-
Desalination, 92 (1993) 281-293 Elsevier Science Publishers
B.V., Amsterdam
281
Mathematical modelling and expert systems integration for
optimum control strategy of MSF desalination plants
B. Fumagalli and E. Ghiazza
Italimpianti Iritecna, Via Di Francis 1, Genoa fltaly)
SUMMARY
On the basis of the experience acquired with the design and
operation of the process control system of Umm Al Nar East
desalination plants (on duty since May 1988), a further development
of this kind of control system is presented. The foreseen
improvements derive from a suitable subdivision of tasks between a
traditional algorithmic system and an expert system based on
Artificial Intelligence techniques. In the first one, calculations
are performed by means of mathematical models to evaluate the main
process parameters, while the actuation of the calculated set
points and the manage- ment of the corresponding plant transient
conditions are carried out by the expert system using the rules in
its knowledge base.
INTRODUCTION
One of the main problems in the automatic control of a
desalination plant is the availability of control algorithms able
to manage all the situations taking place during the load
transients. Due to the close connection between the physical
phenomena governing the flashing of brine, the condensation of
steam outside the tube bundles, the heating of brine in the tubes
and the brine hydrodynamic behavior in the stages, a sequence of
checks on various process variables is required in order to avoid
the following effects:
l brine blow-through due to the low levels in the stages,
corresponding to
001 l-9164/93/$06.00 0 1993 Elsevier Science Publishers B.V. All
rights reserved.
-
282
l brine blow-through due to the low levels in the stages,
corresponding to a high thermal load,
l pile-up of one or more stage, due to the overflow of brine,
corresponding to a low thermal load,
l undesired fluctuations of the sea water temperature at the
outlet of the tubes of the reject section, due to the wrong
sequence and duration of the reject section set-points
increase,
l undesired fluctuations of the t.b.t., due to the wrong
sequence of set- points changes.
The results of these checks bring the system to adjust the
sequence and the duration of the set-points variation steps, until
the final distillate produc- tion flowrate is reached.
IRITECNA developed a control system based on this philosophy, on
duty since 1988 on Umm Al Nar East desalination plant in Abu Dhabi.
The aim of this system is to continuously monitor the variations of
the process parameters of the plant, by means of a steady state
mathematical model. Preliminary data treatment modules provide a
set of true values of the measured variables, obtained by means of
a filtering based on the Lagrange multipliers method. The true
values are then used to calculate the actual fouling factors, which
are among the adapting parameters of the mathemati- cal model. Once
the model has calculated the target set-points, a control algorithm
starts to manage their variation between the actual values and the
final ones.
The complete description of the control system is reported in
111 and [2], while the results of one year of operations on the
plant are reported in [3].
In order to obtain an efficient and reliable operation of the
control system, a continuous tuning of its performances was carried
out for a period of about three months, allowing the process
engineers to calibrate the main parameters and to adapt the control
logic to the desalination plant character- istics.
In recent years the expert systems diffusion in the industrial
field grew up, due to their special features. Unlike traditional
procedural systems, the rules based systems have the following
advantages:
l quite general operating rules can be defined in the system, l
operating rules can be activated only if necessary by the control
algorithm
(inference engine), l qualitative evaluation of events, that is
familiar for the operator, can be
used,
-
l new rules can be easily existing architecture.
283
inserted into the system, without changing the
IRITECNA has already acquired experience in developing expert
system applications in steelmaking plants. An expert system is on
duty in ILVA- DALMINE Multistand Pipe Mill, as a support to
operators for the control of the plant [7], while systems for the
maintenance support in heating furnaces and in blast furnaces are
in the design phase.
For the new desalination plant of Al Taweelah B, a new control
philoso- phy is now under study in IRITECNA, involving an expert
system for the management of the transient operations, while the
data treatment, monitoring functions and set-points calculation are
performed by a traditional algorith- mic system.
A proposal for a system of this kind is presented in the
following sections.
ARCHITECTURE OF THE SYSTEM
The proposed functional configuration of the system is shown in
Fig. 1. The traditional algorithmic functions, such as data
treatment and mathemati- cal models, are allocated in the process
computer, whereas all functions constituting the decisional support
for the system are allocated in the expert system dedicated
computer.
F%OCESSEtGtNEW -51moN
Fig. 1. System configuration.
-
284
The interfaces of the expert system are:
l an operator station, where the system displays all the
information about the plant status, the instructions for the
operator, the alarms. Here the operator inputs the necessary
information, used by the system for the management of
operations,
l the process computer, where the expert system gets all the
information about the target values of the set points, the rough
measurements and their reconciliated values. When an activation of
the mathematical model is required by the expert system, it sends
all the necessary data to the process computer,
l the plant instrumentation, where the expert system downloads
the values of the set points which are automatically updated by the
control system.
