the substitution of the given above values of all

parameters we get the total volume of the natural gas to
be consumed:
V g 2 040 635 m3
The cost of such gas volume in Russia is the following:
C 2040635 1,85 3 775 175 roubles.
Replacement of the traditional hot water source in an

absorption refrigerating unit by the solar collectors for
water heating of the adequate capacity gives a
significant economy. In this case cost of the solar heat
energy varies from 0,04 to 0,30 Euro for 1 kWh
depending on solar energy resources in the particular
region [5].
According to data of Lundh and Dalenbck [4], for
example, to Anneberg (Sweden) conditions (July, 2003)
in period from 12.00 to 18.00 hours the large flat plate
solar collectors array of 240 m2 installed on a
residential building may provide hot water of 60-70 0
temperature that is adequate to 100-140 kW capacity
of the array.
Usage of solar energy is limited in a flow of daytime
and seasons of year and depends on the geographic
latitude of the locality. So, for Central and Northern
Europe the maximum of solar radiation goes to the
period from April-May to September (look Table 2).
3. Usage of solar energy

Alternative option of hot water source is usage of solar
collectors for water heating.
As an example lets examine the single-stage absorption
refrigerating unit using the hot water from autonomous
boiler plant [1]. As a rule, the single-stage absorption
refrigeration units are using hot water with temperature
TW=9098 0 have the following coefficient:
Q0
0,75 ,
Qh
(1)
where Q0 is refrigerating capacity, kW;

Qh input of heat to refrigerating unit from the
heating medium, kW.
Table 2. Value of solar radiation in Europe, kWh/m2 a day
It is required to burn the certain volume of natural gas in

the boiler plant in order to get the necessary
refrigerating capacity:
Qg
Q0
January
February
March
April
May
June
July
August
September
October
November
December
Annually
(2)
where Q g is heat capacity of the boiler running on

natural gas, kW;
the efficiency of the boiler.
Taking Q0 1000 kW and efficiency 0,75 according to
the formula (2) we get Q g 1333 kW.
24 Q g T k l k md
Northern Europe
0.8
1.5
2.6
3.4
4.2
5.0
4.4
4.0
3.3
2.1
1.2
0.8
2.8
We emphasize that the alternative variant based on solar

collectors allows to get a somewhat lower potential of
hot water TW=70 0C in comparison with traditional
autonomous boiler plant potential (TW=90 0C). One of
the possibilities to raise the temperature of hot water
coming from solar collector is to use another available
relevant heat source for backing up the solar
installation.
Scheme of single-stage absorption refrigerating unit
using the solar collector as hot water source is presented
at Fig. 1.
Taking the boiler plant efficiency 0,92 as it is given in

Table 4 [2], at caloricity of natural gas =35 000 kJ/m3
it is possible to calculate the common consumption of
natural gas during the hot period of year (for example,
200 days):
Vg
Central Europe
1.7
3.2
3.6
4.7
5.3
5.9
6.0
5.3
4.4
3.3
2.1
1.7
3.9
(3)
334
Cooling water (to CT)
Returning water (to the heater)

Hot water (from
the heater)
Cooling water (from CT)

Generator
P2
Solar
collector
Cooling water (to user)
Tank
P1
Evaporator
Cooling water (from user)
Absorber
Pump of cooling
agents recirculation
Pump of weak solution

Pump of strong solution
Fig. 1. Single-stage absorption refrigerating unit using hot water from the solar collectors
In order to upgrade the reliability of operation the

suggested scheme of single-stage absorption
refrigerating unit is additionally equipped with hot
water storage drums with vertical position of the

polyethylene tubes 322,4 mm in the seal laid down to
the 100 meters depth (look Fig. 2).
Returning water (to the heater)
Cooling water (to CT)
Hot water (from

the heater)
Cooling water (from CT)

Generator
P2
Cooling water (to user)
Tank
P1
Cooling water (from user)
Solar
collector
Evaporator
Heater
Heat
storage
drums
system
Absorber
Pump of cooling
agents recirculation
Ground level
Pump of weak solution

Pump of strong solution
Fig. 2. Single-stage absorption refrigerating unit using hot water from the solar collector with heat storage drums system in the seal
under ground; 1 polyethylene tubes in the seal
Another solution of engineering, which can be used for

the utilization of solar energy in order to diminish the
expenditures for heat energy necessary to run the singlestage absorption refrigerating unit, is described in
reference [6]. According to Seleznev patent [6], the
solar energy accumulator made as a reservoir with
mirror walls presumes can be implemented for this goal.
The scheme of such solar heat energy accumulator is

presented in Fig. 3.
Simple construction of the reservoir with integrated
collector was suggested by Kuinskas and Liubarskiy
[7]. The solar collector is combined there with the water
accumulation reservoir. The specific capacity of the
reservoir is recommended approximately 40-50 dm3 per
1 m2 of the collectors area.
335
During the period of low solar radiation the biogas

energy may be used to get a hot water for the absorption
refrigerating unit. In this case the hot water preparation
block is operating in a sufficient way.
One of the possible variants of the scheme for
production of hot water based on the biogas energy is
shown in Fig. 4 [8]. In this case the hot water with
temperature w = 80-850 is discharged in the
absorption refrigerating unit. The scheme includes the
following constructions: methane tank, solar collectors,
block of biogas cleaning, internal combustion engine
and the heat pump. According to data of Rabinovych
[8], capital costs could be returned in two years after the
installation of this hot water system.
Fig. 3. Solar energy accumulator according to the patent [6]
T21
T3
T11
Biogas
T1
T2
B1
4
5
K1
Cool and clean

sewers
Fig. 4. Scheme of installation, which uses biogas and heat of sewage for the heat supply system: 1- methane tank; 2 solar
collectors; 3 block of biogas cleaning; 4 internal combustion engine; 5- heat pump plant; 6 condenser; 7 evaporator; T1
water from cooling system of internal combustion engine; T11 water in heating system, t = 80-85 0C; T21 water from the heating
system, t = 55-60 0C; T3 hot water in hot water supply system, t = 50-55 0C; B1 cool water from the running water supply
system; K1 sewage
3.
4. Conclusions
1.
2.
3.
Some alternative systems using solar and

biogas energy for hot water supply to the
absorption refrigerating unit are proposed in
this paper.
Usage of the vertical tubes laid down up to the
depth 100 meters under ground and the
accumulating reservoirs are suggested here for
the accumulation of hot water.
The biogas energy may be used to provide the
absorption refrigerating unit with hot water
with the temperature w = 80-85 0C during the
period of low solar radiation.
4.
5.
6.
7.
5. References
1.
2.
..
//
. --,
3(19), 2005. p.p. 20-24.
www.wartsila.com (called on 02.16.2009).
8.
336
WRTSIL.W-P 09.2004/Bock/s
Office Waasa Graphics.
Lundh M. Dalenbck J.-O. Solar heated
residential area Anneberg initial evaluation //
LTMA mokslo darbai, 2. Klaipda,
Klaipdos universitetas, 2005. p.p. 142-153.
Kytra S. Atsinaujinantys energijos altiniai.
Kaunas, KTU, 2007. 140 p.
..
. 2267060.
., . c
.
.
2- (2-3
2000 . .-), 2. ., 2000. p.p.
405-410.
Rabinovych M.D. Renewable energy sources
for sustainable development of historical cities.
Environment Protection Engineering, vol.32,
1, 2006. Wroclaw, 2006. p.p.81-86.

POSSIBLE PERSPECTIVES OF COOLING TOWER APPLICATION

FOR ELECTRIC ENERGY PRODUCTION
Aleksandr S. PUSHNOV*, Michail G. BERENGARTEN*, Semyon I. VAINSHTEIN*, Vytautas ADOMAVIIUS**
** The Centre for Renewables Energy Technologies at Kaunas University of Technology, Lithuania
Abstract: Possibilities of usage of the cooling tower

combined with non-traditional power generating units
were examined on the base of analysis of the
aerodynamic situation in the cooling towers with natural
draft. One of the schemes presumes the location of axial
air flow turbine in alignment to air outflow on the axis
of tower symmetry at its upper part. Another possible
scheme presumes the location of several small wind
turbines at the top of the cooling tower. The half of
outer cover of the cooling tower oriented to south is
proposed to blacken artificially in order to increase the
power of the air flow inside the cooling tower. The
discussion also deals with the installation of
photovoltaic modules on the bulk of cooling tower:
electric energy produced by them could support the
feeding of cooling towers pump equipment, which is
used for the cooling of recycling water under normal
exploitation mode.
Fig. 1. General view of counter-flow cooling tower HAMON

(France)
Keywords: cooling towers with natural draught, nontraditional energy sources, wind turbines, electric
energy.
1. Introduction
Ecological, economical and energy crisis currently
taking place in the world makes us to look for new
possibilities and new approaches in order to find out
new energy sources for heat and power generation. First
of all, we look for usage of renewable energy sources,
which are universally available, free from payment and
ecologically clean.
Some possible ways to use the cover of acting or
suspended cooling tower for the installation of small
scale wind turbines and (or) photovoltaic modules are
described in this paper. The suggestions discussed
bellow are based on the analysis of peculiarities in
aerodynamic and heat situation in the cooling towers of
various types.
General view and scheme of counter-flow cooling
towers with natural draught are shown at Fig. 1 and Fig.
2.
Fig. 2. Scheme of cooling tower
2. Aerodynamic situation in the cooling tower

As it is shown in a number of experimental and
theoretical researches [15], aerodynamic situation in
337
the cooling towers with natural draft is characterized by

serious irregularities in cross view. There is also
stressed a transformation of air speed field by the height
of cooling tower.
The following peculiarities of speed field in the cooling
tower are important from a viewpoint of possible use of
this wind energy potential:
3. Possible engineering solutions on the cooling

towers application for electric energy production
Lets examine the constructions and parameters of large
ferroconcrete cooling towers with natural draft that are
widely applied in power engineering. The capacity of
such cooling towers on cooled recycled water is
100.000150.000 m3/hour.
Main geometrics of Russian-made and foreign modern
counter-flow evaporative cooling towers are shown in
Tables 1 and 2.
symmetry of speed field with reference to the

cooling tower blocks axis of symmetry,
presence of stabile wall area of increased
magnitudes of air speeds vertical constituent
in the narrowest part inside the cooling tower;
as this takes place, the said area of increased
magnitudes of air speed near the tower walls
survives by height becoming a little bit
narrower at the point of the outgoing opening
of the cooling tower.
Table 1. Geometrics of some constructions of cooling towers
The air flow pattern in the cooling tower is illustrated on

the diagrams below and is taken from theoretical work
of Gonzles [3] with application of k- turbulence
model (Fig. 3, 4). The design values of air speeds
vertical and horizontal constituents in the cooling tower
with height 150 m there is shown according to data [3].
Reference
[7]
[4]
[6]
Gage
diameter, m.
125
Height of
offtake
windows, m.
10
Height of
tower, m.
150
120
[13]
104,5 64,4
80
82
97
108
90
102
110
130
On the
upper level
of nozzles
At the
outgoing
opening
Height of offtake
windows, m.
Diameter of tower,
m.
-576-65
40
25,2
17,6
3,3
Steel
-725-66
39
32,5
21,8
3,55
//
-900-66
43,4
31,2
21,8
3,3
//
-1520-70
53,5
47,6
25,6
3,3
Ferroconcrete
-1200-70
48,3
40
26
3,3
Steel
-2100-70
64,8
52,2
33
3,8
//
-2600-70
71
58,2
37.9
4,3
//
-3200-73
81
64,4
40,4
//
-4000-71
91,5
72
41,6
5,5
//
No. of
project
Fig. 3. Results of steady-state calculation of magnitude field

with the k- turbulence model using wall functions for
uniform vertical flow of moist air. Geometry of the tower is
considered from the cooling pack position [3]
Height of the
cooling tower, m.
Table 2. Features of cooling towers by data [3]
Material of
tower frame
3.1. Electric energy production by turbine placed in

the cooling tower
A natural draft of air flow inside the cooling tower is a
matter of primary interest from the standpoint of cooling
tower usage. According to data of Ponomarenko and
Arefyev [6] the air flow speed in the narrowest part of
the cooling tower may be from 45 up to 1213 m/s.
Potential of this air flow may be used for generation of
electric energy by the wind turbine placed in the upper
part of the cooling tower. Possible engineering solutions
Fig. 4. Results of steady-state calculation of magnitude field

with the k- turbulence model using wall functions for
uniform vertical flow of moist air. All geometry of the tower
is considered
338
on positioning of one or several wind turbines of

various types in the upper part of the cooling tower are
shown at Fig. 5a-5d.
Its well known that some limits on air flow speed are
determined by the demand to minimize the drop liquid
drag-out from the cooling tower by upward air flow.
The increase of air flow speed is integral with increase
of such important cooling tower feature as depth of the
recycled water cooling because of the intensification of
heat and mass exchange processes [6].
Fig. 5d. Scheme of the cooling tower with double-deck

horizontal positioning of small wind turbines: 1 cooling
tower; 2 up-level turbines; 3 low-level turbines.
3.2. Possibilities to increase energy potential of air

flow in the cooling tower because of increase of the
natural draft intensity
Its known that a natural draft of the cooling tower is
determined by:
Fig. 5a. Scheme of the wind turbines positioning in the upper

part of the cooling tower: 1 the exhaust tower; 2 the water
catcher; 3 the water distributing system; 4 the irrigation
ditch; 5 the air-regulating plant; 6 the drainage; 7 the
load-carrying reference framework; 8 the wind turbine.
the towers height and form;

the level of the natural atmospheric air wind
force in the cooling towers surrounding space.
Nevertheless, at present a factor of possible increase of

speed of air in the cooling tower following from
additional heating is not taking into account. Let us
examine this aspect more seriously because a few
amount of heat air volume-up may become useful in
order to increase the wind energy potential inside the
cooling tower.
In reference [3] there are shown design data of air flow
temperature in the cooling tower by its height (look at
Fig. 6 and 7). It is shown that air flow temperature
depending on the cooling tower height from 10 to 150
meters is from 290 to 320. Its possible to provide
additional heating of air flow inside cooling tower and
increase of draft by artificial blackening of the outer
cover of south-eastern side of the towers ferroconcrete
block.
During the hottest days in periods from 1130 to 1600 in
summertime the blackened outer cover of the cooling
tower may be heated up to 90 0. Preliminary estimate
analysis shows that the temperature of inside cover of
cooling tower may be about 700 in this case. The air in
the cooling tower will be heated up under the air flow
speed W0 = 412 m/s. The heated up air will increase
the natural draft (wind speed) inside the cooling tower
what will contribute to the enhancement of the power of
small-scale wind turbines, which could be installed in
the tower.
Fig. 5 b. Scheme of three wind turbines positioning in

outgoing opening of the cooling tower
Fig. 5c. Scheme of cooling tower with small wind turbines of

two types: 1 cooling tower; 2 two wind turbines; 3 two
smaller wind turbines.
339
Q1 Q1T Q1C ;
(1)
Q1T K g F 10 3 K g F t H t B 10 3 , (2)
where Q1 , Q1T and Q1C are the heat flows explained

above, kW;
t H = 90 0 is the temperature of the cooling
tower outer wall;
t B 32 0 is the specified temperature of air
flow inside the cooling tower according to
Gonzales [3] calculations never accounting the
heating of the wall by solar radiation;
F 35796 m2 is the area of the cooling tower
outer wall;
K g 3,56 W/(m2K) is the heat transfer
Fig. 6. Calculation results for the test case under summer

atmospheric conditions at several values of heat load
(variation of temperature)
coefficient of the cooling tower wall

Kg
1

B
(3)
where 0,3 m is the cooling tower wall thickness;

H 23 W/(m2) is the outer heat transfer
coefficient [10];
2,04 W/(m) is the heat conductivity
coefficient (ferroconcrete) [10];
B 11 W/(m2) is the inner heat transfer
coefficient [10].
K g 3,56
Q1C K g FC tC 10 3 ,
(5)
transparent covering of the blackened cooling towers

wall, W/(m2);
FC is the area of the blackened cooling tower
wall irradiated by sun, m2;
tC is the excess difference of temperatures
characterizing the impact of solar radiation in
summertime, 0.
tB
150 m
The following values are taken for the calculation:
the
where K g is the coefficient of heat transfer through the
75 m
after
The heat gained by the cooling tower bulk due to the

solar radiation:
Lets determine the heat leakage through the walls of

cooling tower Q1 compensating difference between the
temperature of outer and inside covers of the cooling
tower walls - Q1T, and the heat leakage as a result of
solar radiations wall heat gain - Q1C. The calculated
scheme is shown at Fig. 8.
W0
calculated
F R r l 62,5 32,5 120 35796 m2. (4)
3.2.1. Calculation of heat leakage inside the cooling

tower
Q1
was
substitution of the indicated above values into the

equation (3).
The area of the cooling tower outer wall
Fig. 7. Calculation results for the test case under summer

atmospheric conditions at several values of heat load
(variation of water temperature)
tH
W/(m2K)
125 m
FC 0,5 F 0,5 35796 17896 m2;
tC 90 0 200 = 70 0;
Fig. 8. Scheme of the cooling tower for calculations
340
K g 3,56 W/(m2 ) (look above).
Table 3. Values of coefficient
Eventually, using the data given above, all heat flows

(1), (2) and (5) can be calculated:
Altitude, m
Q1 Q1T Q1C 11851,4 W.
3.3. Wind in bottom layer

Analysis of the meteorological parameters in bottom
layer shows that its vertical dimension depends on
intensity of turbulent baffling: in summertime under the
well developed turbulence in may reach 50-90 m., in
winter it decreases to 10-20 m.
Under equilibrium stratification of atmosphere, the wind
speed profile in bottom layer is described by the
following proportion from Karman and Prandtl theory
of turbulence:
ln z 2 Z 0
,
ln z1 z0
Proportion of the wind speeds for vertical profile is the

following [15]:
m
m 0,6 V 0,77 .
(8)
where
The cooling tower block is blown by wind with speed V.

