THE AMERICAN SOCIETY OF MECHANICAL ENGINEERS

345 E. 47 St., New York, N.Y.10017

The Society shall not be responsible for statements or opinions advanced in papers or in
discussion at meetings of the Society or of its Divisions or Sections, or printed in its
publications. Discussion is printed only if the paper is published in an ASME Journal.
Released for general publication upon presentation. Full credit should be given to ASME,
the Technical Division, and the author(s). Papers are available from ASME for nine months
after the meeting
Printed in USA .
Copyright © 1985 by ASME

A Feasibility Study of the Application of Artificial

Intelligence Techniques for Turbomachinery Diagnostics
CYRUS B. MEHER-HOMJI
Boyce Engineering International, Inc.
Houston, Texas

ABSTRACT
This paper presents a study into the application of Recently, users of turbomachinery have become interested
Artificial Intelligence (AI) techniques (specifically Expert in the use of computer based systems for health mon-
Systems) to the problem of turbomachine diagnostics. The itoring. These Systems obtain both mechanical (vibration)
diagnostic process is described and a model for computer data as well as aerothermal (temperatures, pressures and
implementation provided. Some diagnostic examples are flows) in order to firm a prognosis and diagnosis of machine
provided to explain the methodology. Expert system health. [1][2][3]. While these systems provide a high level
development is underway at the author's corporation for of diagnostics for specific operating problems the system
incorporation in online monitoring and diagnostic systems. cannot be classified as an "expert" system. This paper
examines the concepts and feasibility as to how existing
INTRODUCTION AND NEED advanced monitoring and diagnostic systems can be inte-
grated within an expert system framework. The expert
The increasing sophistication of modern high speed turbo- system would then act as an aid to the troubleshooting
machinery with higher horsepowers and speeds has been engineers. While it is thought by some that expert systems
accompanied with a wide range of operating problems. are designed to "out think experts" the author envisions the
With critical machines in the petrochemical industry, down expert system to act as an aid to the troubleshooting
time costs (which can exceed $1 million/day) are pro- engineer. Moreover, expert systems for turbomachinery
hibitively high creating need for rapid means for diagnosis diagnostics would be a valuable training aid for young
and troubleshooting. Moreover, with increasing fuel costs engineers enabling them to "think through" complex op-
the need to operate units efficiently is paramount. erating problems via a keyboard.
Traditionally, troubleshooting and diagnostics of complex An "Expert System" is an intelligent computer code that
turbomachinery problems calls for an individual with both uses knowledge inference procedures to solve problems that
analytical skills and a wealth of experience so that he can are difficult enough to require significant human expertise
draw and integrate the complex data available to him, form for their solution. The knowledge base of an expert system
preliminary hypotheses, suggest tests and ultimately diag- consists of facts and heuristics. The facts constitute a
nose the operating problem. It is a practical reality that body of information that is widely shared and generally
there is always movement of these experts and often agreed upon by experts in a field. The heuristics are rules
vacuums are created when an experienced machinery of judgement and rules of good guessing that characterize
engineer leaves or is reassigned. In cases like this, young expert-level decision making in the field. An overview of
engineers have to build up experience, and develop heur- expert systems is provided by Hayes-Roth Et al [4] and
istics relating to diagnostics the hard way. Several large Davis [5]. Expert systems utilize IF-THEN rules (also
petrochemical organizations have developed some excellent known as "production" rules to codify knowledge.
maintenance manuals incorporating years of experience in
an attempt to avoid a "knowledge vacuum". Once again BASIC LOGIC FOR DIAGNOSTICS AND TROUBLESHOOTING
practical reality shows us that these manuals are either not OF TURBOMACHINES
consulted or specific information can not be located within
a short time frame. Current machinery health monitoring and diagnostic sys-
tems employ Boolean logic operations for diagnostic pur-
This paper examines the feasibility of a computer based poses. Considerable work has been also accomplished in
trouble-shooting/diagnostics procedure utilizing techniques the area of Vibration Diagnostics and in the integration of
and methodology developed in the Artificial Intelligence vibration and aerothermal diagnostics for gas turbines and
(AI) field. centrifugal compressors [1][2].

