Copyright'"' IFAC Automation in Mining.

Mineral
and Metal Processing. Tokyo. Japan. 2001

CFD MODELLING FOR CONTROL OF A CHEMICAL WASTE

ROTARY KILN INCINERATOR

YongIiang Yang and Marm. A. Reuter

Section ofRaw Materials Technology

Department ofApplied Earth Sciences
Delft University of Technology
Mijnbouwstraat 120,2628 RX De/ft, The Netherlands

Abstract: Chemical waste processing forms part of a complex network for materials
processing. The rotary kiln for the incineration of chemical waste is a poorly defined
reactor in this network, and should ideally be controlled in a feedforward manner rather
than in a feedback manner as is in the practice. However, the process dynamics and the
poorly defined nature of the feed (including poorly sampled chemical waste in drums)
bring significant challenges in the feedforward control of such high temperature
furnaces. To optimally control such systems with very poorly defined parameters, a
well-structured database should be built to assist in the control. This concept has been
applied to the case of the rotary kiln incinerator through computational fluid-dynamics
(CFD) modelling as an off-line tool. in view of process control. The integration ofCFD
modelling into the process control system, generation of the CFD database and linking
the CFD predictions to the actual process control situation, are discussed in the paper.
The database generated serves as direct or indirect support to the operator in making
kiln-operating decisions and to improve the process control. Copyright C 2001 IFAC

Keywords: feedforwad control, computational fluid dynamics, database structures,

rotary kiln incinerator

1. INTRODUCTION process control heavily depends on thermocouple

measurements taken after the kiln exit plane, because
Rotary kilns are widely used in high temperature the inside of the kiln cannot be accessed This means
processing of materials, e.g. Waelz kiln for processing that the incineration process can only be controlled in a
of zinc containing materials, partial direct reduction of feedback manner. Therefore, it is difficult or even
laterite and chromite ores, calcination of limestone and impossible to anticipate great variations, which results
production of cement clinker. to name but a few in poor process control. poor conversion of CO to COo
applications. Rotary kilns are also used in hazaIdous destruction of dioxin, etc. For a poorly defined system,
waste-incintntion, due to its versatility and flexibility, it would ideally have to be controlled in a feedforward
as illus1rated in Figure 1. The rotary kiln incinerator is manner to ensure that complete destruction of the
like metallurgical furnaces a poorly defined system. In chemical waste is achieved at the same time ensuring
general, there is a lack of information about the that the reactor main1aim its integrity.
chemical waste input and a great variability of the waste
composition due to poor sampling possibilities. The

403
 can assist in the process control of the chemical waste
incineration kiln. If simulation results are introduced in
the process control in an appropriate way, this will
provide a means that would ultimately enable
feedforward control of the incineration process. The use
of CFD modelling has great potential to improve the
process control and to support to the operator in making
kiln~g decisions. A better control of the
incineration process is essential in the following issues:

• maintaining the desired incineration temperature

of about 1300"C inside the rotary kiln to ensure
complete destruction of the hazardous chemical
waste (e.g. dioxin);
• improving the refractory lifetime, which can be
achieved when the temperature can be better
Fig. 1. General structure of the kiln and the controlled and high temperature peaks be
secondary combustion chamber (SCC). avoided;
• monitoring the residence time at sufficiently
The physical/chemical processes and dynamic high temperatures to ensure that the residence
behaviour inside the kiln are as complicated as in time of the combustion gases after the last
metallurgical processes, because of poorly defined injection of combustion air exceeds 2 seconds at
characteristics of feed materials (different types of llOO"C as dictated by the European Directive on
chemical wastes in poorly sampled drums). the Incineration of Waste (since June 2000).
Simulation of the transport phenomena, e.g. through
computational fluid-dynamics (CFD) modelling, in • increasing throughput of low calorific chemical
such a process becomes a very challenging task. waste and decreasing the throughput of fuel oil
Attempts have been made by many researchers in the and high calorific chemical waste, to improve
past 20 years to build up different types of models to the economic performance of the incinerator.
simulate the flow, heat transfer and combustion
processes inside the rotary kiln for chemical waste It is important to note that, for the continuously
incineration. More information can be found, among changing conditions at field-scale, CFD models
many others, in the following references by Jenkins cannot be used directly in on-line process control.
and Molar (1980), Leger et al. (1993), Jakway et al. The computation time for one simulation run is much
(1996), Veranth et al. (1997), Wardenier and Van too long for direct control purpose. Therefore, an
den Bulck (1997), and more currently in authors important question is how the CFD modelling can be
group (Rakhorst et al., 1999). integrated in the process control system and how the
CFD predictions can be linked to the actual process
CFD is a mathematical modelling tool to simulated control situation. A concept for these issues is the
fluid flow related processes. The main objectives of main point in this paper. Here an example is shown
CFD modelling were to get understanding of the on using CFD in simulating a high temperature
velocity and temperature distribution, and how the chemical waste-incineration process operating at
performance of the rotary kiln would be influenced AVR-Chemie in The Netherlands, which is very
by basic design modifications and operational similar to pyrometallmgical furnaces. The concept is
parameters, such as thermal energy input, leakage air, well applicable to pyrometallurgical processes. The
etc. In none of the studies it was investigated how potentials and challenges as well as its limitations of
CFD modelling could be more directly used as a tool using such a tool in process control will also be
in the process control of a chemical waste discussed.
incinerator, which would be one of the major
engineering goals in industry. Consequently, there is
no public knowledge in this area. 2. BASIC INFORMATION FROM CFD
SIMULATIONS
This poorly defined chemical waste incinerator plant is
situated in a complex web of material flow (Vechoef et CFD models for computing gas flow and temperature
al., 2(00). In order to optimise this whole system it is distribution for chemical waste incineration have
obviously imperative that each of the planWunit been established within a framework of a commercial
operations should be optimally controlled so that the CFD code PHOENICS. The governing partial
system as a whole is nm environmentally optimally. differential equations for conservation of mass,
Therefore, the main objective of this paper is to momentum, and energy in a turbulent flow system
illustrate how the chemical waste incinerator can be were solved simultaneously for the given incinerator.
optimally controlled. It is shown how CFD modelling