Wind pressure , to the cover positioned in
perpendicular to velocity vector V is equal to:
Pv 0,5 V 2 .
200
2,6
The wind pressure at the altitudes more than 100 m is

higher in comparison with the heights of the majority of
small and medium scale wind turbines [9]. Therefore
this feature of atmospheric air flows at the altitudes in
range of 80-140 m can be applied for the generation of
power. For the said purpose the wind turbines of various
types can be positioned for power generation at the
adequate point of ferroconcrete construction of the
cooling tower. One of the possible solutions of single
wind turbine positioning on the top of cooling tower is
shown at Fig. 9. Positioning of several wind turbines on
the top of cooling tower is shown at Fig. 10. Towers of
the wind turbines installed in one line should have
different heights in order to mitigate the obstruction of
wind for the next turbine.
Positioning of wind turbines at the altitude 100
meters significantly increases its energetic potential.
Simultaneously there is a decrease of ecological impact
to the environment that is traditional in case of standard
height of wind turbines tower [9].
Increase of the wind power potential following the
growth of wind pressure under the exploitation of wind
turbine at high altitudes is illustrated by data of S. Kytra
[14]. Dependence of the wind power capacity on the
wind turbines hub height is given in Table 4.
where V1 V2 is the wind speed at altitudes z1 and z2,

z0 altitude, where the wind speed goes to zero.
(7)
100
2,1
3.4. Energetic usage of atmospheric flows at the

altitudes more than 40 meters
(6)
z
V1 V2 1 ,
z2
40
1,55
So, by increasing the altitude from 10 to 200 meters the

wind dynamic pressure increases 2,6 times.
For example: according to the conditions of the Jonava
area (Lithuania), the calculated wind speed at the 10 m
altitude is 20,6 m/s. Normative wind ram at the 10 m
altitude is 27,0 kg/m2. Medium wind speed at the said
altitude is 4,9 m/s in January and 4,3 m/s in June.
Q1C 4460,2 W; Q1T 7391,2 W;
V2 V1
10
1,0
(9)
150 m
Taking into account compensation factors used due to

the non-perpendicularity of velocity vector to the
slipped area, the wind pressure equation (9) will be the
following:
1
k k V 2
Pv 1 2
K,
2
2
(10)
where air density, kg/m3;

k1, k2 aerodynamic coefficients from the
windward and wind protected sides of the
cooling tower block respectively;
coefficient that accounts increase of the
dynamic pressure by altitude.
Wind
Values of coefficient K in equation (10) are shown in

Table 3.
Fig. 9. Positioning of a single wind turbine on the top of

cooling tower
341
2.
3.
1
4.
Wind
5.
Fig. 10. Positioning of four small scale wind turbines on the

top of cooling tower
6.
Table 4. Dependence of the wind power capacity on the wind

turbines hub height and class of the localitys roughness [14]
Altitude,
m
10
50
100
Parameters
Wind speed, m/s
Capacity, W/m2
Wind speed, m/s
Capacity, W/m2
Wind speed, m/s
Capacity, W/m2
7.
Values of parameters
depending on the localitys
surface roughness
0
1
2
3
7,6
5,3
4,7
3,7
456
181
119
58
8,9
7,3
6,6
5,8
702
400
309
204
9,5
8,5
7,8
6,9
890
595
462
319
8.
9.
4. Conclusions
1.
2.
3.
10.
Constructions and parameters of large

ferroconcrete cooling towers with natural draft
are presented.
Possibilities of usage of the cooling tower for
power generation by means of non-traditional
power generating units were examined on the
base of analysis of the aerodynamic situation in
the cooling towers with natural draft.
Several options of small scale wind turbines
positioning in the cooling tower with natural
draft for electric energy production are
proposed.
11.
12.
13.
14.
5. References
1.
15.
.., ..,
.. // .:
, 1976. 210 p.
342
..

//
. .. , , 236, 2000.
p.p. 242-247.
Israel Ziga-Gonzles. Modeling heat and
mass transfer in cooling towers with natural
convection. Summary of dissertation. Czech
Technical University in Prague, faculty of
mechanical engineering. Prague. 2005. 23 p.
Hoffmann J.E., Krger D.G. Analysis of heat
mass and momentum transfer in the rain zone
of a natural draft counter-flow cooling tower.//
7th COOLING TOWER AND SPRAYING
POND SYMPOSIUM. Leningrad, USSR, May
29-June 2, 1990. p.p. A10-(18).
Grange J.L. A computer code for particular
operating conditions of wet cooling tower// 7th
COOLING TOWER AND SPRAYING POND
SYMPOSIUM. Leningrad, USSR, May 29June 2, 1990. p.p. 8-(111).
.., ..
. . //
.. . .:
. 1998. 278 p.
.., ..
2005.
Kugeleviius J.A., Kuprys A., Kugeleviius J.
Aplinkosaugini ir energetini energijos
gamybos rodikli vertinimai // Journal of
Environmental Engineering and Landscape
Management, 2005, Vol. XIII, 4. p.p.
192a-199a.
Katinas V., Tumosa A. The possible usage of
wind energy in Lithuania. Energy Agency,
Directorate of the Energy Saving Programme,
Kaunas, 1996. 27 p.
..

. 3-
., . .:
, 1989. 223 p.
http://www.chv-praha.cz
(called
on
02.02.2009)
http://www.uniserv.com.pl
(called
on
02.16.2009)
.., ..

: // .:
, 2007. 52 p.
Kytra S. Atsinaujinantys energijos altiniai.
Kaunas, 2007. 140 p.
..,
..
: ,
, . .: ,
2005. 264 p.

APPLICATION OF RENEWABLE ENERGY SOURCES FOR FEEDING

OF WATER RECYCLING SYSTEMS TECHNOLOGICAL EQUIPMENT
Michail G. BERENGARTEN*, Semyon I. VAINSHTEIN*, Aleksandr S. PUSHNOV*, Vytautas ADOMAVIIUS**
** The Centre of Renewable Energy Technologies at Kaunas University of Technology, Lithuania
Abstract: The principal possibility to use non-traditional

energy sources as a drive for water recycling systems
technological equipment at large chemical enterprise is
shown by example of the cooling tower with natural
draft and cooling tower of type CK-400 having a single
fan.
Analyzing the data given in Table 2 we can to note that

a significant feature of years 2003 and 2005 was a
serious upgrade tendency in start-up of water recycling
systems in Russian Federation. Supply of the recycled
water by the industry branches according to the data [3]
for the major Russian enterprises is shown in Table 3.
Keywords: Water recycling systems, renewable energy

sources, PV power system, cooling towers with natural
draft, cooling towers with fan.
Table 3. Supply of the recycled water in major Russian

enterprises
1. Introduction
Industry
Water recycling systems of Russian Federation are

using up to 160109 3 of recycled water. The said
volume includes the usage of recycled water after
cooling or purification. Up to 136109 3 of recycled
water in Russia is used for cooling of various types of
technological equipment. Lets examine the alternation
dynamics of the whole volume of recycling and serial
water supply in Russia and Belarus that is shown in
Table 1.
Oil refining industry

Chemical and
petrochemical industry
2000
2003
2004
2005
133,5
135,6
135,0
135,5
6155
5842
6391
6369
Data on start-up of the recycling systems for the

recycled water in Russia and Belarus are shown in
Table 2.
Table 2. Start-up of the water recycling systems according to
the data [2] (thousand of cub. meters a day)
Year
Russia
Belarus
2000
135
0,05
2003
1130
163
2004
786
8,4
100
Temperature of
recycled water 0
Cooling
tower in
Cooling
tower
out
4045
25-28
40-45
25-30
Cooling of the recycled water at major enterprises of oil

refining, petrochemical and chemical industries is
realized at sectional and single fan cooling towers and
also at cooling towers with natural draft.
The most part of energy consumption in recycling
systems with fan cooling towers goes to the pumping
equipment. In the fan type cooling towers energy is used
also by fan drives.
The capacity of pumping and fan equipment of the
cooling towers is shown in Table 4 according to the data
given in reference [3].
The possibilities to use solar radiation by means of
positioning of the photovoltaic (PV) modules mounted
on the cooling towers wall in order to produce electric
energy are analyzed below.
The electricity produced by the photovoltaic power
system can be used for baking up the main power source
electric grid. Consumption of electricity from the grid
for feeding of fan and pump equipment of the tower
could be reduced in this case. The best option in this
case would be to install the grid-connected photovoltaic
power system.
Table 1. Total volume of recycling and serial water supply in

Russia and Belarus [2]
Year
Russia,
billions of
cub. meters
Belarus,
millions of
cub. meters
Supply of
recycled
water in
major
enterprises,
thousands
m3/hour
100
2005
1090
0,5
343
Table 4. Features of cooling tower fans

Fan mark
3VG25 (Russia)
2VG50 (Russia)
2VG70 (Russia)
1VG104 (Russia)
HEMA (Germany)
fan diameter:10,4
Balke (Germany)
fan diameter: 10,4
20,0
06-300 8 (Russia)
06-300 10 (Russia)
06-300 12,5 (Russia)
3. Cooling towers with natural draft
Electric drive
capacity,
kW
11
30
75
200
Rotation
frequency,
min-1
365
178
178
110
250
91
250
800
3
3
4
127
65
1410
950
730
The wall of cooling tower in some cases may become a

perspective construction to mount the PV modules. The
scheme of possible lining of PV modules on eastern and
south sides of the tower wall is shown at Fig. 4. In this
case up to 50 % of total tower wall area may used for
mounting of PV modules.
If half of the cooling tower (diameter 125 meters,
height 150 meters) would be covered by the PV
modules, the capacity of PV power system could make up
about 1 000 kW. As it is shown at Fig. 4, axes of
symmetry of all mounted PV modules are oriented
vertically in alignment to the cooling tower axe of

symmetry.
2. Fan cooling towers

Single fan cooling tower of type CK-400 have the
largest area of the outer tower wall among the fan
cooling towers. General view of such cooling tower and
its scheme are shown at Fig. 1 and Fig. 2.
The scheme of possible positioning of the PV modules
on the tower wall is shown at Fig. 3. In case of using
efficient PV modules, the PV power system could
supply over 100 kW power when sun is shining. This
amount of power would be sufficient for running of
some fans (Table 4) and would cover considerable part
of power necessary for the most powerful fans.
It is appreciable that application of PV power systems is
limited by weather conditions. But taking into account
the fact that the hottest period of the year is also the
most extreme one for exploitation of the cooling tower
such an alternative to traditional energy sources may be
rational in some situations. In any case, the use of PV
power systems can help to reduce the load to traditional
sources of power energy.
1
38 m
22 m
2
20 m
40 m
Fig.3. .Scheme of positioning of solar collectors on the wall of
CK-400 single fan cooling tower: 1 the cooling tower wall; 2
PV modules
Fig. 1. Group of CK-400 cooling towers
Fig. 4. Positioning of PV modules on the wall of reverse-flow

cooling tower: 1 the cooling tower wall; 2 PV modules
The parabolic concentrators that are positioned on the

outer cover of cooling tower in several decks by height
also may be used there for increasing of efficiency.
Analysis of operation of the parabolic-cylinder
concentrators of solar radiation was made.
Fig. 2. Scheme of CK-400 cooling tower
344
Symmetrical parabolic concentrator was made in

Russian institute of agriculture electrification with
application of technologies based on the principles of
non-image forming optics. The unit belongs to the class
of solar radiation reflecting concentrators.
The concentrator may be defined as parabolic
architecture formed by two parabolic curves branches
reflector. Usage of solar collectors positioned on the
wall of tower provides the electric energy supply for
operation needs of technological systems of industrial
enterprises. In focal sphere of the reflector there is

positioned a solar module with double-sided working
cover that consist commutated double-sided solar
elements. The face side of module can reflect either
non-concentrated or concentrated solar radiation in
dependence on the angle between mid-plane of
concentrator and Sun height. The wire side reflects
concentrated flow of solar irradiation resulted by multireflection from the concentrators reflecting square.
2
55
1230
1230
33
33
Fig. 5 Scheme of solar module operating activity with symmetric parabola-cylinder concentrator in stationary position under
various Sun height: a) midday, June 1st , b) midday March 16 or 8.30 and 16.30, Moscow latitude; 1 symmetric concentrating
reflector; 2 double-sided photoelectric module; 3 solar rays path.
During the experiments conducted in cooperation with

Russian institute of agriculture electrification the work
of solar unit under various sun heights was modeled
by changing of the plain angle in relation to horizon
plain. The modeling sun height was:
where h0 = - , (look Fig. 6);

is angle of concentrators middle
positioning in relation to solar rays path,
grade;
is angle of concentrators middle
positioning in relation to horizon plain,
grade;
h is modeling sun height, angle grade;
h0 is actual sun height, angle grade.
(1)
145
plain
angle
h0
37
plain
angle
1230
h = h0 + (90 - ),
Accounting the said above:

h = - + (90 - ) = 90
(2)
Position of solar unit symmetric parabolic-cylinder

concentrators middle plain in relation to horizon plain
is shown at Fig. 7.
Fig. 6. Basic vertical position of the concentrators middle

plain in relation to horizon plain
345
2,5
I,
1,5
hm
145
0,5
h0
37
0
0
10
1230
U,
Fig. 8. Current-voltage characteristic of solar module with

concentrator under = 78 angle grade
4. Conclusions
1.
Fig. 7. Change of the concentrators middle plain position in
relation to horizon plain under experimental modeling of
various sun heights
2.
Electrical parameters of the solar unit were determined

by design way under data got in Russian Institute of
agriculture electrification in condition of natural solar
radiation.
Overall geometry of the unit: 1.25x0.73x1.23 m3,
overall geometry of the solar module double glass
1.2x0.145x0.037 m3.
Design value of aperture angle of asymmetric parabolic
concentrator unit A= 24 angle grades.
Ability of the said concentrators to use the diffused
radiation is in interest. So there were taken off the data
in absence of direct solar rays. In this experiment there
was applied solar module with solar elements with
double-sided working surface. The research was
guided separately with every side of solar module with
opposite side closed.
The current-voltage characteristic of solar module
with transverse width 0,145 m under = 78 angle
grade is represented at Fig. 8.
Application of PV modules positioned on the

face side of cooling tower provides electricity
supply for technological equipment of
industrial water recycling.
Parabolic concentrators that are positioned on
the outer cover of cooling tower in several
decks by height also may be used there for
efficiency increasing.
5. References
1.
2.
3.
346
..

// .2004,
10, C. 31-33.
Pushnov A., Berengarten M., Ryabushenko A.
Influence of Aerodynamic Conditions I
Cooling Tower or Losses of Water in System
of Defensive Water in System of Defensive
Water Supply. May 22-23, 2008. The 7-th
International
Conference
Environmental
Engineering, VGTU Press Technika. Vilnius, 2008. - P.290-296.
.., ..
: /
.. . - .:
, 1998. - 376.

WIND POWER PLANT WITH REDOX FLOW BATTERY INVERTER SYSTEM

CONVERTER ENERGY CONVERSION PROCESSES RESEARCH
Povilas BALINAS, Povilas NORKEVIIUS
Kaunas University of Technology, Faculty of Electrical and Control Engineering,
Department of Electric Power Systems, The Centre of Renewable Energy Technologies, Lithuania
shaft with sail wing type turbine WT rotates permanent

magnet excitation system synchronous three phase
generator SG. Generated stochastic character voltage
system (U=const., I=var.) electrical energy is rectified
by AC/DC inverter and transmitted to semiconductor
ISC.
Abstract: Inverter-system converter (ISC) steady state

and transitional character energy conversion processes
of wind power plant with redox flow battery (RFB) are
researched. Low intensity wind energy potential
resources are fairly high in Lithuania. Such intensity
wind energy potential can be better used installing ISC
at wind power plants. RFB unique features and
characteristics provide additional opportunity to store
wind power plant generated stochastic character
electrical energy effectively. Research data is used to
create ISC prototype for wind power plant with vertical
shaft.
Grid
Wind
CS
Load
ISC
AC/DC
WT
Keywords: wind power plant, redox flow battery,

inverter system converter, energy, steady state,
transient, conversion processes.
SG
AC/DC
1. Introduction
RFB
Nowadays low intensity wind energy potential usage at

wind power plants is problematic [4]. Successfully
solved such problem and additionally effectively stored
stochastic character electrical energy at RFB [1, 2]
would create advantageous possibility to use existing
long term low intensity wind energy potential at
Lithuania in significantly effective method. Possibility
of innovative conception usage is researched for such
problem solution. Here voltage system (U=const,
I=var.) stochastic character electrical energy is
converted to dual current system (I=const., U=var.)
energy [3]. This concept practically has to be
implemented by installing additional electrical energy
conversion module semiconductor inverter system
converter at wind power plant with synchronous
generator structure. Such ISC steady state and transient
operation modes research results are presented in this
paper.
Fig. 1. Wind power plant with ISC and RFB electrical energy
conversion structural scheme
Voltage system DC electrical energy of rectifier AC/DC

is converted to AC dual system (I=const., U=var.)
electrical energy at ISC. AC dual system (I=const.,
U=var.) electrical energy frequency f and phase is
controlled by ISC control system CS.
Converted current system AC electrical energy can be
used autonomously and for charging RFB through
additional AC/DC rectifier simultaneously. Then
stabilized AC voltage system energy is transmitted to
autonomous load.
RFB can be connected to ISC input and transmit
electrical energy to load continuously when wind
turbine does not operate.
Modified wind power plant can be connected to grid
and transmit electrical energy to load and grid. In this
case Redox flow battery is disconnected. Wind energy
2. Wind power plant with RFB and ISC structure

and features
Modified wind power plant structure with new features
and functional possibilities is showed in Fig. 1. Vertical
347
potential is used at the maximum when wind power

plant operates autonomously or connected to the grid.
and AC/DC inverter is simulated as subsystem block,

load R as active resistance block, grid as AC source
with internal active resistance block, currents and
voltages are measured with Io, Uo, I, U blocks.
Implemented super capacitor is used for excessive
electrical energy storage from ISC semiconductors
elements during transient processes.
ISC steady state energy conversion processes simulation
with connected RFB is performed changing wind power
plant generated voltage Uo from 5V to 120 V. RFB
rated voltage is 230 V.
with connected R is performed changing
R from 10 to 500 and Uo levels at 5 V, 24 V, 60 V.
with connected grid is performed changing Uo from 5 V
to 120 V.
After steady state processes simulation of ISC with
connected RFB performance Ired, Io= f(Uo), = f(Uo)
dependences are compiled which showed at Fig. 4. and
Fig. 5.
3. Steady state energy conversion processes

simulation results
Simplified structural scheme for ISC energy conversion
processes research is showed at Fig. 2.
IR
Uo,Io
Wind
power plant
R (Load)
ISC
AC/DC
Ired
RFB
Grid
Ug
Fig. 2. Structural scheme of ISC energy conversion processes

research
Energy conversion processes simulation is performed

when ISC output is connected to active load R or grid or
through AC/DC converter to RFB. Here Io, Uo wind
power plant generated current and voltage, IR active
load R current, Ired RFB current, Ug grid voltage is
showed at Fig. 2.
Energy conversion processes simulation is performed
with MATLAB, Simulink software [5]. Simulation
model is showed at Fig. 3.
Discrete,
Ts = 5e-005 s.
powergui
Io, Ire d (A)