The process computer has a VDU interface for a technologist or a
process engineer. Through the VDU the adjusting parameters for the
mathematical model and the data treatment modules can be updated,
and the process variables such as the fouling factors can be
displayed, together with their trends.
The main operating modes of the system are:
l operator guide, when the control system gives information
about the status of the plant and the operations to be
performed,
l automatic operations on the plant instrumentation, in close
connection with the information given by the operator,
PROCESS COMPUTER
Aim of the process computer is to calculate the set points of
the main variables for the control of the production, on the basis
of the different operating conditions of the plant. Within the
process computer software architecture, the following main
functions can be identified: data acquisition, data treatment and
reconciliation, process parameters calculation, set-points
calculation, and data exchange with the expert system.
The block diagram in Fig. 2 shows the main relationships and
data exchange among the main functional blocks of the system.
The diagram shows one of the possible solutions for the software
configuration of the system. In particular, the calculation and
updating of the process parameters, used in the mathematical model,
can be performed by a separate module, using the reconciliated
values of the measurements, or it can be included within the
reconciliation problem, thus considering the heat exchange
equations of the stages as system constraints.
-
l-
Fig. 2. Block diagram.
285
The functions of each block are explained in detail the
following subsec- tions.
Data acquisition and treatment
The measured values of the main variables are taken from the
plant instrumentation, with a predetermined frequency. Usually
these values are already filtered, since normal data treatment
procedures such as check of the limit values, integration,
conversion to engineering units, linearlization of characteristics
are already present in the instrumentation and basic automa- tion
system. The treated values, however, are not available for the
process calculations, since they dont IX71 the system constraints,
e.g. the heat and mass balances and heat exchange equations of the
plant subsystems. The information obtained are not completely
reliable, due to the stochastic errors and to the biases usually
present in the measurements. Moreover, not all the process
variables can be measured.
In order to have a set of process variables, fulfilling the
system con- straints, and an estimation of unmeasurable variables,
data reconciliation methods are available. The aim of data
reconciliation is:
l to improve the reliability of measurements, l to estimate
unmeasurable variables, l to estimate instrumentation biases.
A mathematical approach to the data reconciliation is detailed
in Appen- dix I.
-
286
Process parameters updating
The behavior of the desalination plant is affected by a number
of external parameters, yielding the necessity of modifying the set
points of the plant during its operating life in order to keep the
same production rate. These parameters can be divided as
follows:
variations in the sea water temperature, that are a seasonal
perturbation of the operating conditions, and can be detected by
the temperature measurement, increasing of the fouling in the
evaporator and brine heater tubes for the same production rate, due
to the presence of dissolved solids in the sea water.
Since the mathematical model of the desalination unit is based
on the stages heat exchange equations, it is necessary to know the
fouling degree in each stage. There is no means to know the amount
of the fouling in the plant, but for some empirical correlations,
or mathematical methods involving the heat exchange of each
stage.
Two approaches to the problem are possible: the former is an
iterative calculation, based on the reconciliated values of the
plant measurements. The latter is to include the heat exchange
equations of the stages in the data reconciliation problem, thus
considering the heat exchange coefficients as unknown
variables.
Mathematical model
The set point values to use in order to keep the desired
distillate produc- tion are calculated using a steady state
mathematical model.
The basic equations describing the behavior of the desalination
unit are the heat and mass balance equations and the heat exchange
equation of each stage. the variation of the fouling degree in each
stage, affecting the value of the set points, is considered in the
calculation. A description of the equations involved in the
calculation is reported in Appendix II.
The following variables mainly affect the distillate product
flowrate: brine top temperature Tm, and brine recirculation
flowrate W,. The same value of distillate flowrate can be achieved
with different combinations of T ,nex and W,, for a given plant
condition. The choice of the values to be set affects the behavior
of the desalination plant, yielding the need to define a rule for
this choice. The highest value of the performance ratio of the
plant is achieved with the maximum value of T,, and the minimum
value of W,.
-
287
The necessity to avoid blowthrough between contiguous stages
requires to define some constraints for this choice. Other
constraints are the maximum operating temperature of the antiscale
in use in the plant and the operating limits of the mechanical
equipment, such as the pumps.
A solution of the problem is to insert in the model the
reference operating curves of the plant, that can be modified by
the model itself to account for the modifications of external
parameters. This solution is easy to be imple- mented, but it takes
time for the tuning of the best values to be inserted in the
operating curves. Moreover, no optimization of plant operations is
achieved with this method.