Presented at the Gas Turbine Conference and Exhibit

Houston, Texas — March 18-21, 1985

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 12/17/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 When confronted with an unusually difficult problem in a TURBOMACHINERY TROUBLESHOOTING
specific area involving diagnostics of a turbomachine
problem, the field engineer will turn to a specialist There have been several excellent papers published in the
(possibly an external consultant) who has, within his area of turbomachinery diagnostics [7], [8] and [9]. Most
experience, a subset of the general mechanical and aero- publications tend to focus on specific areas, i.e., rotor
thermal data base that the field engineer did not have. dynamic problems, performance problems or specific pro-
Because computers can store massive amounts of infor- blems that plague a specific class of turbomachine. Very
mation, and as they can be programmed to manipulate data little work has been published on the thinking process that
according to logical processes, there is a possibility of their several experts use when diagnosing a problem. Sohre [10]
being used in the area of machine diagnostics. A basic brings out an excellent point that engineers attempting to
representation of the diagnostic process is shown in troubleshoot machines often plunge into sophisticated an-
Figure 1. alysis and modeling techniques and may lose a 'feel' for
machinery neglecting the important qualitative data of
sight, smell and touch which are more than often most
valuable in a troubleshooting system. This often results in
thousands of dollars being spent on analysis while the fix
simply involves an "obvious" solution such as tightning a
loose machine support or bearing cap.

History &
Walk around Inspection
. Vibration Analysis
Low Level Discriptors
(Signs & Symptoms)
Two fundamental problems in setting up expert systems for
Tests . Log-Sheet Review troubleshooting diagnosis turbomachinery are:
. Lube Oil Tests . Efficiency Drop
. Maintenance Review . Vibration
. Data Analysis, etc. . Noise 1. The wide variety of machine types of varying aero-
. Operating Behavior
dynamic characteristics. Even machines of a similar
-Y type (say centrifugal machines) are so widely dif-
Sensors &
Probabilistic Description Observations "-/ ferent e.g., a gas compressor vs. a centrifugal pump).
of Initial State St
(Generate set of Machine
hypothesis as to Problem 2. Continuous changes in design concepts. Different
what problem is) designs of impellers, bearings and auxiliary systems
Heuristic Search to
apply best Ti
create several idiosyncracies that are more than
often not predictable by the use of analytical tech-
— Run Under Increase..
Surveillance
niques.
Repair
Change Lube Oil Temp. In addition to these two fundamental problems several
[Ti Realign minor problems exist:
Balance
(a) Complex interrelationships between mechanical
and aerothermal problems
_ Etc.

(b) Existance of multiple faults

Transformation Set

(c) Machine-to-machine variation even with "ident-

ical" models
FIG. 1 Representation of a Machinery Diagnostic Problem At this point-in-time, no expert system can be feasibly
developed to encompass the whole area of turbomachinery.
It is, however, possible to break up machines into clas-
sifications that have similar aerodynamic and physical
attributes. (i.e., machine type) or by specific broad
problem area. This is shown in Figure 2.
Diagnostic problem solving based on Baye's rule made For the near future, any expert system developed for
simplistic assumptions that the problems were mutually troubleshooting (after the event) would have to interface
exclusive and problem manifestations were independent. through an engineer even if online monitoring existed.
Modern AI type programs do not have these limitations and Some basic diagnostics can however be provided on an
are consequently of some value. The reasoning process online basis. The expert system software would have to be
used by an expert is very complex. Notions of likelihood interactive-initially querying the machinery engineer as to
play an important part as well as some kind of logical machinery description, generic type, bearing and seal type,
thinking as well as a use of his data base. Errors in coupling information, operation conditions, presence of
diagnostics of machinery result more frequently from sounds or visible signs, and operating history, etc. Inputs
errors of omission than from other sources. This has been would be both vibration analysis information and aero-
amply demonstrated in a wide variety of published case thermal information. The software could then provide
studies. The most successful troubleshooters are those who likely problem areas, suggested tests, or even request more
remember and consider the most possibilities. information for further analysis. The system could draw on
a large data base of case histories. It is possible and
It is the author's opinion that an expert system can form feasible to have an online monitoring system linked to an
a valuable tool to help a less experienced engineer diagnose expert system on a real-time basis. In this case, an expert
or troubleshoot a problem. This view is also held by Tuft system could aid engineers in improving availability and in
[6]. The Expert System is not (at least for the foreseeable pinpointing areas where distress was indicated.
future) going to replace experts but just forms a tool for
improving the quality of decisions that are made related to
Turbomachinery diagnostics. Expert systems can also be
incorporated within online monitoring systems.
2