0,30
0,25
Ired=f(Uo)
0,20
Uo
0,15
Io
0,10
Io=f(Uo)
0,05
0,00
c
1
Breaker1
Out1
20
40
60
80
100
Conn 2
Clock1
120
Uo (V)
Out2
W ind power
plant
Super
capacitor
Switch
Fig. 4. Wind power plant generated current Io, RFB charging

current Ired dependence on wind power plant generated voltage
Uo, Ired, Io= f(Uo)
Conn 1
Conn 3
Step2
WPP internal
resistance
Conn 4
ISC
1,00
0,80
Clock2
Switch1
Breaker3
Breaker4
0,60
c
1
c
1
Step1
U
c
Step3
I
0,40
Step4
Breaker2
0,20
Grid1
R
(Load)
Out1
Conn 2
Out2
RFB
Conn 1
Conn 3
Conn 4
AC/DC
0,00
0
Grid
20
40
60
80
100
RFB
internal
resistance
120
Uo (V)
Fig. 5. ISC efficiency factor dependence on wind power

plant generated Uo voltage, = f(Uo)
Fig. 3. ISC energy conversion processes simulation model
Fig. 4. shows that innovative ISC unique energy

conversion characteristic allow to use low intensity
wind energy potential. RFB charging process is possible
Wind power plant and RFB is simulated as DC voltage

system source with internal active resistance block, ISC
348
when wind power plant generated Uo is equal to any

value.
Current IRED increases linearly when wind power plant
generated Uo increases. Here RFB charging duration is
shorter when Uo increases. Wind power plant generated
current Io value does not depend on variation of voltage
Uo from 5V to 120V. This dependence character shows
that ISC converts wind power plant generated voltage
system energy (U=const., I=var.) to dual current system
(U=var., I=const.) energy.
Fig. 4 shows ISC efficiency factor dependence on
wind power plant generated voltage Uo character. ISC
high efficiency factor value is near 0,9 and decreasing
insignificantly when voltage Uo increasing. Simulation
results shows that ISC has low effect on wind power
plant energy conversion structure efficiency.
After steady state processes simulation of ISC with
connected R performance IR=f(R), =f(R) dependences
are compiled which showed at Fig. 6. and Fig. 7.
Uo=60V
0,8
0,4
0,2
0,0
0
200
300
400
500
Fig.7. ISC efficiency factor dependence on load R, =f(R)

at wind power plant generated different voltage Uo levels.
Grid voltage Ug dependence on wind power plant

voltage Uo, Ug=f(UR) is showed in Fig. 8. Stochastic
character wind power can be steadily transmitted to grid
after ISC and grid frequency f and phase
synchronization when grid voltage Ug is equal to
standard grid voltage RMS values (Ug=230 V; 400 V).
1,4
1,2
Uo=24V
Ug (V)
500
0,6
Uo=5V
0,4
100
R ()
1,6
1,0
0,8
Uo=5V
0,6
Uo=60V
IR (A)
1,8
Uo=24V
1,0
400
0,2
300
0,0
0
100
200
300
400
500
R ()
200
Fig. 6. Load current IR dependence on load R, IR=f(R) at wind

power plant generated different voltage Uo levels
100
0
Fig. 6 shows that ISC output has dual current system

(U=var., I=const.) character. ISC converted current IR
has high stability factor. IR character does not depend on
R, Uo values when Uo varies 12 times and R 50 times. IR
stability factor when load varies from 10 to 500
when Uo=5 V, =2 %; Uo=24 V, =5,8 %, Uo=60 V,
=5,8 %.
Fig. 7 shows that ISC operations is most effective when
load R values is higher than 50 and wind power plant
generated voltage Uo value is higher than 24 V. ISC
efficiency factor is high when load is varying at wide
range.
Steady state energy conversion processes results shows
that RFB connected to ISC input could determine
autonomous load R RMS voltage value by selecting
RFB rated voltage. R voltage can be steady when R
value is changing.
20
40
60
80
100
120
Uo (V)
Fig. 8. Grid voltage Ug dependence on wind power plant

voltage Uo, Ug=f(Uo)
4. Transient energy conversion processes

simulation results
Transient process energy conversion simulation results
are showed at Fig. 9, Fig. 10, Fig. 11, Fig. 12, when
RFB permanently connected to ISC and wind power
plant is connected at 1s of simulation time. RFB rated
voltage is 230 V.
ISC reactive element voltage UReak dependence on time
t, UReak=f(t), when wind power plant is connected to ISC
at rated DC voltage UoR=24 V is showed at Fig. 9. and
when wind power plant is connected to ISC at voltage
Uo=12V (Uo=0,5UoR) is showed at Fig. 10.
349
Fig. 9. ISC reactive element voltage UReak dependence on time

t, UReak=f(t), when connected wind power plant rated voltage
UoR=24V
Fig. 12. ISC semiconductor element current IT1 dependence on

time t, IT1=f(t), when connected wind power plant rated
voltage Uo=12V
Peak value of transient process IT1m is 1,2 times less

than steady state value UReakm at 0,5UoR and UoR.
IT1 transient process duration at 0,5UoR is 1,8 times
longer than at UoR.
At Fig. 13., Fig. 14. Fig. 15., Fig. 16 is showed
simulation results when specified value load R=338 is
permanently connected to ISC output and wind power
plant is connected at 1s of simulation time. Here load R
steady state effective voltage is standard value and equal
to 230 V , wind power plant generated rated voltage
UoR=24 V.
ISC reactive element voltage UReak dependence on time
t, UReak=f(t), when wind power plant is connected to ISC
at rated voltage UoR=24V is showed at Fig. 13. and
when wind power plant is connected to ISC at voltage
Uo=12V (Uo=0,5UoR) is showed at Fig. 14.
Fig. 10. ISC reactive element voltage UReak dependence on

time t, UReak=f(t), when connected wind power plant voltage
Uo=12V
Peak value Ureakm=231,7 V of steady state process when

wind power plant rated voltage is UoR and Ureakm=230,6
V at 0,5UoR.
Peak value of transient process Uom is 0,8 times less
than steady state value UReak at 0,5UoR and UoR. Ureak
transient process duration at 0,5UoR is 1,6 times longer
than at UoR.
ISC semiconductor element current IT1 dependence on
time t, IT1=f(t), when wind power plant is connected to
ISC at rated voltage UoR=24V is showed at Fig. 11 and
when wind power plant is connected at 0,5UoR is
showed at Fig. 12.

time t, UReak=f(t), when connected wind power plant rated
voltage UoR=24V

voltage UoR=24V
Peak value IT1m =0,82 A of steady state process when

wind power plant rated voltage is UoR and IT1m =0,79 A
at 0,5UoR.

time t, UReak=f(t), when connected wind power plant rated
voltage Uo=12V
Fig. 13., Fig. 14., shows that Ureak transient process

duration does not depend on wind power plant
generated voltage Uo.
ISC semiconductor element current IT1 dependence on
time t, IT1=f(t), when wind power plant is connected to
350
ISC at rated voltage UoR=24V is showed at Fig. 15 and

when wind power plant is connected at 0,5UoR is
showed at Fig. 16.
autonomous load R RMS voltage value by selecting

RFB rated voltage. R AC voltage can be steady
when R value is changing.
6. ISC operation is steady when it operates
autonomously or connected with grid. Electrical
energy is transmitted steadily and uninterruptedly to
grid or load when wind power plant voltage Uo
varies at wide range.
7. ISC reactive element voltage peak value of transient
process Uom is 0,8 times less than steady state value
UReak at 0,5UoR and UoR. Ureak transient process
duration at 0,5UoR is 1,6 times longer than at UoR
when RFB is permanently connected and wind
power plant connection performed to ISC.
8. ISC semiconductor current peak value of transient
process IT1m is 1,2 times less than steady state value
UReakm at 0,5UoR and UoR. IT1 transient process
duration at 0,5UoR is 1,8 times longer than at UoR
when RFB is permanently connected and wind
power plant connection performed to ISC.
9. ISC reactive element voltage and semiconductor
current peak values of transient process increases
exponentially to steady state values when ISC is
connected to load R and wind power plant
generated voltage varies from UoR to 0,5 UoR.
10. Semiconductor element current IT1 and reactive
element voltage UReak transient process duration
does not depend on wind power plant generated
voltage variation from UoR to 0,5 UoR when R is
permanently connected and wind power plant
connection performed to ISC.

voltage UoR=24V

voltage Uo=12V
Fig. 15., Fig. 16., shows that IT1 transient process

duration does not depend on wind power plant
generated voltage Uo.
Fig. 13., Fig. 14. Fig. 15., Fig. 16 shows that transient
process Ureakm, IT1m values increases exponentially to
steady state values independently on wind power plant
generated voltage Uo value.
6. Research results estimation and perspective

activity
1. ISC energy conversion steady state, transient
process analysis is presented at this paper.
2. ISC possible operation modes control algorithm has
to be created for prototype model transient
processes optimization.
3. ISC structure has to be optimized for transient
process negative effect reduction on researched ISC
structure elements.
4. ISC prototype model has to be created for
theoretical research results adequacy verification.
5. Conclusions
1. Research results confirmed voltage system
(U=const., I=var.) electrical energy conversion
possibility to current system (U=var., I=const.)
electrical energy. Converter operates at inverter
system converter mode.
2. Wind energy generated voltage value variation has
low effect on ISC efficiency factor at wind power
plant with RFB.
3. Stochastic character wind power can be steadily
transmitted to grid after ISC and grid frequency f
and phase synchronization when grid voltage Ug
is equal to standard grid voltage RMS values
(Ug=230 V; 400 V). Electrical energy steadily is
transmitted to grid when Uo variation range is -5
100%.
4. ISC can operate autonomously. In this case RFB
have to be connected to ISC for energy storage.
RFB set autonomous load AC voltage effective
value.
5. Energy conversion processes results shows that
RFB connected to ISC input could determine
7. References
1.
2.
3.
4.
5.
351
P. Balinas. P. Norkeviius. Wind micro power

plant with redox flow battery conversion processes
research. Energetika. 2009. Nr.1
Cigre technical brochure on modelling new forms
of generation and storage. 2000. 139 p.
P. Balinas. Aukto danio tampos srovs
galios keitikli teorijos sintez (monografija).
Kaunas. Technologija. 1994. 272p.
C. Abbey, J. Chahwan, M.Gattrell, G.Joos.
Transient modelling and simulation of wind
turbine generator and storage systems. 21,rue
dArtois, F-75008 PARIS.
SIMULINK. Model-Based and System-Based
Design. 2001. 846p.

HYBRID WATER HEATING SYSTEM BASED ON SOLAR COLLECTORS AND

HEAT EXCHANGER FED FROM THE CENTRALISED HEAT SUPPLY GRID
Vytautas ADOMAVIIUS*, Gytis PETRAUSKAS**
* The Centre for Renewable Energy Technologies at Kaunas University of Technology, Lithuania
** Department of Control Technology at Kaunas University of Technology, Lithuania
Abstract: Water heating is widely spread technology
used for domestic and industrial purposes. Presently
water heating is still based mostly on the fossil fuels
although mature and cost-effective technologies running
on the renewable energy sources already exist for this
purpose. Water heating systems can run on renewables
purely or partially. This paper describes concept of
hybrid water heating systems based on solar collectors
and heat exchanger fed from centralized heat supply
grid, structure and sizing of the proposed system and
principles for control of solar collectors.
options for water heating due to solar energy universal

availability and mature of this technology. However, as
it is known, this source of renewable energy has one
considerable disadvantage: energy of solar radiation is
not constant over the time. Combined (hybrid) water
heating systems can be used in order to compensate this
shortage. They can be purely renewable or mixed when
another water heater is based on the fossil fuel. There
are many various sources of energy to choose for
backing-up the solar collectors: wind [1], biomass,
fossil fuels, centralized heat, electricity and other.
Keywords: renewable energy, solar energy, solar

collectors, water heating, hybrid systems.
2. Concept of hybrid water heating systems based on

solar collectors and backed-up on centralised heat
1. Introduction
Often it is reasonable and convenient to use the water

heating systems, which already exist in condominiums,
enterprises or other users of hot water, and to combine it
with solar collectors. The solar collectors could allow
cutting bills for heat energy significantly.
As it is shown in Fig. 1, solar irradiation is not
distributed evenly over the year in Lithuania. Therefore
only in summer months solar collectors could cover all
demands of hot water. In late autumn, winter and early
spring the main producer of hot water will be the
existing water heating appliances, which will cover
shortage of solar energy in this period of year. Input of
solar collectors then will not be significant (Fig. 1).
The last dramatic events in the crude oil market in 2008

when oil price increased up to 147 USD per barrel and
in the natural gas market on the beginning of 2009 when
Russia cut its gas supply for more than three weeks to
many European countries have evidently proved the
necessity to shift on more reliable and more predictable
sources of energy in scale as large as possible and
reasonable. First of all it could be the non-polluting
renewable and other available local energy sources.
Water heating is widely spread technology used in the
domestic sector and for the industrial applications
including many small and medium enterprises.
Considerable amounts of energy are used for this
purpose. Presently water heating installations mostly are
running on the fossil fuels or sometimes on electricity,
which mainly is based on the fossil fuels as well.
Substitution of the fossil fuels currently used for water
heating could give the well-known benefits: increased
energetic independence due to the reduction of imported
fossil fuels, creation of more jobs and stimulation of
local economic development, positive impact on
mitigation of climate change and some other.
A huge and rather easily realizable potential for water
heating by means of the renewable energy usage instead
of fossil fuels exists in Lithuania and other countries all
over the world. Solar collectors are one of the best
180
160
140
120
100
80
60
40
20
0
EH,
kWh/m 2
10 11 12
Months
Fig. 1. Distribution of solar irradiation in horizontal plane over

the year in Kaunas, Lithuania
352
expensive in comparison with energy from centralized

heat supply grid. Therefore in many cases the best
solution could be the hybrid water heating system based
on solar collectors and the existing heat exchanger fed
from the centralized heat supply grid. Structural scheme
(Fig. 2) and description of this hybrid water heating
system based on solar collectors and centralized heat
source are presented bellow.
Presently water heating systems running on centralized

heat or natural gas prevail in cities and towns of
Lithuania and other countries of Central and East
Europe. In Lithuania water heating systems fed from the
centralized heat supply grid are used mostly.
Currently one type of hybrid water heating systems with
solar collectors already exists. Solar collectors there are
backed-up by means of electric water heaters [2].
However, electric energy is considerably more
P2
E, W/m 2
T2
T3
DHW HEAT
EXCHANGER
CHSS
outlet
DHW users
P2
Preheated water
T1
SC
CU
CV
P1
P1
PREHEATED Cold water

WATER
S TORAGE
TANK
CHSS
inlet
Fig. 2. Structure of hybrid water heating system based on solar collectors and centralized heat supply grid
which can be activated and customized online without

the need for additional tools.
In sunny days major part of heat energy, which is
necessary for water heating up to the standard
temperature (for example, 50-55 C), will produce solar
collectors.
If temperatures of water measured with thermometer T1
will be bellow the standard level, the DHW heat
exchanger will boost it up to the necessary value
according to the set up programme. In cloudy and dark
days, especially in winter months, solar collectors SC
will not develop sufficient heat capacity and therefore
the main energy source for water heating will be the
centralized heat supply grid.
Implementation of this hybrid water heating system
would eliminate blackouts of hot water supply in every
summer when heat energy from the centralized grid is
cut due to the repairs in the grid. Users hated snake
tax for the serpentine pipe, which heats bathrooms,
would be eliminated in summer months as well and
would be reduced in other months.
Efficiency and cost-effectiveness of the described water
heating system are the most important targets to be
achieved during the process of its implementation.
3. Structure of the proposed hybrid water heating

system
As it is shown in Fig. 2, the proposed hybrid water
heating system is based on the solar collectors and the
heat exchanger fed from the centralized heat supply
grid. The system consist of the solar collectors SC, the
preheated water storage tank, the collector pump P1, the
sensors of temperature T1, T2 and T3, the ethylene
glycol pressure switcher P1 and the DHW (domestic
hot water) pressure switcher P2, the DHW pump P2,
the DHW exchanger, the controlled valve CV and the
control unit CU (integrated in to the buildings
management system). As a contemporary CU could be
recommended the DDC4000 group controllers. It is a
CU, which is used in building service systems for
measuring, regulating, controlling, optimizing and
monitoring. It is based on a 32-bit processor and
functions according to the multitask principle. Its
operating system is Linux. DDC4000 CU communicates
via Ethernet (TCP/IP, BACnet). The DDC4000 is
equipped with object oriented software and hardware
modules, making available a large number of functions,
353
Successful achievement of these targets mainly depends

on the following key factors:
o
o
o
where
during the one summer month, kWh,

k d = 0.278 coefficient for matching the
Type of chosen solar collectors,

Correct sizing of the system,
Correct (optimal) control of the system.
dimensions,
c = 4190 the specific heat of water, J/kg .K,
Vacuumed tubular or flat solar collectors can be chosen

as the main heat suppliers for summer months. Industry
currently produces both of them. However, superiority
of the vacuumed tubular solar collectors is proven and
well-known. Comparison of efficiency of vacuumed
tubular solar collectors against flat collectors is
presented in Fig. 3. As it can be seen on this figure, the
vacuumed tubular solar collectors are approximately
from 1.5 (in summer months) up to 3 (in winter months)
times more efficient than the flat ones. Therefore it is
recommended to use vacuumed tubular solar collectors
in order to get better efficiency of solar energy
conversion into heat.
140
120
E1S is the energy required for water heating
V1S the water demands in one summer month,

m3,
T1 the temperature of hot water, K (C),
T2 the temperature of cold water, K (C),