Another solution is to build up a cost function for the system,
and to choose the set points of T,, and W,, which minimize the
function. This approach to the problem has the advantage of being
more flexible because several variables can be considered in the
cost function. On the other hand, the mathematical approach is more
complex, the calculations are time consuming, and a powerful
machine has to be provided as a hardware support. This problem
becomes more and more critical if other variables than T,, and W,
are involved in the calculation.
EXPERT SYSTEM
The aim of the expert system is the supervision of the operation
and the on line control of the plant, supplying the degree of
knowledge of a plant engineer or of a skilled operator. In such a
way elements of knowledge derived from experience - which can
hardly be handled using traditional logic sequences and algorithms
- become part of the control strategy.
The functions performed by an expert system depend on a set of
operating rules stored in the system knowledge base. The knowledge
base is built with the aid of the knowledge engineer (e.g. a
process engineer), who supplies all information about the actions
to be taken in the various plant conditions.
An important element of the rules is the possibility to include
qualitative evaluation of the phenomena involved in the plant
behavior.
The main functions foreseen in the expert system for the
desalination plant are:
l plant operations management. The system detects events which
may produce a change in the plant operating environment, and
supplies information about the proper variation the operating
targets and the boundary conditions,
l transients management, l trouble shooting.
-
288
The plant operations management includes checks and actions of
the following kind:
l the sea water temperature is changing, l the 1.~. steam
availability to the desalination unit is changing, l the sea water
salinity is changing, l pump trouble is arising.
The expert system has to evaluate the new constraint values for
the involved plant sections, in order to supply the mathematical
model with the new limits. Moreover, the targets of the optimum
control strategy are changed too as a consequence of the checks
carried out of the plant operating conditions.
The transients management is invoked in one of the following
cases:
l the production target has been changed by the operator, l the
production target cannot be kept any more, due to the change of
the
boundary conditions of the system, l the set point values
necessary to keep the production target have changed
due to a change in the external parameters.
The expert system has to manage the plant transients, consequent
to the set points changes, executing all the checks and operations
in order to guarantee the necessary safety of equipment. Typical
problems arising in this case are:
for a load increase, an interaction exists between the variation
of the brine top temperature set point and the brine recirculation
set point. Due to the different responses of the system to the
change of these set points, an increase of the brine recirculation
may lead to a decrease of the tempera- ture at the outlet of the
brine heater, even in the presence of a t.b.t. set point increase.
As a consequence, the heating of the brine is delayed, and an
excess of steam to the brine heater is necessary, for a load
increase, a fast increase of the sea water flowrate can lead to a
decrease of the sea water temperature at the reject outlet, and of
the makeup flowrate as a consequence. In order to avoid this
problem, a suitable interaction between the sea water flowrate and
the makeup flowrate increases is suggestible.
The expert system can solve these problems selecting the proper
sequence of steps for the change of the set points.
-
289
The trouble shooting gives the operator information about the
abnormal behavior of the process variables involved in the
automatic operation of the plant. The expert system analyzes the
main process variables, getting information from their values,
their trends and their relations, in order to recognize anomalous
situations. The results of the analysis are also used to take the
correct control actions on the plant.
The expert system configuration does not include the automatic
control of the overall plant, because this target would require the
design of the instrumentation and control system in close
connection with the expert system. Only the plant subsystems
involved in the control of the distillate production are analyzed.
The possibility to use results of the existent mathematical models,
however, is a powerful tool giving the system the possibility to
correlate the variables involved in the analysis.
The functions of the expert system are carried out following the
operating rules stored in the system knowledge base. The main
relationships among the variables involved in the investigation may
be of the following kinds: logical, derived from the application of
formulas, derived from the calcula- tion of a mathematical model,
and derived from statistical analysis. The expert system gets
further information from qualitative evaluation, such as the
following: the brine level is normal, the brine temperature in the
stage is increasing, the load increase is in the starting
phase.
On the basis of this information, the proper actions are taken.
Examples of rules connecting the previous information and the
actions that follow:
l the level of first-stage is normal, l brine temp. is
increasing, l load increase is in_starting~hase.
/ _______._.
i set the setpint of brine recirczation to set-point +0
l the level of last-stage is normal l the level of first-stage
is increasing
1 ________.*
l the flowrate of brine recirculation is increasing I
1 set the setgoint of level_control of last-stage to set-point +
delta
CONCLUSIONS
An alternative way for the control of a desalination plant can
be based on Artificial Intelligence techniques. An expert system
provides the designer with tools to insert into the control system
qualitative knowledge rules, usually followed by skilled operators
in the conduction of the plant. The presence in the control system
of quantitative relations between the process
-
290
variables is however useful for optimization purposes, and for
the compari- son between the actual and the desired values. A
particular advantage in the use of an expert system is the
possibility of easily update the knowledge base by adding new rules
during the operation of the plant. The figure of the knowledge
engineer plays an important role in the creation and updating of
the knowledge base.