404
Different turbulence models such as the standard k-e, I. '
~- - -- ~N• • 1.34
RNG k-e, and Chen-Kim k-e models were tested
[Rakhorst et al., 1999], and was found that the
1. 2 \ VN,. -1 .28
I- El ltt
- -- -. £2( 81/
11

standard k-e model is well applicable and this model Back· miled volum. :V.. N
0.'
was used in the current modelling work. For ~
W
0 .79 for E1 (1f)
0 .1 5 for E2(9 )
combustion modelling, a global chemical reaction 0.' O.ad volume : VJVz 1-V..N
0 .21 lor El (9)
scheme (SCRS) for chemical waste combustion O. 0 .2 5 'or E2(e )

reactions has been included in the models, and 0.2

reasonable results for temperature distribution have
been obtained and validated with certain field-scale
0 ii 0 .5 15 25 3 .5

temperature measurement data.

As a result, distributions of velocity, temperature and Fig. 4. RID curves of the main burner stream (r-=16
other useful variables (e.g. fuel mass fraction) are seconds) and load chute air (t=23 seconds).
obtained after the simulation. In addition, residence
time distribution was also simulated through The information regarding the residence-time,
numerical tracer tests in a solved CFD model. temperature distribution is very useful to characterise
Statistical RTD analysis was made through the different feed types once the model is suitably
obtained integral and differential curves. calibrated. This information then can be used to
Furthermore, residence time distribution of fluid optimise:
particles after entering the furnace was tracked,
which gives further information on residence time • the chemical waste destruction,
distribution. Figure 2 illustrates the velocity
distribution across a few regions of the incinerator, • the energy recovery and hence also the off-gas
and a clear view of the complexity of the flow pattern composition,
and mixing behaviour could be seen. Figure 3 • the slag composition, designed for different
illustrates the computed temperature and residence chemical waste types in view of their Iiquidus
time distribution together with the combustion temperatures and capacity to contain minor
modelling. Figure 4 shows the predicted RTD curves. elements, and
The mean residence time, dead volume and mixed
flow volume could be calculated. • the wear of refractories, affected by temperature
(spikes especially), slag and minor elements.

3. INTEGRATION OF CFD MODELLING IN THE

PROCESS CONTROL SYSTEM

The idea for a feedforward control structure for

metallurgical furnace has become a well-established
concept in the metallurgical industry nowadays. This
will be translated in this paper for the rotary kiln
incinerator, from which it will become apparent that
a well-structured database could form the pivot
around which the supervisory control structure
should be built (this is in addition to the low level
Fig. 2. Flow pattern inside the rotary kiln and SCC. PLC and other low level feedback control loops)
(Figure 5).

Most elements of the control structure interact with

each other and databases play an important role. It is
essential that a good process information database is
available, which contains the chemical waste input
data and temperature measurements that can be
correlated with chemical waste input data. This
information is needed for the simulation of different
kiIn-operating conditions. The simulation results
should be fed back to a CFD database that can be
used by the operating personnel and the supervisory
Fig. 3. Temperature and entry time distribution control system.
across load-chute plane of the incinerator.

405
 Fig. 6. Oftline CFD simulations and continuous
model validation with temperature measurements.

Fig. 5. A control structure of a chemical waste

incineration facility with a CFD model database.

The various process data can then for example be

matched with the CFD results to provide the operator
with useful process information that matches the
poorly defined feed with various operating
conditions. Various intelligent process control tools
can be used to achieve this (Reuter et al., 1996; Fig. 7. Post-processing of CFD predictions and use of
Reuter 1999). CFD database in the process control system.