We consider that this required energy for water heating
in summer months must be covered mainly by solar
collectors. Enlargement of solar collectors area in order
to have enough energy to meet all DHW demands in
spring and autumn can not be justified because it would
lead to the underused capacity in summer and to
significantly increased expenditure. Only a small
enlargement of solar collectors area can be acceptable.
Heat energy produced by the solar collectors array over
the one chosen summer month approximately can be
calculated in this way:
E, kWh/m 2
100
80
E S C = E1h S SC ,
60
40
where
20
2
9 10 11 12
Months
Fig. 3. Comparison of efficiency of solar collectors: vacuumed

tubular (black columns) and flat (white columns)
irradiation over the one chosen summer month

to the horizontal plane, kWh,
S SC the required area of the solar collectors
The main principles of systems sizing and optimal

control are presented below under the subsequent
headings of this paper.
array, m2.
4. Sizing of the system
The required area of solar collectors array

approximately can be expressed in shape of the
following inequality by using the equations (1) and (2):
Determination of the optimal area of solar collectors

array and volume of the preheated water tank are the
main tasks of the proposed hybrid water heating system
sizing. Their sizes must correspond with capacity of the
existing heat exchanger and DHW demands.
Capacity or area of solar collectors array depends on
the required volumes of hot water. If the system of hot
water supply is being renovated and number of users
remains the same, the volumes of necessary hot water
can be determined by checking records of hot water
consumption over the past years. In case of installing a
new system in a new building, the necessary volumes of
hot water should be calculated on basis of expected
DHW users and their demands.
The energy required for water heating during the one
summer month can be calculated by means of this
equation:
E1S = k d c V1S (T1 T2 ) ,
E S C is the energy produced by the solar

collectors array during the one summer
month, kWh,
the average efficiency of water heating
system based on solar collectors,
E1h the average perennial global solar
0
1
(2)
S SC
k d c V (T1 T2 )
E1h
(3)
In general, volume of the preheated water storage tank

depends on the average DHW daily demands and on the
possible number of dark (cloudy) days in turn.
Possibility to use another heat source in the hybrid water
heating system allows reducing of the tanks volume,
however, the desire to maximize solar energys share
leads to the increase of this volume. Optimal decision
can be found taking into consideration all necessary
information including cost-effectiveness. Roughly
volume of the preheated water storage tank can be
chosen equal to the average DHW daily demands.
(1)
354
6. Conclusions
5. Principle of solar collectors control

1.
Usually the control unit CU gives the signal to switch

on the pump P1 (Fig. 2) when the difference between
the temperatures T2 and T1 reaches the value T set in
advance. We suggest the variable T depending on the
solar irradiance E and the temperature of water in the
preheated water storage tank T1 in order to reduce
consumption of power for feeding of the pump P1 and
increasing of solar energys share in the proposed
hybrid water heating system. The calculation of the
variable T carries out the control unit CU.
Calculation of the referable difference of temperatures
T can be based on the following equation:
2.
3.
4.
T
Ton = Tmax .1 1
T1N
where
E

,
E ST
(4)
5.
Ton the difference of temperatures for

switching on the pump P1, C,
Tmax the maximal value of the difference of
6.
the temperatures, C,
T1 the instantaneous temperature in the tank,
C,
T1N the rated temperature in the tank, C,
7.
E the instantaneous solar irradiance, W/m ,

E ST the standard solar irradiance, W/m2.
2
Value of the standard solar irradiance is generally

received and is equal to 1 000 W/m2 (one sun). The
rated temperature in the tank for the DHW can be
accepted 60 C. The maximal value of the difference of
the temperatures T2 and T1 can be accepted 30 C. A
new more convenient shape of the equation (4) can be
deduced after the substitution of the mentioned above
constant values in the initial expression:
Ton =
E (60 T1 )
.
2000
Combination of the solar collectors for water

heating and various other water heating systems
already existing in households,
industrial
enterprises or other hot water users and running on
fossil fuels would be beneficial in many aspects.
Concept of hybrid water heating systems based on
solar collectors and heat exchanger fed from
centralized heat supply grid, principles of its sizing
and systems control are presented.
A huge potential for installation of the described
hybrid water heating systems exists in Lithuania
and neighboring countries.
Simple methodology for determination of the
recommended area of solar collectors array and
volume of preheated water tank is presented.
The proposed water heating system allows
approximately twofold cutting of bills for the
domestic hot water due to the partial substitution of
the heat from centralized supply grid by energy of
solar irradiation.
Installation of the proposed hybrid water heating
system could allow avoiding of hot water supply
blackouts in summertime when heat energy from
the centralized grid is not supplied due to the
repairs in the grid.
Principle of solar collectors control based on the
variable reference value of the difference of
temperatures in the solar collector and in the
preheated water storage tank is proposed.
7. References
1.
2.
(5)
3.
The instantaneous solar irradiance can be measured by

means of the phyranometer.
Values of the difference of temperatures for switching
on the pump P1 can vary depending on values of the
solar irradiance E and temperature in the preheated
water storage tank T1.
Means of control and optimisation of the proposed
hybrid water heating system based on solar collectors
and heat exchanger in order to maximize solar heat
energys share are described in paper [4].
4.
355
Adomaviius V., Watkowski T. Hybrid water

heating systems based on solar and wind energy //
Proceedings of International Conference Electrical
and Control Technologies 2008. KTU, Kaunas,
2008. P. 349-354.
Uken E-A. and Monyane D. Energy savings with a
domestic solar-electric hybrid hot-water cylinder
system. Proceedings of the 9th International
Domestic Use of Energy Conference, Cape Town,
South Africa, April 2001. P.159-162,.
Duffy J. A. and Beckman W. A. Solar Engineering
of Thermal Processes. John Wiley and Sons, Inc.,
2008. 908 p.
Adomaviius V., Petrauskas G. Optimisation of
hybrid water heating system based on solar
collectors and heat exchanger in order to maximize
solar heat energys share // Proceedings of
Technologies 2009. KTU, Kaunas, 2009 (in
press).

OPTIMISATION OF HYBRID WATER HEATING SYSTEM BASED ON SOLAR

COLLECTORS AND HEAT EXCHANGER IN ORDER TO MAXIMIZE SOLAR
HEAT ENERGYS SHARE
Vytautas ADOMAVIIUS*, Gytis PETRAUSKAS**,
** Department of Control Technology at Kaunas University of Technology, Lithuania
countries are good examples for small countries

currently lagging behind. Hopefully, Lithuania presently
is already close to making important decisions in this
sphere by creation the legal basis favourable for
development of renewable energy technologies. Apart
from other well known benefits, the worldwide
expansion of renewables could lead to decrease of
demands of fossil fuels and to keeping their prices on a
moderate level. It could allow hoping to diminish
probability of the new global financial-economic crisis
similar to what we experience now.
Think globally, act locally - this is a popular
catchword of the green movement currently ongoing
worldwide. There are many possibilities to act locally
in Lithuania and in every other country. Every costeffective way to reduce our dependence on dirty fossil
fuel should be seriously considered.
We suggest installing of the hybrid water heating
systems based on solar collectors and heat exchanger
running on the centralized heat supply grid in Lithuania
and other countries having similar DHW systems. In
Lithuania the heat exchangers usually exist in
condominiums and other big buildings. Concept of this
hybrid water heating system is described in paper [3].
Subject of the present paper is optimisation of operation
of the proposed water heating system in order to
maximize free solar energy share.
Abstract: In principle, one of the main tasks for control

of solar collectors in the hybrid water heating system,
which is based on their combination with heat
exchanger fed from the centralised heat supply grid, is
to arrange operation of solar collectors in the way
maximizing the capture of solar irradiation. Solution of
this problem by means of programme packet for the
designing of control systems Planning System 4000 is
described in this paper.
Keywords: solar energy, solar collectors, water heating,
hybrid system, optimization of operation.
1. Introduction
The world is making huge and rapid leaps towards a
clean energy economy. Germany, Denmark, Spain and
other countries were leaders in this sphere over the
many years. Last time a large number of countries in
Europe, America and Asia became seriously engaged in
this process as well.
A Solar Grand Plan for the USA was proposed by Ken
Zweibel, James Mason and Vasilis Fthernakis [1].
According to this plan, a massive switch from fossil
fuels to solar power plants is anticipated in the nearest
future. Solar power plants could supply 69 % of
electricity and 35 % of total energy consumed in the US
by 2050. 420 billions USD should be allocated to fund
the realization of this plan.
Another bold plan is initiated by Nobel Prize winner
former Vice-president Al Gore to repower America
with 100 % clean electricity within 10 years [2]. Under
this plan it is intended to invest hundreds of billions
USD for implementation of this idea. These investments
will create high-paying new jobs for 5 millions
Americans. It also will demonstrate the readiness of the
decision makers of this nation to pave the way for clean
energy across the country and provide long overdue
solutions to the climate crisis.
The mentioned above and other steps in right direction
to sustainable development made by the developed
2. Object of research
The suggested experimental hybrid water heating
system is intended to install in condominium (Fig. 1),
which is located in Kaunas, Lithuania. The main
technical characteristics of this condominium are given
in Table 1.
Presently domestic hot water (DHW) is being prepared
in the condominium only by means of heat exchanger
running on the centralized heat supply grid. The main
disadvantages of the existing DHW system are
considerable price for water heating and blackouts of
hot water supply in summertime when the system can
356
dwellers of condominium in summer months. The heat

exchanger could be switched out during this period of
year. The next 8 months both parts of the hybrid water
heating system have to operate solar collectors and
heat exchanger.
not operate due to the cutting off heat energy supply

from grid because of the planned repair works. Usually
it happens one time per year in May or June and
duration of the blackout is about 7-15 days.
450
V,m3
400
381
324 322 324 355
350
302
300
290
290
255
273
227
250
200
150
85
100
50
0
1
10
11
12
13
10 11 12 Months
Structure of the proposed hybrid water heating system is

given and described in paper [3].
With reference to the technical characteristics (Table 1)
of the condominium it can be accepted the following
parameters for sizing of the experimental hybrid water
heating system:
Table 1. Technical characteristics of condominium
3. Parameters of the system under research
Average monthly demands of domestic hot water in the

condominium are presented in Fig. 2. Temperature of
cold water coming from running water supply grid
varies in Kaunas from 6 C (in spring) to 10 C (in
autumn). Temperature of DHW at the outlet of heat
exchanger is kept about 55 C (standard value).
Fig. 2. Average monthly consumption of domestic hot water

in the condominium
Fig. 1. The condominium where the experimental hybrid

domestic hot water system is intended to install
Nr.
1
2
3
4
5
6
7
Parameters of house
Units
Values
Number of apartments
75
Number of stairwells
7
Number of stories
6
Number of dwellers
270
Total heated area
m2
6 750
Total heated space
m3
16 500
Length x width x height
m
60 x 20 x14,60
Annual consumption of
m3
3 428
DHW
Average monthly
m3
285,67
consumption of DHW
Average daily
m3
9,39
consumption of DHW
Annual consumption of
MWh
175
heat for DHW
Average amount of heat
energy for production of
kWh/m3
51
1 m3 DHW, heat losses
and 75 serpentine pipes
Tariff for water heating
Lt/m3
12,19
Rate of Litas against Euro
is constant and equal to 3,4528 LTL for 1
o
o
Average daily demand in summer (May

August) months is 10 m3/d;
Average DHW demand in summer months is
300 m3 per month;
Capacity of the preheated water tank is equal to
the average daily demand 10 m3.
Technical characteristics of heat exchanger are given

below in Table 2.
Table 2. Technical characteristics of heat exchanger
Nr.
1
2
3
4
5
6
7
Parameters of heat
Units
Values
exchanger P 0,15 - 200
Producer Joint Stock Company
UAB Pergals koncernas
Rated thermal capacity
kW
200
Rated water flow
m3/min
0,15
Operating pressure
MPa
0,6
Testing pressure
MPa
1,25
Maximal operating
C
120
temperature
Length x width x height
cm
37 x 28 x 93
Weight
kg
280
The required number of vacuumed tubes of solar

collectors array can be calculated according to the
formulas given in paper [3]. The necessary number of
tubes, which length is 1,80 m and diameter 58 mm, in
the array makes up 900 tubes. Orientation of the solar
collectors array is southward, tilt angle 45 degrees.
Total length of array will make up approximately 75 m.
The condominium has a flat roof, which is very

convenient and proof enough for the installation and
maintenance of solar collectors array. The area of the
roof makes up over 1 000 m2 and this is more than
enough for mounting the necessary array of solar
collectors meeting the demands of DHW supply to the
357
is based on a 32-bit processor and operates according to

the multitask principle. The operating system is Linux.
As it was described in [3], the optimal control of solar
collectors operation is based on the instantaneous
calculations of the reference value T the difference
between the temperatures T2 (the temperature of the heat
carrier in the solar collector) and T1 (the water
temperature in the preheated water storage tank).
According to the formula (5) proposed in [3] for the
calculation of T, the reference value of T will be high
It is recommended to split the array into the smaller

subarrays in order to place them conveniently for
mounting and maintenance.
4. Means of solar collectors optimal control
The DDC4000 control unit (CU) can be used for control
of the proposed DHW system. This type of CU is often
used for measuring, regulating, controlling, optimizing
and monitoring purposes in building service systems. It
Fig. 3. Block diagram of control process for the hybrid water heating system
358
(over 20 C) and frequency of switching operations of

the pump P1 [3] will be low when the irradiance is high
(close to 1 000 W/m2) and the temperature in the
preheated water storage tank is low (about 10-15 C). In
the opposite case the reference value of T will be low.
Formula (5) can be made more exactly after the natural
experiments. Optimal control of solar collectors can
enlarge the share of solar heat energy in this system up
to three times in comparison with not optimised cases of
operation.
The DDC4000 CU is able to realise the described
calculations and optimal control of water heating
process. It communicates via Ethernet (TCP/IP,
BACnet). The DDC4000 is equipped with object
oriented software and hardware modules, making
available a large number of functions which can be
activated and customized online without the need for
additional tools.
Using BACnet protocol DDC4000 allows the hybrid
water heating system to be integrated in to the Building
management system (BMS). The mentioned above
software and hardware objects are summaries of control
functions in a block with the input and output values.
The following DDC4000 software and hardware objects
are used for the optimal control:
The hardware object H614 "Valve continues" controls a

modulating valve actuator with target setting of S238 Y
output 0-100%. Also hardware object supports manual
influence from DDC in parameter 5 Manual or from
BMS in parameter 13 Z-influence.
The software object Arithmetic carries out
mathematic calculation of object sources subtraction
the temperature of water in PWST and the temperature
of heat carrier in SC [3]. Parameter 2152 result
depicts the result of the arithmetic calculation.
Analogous values (measured or calculated values) can
be monitored for going above or below limiting values
with the DDC software object S066 Limiting value.
The value to be monitored is set in the 4839 Source
limit parameter. The stipulation of the min and max
limiting value is made in the 8615 Selection (MinMax-infringement) parameter. If the value selected in
the parameter 8610 limiting value (set the difference
between the temperature of water in PWST and the
temperature of heat carrier in SC) has reached an
internal contact, parameter 2 G Result limiting value
is set. It is the main premise for collectors pump
switching ON (Fig. 3). The switch back takes place with
a switch back difference in line with parameter 8612
Xsd. The parameter 8610 limiting value is used to
define the instantaneous difference between the
temperature of heat carrier in collector and the
temperature of water in PWST for turning collector
pump on. The optimal difference of the temperatures
has to glide depending on the solar irradiance in order to
have the optimal hybrid water heating system. It is
essential for the proposed hybrid water heating system
control.
The software object S066 makes it to glide depending
on the analogous command value the solar irradiance,
parameter 4832 Q Guide size. This guidance signal
(solar irradiance) lifts or lowers the limiting value set
the difference between the temperature of heat carrier in
collector and the temperature of water in PWST. Impact
of the command value only results in a change of the
target value within a glide range. The glide range is
stipulated by the parameters 8613 Glide START and
8614 Glide END, which are adapted to the actual
hybrid water heating system. The influence of the
guidance signal on the limiting value change is
determined with the parameter 8611 EF/GW (Fig. 4).
The current limiting value is indicated in the parameter
3 XS as actual limiting value (Fig. 5). The value could
be registered as a trend curve in DDC4000 control unit
(Fig. 9). If gliding of limiting value is not active
(command value invalid or EF/GW=0), the limiting
value and 3 XS actual limiting value are identical.
S238 - Basic program PID;

S083 - Arithmetic;
S066 Limiting value;
H613 Valve 3-point or H614 Valve
continues (optionally);
H901 Pump single stage or H905 Double
pump (optionally).
Block diagram of control process for the experimental

hybrid water heating system is given in Fig. 3. Software
object S328 - basic program PID control is used for the
control valve position calculation. The control variable
sensor (DHW temperature) is defined in parameter
5102 source control variable. In parameter 5100 XS
the desired target value for the fixed value regulation is
set the DHW set point. The calculation of the Y output
is used for CV control. The value of the Y output
calculated by the basic program can be overwritten by
manual intervention (7801 manual intervention), BMS
intervention (h01).
The current value of the calculated Y output is displayed
in the parameters 5110 Valve current value. PID
parameters 5120 P share, 5106 I share, 5125 D
share are available on operator level (Fig. 3). The value
of calculated Y output is used by hardware objects H613
"3-point valve" and H614 "Valve continues" optionally.
It depends on the kind of valve actuator three points
or continues modulating.
The hardware object H613 "3-point valve" controls a 3point valve actuator using an analog signal of S238 Y
output (0-100%). If the value of the Y output is to be
increased, an OPEN pulse is created. The length of the
OPEN and CLOSE pulses is calculated from the value
of the change in the Y output and the actuator motor
operating times tMotAuf or tMotZu. Also hardware
object supports manual influence from DDC or BMS.
359
Fig. 6. View of signaling lamps on the CU screen

Fig. 4. Values of solar irradiance on CU screen for gliding of
the difference of temperatures
The deflection of parameters is represented as alarm

message on CU screen and is accessible in users and
service levels (Fig. 8).
Fig. 5. Actual values of the difference of temperatures on CU

screen
Fig. 7. The value setting view on the CU screen
Hardware object H901 "Pump single stage" and H905

"Double pump" necessary for the collector pump and
for the DHW circulation pump control are used
optionally. It depends on the kind of pump. The
hardware object H901 controls a single-stage pump.
H901 supports:
o
o
o
o
o
Switching delays;
Pump blocking protection;
Operating hours / limiting value;
Command execution check;
Malfunction catch / malfunction handling.
The hardware object H905 controls a switchable double

pump (Fig. 2). It supports the same functions as H901
and pump switching function as well (varies with
operating hours, per switching parameter or for
malfunctions) [4].
For the visualization of situation in control process
some indication lamps are set on CU screen (Fig. 6).
DHW parameters are accessible in user or service
levels. The limit of difference in temperatures used as
signal for starting of solar collectors pump P1 [3] is set
on the CU screen only in service level (Fig. 7).
Fig. 8. View of CU screen in case of alarm message - collector

pressure drop below the allowed value
360
3.
4.
5.
6.
Fig. 9. View of trend curve on CU screen for actual difference
of temperatures
The main decisions regarding this work were based on