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 12/17/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Classification Scheme Alternatives For
Expert System Diagnostics
2. M (P, M) This is a reverse association between
Problem P and Manifestation M.
3. FORM-OF (P1, P1.1) which implies that problem of
Category P1.1 is a form of Problem Category P1.
By Machine/Type

EXHAUST
• Coupling Problems

• Rotor Bearing problems GAS GENERATOR

TURBIN
Generic Type • Lube Seal Systems
• Controls/Auxiliaries
. Centrifugal Compressors
• Gear Box Problems
. Steam Turbines-Utility M.D.
• General Vibration
. Gas Turbines - Single-Shaft Aerodynamic Problems
Split-Shaft
Blading Problems
. Pumps
. Electric Motors
. Gear Boxes
S//P

LOA
By Model Type

FUEL
COMBUSTOR
POWER TURBINE
OD\ PRESSOR

I/AR/gaze Apr

FIG. 3 Split-Shaft Gas Turbine Showing Station Designations

DIAGNOSTIC LOGIC MODELS APPLIED TO TURBOMACHINES

There are several important points that emerge from a Figure 4 shows how abductive logic can be applied in a
study of general troubleshooting methodology. Firstly, troubleshooting/diagnostic framework. The rough proce-
problem manifestations are statistically dependent and not dure would be as follows:
mutually exclusive; thus, diagnostic methods based on
Bayes rules are not always valid. For example, in the case 1. Obtain information (via keyboard). This would be an
of gears, misalignment may be accompanied by both tooth interactive query type session. In certain cases,
fracture and surface distress. Secondly, experienced online data acquisition can be used.
troubleshooters think in a "diffuse" manner, i.e., their
reasoning is not necessarily "step-by-step". Diagnosis is 2. For each manifestation, the EVOKE operator would
made in general terms and then progressive refinements evoke a problem Set Pi.
are made until underlying causes have been identified.
Thirdly, problems in machinery may be classified in a 3. For each Problem Pi, lists A,B,C&D would be cre-
hierarchal structure in which certain class of problems may ated. These lists essentially indicate how close the
be a subset of a larger problem category. Moreover, often manifestations "match" the problem under consid-
times, problems of a certain class will give rise to similar eration. If a particular problem has been previously
or even the same manifestation. For example, oil whirl is evoked then the lists A,B,C&D would be updated.
a subclass of subsynchronous instability problems.
4. Hypotheses (regarding problems) are then screened
Abductive logic (method of hypothesis) is a means by which and a top few selected. Dominance of two Hypo-
a set of observations (manifestations of a problem) is used theses H1 and H2 is established by checking if the list
to evoke a set of hypotheses. As certain manifestations "A" for H1, less items explained by previous diag-
could evoke several hypothesis, a weighting procedure noses, is a subset of a similar construct of H2. If it
should be used. For example, in the Gas Turbine shown in is a subset, then H1 is dominant.
Figure 3, a drop in output power could be caused by
compressor fouling, internal seal leakages, external leak- 5. Once hypotheses have been narrowed down, by spec-
ages, drop in turbine efficiency, drop in combustor effi- ific tests or questions, an action recommendation can
ciency, etc. The most common (likely) hypotheses would be be developed by the engineer using the expert system.
compressor fouling, seal leakage problems or turbine
damage. For example, if we consider the case of a split shaft
turbine under constant power, the symptoms [12] shown in
Each of these hypotheses would then generate a set of Figure 5(a) will lead to the conclusion of a fouled
consequences. Then hypotheses would have to be grouped compressor. The case under constant speed is more
together into mutually exclusive subsets corresponding to difficult as the manifestations are qualitatively the same.
specific problem areas. Based on a structure proposed by The manifestations evoke hypotheses of "fouled comp-
Pople [11], a theoretical framework is now developed for ressor" and "increase in the area of the low pressure
diagnostics of a type of turbomachine. Three relations may turbine". In this event the expert system would request a
be defined: quantitative description which would allow discrimination
between compressor fouling and a drop in low pressure
1. E (M, P) This relates to the Evocative association turbine nozzle area. The case of constant speed operation
that Manifestation M has for a Problem P. is shown in Figure 5(b).
3