The supervisory control refers to a situation in which It is well known that CFD modelling generates a
a computer controls the process and the operators and large amount of output data, which is difficult to use
management interact with this supervisory computer. directly in process control. Appropriate post-
However, in the short to medium term, it may be processing of the results should be carried out to
expected that the operators remain making the produce useful data-sets in order to be able to link the
determination of most set-points to the controllers for CFD predictions to the actual process control
the input of chemical waste. In reality, it may be situation, which is shown in Figure 7.
difficult to have the supervisory control system
taking care of all the decision-making due to the The development of this concept may ultimately
nature of the incineration process. Therefore, the line result in a specific indication of the optimum or most
between the supervisory control system and the desirable values for the set-points of the important
controllers is a dotted one. controllers within the incineration facility. These set-
points may be provided simply as information to the
At first, the simulation results should be used to operator, i.e., the supervisory control system may be
relate various parameters and input streams, to open loop, after the operator simply accesses the
predict temperature distribution, residence time, CFD database via the PC in the control room. In this
thermal load, etc. The amount of chemical waste way the CFD database can provide better information
input scenarios should then be increased and than human experiences or guesses. This information
optimised in order to be able to determine the best can be used to suggest set-points.
operating strategy for a specific set of input streams
and to improve kiln performance. After that, the
information should be arranged in such a way, that 4. LINKING CFD PREDICTIONS TO THE
useful incineration control could be performed in a AcruAL PROCESS CONTROL SITUATION
feedforward manner.
As above-mentioned. CFD models provide a lot of
In Figure 6, the dynamic process of running off-line information in a standard form, which cannot be
simulations of different input scenarios on basis of directly linked to the actual process control situation.
chemical waste input data and the model validation Therefore, proper post-processing of the results is
with temperature measurements is illustrated. required to produce useful data sets, which vectorise
Especially in early stages of generating CFD various operating conditions, including the effect of
simulations, model validation is essential for a various (poorly defined) feed materials. Only these
continuous development and improvement of the data sets should be stored in the CFD database.
models. The CFD predictions are eventually stored in Further investigation should be carried out on this
the CFD database. topic, but two potential approaches of how this could
be done, are presented in the following section.

406
 4.1 Averaged temperature and vector approach )I. axi s

Every cross section, which is taken from the model

geometry in order to show the simulation results, will
show to more or less extent a gradient The visual
observation of the CFD predictions may be clear
then, but for storage in the CFD database, appropriate
data sets are required

The temperature gradient at specific locations in the

incinerator must be quantified. For this purpose one
or more cross sections should be taken, which are
supposed to give most information about the Fig. 8. Example averaged temperature and vector
incineration process, i.e. cross sections that are most approach.
useful as support to the operator and to be supplied to
the process control system. The average temperature The center of the average temperature can be calculated
weighted with mass flow rate gives a quantification based on the temperature moment over x and y c0-
of the temperature or enthalpy distribution in a cross ordinates, as is shown in equations (4) and (5).
section and can be supplied to the database. The
.
geometry of the model is built up of many grid cells.
The temperature in each cell is assigned a weight r zd(';'.I)'H.I)') L,x;(,;,;HJ
. (4)
associated with the mass flow rate through the cell.
In this way the average temperature is given by a
weighted linear combination of all cell temperatures
i
r d(';'.I)'H.I)')
;=1

L,(,;,;HJ
#1

in one cross section, as shown in equation (1):

.
L,y;(,;,;HJ
;=1 (5)
;=1
(1)
f(,;,;HJ
;=1

Further processing of the average temperature can be

In this equation T; is the temperature and mi is the illustrated in Figure 9, where average temperature
mass flow rate in kg/s in the i-th cell, having a total profile along the main flow direction of the reactor
number of n cells in the cross section. T; can be can be constructed. This is normally required for
calculated from the specific enthalpy Hi, if it not design and operating purposes of industrial furnaces.
solved directly. The averaged temperature can then However, simple average figures are normally used
be expressed as follows in equation (2): in practice. The data points in Figure 9 also include
the x-y information (hidden), which is obtained from
equations (4) and (5). Then a very good picture of the
i:,(H;.m;) (2)
t_ ,..;p..d = .,l;;-;,!",'._ __ kiln thermal behaviour can be obtained.

Im;·c
i=1
p In addition to the temperature averaging, temperature
distnbution functions along any cross sections within
However, the average temperature does not give any the reactor can also be obtained and further analysed.
information about the location of the different Temperature gradients and the temperature
temperatures. A vector approach may be used to distribution functions can be acquired from further
point out the location, for example, the hot region in post processing of CFD outputs.
the cross section. The cross section should then be
attached to a co-ordinate system.