[4, 5].
7.
5. Discussion
8.
This paper was targeted for the preparation to install the

proposed water heating system for a common DHW
user in order to have a certain theoretical background
before the renovation of the existing system. The
experimental system is intended to install at first in one
condominiums of Kaunas city. After this the research of
the proposed hybrid water heating system is intended to
continue at natural conditions. If the anticipated positive
results will be achieved and disseminated, further
implementation of these systems could be considerable.
A large potential for implementation of the proposed
hybrid water heating systems exists not only in Kaunas
and Lithuania but also in other many countries,
especially in Central and East Europe countries, where
the centralized heat supply grid and buildings with flat
roofs are common.
9.
7. References
1.
2.
3.
6. Conclusions
1.
2.
reduce the bills for domestic hot water in

condominium.
One of the main tasks for control of solar collectors
in the proposed hybrid water heating system is to
arrange operation of solar collectors in the way
maximizing the capture of solar irradiation.
Optimization of the proposed condominiums
hybrid water heating system is carried out on the
control unit DDC4000.
The object oriented software and hardware objects
of DDC4000 make available a control program to
be formed in way of objects activation and
customization.
The optimal control of solar collectors operation is
based on the instantaneous calculations of the
reference value of the difference between the
temperature of the heat carrier in the solar collector
and the temperature of water in the preheated water
storage tank.
The gliding limit value for collector pump
switching allows optimize control for maximum
solar energy utilization.
Optimal control of solar collectors can enlarge the
share of solar heat energy in this system up to three
times in comparison with not optimised cases of
operation.
The experimental hybrid water heating system
described in this paper is intended to install at first
in one of Kaunas city condominiums.
Presently there are many possibilities for

substitution of the technologies based on dirty fossil
fuels by technologies running on the renewables
partially or entirely.
The hybrid water heating system based on solar
collectors and heat exchanger fed from the
centralised heat supply grid is proposed in order to
shift partially on renewable solar energy and to
4.
5.
361
Zweibel K., Mason J. and Fthernakis V. A Solar

Grand Plan // Scientific American Magazine.
December 16, 2007.
Repower America 100 % clean electricity within 10
years. www.RepowerAmerica.org (called on
19.01.2009).
Adomaviius V., Petrauskas G. Hybrid water
heating system based on solar collectors and heat
exchanger fed from the centralized heat supply grid
// Proceedings of International Conference
Electrical and Control Technologies 2009.
KTU, Kaunas, 2009 (in print).
Kieback&Peter GmbH & Co. DDC4000
engineering manual, 2009. P. 73-85.
Duffy J. A. and Beckman W. A. Solar Engineering
of Thermal Processes. John Wiley and Sons, Inc.,
2008. 908 p.

CONTROL OF ENERGY FLUXES IN THE GRID-TIED

DOMESTIC POWER SYSTEM BASED ON WIND ENERGY
eslovas RAMONAS*, Vytautas ADOMAVIIUS**, Vytautas KEPALAS*
*Department of Control Technologies at Kaunas University of Technology, Lithuania
** The Centre for Renewable Energy Technologies at Kaunas University of Technology, Lithuania
Abstract: Control of energy fluxes in the grid-tied

domestic power system based on wind energy is
described in this paper. One object of control is the wind
turbine itself. The load of wind turbine must be
automatically controlled in order to maximize the power
output. The produced electric energy must be divided
between the electrical appliances, the power storage and
the heat storage tank. Surplus of electricity can be
supplied to the electric grid. Power from electric grid
must be delivered to all domestic energy users in case of
wind turbine breakdown or unexpected long term
doldrums. Control system must safeguard energy
distribution in all mentioned cases.
some obstacles blocking the rapid expansion of

renewable energies and that we have to join forces to
overcome them. It will be done because the time is
working in favour of renewables: due to technological
advances, renewable energies are often already a
competitive alternative to conventional energy sources.
And what's more, technological progress means they are
cheaper from year to year. Renewables are developing
into an important economic sector. In 2008 over 150
billion dollars were invested in renewable energies
worldwide [2].
The foundation of IRENA will promote the expansion
of renewable energies internationally. IRENA will
encourage both industrialised and developing countries
from all over the world to create their political and legal
frameworks for the successful sustainable development.
Creation of the right incentives and securities for
investment are very important tasks of new
organization. This process needs to be steered by
governments, since a distorted market is not capable of
initiating the transformation of energy systems. IRENA
will act as a catalyst to facilitate technology and
knowledge transfer, and to support capacity building
[2]. Supporters of sustainable development now can
expect to watch the acceleration of this process.
Keywords: control system, wind energy, wind turbine,

domestic power system, grid-tied inverter.
1. Introduction
Up to this year (2009) International Energy Agency
(IEA) and International Atomic Energy Agency (IAEA)
were the main international organizations in sphere of
energy affairs. However, mostly they were not among
the diligent supporters of sustainable development.
Therefore new international organization was necessary
to coordinate and promote development of the
renewables. Germany, Denmark and Spain were the
leading promoters of this idea. The Conference on the
establishment of the International Renewable Energy
Agency was held in Bonn, Germany on 26 January,
2009 [1].
As it was reported in the conference, the International
Renewable Energy Agency (IRENA) was founded to
press ahead with the expansion of renewable energies in
an even more comprehensive and targeted way [2].
German Federal Environment Minister Sigmar Gabriel
declared that IRENA is an expression of our conviction
that in future we can cover the main share of our energy
consumption with renewable energies and an
expression of our awareness that renewable energies
offer huge potential and that the technologies for their
use are available. But he also recognized that there are
2. Research object
Small scale efficient wind turbines [3] of sufficient
capacity can be used for the heating purposes as well
[4]. Simplified scheme of wind turbines power
conversion system used for all domestic energy needs is
shown in Fig. 1. The wind turbine (WT) has a
permanent magnet synchronous generator SG. As it is
shown in Fig. 1, the synchronous generator SG is
connected to the electric grid over the rectifier, the
shorting circuit, the inverter, the transformer and the
grids automatic switch 1Q,F. Shorting circuit consist of
shorting transistor Vt, reactor Ld and disjunctive diode
Vat. Wind energy converted by the WT is used for the
domestic electrical appliances, for charging the battery,
for the domestic hot water (DHW) preparation and
362
space heating. Surplus of electric power can be supplied

into electric grid. Appliances for the DHW and space
heating can be connected directly to the rectifier by
means of the automatic switch 2Q,F. Capacity of the
inverter can be reduced in this case and therefore the
overall cost of the whole system will be reduced as well.
The grid-connected inverter fed from the battery B some
time can supply electricity for the most indispensible
users (lighting, refrigerator, radio, TV) in case of

blackout and absence of wind simultaneously.
Compatibility of the battery voltage and the grid voltage
is an important task of the systems design. The voltagematching transformer connected as the autotransformer
is used in this case (Fig. 1). The required voltage of the
transformers secondary windings can be calculated by
using this equation:
Fig. 1. Simplified scheme of wind turbines power conversion system used for the domestic energy needs
U 2 p U1p
UB
3 ks
(1)
IR
where UB the battery voltage; U1p the phase voltage

of the electric grid; U2p the phase voltage of the
transformers secondary winding; ks factor of the
inverters scheme.
The WT is controlled from its control system (Fig. 1).
The control system generates pulses for the inverter
control (ICP) and for the transistor of shorting circuit
(SCP), performs discrete control of the heating units
(HCU), the battery (B) and control of other units and
appliances. Apart from this, the control system performs
control of the WTs load over the shorting circuit.
Control system of the generators load consist of the
torque signal former TSF, the reference current signal
former CSF, the shorting control pulse former SCPF
and the current controller Wsr. The following signals
operate in the control system: g signal of the angular
speed, Tg signal of the torque, Ir signal of the
reference current, Ig signal of the rectified current, Uk
the control voltage, vw signal of the momentary wind
speed. The reference current can be calculated in a
similar way as in paper [5]:
Pi
,
kis k m g
(2)
In our case we have the only difference the power

share supplied into electric grid (Pi) is taken here
instead of full power of the generator because we have
to take into consideration that generator is (or can be)
already loaded by heating units and (or) other electrical
appliances. The generators power share Pi supplied into
electric grid can be described mathematically in this
way:
Pi Ptm
U d2
,
Rh
(3)
where in equations (1) and (2): km torque factor; Pi

the generators power share supplied into the grid; Ptm
the power produced by wind turbine; kis - is coefficient
of rectifiers scheme (for the three-phase six-pulse
rectifier kis0.78); Ud the rectified voltage of the
generator; Rh the active resistance of the heating unit.
The power produced by wind turbine Ptm can be
calculated by using power curve, which is given in the
363
As it can be seen in Fig. 1, the load of WTs generator

can be controlled in discrete mode by changing the
number of electricity users and gradually by adjusting
the power supplied into electric grid. Algorithm of wind
turbines load control presented in Fig. 2 explains how it
can be done.
WTs technical documentation. As a rule, polynomial of

the n-th order is used for approximation of the power
curve on purpose to have a sufficient adequacy of the
mathematical description. So, the power curve of WT
mathematically can be described by this equation:
Pt an vwn an1 vwn1 an2 vwn2 ...

a1 vw a0
(4)
Fig. 2. Algorithm of wind turbines load control
364
The presented algorithm of wind turbines load control

allows maximizing the utilization of power produced by
the WT. As it is shown in Fig. 2, at first the wind speed
vW is measured and the power Pt produced by the WT is
calculated. If the condition Pt > 0 is met, the control
system allows switching on the domestic electrical
appliances according to the presented algorithm. If the
Pt = 0, the operation of the electric grid is checked (the
grid failures can occur due to the blackouts, repair
works, etc.). If the grid failure is not found, supply of
electricity from the grid is switched on.
If the failure of grid exists, then the control system
checks the level of the batterys charge. If the battery is
charged enough (Ub > Up1), the power supply from
battery to the domestic electrical appliances is allowed.
In contrary some domestic electrical appliances are
switched off and the indicator Low battery is switched
on. The minimum permissible level of the battery
discharge Up2 is also checked: if this level is reached, all
domestic electrical appliances are disconnected.
Fortunately, this case is hardly possible because the
power supply is 3 times reserved (3 sources are used
the WT, the power grid and the battery. If the wind rotor
is running (Pt > 0), then the control system checks the
condition Pt > Pda (capacity of the electrical
appliances). If yes, further on the battery is checked if it
is fully charged. If voltage of the battery is below the
maximum value Ubm, the charging is switched on.
The water heating process is controlled as well: if the
power of WT is high enough (Pg < Pt) and if the
temperature of water is bellow the limit of the
permissible value TWT, the water heating is switched on.
When the limit value is reached, the heating of water is
switched off.
If the strong wind is blowing and a certain part of the
WTs power is unused when the condition Pg < Pt is
met, the grid inverter is switched on and the surplus
power is supplied into the electric grid.
If the wind is not blowing (Pt = 0) and grid failure
exists, electricity is supplied from the battery only for
the domestic power appliences. Two levels of the
battery charge are checked: if the battery is full, all
appliances are fed. If the battery charge level is below
Up1, some electric appliances are switched off and the
indicator Low battery is switched on.
presented in Fig. 3 will be used as the equivalent of

wind turbines circuitry (Fig. 1).
Ld
s2
id
rd
L1
s3
iL
r1
Rh
eg0
ih
PMSG
s4
s1
+
b
ii Lf
rf
iv
rk
rb
Vat1
ik
ib
iv
Ck Eb
Vat2
Lt
ii
rt
ei
SIC
Fig. 3. Equivalent scheme of the system
Like it was described in reference [6], system of

equations for the power inverting mode can be made up
as follows:
1
did
dt L u d u ab rd id ;
d
dii 1 u e r i ;
dt Li ab i i i
du
1
c U c0
ik ;
dt
C
k
u e u u r i i ;
g0
g
K
1
h
d
d
d ih id
u g 2 r1 ih id 2 L1
;
dt
m xK
ih id ;
u K
2
U c 0 k s Eim ;
u u r i ;
c
k k
ab
1
ib r u ab Eb ;
b
ik id ib ii .
(5)
where ei the counteracting electromotive force of the

grid; Eim the maximum value of the grids delta
voltage; rirt+rf the active resistance of the inverter
circuit; rt the active resistance of the grid; rf the
active resistance of the filter; LiLt+Lf the inductance
of the inverters circuit; Lt the inductance of the grid;
Lf the inductance of the filter; ii the inverted current;
uk the voltage of the storing capacitor Ck; ik the
current of the storing capacitor Ck; ih - the current of the
heating unit; ib - the current of the battery; id - the
current of the reactor; ks coefficient of the rectifiers
scheme.
The counteracting electromotive force of the inverter ei,
electromotive force of the generator eg0 and other
parameters can be calculated by using the method
described in the preceding works of authors [5, 6, 7].
Synchronous generator with permanent magnets
(PMSG) can be described like it was done in paper [5]:
3. Mathematical description of the proposed power

system
Mathematical description of the systems power storage
and inverting circuits (SIC) is set out well enough in the
previous papers published by authors [5, 6], therefore it
can be skipped in this paper. Only singularities of
mathematical description depending on switching mode
of the heating elements and the battery here will be
presented. The average values of currents and voltages
can be used for the description of these processes and
non-stationary processes of the electromagnetic field
alternations can be neglected [5, 6]. One-line scheme
365
d g
1
Mt M g M f ;
J
dt
Ptm
M t ;
g
M g k m iL 2 ;
3 3
M k .
F
g
f
6
5
Pt 0.0461 v w
2.1317 v w
36.5307 v w4
3
271.5777 v w
731.0533 v w2 531.6232 v w
(7)
8.5174.
(6)
Mathematical model of the wind-based domestic power

system consists of three functional parts: the converter,
which includes inverter, power storage circuit and
control pulse former, the generator and the wind turbine.
The control pulse former consists of the counteraction
electromotive force unit and modulator. The TTVS unit
performs the gradual control of the generators load.
The load of generator is controlled by means of varying
value of the stators rectified current. The reference
rectified current is calculated in the models unit WT
according to the (2) and (3) formulas.
The reference current iLR reaches the TTVS unit through
the inlet iref and the measured rectified current
through the inlet id. The control pulse formed by the
controller TTVS on its output Cp-Vt controls the
rectified current of the generator.
where Mg the torque of generator; Mt the torque of

wind rotor; Mf the combined torque of viscous friction
of rotor and load; J the combined inertia of rotor and
load; g the angular velocity of the generators rotor;
km torque factor; kf friction factor; Ptm the power
produced by wind turbine; the angle of the rectifier
diodes commutation.
4. Mathematical model and simulation
Mathematical model of the proposed power system is
shown in Fig. 4.
5. Results of research
As it was shown in Fig. 2, according to the proposed
algorithm the system can supply the wind energy
converted into the power to the water heating unit or
into the electric grid, if the heat storage tank is fully
charged. If generator produces more energy than heating
units require, the surplus power can be supplied to the
electric grid as well.
Quality of the load control in the proposed system was
researched by means of mathematical simulation.
Results of the research are presented below in Fig. 5 and
Fig.6.
Fig. 4. Mathematical model for the WTs power system
It was made up using the systems of equations (5, 6) and

the equations (24) as well. The storage chain Storage
Chain and the inverting circuits Inverter of the
mathematical model (Fig. 4) are made up by application
of the previously published works of authors [5-7]. The
mathematical model of wind turbine was worked out
using the WTs power curve, which can be described by
means of the following equation:
Fig. 5. Curves of the researched wind power systems

parameters at the following jumps of the action signals of
generator: wind speed v = from 9 to 5 m/s, at t = 6 s;
switch off heating unit at t = 5 s
366
Experimental curves are presented in Fig. 5 for the case

when WT charges the heat storage tank and supplies
electric energy into the electric grid. The process of
control there is shown when the heating unit was
switched off at the time 5 s and the wind velocity
jumped down from 9 m/s to 5 m/s at the time 6 s.
As it is clearly shown on Fig. 5, the power of generator
remains practically the same after the disconnecting of
heating elements and the same share of power used for
heating is then supplied into the electric grid (all power
of the generator is supplied into the electric grid after 5
s). The rectified current of the generator slightly
decreases and the rotational speed of the generator (and
wind rotor) slightly increases due to the change of the
current circuit configuration. Power of the WT sharply
decreases after 6 s due to the wind speed jumping down.
Therefore the angular speed of the wind turbine and the
generators rectified current accordingly decreases.
operation and more exact specifying of the power

circuits parameters (Ld, rk, Ck).
6. Conclusions
1.
2.
3.
4.
5.
Control of load and power fluxes in the grid-tied

domestic energy system based on wind turbine is
researched and algorithm of wind turbines load
control is worked out.
The proposed wind turbines control system
allows the meeting of electricity demands not
only in case of low wind speed but also in cases
of the blackouts and emergency.
The proposed wind turbines control system also
allows reducing capacity of the inverter and
therefore the overall expenditure of the full
system because only a part of power is supplied
into the electric grid.
The presented mathematical model of the
described energy system allows researching the
systems modes of operation and more exact
specifying of the power circuits parameters.
The presented experimental curves confirm the
correct operation of the researched domestic
power system based on wind energy.
7. References
1.
2.

generator: wind speed v = from 9 to 5 m/s, at t = 6 s;
switch off heating unit at t = 5 s
3.
Alteration of the wind-based domestic power systems

parameters depending on the change of wind turbines
operation modes is shown in Fig. 6. There it can be
noticed that at first the inverted current is not very
strong because the major part of power produced by the
WTs generator is used for the heating. After the
disconnection of heating unit at the instant 5 s the
voltage on the terminals of inverter slightly increases
and sharply increases the inverted current. The load
torque of the wind turbines practically remains the
same. At the instant 6 s, when the wind speed jumps
down, the rotational speed, the power of wind turbine
and the inverted current decreases. It can be concluded
that the experimental curves of the researched wind
power system correctly reflect the electromagnetic
processes taking place in the systems power circuits.
The worked out mathematical model of the wind-based
energy system allows researching the systems modes of
4.
5.
6.
7.
367
German Federal Ministry for Environment,

Nature Conservation and Nuclear Safety. A
milestone for future-oriented energy supply.
Press release No. 023/09. Berlin, 26.012009.
Gabriel S. A pillar of clean energy supply.
Founding Conference of the International
Renewable Energy Agency (IRENA). Bonn,
January 26, 2009.
Martinez F., Herrero L., Santiago P. and
Gonzales J. Analysis of the efficiency
improvement in small wind turbines when speed
is controlled. 1-4244-0755-9/07/$20.002007
IEEE.
efik M. Bajmak. Possible applications for the
wind energy in the heating and air conditioning
systems. Facta Universitatis. Series Mechanical
Engineering Vol. 5, No.1, 2007. pp.71-78.
Adomaviius V., Ramonas ., Kepalas V.
Control of wind turbines load in order to
maximize the energy output // Electronics and
Electrical Engineering, 2008. Nr. 8(88). P.7176.
Kepalas V., Ramonas ., Adomaviius V.
Connection of the RES-based Power Plants into
the Electric Grid // Electronics and Electrical
Engineering, 2007. Nr. 8(80). P.67-72.
Paulauskas M., Kepalas V., Ramonas .
Platuminio asinchroninio ventilinio kaskado
dinamika // ISSN 1392 - 1215 Elektronika ir
elektrotechnika, 1996. Nr. 4 (8). P. 56-58.