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 12/17/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 I Accept Manifestations (Mi) and Symptoms Via keyboard

QUERY RE: ATTENDANT CIRCUMSTANCES:

Process changes, maintenance, machine history; maintenance; history of bearing/oil temperatures,
vibration trends, performance trends, qualitative observations — noise , smoke past failures,
tube oil analysis data, etc.

For Each Mi use E Ope ator to Evoke Problems

Pi (Hypotheses)

or EgVir','eYnt'res

C. List of Mi expected D. List of Mi expected

A. List of Mi observed B. List of Mi observed
for Pi but not but not investigated
but not explained with evoking strengths
observed yet
by Pi that match Pi

Yes If Pi previously evoked,

Update
update Lists A,B,C,D.
Lists

Assign weights for each hypothesis

Positive Weight Negative Weight Bonus Positive Weight

If large number of manifestations If data expected is found absent If problem is casually linked to
are explained or data not explained to problems already confirmed

Generate set of evoked hypotheses Hi based

on weights and establish dominance

IF

[Items explained 1 Items explain

by previously e. Place 111 at
LIST AHI
confirmed diagnosis LIST AH2 - by previously top of list
confirmed dingy
—( HI 1-12

Based on the top ranked hypotheses

use specific questioning to narrow choice Indicate Diagnosis

Develop Action Recommendations

Change process
Run Machine Under Repair at Scheduled Routine Shutdown ..... conditions—flow, speed,
Close watch time temperatures, etc.

FM. 4 Diagnostic Logic Model

Figure 6 shows a conceptual hierarchal tree for a M.D. 2. Load decreased 16.5% (M2)
steam turbine showing how a manifestation of "insufficient
power" can be used to evoke a variety of problems. It can 3. First Stage Pressure increased 21.2% (M3)
also be seen that low efficiency is a problem caused by
another problem, fouling, gland seal leakage, etc. 4. HP Section efficiency dropped 12.2% (M4)
For example, based on a case reported by Cotton &
Schofield [13] is now presented. A turbo-generator started
losing output at a rate of 1.2 MW/day after operation for
21 months. All changes noted below were gradual:
1. Throttle flow rate dropped 17.2% (Manifestation M1)

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 12/17/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 FIG. 5(a) Split-Shaft Gas Turbine Diagnostics at
Constant Power (Symbols refer to FIG. 5(b) Diagnostics under Constant Speed Condition
Fig. 3).