An example of this idea is given in Figure 8. By

means of integration over the whole cross section for
the enthalpy and mass flow rate, the centre of
enthalpy may be calculated and coupled to a
temperature vector. This will make the average
temperature a function of the x and y co-ordinates, in
addition to a function of the enthalpy and mass flow o Main flow direction
z
rate per cell:

Fig. 9. Dlustration of average temperature profile

t = f(H,m,x,y) (3) along the main flow direction in the reactor.

407
The information of Figures 8 and 9 can be stored in should be optimal. Due to the complexity of the
the database and matched to the appropriate various material streams it is often very difficult to
operating conditions. This implies that, given certain optimally control the individual reactors. This paper
feed types, the operator can obtain an immediate presented an approach how, in a feedforward manner,
picture of the distribution characteristics CFD can be applied to deal with the processing
f::; f(H,m,x,y,z) to optimise the various aspects complexity.
mentioned. Also, this will enable the operator to
select the correct mix of feed materials to ensme a
"correct" distnbution as presented by Figure 8 is ACKNOWLEDGEMENTS
maintained andIor reached. This approach then
becomes a soft sensor for the operation. Financial support and permission to publish the data
of the kiln operation from A VR-Chemie are greatly
acknowledged. Special thanks are due to Mr. J.
4.2 Direct temperature approach Verwoerd and Mr. A. Quak from AVR-Chemie for
their constant interests in and support to this work.
The incineration process inside the rotary kiln
heavily depends on thermocouples that are installed
after the kiln exit plane. This means that the process REFERENCES
can only be controlled in a feedback manner.
Therefore, it is very difficult to anticipate great Jenkins, B.G., F.D. Molar (1980). Modelling of heat
variations, which results in poor process control. transfer from a large enclosed flame in a rotary
kiln. Transaction of the Institution of Chemical
The CFD predictions, that give the temperature in the Engineers, 59, pp. 17-25.
location where normally a thermocouple records the Jakway, A.L., A.M. Sterling, V.A. Cundy, C.A. Cook
temperature, could replace these measurements. In (1996). Three-dimensional numerical modelling
this way it is possible to a priori have information of a field-scale rotary kiln incinerator.
about potential temperatures that may be reached EnvironmentDl Science & Technology, 30, pp.
with different kiIn-operating scenarios. This implies 1699-1712.
that the CFD results in the database act as a soft Leger, C.B., V.A. Cundy, AM. Sterling (1993). A
sensor, and provide information useful for process three dimensional detailed numerical model of a
optimisation and control. field-scale rotary kiln incinerator. Environmental
Science & Technology, 27, pp. 677-690.
Rakhorst, J., Y. Yang. M.A. R.euter, J.H.L. Voncken
(1999). Computational modelling of the gas flow,
5. CONCLUDING REMARKS mixing, and temperature distnbution in a rotary
kiln hazardous waste incinerator. The Phoenics
In this paper, a concept was proposed for using CFD Journal, 12(3), pp. 278-292.
simulation results in a supervisory control system Reuter, M.A. et al. (1996). Intelligent control of
through a rotary chemical waste-incinerator. It is submerged-arc furnaces for ferroalloys, Journal
highlighted that a feedforward control is highly ofMetals, 48(12), pp. SI-53.
needed for such poorly defined furnace systems. To Reuter, M.A. (1999). The simulation of industrial
provide the data for the feedforward control system, ecosystems. Minerals Engineering, 11, pp. 891-
distnbuted information from CFD simulations after 917.
proper processing could be very useful for building Wardenier, K. and E. Van den Bulck (1997). Steady-
up part of the control database. slate waste combustion and air flow optimisation in
a field scale rotaIy ~ Environmental Engineering
Presently the combustion models for the incineration Science. 14(1), pp. 43-54.
system are under further refinement, and at the same Veranth, I .M., GD. Silcox, D .W. Pershing (1997).
time, different operating and feed material scenarios Numerical modelling of the temperature
are being simulated, and a comprehensive CFD distribution in a commercial hazardous waste
database will be constructed and tested at AVR- slagging rotary kiln. Environmental Science &
Chemie in their hazardous chemical waste Technology. 31, pp. 234-253.
incineration process. This is matched with Verhoef, E.V., G.P.J. Dijkema and M.A. Reuter
temperatme measurements to validate the results. (2000). The complexity of material cycle
The various distnbutions provide the operator with simulation and design - co-incineration. In:
different valuable diagnostic information, which can Proceedings MINPREX 2000, 1nl congress of
be used to optimise the plant operation. mineral processing and extractive metallurgy,
publication series 512000, (p. Griffiths and A.
The complexity of the interaction between various Spry EeL). pp. 553-558. Melbourne, Australia.
chemical waste and material processing facilities
becomes apparent If this complex system is to be
optimised, the control on the individual plant level

408