RESEARCH OF THE POWER CONVERSION PROCESSES IN THE SYSTEM OF

POWER SUPPLY FROM A NUMBER OF WIND TURBINES OVER THE ONE
GRID-TIED INVERTER
eslovas RAMONAS*, Vytautas ADOMAVIIUS**, Vytautas KEPALAS*
* Department of Control Technology, Kaunas University of Technology, Lithuania
** The Centre for Renewables Energy Technologies, Kaunas University of Technology, Lithuania
Last year Lithuania started to produce small wind

turbines as well. Number of installed small wind
turbines in Lithuania is not booked. Approximately it
could be about 10-20 installations per year during the
last 2-3 years but considerable increase is predicted.
The degree of maturity of various wind turbines varies
in a wide range. Manufacturers still have many
possibilities to improve design of this class wind
turbines in order to ensure operation capability in lower
and higher wind speeds, to apply more advanced blade
design an manufacturing methods, to decrease noise
level, to made the wind turbines more visually
attractive, to reduce their prices, etc. German
researchers have a target to reduce price of small wind
turbines up to 1 euro per 1 watt of capacity.
Design of power electronics devices can be improved
as well. First of all, the devices could be designed to
meet stronger safety and reliability standards. The same
model of wind turbine could be adaptable to either ongrid or off-grid use.
Abstract: Only one mutual inverter can be used in the

power conversion system for a number of wind turbines
having different capacities and installed close to each
other. This is applicable both for stand-alone and gridtied cases of wind turbines and for wide range of
capacities. The proposed electrical scheme for power
supply from the wind farm into electric grid allows
using wind turbines of different types and capacity.
Results of research of the power conversion processes
in the system of power supply from a number of wind
turbines over the one grid-tied inverter are presented in
this paper. The main advantage of this scheme is costeffectiveness of this power conversion system.
Keywords: wind farm, mutual grid-tied inverter, power
conversion system.
1. Introduction
The wind-powered installations for electricity
generation with rated capacity up to 100 kW are
considered as small wind turbines. Small wind turbines
are produced in more than 25 countries and over 300
different models already are available worldwide [1].
The global market of small scale wind turbines is led by
the USA, the United Kingdom, China, Canada,
Germany and other countries.
Presently price of small wind turbines usually varies
between 3 and 5 USD per watt of capacity. However,
sometimes price of this class wind turbines is very high
[2]. Price of electricity generated by the small wind
turbines usually is about 10-15 cents (US).
Rate of development of small wind turbines global
market is not as high as in case of large wind turbines
the annual increase makes up about 15-20 %. However,
it is considered that business for small scale wind
turbines can be big as well and that it was too long
overlooked [3]. Indicators of small wind turbines
market for 2008 year confirm a considerable leap in this
sector indeed.
2. Research object and method

Sometimes many small scale wind turbines (WT) of the
same type can be mounted on a parapet of big building
or on the ground in order to produce more electricity
[3]. In this case it would be reasonable to use only one
inverter in the power conversion system on purpose to
reduce expenditures on the power converting system.
Research of the proposed system for power supply from
the farm of small scale wind turbines into electric grid
(EG) over one mutual inverter is described in this paper.
Wind farms, which have wind turbines of different
capacity and type, are applicable as well.
Elaboration of the proposed power conversion system
for small wind turbines is based on the researches
described in papers [4, 5, and 7].
Simplified electrical scheme of the system for power
supply from the farm of small scale WTs into EG is
presented in Fig. 1.
368
Fig. 1. Simplified scheme of power supply from park of small-scale WTs into EG over the one inverter
the WTs angular speeds, Tgn the signals of the

torques of the generators, Irn the signals of the
reference currents of the generators, Ign the signals of
the rectified currents of the generators, Ukn the control
voltages for the generators, Vwn the signal of the
momentary wind speed of the generators.
Mathematical simulation (MATLAB/SIMULINK) was
used as a research method to check the operation
abilities of the proposed electrical scheme. Initially the
task was simplified by using only two wind turbines,
two shorting circuit, one capacitor for power storage
and one mutual inverter. The same method can be used
for research of the system representing the wind park
with any number of WTs.
The suggested scheme has the power circuitry and the

control system.
The systems power circuitry consists of n
asynchronous and synchronous generators SG-1, SG2 SG-n, n rectifiers, n shorting circuit units
consisting of power storage reactors Ldn, transistors Vtn
and disjunctive diodes Vatn, one mutual capacitor for
power storage Ck, one mutual inverter, inverters switch
IS, transformer and of some other known elements.
Control system of wind turbines operates together with
n shorting circuit units, control pulse formers SCPFn,
current signal formers CSFn, torque signal formers TSFn
and one inverter pulse former IPF. The following
signals operate in the control system: gn signals of
369
is being substituted by the equivalent one phase scheme.

Power switches in this scheme are being considered as
ideal.
Equivalent scheme of the power converter presented in
Fig. 1 is given in Fig. 2. The unmarked diodes show the
directions of conductivities in this scheme. This scheme
has the same n equivalent circuits of rectifiers, one
mutual capacitor for power storage and one inverter.
3. Mathematical description and model of the

proposed wind energy conversion system
Exact description of electromagnetic processes in
circuits of power conversion system given in Fig. 1 is
rather complicated task. Therefore some simplifications
used in theory of converters circuits and facilitating its
mathematical description will be applied [4, 5, 7].
Power converter having electrical circuits with n-phases
Ld1
Ld2
rd1
Ldn
rd2
Lg1
S1
rf
Lg2
S2
Vat
Lgn
ign
rg1
rg2
rgn
ug2
Sn
+
ugn
SG2
Lf
ii
ig2
SG1
Si
rdn
ig1
ug1
b
SGn
rk
Vat
ik
iv
Ck
Lt
ii
rt
ei
SIU
Fig. 2. Equivalent scheme of the system
where ugn the voltage of the generator n; ign the

rectified current of the generator; rnrgn+rdn the active
resistance of the generator circuit; rgn the active
resistance of the generator; rdn the active resistance of
the reactor; LnLgn+Ldn the inductance of the
generators circuit; Lgn the inductance of the
generator; Ldn the inductance of the reactor.
When the generators operate in the inverting mode, the
system of equations for calculating its power currents
can be made up as follows:
When the wind turbines generators ug1, rg1 ugn, rgn are
switched into the scheme of short circuit by means of
the switches S1 Sn (in case of AC the rectifier is
necessary), the power is being stored in the reactors Ld1
Ldn. When the switches S1 Sn are switched out, the
power flows into the storing capacity Ck. If the Si is
switched on, the inverter I supplies the power into
electric grid 7. Pulse width modulation realised by
means of the switch Si is used in order to adjust voltages
of the wind generators and the electric grid. The grid
voltage is used for the modulation. In this way
operation of converter is being synchronized with the
electric grid. Carrier frequency of the modulated signal
usually makes up from few to 20 kHz. Operation of the
switches Sn can be synchronized with the electric grid or
not. In our case switches S1 Sn operate independently
from the electric grid voltage.
Mathematical description of the systems power circuits
was made up with reference to the equivalent scheme of
the proposed system (Fig.2).
Every generator (Fig. 2) can operate in the two modes:
shorting mode when the switches S1 Sn are switched
on, and power inverting mode when the switches S1
Sn are switched out. When the generators are operating
in the shorting mode (S1 Sn are switched on), their
power circuits are not related electrically (independent).
Then currents of the generatorspower circuits can be
deduced from the following independent equations:
dign
dt
1
u gn rn ign .
Ln
dig1 1
dt L u g1 uab r1 ig1 ;
1
dig 2 1
u g 2 uab r2 ig 2 ;
L2
dt

1
dign
u gn uab rn ign ;
dt
Ln
di
1
i uab ei ri ii ;
dt Li
U k E ;
s
im
c0
du
1
c
U c0
ik ;
dt
Ck
u u r i ;
ab n c k k
i i i .
k i 1 gi i
(1)
370
(2)
The second generator SG2 is synchronous machine with

permanent magnets. Its type GL-PMG-5, rated
capacity 5 kW. The power curve of this wind turbine
of vertical axis mathematically is described by the
following equation:
where ei the counteracting electromotive force of the

grid; Eim the maximum value of the grids linear
voltage; rirt+rf the active resistance of the inverter
circuit; rt the active resistance of the grid; rf the
active resistance of the filter; LiLt+Lf the inductance
of the inverters circuit; Lt the inductance of the grid;
Lf the inductance of the filter; ii the inverted current;
uk the voltage of the storing capacitor Ck; ik the
current of the storing capacitor Ck; ks the coefficient of
the rectifiers scheme.
The counteracting electromotive force (EMF) of the
inverter can be calculated by using the method
described in the preceding works of authors [4, 7].
Apart from the shorting and inverting modes of
operation, the proposed system of wind power
conversion has the power storage mode of operation
when the power from the generators is supplied into the
storing capacity Ck. This mode of operation can be
described by using the system of equations (2) without
the fourth equation (counting from the top) taking into
consideration that the inverting current is cut off (ii0).
Mechanical part of the system (Fig.1) can be described
by independent equations as follows:
d gn
dt
1
Jn
Ptn
Tgn ign ,
i
gn n
Pt 16.517 10 5 v w5 0.01881 v w4
0.6797 v w3 9.1408 v w2 27.8 v w 4.
Mathematical model (Fig. 3) for the wind power

conversion system with 2 generators and one mutual
inverter was elaborated on basis of the previous
researches of authors [4, 5 and 7] and on the systems of
equations (1 5).
(3)
where Tgn the torque of the n-th generator; Ptn the

power produced by the n-th wind turbine; Jn the
moment of inertia of the n-th wind turbine; ign the
rectified current of generator; in multiplication factor
of the gear box of the n-th wind turbine; gn the
rotational speed of the n-th generator.
Each generators torque Tgn depends on its rectified
current ign. More detailed mathematical description of
the generators of various types, including asynchronous
and synchronous, is presented in the preceding authors
works [4, 5].
Mathematical description of the WTs capacity
depending on the wind speed is based on their power
curves, which usually are given in the wind turbines
technical documentation. Polynomial of the n-th order is
used for the approximation of the power curve on
purpose to have a sufficient adequacy the mathematical
description.
Number of wind turbines and generators in the
mathematical model of the system for power supply was
limited up to two in order to facilitate the researches.
The
slip-ring
asynchronous
generator,
type
4AHK355M12Y3, rated capacity 110 kW, rated
rotational speed 500 rpm was taken as the first
generator AG1. This generator is rotated by the wind
rotor of wind turbine N27/150. The power curve of the
wind turbine N27/150 mathematically is described by
this equation:
Pt 16.517 10 5 v w5 0.01881 v w4
0.6797 v w3 9.1408 v w2 27.8 v w 4.
(5)
Fig. 3. Mathematical model for the power supply system with

two different wind turbines (generators AG1 and SG2)
The same units can be found in the mathematical model

of the proposed power supply system as in the electrical
scheme (Fig. 2). Mathematical model of the power
supply system with two different wind turbines consist
of three functional parts:
o
o
o
the converter, which includes inverter and

power storage chain,
the control pulse former (counteracting
electromotive force unit and modulator),
the two units of WTs.
The WTs units include control pulse formers used to

control their short-circuiting transistors.
Mathematical model (Fig. 3) of the storage and
inverting circuits (Storage chain, Inverter) was
elaborated by the application of the same researches
previously carried out by authors [4-7].
(4)
371
As it can be seen on Fig. 4 and Fig. 5, load of the

generator AG1 suddenly starts to increase after the
jump of the control signal of this generator (AG1) at the
moment of time 4 s. Therefore the inverted current and
voltage on the inverter terminals begin to increase after
this moment. After this some impact originates on the
operation of the second generator (SG2) the angular
speed of the SG2 slightly increases due to the increase
of the voltage on the inverter terminals.
Additionally, as it can be seen on Fig. 4 and Fig. 5,
changes of operation mode of small scale generator SG2
(rated capacity 5 kW) practically do not have any
impact to the operation of the generator AG1 (rated
capacity 110 kW).
Capacity of the power storing capacitor has a significant
impact into the character of the voltage on the inverters
terminals and on the inverted current: their
electromagnetic oscillations decrease considerably at
high capacity Ck (Fig. 4 is for Ck capacity 2000 F and
Fig. 5 for 0.2 F).
Inverted current in Fig. 6 has not a shape of real sine
it has a pulsating character and therefore the filter is
necessary.
4. Research of the energy conversion processes in

proposed wind energy conversion system
Processes of energy conversion in the system of power

supply from park of small scale WTs into the EG over
the one inverter were researched by means of
mathematical simulation. Results of the researches are
presented below in form of the experimental curves.
Modes of operation of very different generators by type
and capacity were researched when the one grid-tied
inverter served for both of them.
Curves of response of the researched wind power
systems parameters at the jumps of the action signals
(explained below) are given in Fig. 4 and Fig. 5.

generators: AG1 vG1=1014 m/s, at t1=5 s; UG1=31 V,
at t2=4 s, SG2 vG2=912 m/s, at t1=7 s; UG2=3.0 V, at
t2=0 s; (Ck2000 F)
Fig. 6. Curves of the AG-1 generator parameters at the

following jumps of the action signals of generators: AG1
vG1=1014 m/s, at t1=5 s; UG1=31 V, at t2=4 s, SG2
vG2=912 m/s, at t1=7 s; UG2=3.0 V, at t2=0 s
There vG1, vG2 jumps of wind speeds accordingly

for the first and the second wind turbines; UG1, UG2
jumps of control signals of control pulse formers for
short-circuiting transistors of the corresponding circuits.
Curves of the rectified current of the first wind turbines

generator (AG1), angular velocity and torque of the first
wind turbine are presented in Fig. 7.

generators: AG1 vG1=1014 m/s, at t1=5 s; UG1=31 V,
at t2=4 s, SG2 vG2=912 m/s, at t1=7 s; UG2=3.0 V, at
t2=0 s; (Ck200000 F = 0.2 F)
Fig. 7. Curves of the AG1 generators parameters at the

following jumps of the action signals of generators: AG1
vG1=1014 m/s, at t1=5 s; UG1=31V, at t2=4 s, SG2
vG2=912 m/s, at t1=7 s; UG2=3.0 V, at t2=0 s
372
The Fig. 7 shows how parameters of the generator AG1

are changing after the increase of its load. It also can be
seen that after the sudden increase of the wind speed
suddenly increases the torque generated by the turbine,
but the angular speed of the wind turbine and the
rectified current of the generator increase slowly due to
the considerable moment of inertia.
Curves of the rectified current, angular velocity and
torque of the WT shaft for the second wind turbine
(generator SG2) are presented in Fig. 8.
3.
4.
5.
6.
7.
capacities of wind turbines generators any types

of wind turbines can be included into the system.
Voltage of the wind turbines generators in the
proposed scheme can adjust to the voltage of
electric grid automatically.
Capacity of the power storing capacitor has a
significant inverse impact into the amplitude of
electromagnetic pulsations of the voltage on the
inverters terminals and of the inverted current.
As it was researched, the impact of one generator
into the operation of other generators can be
neglected.
The main advantage of the proposed scheme for
power supply from the wind park is costeffectiveness of the power conversion system.
The made up mathematical model allows to
research effectively the multi-generational systems
of power supply from park of WTs into EG over
the one inverter in order to find the ways for the
improvement of such systems.
6. References
1.
2.
3.
Fig. 8. Curves of the SG2 generator parameters at the

following jumps of the action signals of generators: AG1vG1=1014 m/s, at t1=5 s; UG1=31 V, at t2=4 s, SG2 vG2=912 m/s, at t1=7 s; UG2=3.0 V, at t2=0 s
4.
As it can be seen on Fig. 8, the change of the more

powerful wind turbines (AG1) operation mode has
impact into the operation of the smaller wind turbine
(SG2). However, this impact is not considerable (it does
not exceed 10 %) and it can be easily compensated by
the adjusting of control signal Uk2 in the circuitry of the
generators SG2 current control.
5.
6.
5. Conclusions
1.
2.
The proposed conversion system for power supply

from the park of wind turbines into electric grid
allows using only the one mutual grid-connected
inverter.
The described power conversion system with one
mutual inverter is friendly to various types and
7.
373
AWEA Small Wind Turbine Global Market Study

2008. Published by the American Wind Energy
Association, June 2008. 26 p.
http://www.allsmallwindturbines.com (called on
07.02.2009).
Murray D. Big Opportunities for the Small Wind
Industry. Go Green Energy, LLC. 2008. p.5.
(http://www.gogreenenergyllc.com called on
03.04.2008).
Kepalas V., Ramonas ., Adomaviius V.
Connection of the RES-based Power Plants into
the Electric Grid // Electronics and Electrical
Engineering, 2007. Nr. 8(80). P.67-72.
Adomaviius V., Ramonas ., Kepalas V. Control
of wind turbines load in order to maximize the
energy output // Electronics and Electrical
Engineering, 2008. Nr. 8(88). P.71-76.
Adomaviius V., Ramonas . Power supply from
the park of wind turbines over the one grid-tied
inverter //Proceedings of the International
Conference Biosystems Engineering and
Processes in Agriculture. - No. 13, 2008. P.7780.
Paulauskas M., Kepalas V., Ramonas .
Platuminio asinchroninio ventilinio kaskado
dinamika // ISSN 1392 - 1215 Elektronika ir
elektrotechnika, 1996. Nr. 4(8). P. 56-58.