P1.2 Gauges not reading right

Instrumentatioi P2.1 Low steam pressure

Problems

P2.2 Unit Overloaded

P2.3 Low Steam Temperature

Process
Problems
P2.4 Low Vacuum Noise
P2

Insufficient P2.5 High Back Pressure

Power

Governer P3.1 Governer Linkage Problems

Problems
P3
P3.2 Low Governer Oil Pressure

P3.3 Valve Sticking

Aerodynamic
Problems
P4.1 Damaged Blades St Vanes
P4

FIG. 6 Steam Turbine Diagnostics

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 12/17/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Figure 7 shows the struture of the diagnostic problem Thus an attempt is made to determine that the probability
showing the Evocative (E) and Manifest (M) routes. It can P (that Problem Pi exists for evidence set E) is greater
be seen that the diagnostic of Fouling is the only one which that some number x (a level of confidence say .95). An
manifests itself in Ml, M2, M3 and M4. Also, note that alternative approach includes an important development by
Efficiency Drop (M4) is also a Problem P5 evoked by the Zedeh [14], [15] of Fuzzy sets. Fuzzy set theory was
manifestation of Load Drop (M2). developed in 1965 in order to help formalize the human
thought process which is too imprecise and vague by
classical mathematical standards. This is exactly the case
in the diagnostic process - relationships between symptoms
and problems are qualitative because of the absence of
sharp boundaries characterizing attributes of a particular
problem. Moreover, one of the most important things that
an expert diagnostician does is to summarize information
into Fuzzy sets which bear an approximate relation to the
primary data. Also, expert diagnosticians do not step
"crisply" through "logic trees" to come to conclusions -
vagueness and inexactitude actually help in hypothesis
generation. Some typical simplified Fuzzy Production rules
Ml are:
Throttle Flow
Drop
IF (Gear Loading is not small) THEN (Gear Mesh
Amplitude is large)
IF (N2 increases) THEN (Vibration increases a lot)
IF (X is small) THEN (Y is medium) ELSE (Y is small)
M2
Load Drop IF (FLOW is near SURGE) THEN (BPF Amplitude
Increases)
APPLICATION TO ONLINE MONITORING SYSTEMS
The incorporation of expert systems on an online mon-
itoring and diagnostic system is currently under develop-
ment. The system under development will monitor and
analyze online data from four large industrial, two-shaft
gas turbines which drive centrifugal compressors. Data
NOT FOUND, collected will include aerothermal (pressure, temperature,
REJECT I flows, speeds) data and mechanical data, including vibration
(proximity probe and accelerometer), and bearing temp-
eratures. Figure 8 shows the general configuration that
will be provided. The diagnostics will involve the use of
expert system production rules to provide diagnostic and
Area udden Change
troubleshooting aid. Some diagnostics for gas turbines
Reduction Performance include:
in 2nd or
subsequent
stages NOT FOUND Compressor Section. Compressor analysis is done by
REJECT 7 monitoring the inlet and exit pressures and temperatures,
the ambient pressure, vibration and the pressure and
temperature of the lubrication system. Monitoring these
parameters allows the detection of the following problems.
Production rules for the first two problems are provided
FIG. 7 Diagnostics for Steam Turbine-Fouling Problems with the rest being treated descriptively.
Clogged Air Filter. A typical production rule would
read: IF (Air Flow Drop) AND (Pressure Drop Increase
Over Exceedence) THEN ASSERT "Clogged Air Filter".
Compressor Fouling. A typical production rule would
read: IF (Pr Ratio Decrease) AND (Exit Temp. Increase)
FUZZY SET CONSTRUCTS FOR MACHINERY DIAGNOS- and POSSIBLY (1xRPM Vibration Increase) AND (Comp.
TICS Eff. Drop) THEN ASSERT "Compressor Fouling". A
relationship developed by ZabaF1T exists linking quan-
A fundamental problem in diagnostic theory relates to titatively the change in compressor efficiency and vol-
determine what set of "evidence" (Ed can be represented umetric flow rate:
by Pi, i.e. to find an evidence set E such that
• Anc
AV
P (Pi/E) < x (1) ze (2)
V nc
•
(Where V =Compressor Volumetric Flow Rate, n c
Compressor Polytropic Efficiency and Z c = Compressor
Flow Change Factor).
6

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 12/17/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Bearing Failure. Symptoms of bearing trouble include
a loss of pressure, an increase in the temperature dif-
ference across the bearings, and an increase in vibration.
If oil whirl is present, there will be a vibration frequency
4 GAS TURBINE DRIVEN COMPRESSOR TRAINS corresponding to one half the running speed. Online
DYNAMIC

ANALOG
mechanical analysis permits detection of this.
DicrrAL.
INPUT INPUT INPUT
Additional diagnostics are available for compressor blade
MP MODEL DMITA
PEWMC DCA
3487/3401/34.17
ME= DCA
OD.WETZ SYSTD, 22
SECKIEVX OP NODS
loss, icing, rotating stall, intercooler problems, and fluid
xansrcbt oomsmost MEd
RE20100.1 slugging in load compressors.
Combustor Section. In the combustor, the only two
parameters which can be measured are fuel pressure and
W WOO. SOCA
C COCO(
evenness of combustion. Due to limited probe life, turbine
SPE222JD AWL1202
inlet temperatures are not measured, but may be com-
puted.