REFERENCE
S. Arazzini and D.M.K. Fareigh, Desalination, 55 (1988) 91-106.
R. Cirelli, B. Fumagalli, E. Ghiazxa and E. Longoni, Control10 di
process0 di un impianto di dissalazione, 21 st Convegno
Intemazionale, BIAS, 1987. S. Rebagliati, E. Ghiazza and K. S.
Abueida, One year operational experience on the process control
system at UANE MSF desalination plant, IDA Congress, Kuwait, 1989.
McGhee, Grimble and Mowforth, Knowledge based systems for
industrial control - IEEE Control Engineering Series 44. R.S.H. Mah
and A.C. Tamhane, AIChE J., 28 (1982) 828. S. Narasimhanand R.S.H.
Mah, AIChE J., 33 (1987) 1514. A. Batistoni Ferrara, P. Fontana, E.
Longoni et al., An expert system for operators support in the
control of a multistand pipe mill plant, Automaxione e
strumentazione, Nov. 1989.
APPENDIX I: A MATHEMATICAL APPROACH FOR THE DATA
RECONCILIATION
PROBLEM
Let us suppose to have a vector of n measurements expressed as
follows:
Y =Db+r
where y is the vector of measurements, b is the vector of the
state variables of the system, and E is the vector of the noise,
with normal distribution and null average value. D is the matrix of
functions which link the state variables to the measured variables.
The aim of the reconciliation problem is the calculation of a
vector of state variables 6, which are different from the state
variables b taken from the measurements, and tWil1 the system
constraints. The error between the value of y and the corresponding
reconciliated value y is given by the following expression.
-
291
e=y-7 =(I-DM)y-DNc
M = (I- NA)(D~Q-~D)-~D~Q-I
N = (D~Q-~D)-%~[A(D~Q-~D)-~A~]-~
The calculated value of error e has null average and variance
given by the following
V = (I-DM) Q(I-DM)T
The constraint equations are expressed as follows
Ab = c
where the coefficients of the matrix A are constant in the case
of linear constraints, and functions of the unknowns in the case of
non linear con- straints.
In the case of linear constraints, the values of the vector b
are given by the following equations:
~;=&+(D~Q -lD)-lATIA(DTQ-lD)-lAT]-l (c -Ai&)
b, = (DTq-lD)-l DTQ-ly
In the case of non linear constraints, an iterative calculation
is necessary. A first attempt value is supposed for the
coefftcients of the matrix A, then the values of b are calculated,
and they are used for a better estimation of matrix A. The
procedure is repeated until the convergence is reached.
-
292
APPENDIX II: MATHEMATICAL MODEL EQUATIONS
The mathematical model is based on the resolution of a system of
linearized equations, as follows (see Fig. 3):
WVi
I i wvi+l
hi i+ 1
Fig. 3.
l Overall heat balance
Recovery section
WS~. CpSi.TSi - WSi+~ .cpSi+l. Tsi+l+ Wdi*cpdi.Td, -
Wdi+l*cpdi+l*Tdi+l- Wr.cpri.Tsi + Wr.cpi+l.Tsi+l = Wvi+l*Hvi+l -
Wvi*Hvi
Reject section
Wsi+~.cpSi+~.Tsi+~-Wsi+~.c~i+~.TSi+~+ Wdi+l.cpdi+l*Tdi+l -
Wdi+2.Cpdi+l.Tdi+2- Wf.cpri+l.Tsi+l + Wf*cPTi+2*Tsi+z = Wvi+l.
HVi+l - WV, .Hv,
l heat exchange equation of the upper part of the stage
-
293
Recovery section
Wr . spry. Tri - Wr . CIKI+~ . Tri+r = Ui. Are Tri - Tri+r
111 Tc, - Tri+r
Tci - Tri
Rejection section
Wf.cpri+l.Tri+I - Wf.cpTi+2.Tri+2=Ui*Ai. Tri+l - Tri+2
In Tc, - Tri+2
Tc, - Tri+r
with the following boundary conditions:
l in the first stage Ts, = T max l in the last stage Tr, = T SW
l in the last recovery stage and in the first reject stage
l in the last recovery stage and in the last reject stage
where
Ts,* = Ts,,
Tr,* = Tsz2
Ts Tc Tr ws Wr Wf WV Hv U A cps, cpd, cPr
flashing brine temperature distillate temperature recirculating
brine (feed water) temperature in tubes flashing brine flowrate
recirculating brine flowrate feedwater flowrate steam flowrate
steam enthalpy overall heat exchange coefficient stage heat
exchange surface specific heat capacity