COMPARISON OF SMALL SCALE WIND TURBINES PROPERTIES

Vytautas ADOMAVIIUS*, Thomas WATKOWSKI**, Edmundas ILINSKAS***, Alfonsas ADOMAVIIUS****
**Lithuanian-German joint venture UAB TOMETA, Germany
***Joint Stock Company UAB SAULES ENERGIJA, Lithuania
****Kaunas Business Machines Company, Lithuania
Abstract: Manufacturers from all over the world

produce many types of small scale wind turbines, which
parameters are measured under different conditions.
Therefore comparison and choosing of the small wind
turbine are not very simple tasks. The method used for
comparison of small wind turbines by means of
comparative indexes is presented in this paper and
applied for the evaluation of 45 small wind turbines of
horizontal axes and 14 small wind turbines of vertical
axes. Calculations of the comparative indexes of the
selected small scale wind turbines were carried out on
basis of data from internet. The rating of small wind
turbines from both classes according to the comparative
indexes was carried out and the corresponding results
presented in the paper. A distinct general superiority of
small horizontal axis wind turbines against the vertical
axis wind turbines is showed by means of columnar
diagrams.
or space heating [4, 5]. Attractiveness of these systems

is also nonpolluting and power production without any
fuel.
As it is foresighted, business of small wind turbines in
future will keep growing rapidly [6, 7]. This trend will
be actuated due to the decreasing price of small WT.
The target price 1000 for 1 kW (or 1 for 1W) of
installed capacity can be achieved rather soon. The best
small WT already are close to it.
2. Peculiarities of the small scale wind turbines
estimation
Various manufacturers of small WT test them at
different conditions. Rated speed of various WT is very
different as well from 8 m/s up to 17 m/s. Standard
testing conditions for small WT are not established yet.
Some manufacturers are not giving the power curves for
their WT. Sometimes the manufacturer are claiming that
their WT can put out a certain quantity of energy per
month while productivity of WT will mostly depend on
the local wind energy resources in the site of WTs
installation. Therefore choosing of wind turbine is rather
responsible task, which is related with possibility to
suffer considerable financial losses, especially when the
necessary capacity of WT or their number is high.
In the meantime unambiguous standard testing
conditions (STC) are established for measurement of the
photovoltaic cells or modules nominal output power.
They are following:
Keywords: wind energy, small scale wind turbines,

efficiency, cost-effectiveness, comparison.
1. Introduction
Wind turbines (WT) of capacity up to 100 kW are
classed to the small scale. Presently small scale wind
turbines are the fastest growing of any renewable energy
technology. Over 120 manufacturers produce over 300
small wind turbines of various designs and capacities in
all over the world [1-3]. Two varieties of wind turbines
exist: horizontal axis wind turbines (HAWT) and
vertical axis wind turbines (VAWT). Currently the
HAWT prevails number of their types makes up about
85 % from the whole number of the small WT types of
both varieties. Superiority of HAWT is even greater if
the cumulative installed capacity of both varieties is
compared.
Demand of small scale wind electric power systems is
growing rapidly because they are one of the most costeffective home-based renewable energy systems. They
can be applied for various purposes: for feeding of
electrical appliances, production of domestic hot water
o the irradiance level is 1 000 W/m (or one sun) ,

o the reference air mass (AM) 1,5 solar spectral
irradiance distribution,
o temperature of PV cell or module junction is 25C.
The air mass index AM 1.5 correspond the solar zenith
angle equal to 48.19.
Similar unambiguous standard testing conditions could
be established as well for the small WT designed for
operating at low, moderate and high wind speeds. The
independent testing laboratories should perform the
testing and certification of newly designed WT.
374
wind turbine develops at the wind speed 10 m/s per 1 m2

of the swept area S and can be determined as follows:
3. Methods used for comparison of WTs parameters

Comparison of both horizontal axis wind turbines and
vertical axis wind turbines properties can be performed
by means of the comparative indexes. Index of
comparative price of the WT, which evaluates its costeffectiveness, can be used as one of the main indexes. It
can be calculated by dividing the price of WT C
(without price of tower and VAT) by the WTs capacity
at the wind speed 10 m/s P10:
K ew
C
; /W
P10
K wm
K te
(1)
P10
; W/kg
m
K wm P10 2
; W2/m2
K ew S C
4. Comparison of HAWTs parameters

The presented above comparative indexes were
calculated for every WT of the group comprising of 45
small scale HAWTs having different design and
capacity from 108 W up to 45 kW. 5 best and 3 worst
results for every comparative index were selected for
the presentation below in Fig. 1 Fig. 4.
(2)
16,54
Kew, E/W
12,37
1,21
1,29
Proven 15
CB-3,1/200
1,34
13,27
1,47
1,45
ICO-GE 750 WindMax-H5 WindMax-H8 Aerocraft 120 Fortis Espada

2.2
Superwind
Fig. 1. The 5 lowest and the 3 highest prices of small HAWTs for 1 watt
Kwk, W/kg
45,00
40,00
35,00
30,00
25,00
20,00
15,00
10,00
5,00
0,00
42,80
40,6
38,3
36,8
35,4
8,2
Simplex 350 Ampair600- Whisper 200

48
(4)
Further on, calculations of the comparative indexes of

all selected small wind turbines according the formulas
(1) (4) were carried out on basis of data from internet
[1] and other sources [2, 3].
Index of comparative power K wm reflects the technical

efficiency of wind turbine. It shows what power the
18
16
14
12
10
8
6
4
2
0
(3)
We suggest the combined comparative index evaluating

the wind turbines technical efficiency and costeffectiveness. Calculation of this index is not
complicated as well:
Another comparative index can be used for evaluating

the relative power of wind turbine relational with one
kilogram of its weight. It will allow comparing very
different wind turbines as well. It can be calculated by
dividing the WTs capacity at the wind speed 10 m/s P10
by the turbines weight (without tower):
K wk
P10
; W/m2
S
Hannevind
30
Airdolphin
Aerocraft
120
7,4
5,0
BWC Excel Ampair 100

R
Fig.2. The 5 highest and the 3 lowest capacities of small HAWT per 1 kg of their weight (without tower)
375
350,00
Kwm, W/m2
308,64
300,00
256,64
255,56
250,00
244,09
240,33
200,00
150,00
100,00
87,50
85,71
72,43
Superwind
350
GW-11
Aimpair 300
50,00
0,00
Keith_2K
ARI-450
Aerocraft
1002
Airdolphin
Proven 15
Fig. 3. The 5 highest and the 3 lowest capacities of small HAWT per 1 m2 of their swept area
250
Kte, W 2/Em2
198,62
200
173,35
150,08
150
138,83
136,59
100
50
9,29
6,92
5,29
Ampair 300
Superwind
350
0
Proven 15
ICO-GE 750
ARI-450
WindMax-H5 WindMax-H8 Fortis Pasaat

3.1
Fig. 4. The 5 highest and the 3 lowest technical-econimical efficiencies of the 45 researched small HAWT
A power of all wind turbines at the same wind speed 10

m/s was used for calculations of the comparative
indexes instead of the rated values of power because in
contrary the comparison would be baseless. The rated
power values of various WT are determined at the very
different values of wind speed. This value of wind speed
(10 m/s) was chosen because the overwhelming part of
EU territory is covered by areas of low and moderate
wind speeds. E.g., winds of high speed in range 15-17
m/s in Kaunas region are blowing approximately only
few hours per year at the best case.
Analysis of the calculated comparative indexes for the
small HAWTs disclosed very considerable differences
between the best and the worst values. This proves the
necessity of serious research before the selection of the
proper wind turbine in order to avoid possible financial
losses. Only the most efficient and cost-effective wind
turbines can be successfully used for home-based
renewable energy systems.
The main parameters of the 5 best small HAWTs of 45

researched are given in Table 1. The rating was
performed on basis of the calculated comparative index
of technical-economical efficiency (Fig. 4). According
to the results of research, the best rating has small
HAWT Proven 15.
Table 1. Parameters of the 5 best small HAWTs of 45 rated by
the index of technical-economical efficiency
Name of
HAWT
Proven 15
ICO-GE 750
ARI-450
WindMax-H5
WindMax-H8
376
Name of parameter
Rated
Rated
Swept
Price,
wind
power,
area,
speed,
2
W
m
m/s
15 000
9,8
63,58
18 511
750
12
2,54
789
450
11,8
1,54
495
485
13
1,54
450
775
12,5
2,54
750
VAWTs of different types. The swept area of VAWTs

was considered as the axial section of the cylinder swept
by the wind rotor of turbine. 5 best and 3 worst results
for every comparative index were selected for the
presentation below in Fig. 5 Fig. 8.
5. Comparison of VAWTs parameters

Comparison of VAWTs was carried out in a similar way
as it was performed for the mentioned above group of
HAWTs: the comparative indexes were calculated for
every WT of the group comprising of 14 small scale
140
125,00
Kew, E/W
120
97,22
100
89,87
80
60
40
20
2,63
3,00
3,09
3,67
3,78
0
Vertikalrotor Windspire FSW Gyro-5 FSW Gyro FSW Gyro10 kW
Extreme
WS-4B
WS-2B
WS-0.3C
Fig. 5. The 5 lowest and the 3 highest prices of small VAWTs for 1 watt
Kwk, W/kg
18
16,67
16
14
12
10
8
7,60
6,62
4,23
3,82
3,20
2,86
0,93
2
0
WindGen Vertikalrotor
1000
0.3
Turby
FSW Gyro-5 FSW Gyro FSW Gyro 3 FSW Gyro-2

Extreme 5
kW
WS-0.3C
Fig. 6. The 5 highest and the 3 lowest capacities of small VAWT per 1 kg of their weight (without tower)
377
Kwm, W/m 2
450
384,62
400
350
293,71
300
250
181,52
200
169,81
171,43
121,13
150
72,00
72,00
100
50
W
S2B
W
S4B
1
FS
W
G
yr
W
in
ds
pi
re
Tu
rb
y
kW
o5
G
yr
FS
W
FS
W
G
yr
W
in
dG
oEx
tre
en
m
e
10
00
Fig.7. The 5 highest and the 3 lowest capacities of small VAWT per 1 m2 of their swept area
Kte, W2/Em2
90,00
80,00
70,00
60,00
50,00
40,00
30,00
20,00
10,00
0,00
79,60
43,95
W
S0.
3C
W
S4B
0,80
0,74
FS
W
yr
o
10
kW
en
10
00
W
in
dG
0.
3
er
tik
al
ro
to
r
FS
W
e5
yr
o5
1,07
W
S2B
49,37
yr
oEx
tre
m
G
FS
W
57,79
58,74
Fig. 8. The 5 highest and the 3 lowest technical-econimical efficiencies of the 14 researched small VAWT
Analysis of the calculated comparative indexes for the

small VAWTs disclosed vast differences between the
best and the worst values again as in case of HAWTs.
The extreme values of comparative indexes differ
approximately from 5 (see Fig. 7) to 100 times (see Fig.
8) what is a signal urging to perform a serious
examination of the wind turbines properties before the
doing decision to specify or/and purchase one or another
WT. The problem of the WTs selection is topical as
well regarding the choice between the HAWT and
VAWT.
The main parameters of the 5 best small VAWTs of 14
researched are given in Table 2. Number of the
examined VAWTs is smaller than HAWTs because the
number of produced and available in the internet small
WT of vertical axis is overwhelmingly smaller as well.
Table 2. Parameters of the 5 best small VAWTs of 14 rated by

the index of technical-economical efficiency
Name of
VAWT
FSW GyroExtreme 5
FSW Gyro-5
Verticalrotor 0.3
WindGen1000
FSW Gyro-10
Name of parameter
Rated
Rated
Swept
Price,
wind
power,
area,
speed,
W
m2
m/s
500
12,5
1,43
1 550
500
300
1 000
10 000
10,72
12,5
12,5
11,62
3,03
2,5
1,56
37,2
1 700
998
4 675
22 000
The power curves of WTs and their acoustic noise

should be evaluated as well for the more comprehensive
378
research [8]. The power curve gives important

information on the WTs operation in the range of high
wind speeds what can have more or less positive or
negative impact into the total energy output. The
acoustic noise levels may be very important when the
WT is installed in noise-sensitive surroundings.
7. Conclusions
1.
2.
6. Comparison of HAWTs against VAWTs
Kew, E/W; Kwk, W/kg
Comparison of the average indexes of 5 best HAWTs

against 5 best VAWTs is presented in Fig. 9 and Fig 10.
As it is shown in the columnar diagrams, the lowest
average prices for 1 watt and the highest average
capacities per 1 kg of the 5 best small HAWTs
50
40
30
HAWT-5b
VAWT-5b
4.
38,8
5.
20
10
3.
3,22
7,79
1,35
Kew
Kwk
8. References
Kwm, W/m2; Kte, W2/Em2
Fig. 9. Comparison of the lowest average prices for 1 watt and

the highest average capacities per 1 kg of the 5 best small
HAWTs against the 5 best VAWTs
300
250
200
150
100
50
0
261,05 240,22
1.
2.
HAWT-5b
3.
VAWT-5b
159,49
4.
57,89
Kwm
Demand of small scale wind electric power

systems in the world market is growing rapidly
because they are one of the most cost-effective
home-based renewable energy systems.
The comparative indexes were used for the
evaluation of small wind turbines comprising
two groups: 45 small wind turbines of
horizontal axes and 14 small wind turbines of
vertical axes.
Analysis of comparative indexes 45 small wind
turbines of horizontal axes showed very
considerable differences between the best and
the worst values.
Analysis of comparative indexes 14 small wind
turbines of vertical axes showed the same
trends as in case small wind turbines of
horizontal axes.
A distinct general superiority of small
horizontal axis wind turbines against the
vertical axis wind turbines is showed by means
of columnar diagrams.
Kte
5.
Fig. 10. Comparison of the highest average capacities per 1 m2

of swept area and the highest average technical-economical
efficiencies of the 5 best small HAWTs against the 5 best
VAWTs
A distinct general superiority of small horizontal axis

wind turbines against the vertical axis wind turbines can
be seen in Fig. 9 and Fig. 10. The average price for 1
watt of the 5 best small HAWTs is over 2 times lower in
comparison with the 5 best VAWTs. The highest
average capacity per 1 kg of the 5 best small HAWTs is
about 5 times higher if compared against 5 best
VAWTs. Only capacities of small VAWT per 1 m2 of
their swept area are almost equal for both varieties. This
index for the best of VAWTs (384.62 W/m2) is even
better than for the best of HAWTs (308.64 W/m2).
However, the highest average technical-economical
efficiency of the 5 best small HAWTs compared against
the 5 best VAWTs is about 3 times better.
6.
7.
8.
379
http://www.allsmallwindturbines.com (called
on 03.02.2009).
Small Wind Turbine Global Market Study
2008. Published by the American Wind Energy
Association. AWEA, June 2008. 26 p.
Christensen K. Catalogue of Small Windmills
from 5 W to 50 kW, 2006. Nordic Folkecenter
for Renewable Energy. 105 p.
Adomaviius V. Sizing of wind-based energy
system for homesteads power and heat.
Proceedings of International conference
Electrical and Control Technologies 2007.
KTU, Kaunas, 2007. pp. 157-162.
Adomaviius V. The concept of wind energy
consumption for heat and power in the
dwelling houses. Proceedings of the 12th
International conference Technical and
Technological Progress in Agriculture.
Raudondvaris: IAE at LUA, 20-21 September,
2007. pp. 226-230.
Martinez F., Herrero L.C., Santiago P. G. and
Gonzales J. M. Analysis of the efficiency
improvement in small wind turbines when
speed
is
controlled.
1-4244-07559/07/$20.002007 IEEE.
Murray D. Big Opportunities for the Small
Wind Industry. Go Green Energy, LLC. 2008.
5 p. (http://www.gogreenenergyllc.com
called on 02.16.2009).
Migliore P., van Dam J. and Huskey A.
Acoustic tests of small wind turbines. AIAA2004-1185. pp. 14.