> HP MODEL 2199C

E-- Plugged Nozzle. This is indicated by an increase in
CPU
•002.79122
fuel pressure in conjection with increased combustion
D62. DANE
unevenness. This is a common problem when heavy or
1X02 717X
WJC7C PPE UME2
residual fuels are used.
Crossover Tube Failure. This can be detected by
noting the temperatures at the exit and correlating them
with a burner upstream.
OP 2627A W 262■2
OPE/WOWS
CAD
MOLY
ccetsca ODDS= Cracked or Detached Liner. -This is indicated by an
increase in vibration readings and a large spread in exhaust
temperature.
Liquid Entrainment in Fuel. This would cause rapid
changes in Turbine Inlet Temperature and Inter Turbine
Temperature. High Speed Scanning would be needed to
detect this.
Turbine Analysis. To analyze a turbine, it is
FIG. 8 Configuration of Online Monitoring and necessary to measure pressures and temperatures across
Diagnostic System the turbine, shaft vibration, and the temperature and
pressure of the lubrication system. Analysis of these
parameters will aid in the prediction of the following:
Z e is dependant on the nature of the deposits and their Nozzle Coking (Power Turbine). Symptoms
distribution on the blade surface. These computations may would include an increase in Turbine Inlet Temperature, an
be done online. increase in Shaft HP and an increase in Pressure Ratio. A
constant N1 speed is assumed.
Compressor Blade Damage/Flutter. High speed vi-
bration data acquisition techniques can allow transient Nozzle Coking (Gas Generator Turbine). Symp-
excessive vibration to be noted. These may be picked up toms would be a drop in Turbine Inlet Temperature, a drop
by judiciously located accelerometers. Dynamic high in Shaft Horsepower, a drop in Fuel Flow and an increase
response pressure transducers can also be utilized. in Compressor Pressure Ratio. A constant N1 speed is
assumed.
Seal Leakage Problems. This would be indicated by
an increase in inter turbine temperature, an increase in Turbine Fouling. This is indicated by an
fuel flow and an increase in engine pressure ratio. increase in turbine exhaust temperature possibly accom-
panied by a change in vibration amplitude. There will also
Stall and Rotating Stall. This would be reflected by be a drop in efficiency and flow coefficient. Evaluation of
sudden and spasmodic changes in compressor efficiency.
Pressure measurements at the Bellmouth location would be A Kt / A t
required. Zt - Kt (3)
A nt

Balance Piston Distress. In the case of load com-

pressors, balance piston problems can be detected by enables location of where the fouling occurs. Kt is the
correlating changes in Balance Chamber Pressure and turbine flow coefficient, r1 t is the Turbine Polytropic
Thrust Readings (ideally, thrust bearing metal temper- Efficiency and Zt is the Turbine Flow Coefficient Charge
ature). Factor).

Foreign Object Damage. Symptoms would be an Damaged Turbine Blade. This results in a
increase in Turbine Inlet Temperature, a drop in Pressure vibration increase accompanied by an increase in the
Ratio and a reduction in Shaft Horsepower. exhaust temperature.

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 12/17/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Nozzle Bowed. -The exhaust temperature will [5] Davis, R., "Expert Systems: Where are We and Where
increase and there may be an increase in vibration noted at Do We Go From Here?". AI Magazine, Vol. 3, No. 2,
the excitation frequency of the first stage. pp. 3-22, 1982.

Nozzle Erosion (Power Turbine). -This would be [6] Tuft, R. H. "No! Our Condition Monitoring Techniques
indicated by a drop in Turbine Inlet Temperature, a drop in are Not Adequate". Editorial in Sound and Vibration,
Shaft horsepower and a drop in Pressure Ratio. Constant P. 5, April, 1984.
N1 speed assumed.
[7] Sohre, J. S. "Operating Problems with High Speed
Nozzle Erosion (Gas Generator Turbine). Symp- Turbomachinery, Causes and Corrections". ASME
toms would be an increase in Turbine Inlet Temperature, Petroleum Mechanical Engineering Conference, Dal-
increase in Shaft Horsepower, increase in Fuel Flow and a las, Texas, November, 1968.
drop in Pressure Ratio. Constant N1 speed assumed.
[8] Sohre, J. S. "Reliability Evaluation of Troubleshooting
CONCLUSIONS of High-Speed Turbomachinery", ASME Petroleum
Mechanical Engineering Conference, Denver, 1970.
Based on this study, expert systems in the field of
turbomachinery are being developed. These systems will be [9] Broekmate, A., "Diagnosis of Machinery Vibration"
for specific machine types and will be used to assist DSM Research and Patent Dept., Equipment Testing
troubleshooters in decision making. Time estimates for Center, Geelen, Holland.
developing a reasonable expert system for a generic class
of turbomachine is estimated to be about 2 to 3 man years. [10] Sohre, J. S. "Common Sense in Turbomachinery
Troubleshooting". Informal Keynote Address pre-
Expert systems developed, will be useful tools in helping sented at the Vibration Institute, Chicago, Illinois,
engineers troubleshoot a wide variety of problems. AI March 30, 1982.
techniques, coupled with development in areas such as
fuzzy set-theory, have the capability of providing very [11] Pople, H. E., Jr., Et. Al. "DIALOG: A Model of
powerful aids to diagnostic engineers. Diagnostic Logic for Internal Medicine". Proceedings
of the Fourth International Joint Conference on
ACKNOWLEDGEMENTS Artificial Intelligence, September, 1975. pp 848-855.

The author wishes to thank J. E. Broussard and H. T. [12] Saravanamutto, H.I.H. and McIsaac, B.D., "Thermo-
Sethna for their assistance in preparing this paper. dynamic Models for Pipeline Gas Turbine diagnostics".
ASME Paper No. 83-GT-235.
REFERENCES
[13] Cotton, K. C. and Scofield, 0. "Analysis of Changes
[1] Boyce, M. P. and Meher-Homji, C. B. "Aerothermal in the Performance Characteristics of Steam Tur-
Mechanical Health Monitoring and Diagnostics of bines". ASME Journal Engineering for Power, April,
Turbocompressors". Proceedings of the Eleventh 1971, pp. 225-237.
Turbomachinery Symposium, Texas A&M University,
December, 1982. [14] Zadeh, L. A. "Fuzzy Sets". Information and Control,
Vol. 8, pp. 338-353 (1965).
[2] Boyce, M. P., Meher-Homji, C. B. and Mani, G., Th e
Design and Development of Advanced Online Mon- [15] Gupta, M. M., "Fuzzy Information in Decision An-
itoring and Diagnostic Systems for Gas Turbines". alysis". IEEE Conference on Cybernetics and Society,
1983 Tokyo Joint International Gas Turbine Con- October 28-30, 1982, Seattle, Washington.
ference, Paper #83-Tokyo-IGTC-04. November, 1983.
[16] Zaba, T., "Losses in Gas Turbines Due to Deposits on
[3] Boyce, M. P., Et. Al. "Operating Experience With Blading", Brown Boveri Review, 12-1980, p. 716.
Health Monitoring and Diagnostics of MD Steam
Turbines and Centrifugal Compressors". ASME Joint
Power Generation Conference, Indianapolis, Sept.,
1983. ASME Paper #83-JPGC-PWR 28.

[4] Hayes-Roth, F. Et. AL, "Building Expert Systems".

Addison Wesley, 1983.

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 12/17/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use