PECULIARITIES OF WIND POWER PLANTS WORK IN THE SMALL POWER

SYSTEM
Tomas DEVEIKIS*, Enrikas NEVARDAUSKAS*, Renata STANIONIEN**
*Kaunas University of Technology, Lithuania
**The Centre of Renewable Energy Technologies at KTU, Lithuania
During fault period part of the network is disconnected

from the network. The disconnection is possible in the
part of the network as also in the line with one
generator.
Generator or group of generators should work and
maintain nominal voltage and frequency supplying
electric to connected network part during disconnection
and when system is disconnected. Sectional breaker is
disconnected remotely. Small system scheme in the
network is given in fig. 1.
Abstract: The small power system usually is formed

when large power system is destroyed after some
emergencies. Capabilities of small power system with
wind power plants to work as autonomous system are
investigated. The model of small isolated power system
and calculations are proposed according to the real data.
Wind power plant parks are analyzed for reactive power
and voltage control in the small power system.
Keywords: wind station, reactive power, voltage control,
small power system.
1. Introduction
Currently, the number of distributed generators
connected into the electricity system is increasing in
Europe. Small generators may work in a fault mode
generating electricity to small system when fault occurs
in the electricity network.
Small system concept is the system including one or
more generators and consumers which might be
disconnected from centralized system and to work
autonomously.
Local system simulation is performed based by actual
network data with existing distributed generators
working in the network system. Also, the possible
operation of wind power station in local system is given.
The potential of voltage controlling through reactive
power generation or consumption in the latest wind
power station is analyzed. Dynamic stability of local
system is analyzed when local system is disconnected
due to fault in the network system.
Fig. 1. Small system in network system
Lithuanias electricity system splitting to the A, B, C, D

regions with disconnection points is given. Each region
is connected to one or more disconnection points. The
regional disconnection points of electricity network are
given in fig. 2.
The main object in local system is power balance
between load and generation in real time. The balance
must be maintained by the lead generator controller.
This imply to limited penetration of distributed
generators into the network system concerning load
extent. Traditional network can not create a local system
without the connected distributed generators.
Small system mode requires estimation of reactive and
active power balance.
Significant factor to the small system is created power
loss (reactive power) in the network airlines.
Power line or transformer would disconnect of the
electricity network during disturbance in balanced
system.
2. Main principles of small system

Small system is described as one or more power
generators (PG) supplying electricity to certain
consumers group and managing to operate
independently certain period, while fault is being solved
in the network system.
380
Voltage reduction or increment is possible with shunt

reactors or capacitors;
Wind power plants with reactive power capacity
regulation;
Main factors are power generation and load balancing.
Other factors such as electricity network disconnection,
generator controlling and responsiveness, protection
determination is a technical solutions.
Summer minimum and winter maximum regimes have to
be evaluated during small system potential operation
analysis.
Small system operation is no possible, when the load is
greater than the maximum generated power. This is a
basic rule of a small system.
If a small system can not disconnect and work during
disconnection, generators should be able to start
independently and to generate full rated power from
zero regulating voltage and frequency.
3. Reactive power control of wind power plants in
electricity network
The generating wind power depends on the wind speed.
Wind power stations have certain characteristics which
are different to induction and synchronous generators.
There are wind power stations with double fed induction
generator (DFIG). Generator stator windings are
connected to the electricity network and the rotor
windings are connected through the electronic power
converters with current control circuits mainly used for
direct / alternating voltage conversion. In this case, the
electrical and mechanical rotor frequencies are separated
while electronic power converter compensates the
difference between the mechanical and electrical
frequency with additional current and frequency control.
The variable speed control has become possible by
DFIG. This means that the mechanical rotor speed can
be controlled according to certain function, resulting
maximum amount production of energy.
Wind power with DFIG can control reactive power
charging current to the rotor. In this case, there is no
relation between the ratio of reactive power and the
other factors as the rotor speed and active power
generation.
Instead of that, a certain rotor speed and the active
power consumption enable generation or consumption
of reactive power. The generator torque and the
generation of the reactive power directly depends from
the current, which is maintained form the electronic
power converter feeding form the rotor.
Voltage regulation by consuming and generating
reactive power with wind power stations consisting
DFIG enables to control reactive power in certain range.
Reactive power control limit is 0.2 Q/Pn at nominal
power and active power control limit is 0.5 Q/Pn at
above 0.5 nominal active power.
Wind power station V80 can generate or consume
1 Mvar reactive power at the maximum.
Generated reactive power may vary depending on DFIG
type. Vestas manufacturer PQ-type wind power station
characteristic is presented in fig. 3.
Fig. 2. Regional network disconnection points
During disconnection from large system generator or

generators of small system is set to control voltage and
frequency. Generators must provide lacking reactive and
active power in the small network systems. Active
energy power lack can be covered with additional
facilities as diesel generators, gas turbines, kinetic
energy storages, which are connected to small system or
certain regions. Reactive power lack can be covered
with use of wind power station (WPS) with induction
and synchronous generators connected to small system.
The high power imbalance forms immediately after
network disconnection. In addition, the generators
automatic voltage regulator is required to respond and
control reactive power demand of electricity system.
During this period of balancing the voltage and
frequency is unstable and may deviate from the limits.
Short-term transient process depends on the generator
control system regulator functioning.
The control system's effectiveness depends on the
scheme and its settings, which are variable among the
producers.
Synchronous generator and the control unit connected to
a small system have to control active power and the
power factor and to respond to the changing speed and
voltage resulting in disconnected system.
If the generator power range and the excitation are
sufficient, the small electrical system can achieve and
balance power, frequency and maintain a stable no less
than nominal voltage.
Small system active power balance basis:
Power increment or decrement is achieved by changing
turbine generators speed;
Turbine speed regulator sustains the required frequency,
which is directly related to the generators revolutions.
Small system reactive power balancing:
381
Fig. 3. Vestas PQ-type wind power station characteristic
Fig. 4. Enercon PQ wind power station characteristic with

synchronous generator and phase rotor
Wind power stations with wind vane directly connected

to the generator is expressed as wind power plants with
direct gear. Generator may have a synchronous
generator with phase rotor or permanent magnets. Stator
is not directly connected to the network, though through
the digital power converter. Wind power plant with an
electronic transmitter has the option to control generated
power, electrical energy and voltage.
The generator torque and reactive power generation
directly depends on the current, which is maintained
from electronic power converter feeding rotor.
Current necessary to create the necessary torque sets
converter possibility of reactive power generation or
consumption. This means that the generator power
factor and electrical load factor in network side can be
controlled by the converter independently of each other.
Wind power station with a voltage control unit can
control the point voltage by changing the amount of
reactive power generation or consumption. The control
action is performed as follows: the voltage is measured
at a certain point, and the signal is transmitted to the
voltage controller. The controller determines the amount
of reactive power, which might be consumed or
generated according to controller transfer function.
This is the easiest way to enable wind power station to
control only the terminal voltage on the stepping
generator terminals, however the voltage is not
controlled by the network points if reactive power is
uncontrolled. However, this may be not sufficient.
Though, wind power station which can control reactive
power capacity, can support point voltage close to local
consumers nominal voltage. When the measured voltage
is too low reactive power generation is increased, and
vice versa, when the voltage is too high, reactive power
generation is reduced.
Enercon E-82 and E70 maintain power factor cos = 1,
and also does not require and do not produce reactive
power when active power is changing in range from
0 kW to 2000 kW. It also can support power factor
uneven zero. This allows wind power stations to
promote reactive power control and to maintain voltage
in the network. Reactive power range depends on the
wind power station model and configuration. Enercon
E-70 PQ characteristic is presented in fig. 4.
Enercon E-70 reactive power control limit is between
0.5 Q / Pn at the nominal power. Wind power station E
S-70 specifications are S=2050 kVA, P=2000 kW, and
Q=1000 kvar. By choosing additional STATOM system
might be eliminated active power 0.2 (p.u.) load limit of
generation and reactive power consumption or
generation point starts at 0 in wind power park.
Enercon wind power stations are variable modifications

depending on generating power or reactive power
consumption. In particular case active power generation
depends on reactive power consumption. The voltage
must be measured in two or more points to balance
power transformer with tap is used.
4. Model of small electrical system with the wind
power stations
Small system consists of Maeiki power station with
two generators and three wind power parks, which are
connected to the 110 kV network (fig. 5).
Fig. 5. The possible operation of the small system in Zones 3

and 4
Maeiki power station consists of two 100 MW

generators. However, generators are limited to 80 MW
by turbine.
The wind power park of 30 MW total power consisting
of 15 Enercon E-70 type 2 MW rated power wind power
stations is connected to Zone 3 Palanga-ventoji 110 kV
transmission network. Substation power rating is 32
MVA. The wind power park 16.5 MW total power
consisting of 6 Vestas V-100 type 2.75 MW rated power
wind power stations is connected to Zone 4 ventoji-
382
idikai 110 kV transmission network. Substation power

rating is 32 MVA.
Zone 4 to the 110 kV transmission network segment of
the Holy-idikai, Benaiiuose connected to the village
of 6 wind power park Vestas V-100, a power of 2.75
MW each and 20/110 kV transformer substation. 14
MW of wind power ir generated additionally by Enercon
E-82 type wind power stations in Zone 4.
Total capacity of wind power parks is 60 MW.
"Mustang" program is used for simulation. Situation is
simulated in 110kV power line. It is also disconnected
the line from Klaipeda idik substations. Switchover is
performed in line sections Maeiki power stationVarduva and Juodekiai, idikai Klaipda substation.
In this line section is esential consumers such as
Maeiki nafta, Klaipda port, Bting oil terminal and
villages (idikai, Skuodas, Lenkimai, ventoji, Palanga,
Tauraukis).
The main generator is Maeiki power plant maintaining
frequency, voltage, active and reactive power balance.
Voltage must not exceed 10% of the nominal voltage.
Different regimes of minimum, average and maximum
in summer and winter mode of wind power generation is
performed.
Main consumer is Maeiki nafta consuming 55 MW of
active power and 3.5 Mvar of reactive power. Maeiki
power plant one generator instantaneous change is 7%
of 80MW power. Minimal power generation of one
generator is 2% of 80 MW power. Reactive power
generation limits is from 20 Mvar to 85 Mvar at 80 MW
power generation.
Generator voltage range is form 10 kV to 11.5 kV.
Voltage change while wind power parks generate
nominal power and generator voltage on generator
terminals is 10.5 kV is given in figure 6. Voltage change
is 10% range. Reactive power varies from 9 to 13 Mvar
in different modes. The wind power park is connected in
point 5 also maintains voltage in point 6.
During fault in network system small system is
disconnected. Small system voltage fluctuation is
simulated in Maeiki power plant and Varduva line
during short circuit fault nearby Maeiki power plant
substation. Generator power fluctuates during short
circuit but after 10-15 seconds stabilize.
40
P [MW]
30
20
10
0
0
10
12
14
Taime [s]
Generator power 25MW and WPP power according to

area av erage wine speed
Generator power 17MW and WPP power according to
Generator power 40MW and WPP max power
Generator power 25MW and WPP max power
Generator power 17MW and WPP power is 0
Fig. 7. Maeiki power plant power change in small system

after disconnection
After disconnection small system frequency change is

given in fig. 8.
70
60
P [MW]
50
40
30
20
10
0
Taime [s]
Generator power 40MW and
10
12
14
WPP power according to

WPP power max
Generator power 17MW and WPP power max

Fig. 8. small system frequency change after disconnection
Maeiki power plant one generator operation change at

different regimes is given in fig. 9. Second generator
power change is identical.
52
f [HZ]
50
48
46
0
Taime [s]
Fig. 6. Voltage change in small system at different wind

power park generation
10
12
14

WPP power max
Generator power 25MW and WPP power max

After disconnection Maeiki power plant power change

in small system is given in fig. 7.
Fig. 9. Generator power change at different regimes in small

system where short circuit location is Maeiki power plant
Varduva line
383
Voltage change during disconnection while system

generator generates 40 MW and wind power park power
generation is at maximum is given fig. 10.
System power fluctuates during disconnection but after

short period stabilizes. Active power and reactive power
change in 14 MW wind power park point during
disconnection from network system while Maeiki
power plant generators generates 40 MW power each is
given in fig. 13.
140
120
P [MW]
100
80
60
P [MW]
40
20
0
0
Taime [s]
Voltage in Maeikiai power plant point
Voltage in Klaipda port point
10
12
14
Taime [s]
Voltage in Bting oil terminal point
WPP activ e power and WPP power according to av erage

annual wine speed
WPP reactiv e power and WPP power according to av erage
annual wine speed
WPP activ e power and WPP power max
Fig. 10. Voltage change during disconnection while system

generates 40 MW and wind power park power generation is at
maximum
WPP reactiv e power and WPP power max

generation is according to area average wind speed
fig. 11.
Fig. 13. Active and reactive power change 14 MW wind

power park point during disconnection from network system
while Maeiki power plant generators generates 40 MW
power each
140
120
Wind power parks generated power is given in fig. 14

when measurements were performed every 2 minutes in
real time.
P [MW]
100
80
60
40
20
0
0
Taime [s]
Fig. 14. Wind power parks generated power in time

generates 40 MW and wind power park power generation is
according to area average wind
Peak power fluctuations is after 2182 minute which

reaches 6 MW value. This is most complicated
operation moment in 2500 minute cycle.
Wind power park power generation fluctuation up to 6
MW is presented in fig. 15.
Maeiki power plant generator manages to change
power generation up to 1.6 MW per minute, then both
generators may change power generation up to 6.4 MW
per 2 minutes. Maeiki power plant may balance wind
power park power generation fluctuation.

generation is according to area average wind speed
fig. 12.
140
120
P [MW]
100
80
60
40
20
0
0
Taime [s]
Fig. 15. Wind power park generated power during the period

generates 17 MW and wind power park power generation is
according to area average wind
Load forecast model must be applied for this region

after disconnection. Maeiki power plant generators
are loaded based on load forecast model.
384
5. The future concept of small power grid

At present, wind power plants connected to the electrical
network causes balancing problems.
The suggested model allows to use renewable sources
more effectively and to supply heat to certain districts as
well as electrical power by using the electrical and
thermal energy supplied by fuel cells.
The power grid with wind power plants, fuel cells and
energy storage system (shown in fig. 5) could operate in
island conditions. Also, such system could operate
connected to the external power grid and be isolated in
case of accident until the external power system is
restored.
The presented system covers fuel cells, energy storage,
electrolyzers, hydrogen storage, wind power plants and
consumers. Conventional generating units can be
connected as well.
The effectiveness of fuel cells is better comparing to the
conventional power sources and can reach 40-60% for
power generation and up to 90% for power and heat
generation.
The hydrogen could be stored in stationary tank and
used instead of diesel fuel. The hydrogen is produced
from water by electrolysis. The electrical energy would
be supplied by wind power plants in real time or by
electrical storage system. Also, regenerative fuel cells
can generate hydrogen as well. This system operates in
closed cycle to produce hydrogen from water.
The following types of fuel cells are suitable to use in
the power system: SOFC, PAFC and MCFC. The
electrical capacity of such fuel cells could be 50-3000
kW. For quick startup and load following, PEMFC and
DMFC could be used. The startup time of PEMFC
varies from 1 s to couple seconds. The loading could
vary from 10 to 100%.
The energy storage system vanadium redox battery
(VRB) is suggested to be installed in the power grid (fig.
16). The accumulated energy can reach few hundreds
MWh. The large VRB energy storage system can
operate jointly with wind power parks and to supply
power during peak loads. For smoothing the load
diagram, capacity of VRB system must be high enough
(few MW). At present, large VRB systems are used for
industrial objects and energy supply systems and their
capacity is of range 25 kW 10 MW.
Stationary power plants based on fuel cells like
DFC300, DFC1500 and DFC3000 could be connected
in the power grid. To increase the power, these power
plants can be connected to work in parallel. For
hydrogen production, the hydrogen plant like
DFC300MA could be used.
At present, the price of renewable energy generation is
very high (the relative price is 2500$/kW for a fuel cell
and 2000 $/kW for wind power plant). The capital price
of fuel cells could drop significantly if such systems
were installed in the decentralized power grid to work at
basis load. According to World Alliance for
Decentralized Energy reports, decentralized power
plants would significantly decrease investments required
Fig. 16. The concept of power grid with wind power plants,
fuel cells and energy storage system
for power grid development. The distributed generation

would give not only economical benefit but also ensure
the real concurrence between different energy suppliers.
At market conditions, the energy sector would be
developed more intensively and efficiently because the
priority is given to the most effective solutions in the
market.
6. Conclusions
The power system should be flexible and be able to
separate into islands after large disturbances and
emergency conditions.
Small power system or part of large one can work in
island conditions without power supply interruptions.
Operation conditions of a part of the power system
which is intended to be disconnected during emergency
conditions must be analyzed. For disconnection without
large disturbances and without significant frequency
variation, the generating units should be balanced
according to the total load in this area.
In order to avoid system collapse in islanded network
after disconnection of pilot generating unit, the back-up
generator should be installed for system restoration.
The analysis of dynamic stability shows that the small
power system can be successfully disconnected from the
main power grid and sustain the stable operation.
7. References
1.
2.
2.
3.
4.
385
Deveikis, Tomas,
Nevardauskas,
Enrikas
Stanioniene
Renata.
Investigation
of
electromagnetic compatibility of wind power station
and electrical network user // EMD 2008 :
proceedings of the XVIII International Conference
on Electromagnetic Disturbances, September 25-26
2008, Vilnius, Lithuania. - ISSN 1822-3249. Vilnius. - 2008, p. 155-160
Assessment of Islanded Operation of Distribution
Networks and Measures for Protection. ETSU
Report ETSU K/EL/00235/REP. 2001.
www.enercon.de
www.vestas.com

Proceedings ECT 2009

Uploaded by

Copyright:

Available Formats

Proceedings ECT 2009

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Proceedings ECT 2009

Uploaded by

Copyright:

Available Formats

ISSN 1822-5934

ELECTRICAL AND CONTROL

ELECTRICAL AND CONTROL

KAUNAS, TECHNOLOGIJA, 2009

UDK 621.3(474.5)(06); 681.5(474.5)(06)

The 4th International Conference on Electrical and Control Technologies

The 4th International Conference on Electrical and Control Technologies ECT-2009 is

For information write to:

Kaunas University of Technology, 2009

University of Applied Sciences Schmalkalden, Faculty of Electrical Engineering, Germany

University of Applied Sciences Schmalkalden, Faculty of Electrical Engineering, Germany

Investigation of the influence of fuzzy set structure to the color segmentation

Department of Control Technologies, Kaunas University of Technology, Lithuania,

Development and investigation of an efficient algorithm for alternative cursor control

Department of Control Technologies, Kaunas University of Technology, Lithuania

Optimal control of robots by dynamic programming

University of Applied Sciences Schmalkalden, Faculty of Electrical Engineering, Germany

Map drawing of unknown environment using autonomous mobile robot

Department of Mathematics Research in System, Kaunas University of Technology, Lithuania

Influence of dimensions of a mobile robot on navigation using vector marks

Kaunas University of Technology, Lithuania

Optimization of hexapod robot locomotion

Faster autonomous mobile robot learning using decision correction

Kaunas University of Technology, Lithuania

Voice controlled telephony services

Kaunas University of Technology; Vilnius University, Kaunas Faculty, Lithuania

Data acquisition in OPC-based industrial it systems

Synthesis of program modules for typical boiler-house applications

Kaunas University of Technology, Lithuania

Distributed substation units control system in district heating network

Kaunas University of Technology; AB Axis Industries

Estimation of typical safety circuits for medium power industrial boilers

ABB UAB, Lithuania; Kaunas University of Technology, Lithuania

Positioning of stepper motors in graphical data acquisition system

Visibility of text in multimedia presentation

Kaunas University of Technology, Department of Electrical Engineering, Lithuania

Automatic PLC-source-code generation of C4.5 decision trees

University of Applied Sciences Schmalkalden, Germany; Kaunas University of Technology, Lithuania

University of Applied Sciences Schmalkalden, Faculty of Electrical Engineering, Germany

Correction of measuring track processing errors in vibration measurements of coal wagon

Regression modelling using adaptive basis function construction

Institute of Applied Computer Systems, Riga Technical University, Latvia

Categorizing sequences of laryngeal data for decision support

System parameters identification in the state space

Model of current space vector control

Parameter estimation of induction motor at locked rotor

Vilnius Gediminas Technical University, Lithuania

Modelling of ammonia production process

Process Control Department, Kaunas University of Technology, Lithuania

Kaunas University of Technology (KTU), Lithuania

Analysis of efficiency of a pump set operating in an irrigation system

Technical University of Radom, Faculty of Transport and Electrical Engineering, Poland

Application of simulation tools for a fertilizer granulation circuit

Process Control Department, Kaunas University of Technology, Lithuania

Innovative moisture control method for concrete mixing systems

Department of Electronics and Measurement Systems, Kaunas University of Technology, Lithuania

The independent fuzzy control with correction of two dimensional process

Department of Systems Analysis, Vytautas Magnus University, Lithuania

Investigation of adaptive pH control system

Process Control Department, Kaunas University of Technology, Lithuania