SUPERVISORY FUZZY LOGIC CONTROLLER USED FOR PROCESS LOOP CONTROL IN DCS

SYSTEM
M. ABD EL-GELIEL .Prof M. A . El-Khazendar
Arab Academy for Science and Technology Tanta University
D e p m e n t of Electric and control Eng. E e c .Power &Machine Dept.
P.O.1029 Miami, Alexandria, Egypt Tanta, Egypt
[email protected]
- m a khazendar@,hotmail.com

Abstract - In this paper, the designed schemes for two dependant on the experience of the individual operator [IO].
Mamdani fuzzy controllers, employing the scaling factor The development of controllers capable of generating tuning
tuning are proposed. The first fuzzy logic controller, is a parameters of fuzzy controller to obtain the desired dynamics
normalized controller used to control the process loops, for the plant is of a great importance. There are many
the tuning for its input and output-scaling factors is done parameters can be adapted on line in fuzzy controller
through the second fuzzy controller (the supervisory (universe of discourse, FIS, FAM, etc.) some of the
controller). The supervisory fuzzy controller tunes the adaptation methods (proposals) in [4,9,11,12]. Adjustable
normalized fuzzy controller based on the variation of the scale factor of normalized fuzzy controller [4] are one of the
loop response. adaptation parameters. Some of the scale factor selection
The normalized fuzzy controller and the supervisory methods are explained in [4,13,14,15]. In this work, a fuzzy
fuzzy controller are organized with specific experience controller that self-adapts the parameters, mainly the input
information ahout a nonlinear thermal process. The and the output gain coefficients is proposed, using only
proposed fuzzy controllers are applied practically to qualitative knowledge of the plant. The controller will start
control the thermal process. That controller is applied to with a set of fixed parameter (normalized fuzzy controller
another process (level control process), where the with input scale factor and output scale factor) and through
controller gains are selected based on the process the supervisory fuzzy controller, the scale factors of the
conditions and limitations. normalized fuzzy controller are adjusted or adapted. The
The great advantage of that contribution is the ability adaptation of scale, factors are done by two methods, the
to he generalized for different process variable loops in output-input scale factors are tuned due to the error between
Multi-inpuuoutput system o r any DCS system. The the reference input and actual output firstly. Finally input-
supervisor here make a decouple between the process output scale factors are tuned according to the error between
Keywords: Normalized fuzzy controller, supervisory the desired response ad the actual response (model reference
fuzzy controller, scaling factor tuning, and DCS systems adaptive technique)[l7]. This adjustment is accomplished in
control loop. continuous time. The analysis of the robustness of the
proposed algorithms is performed for different real-time
1- INTRODUCTION situation of the case study (Thermal process).
The same control algorithm is applied to control level
The dynamic characteristics of most control system are process with the same tuning parameters (with-out retuning
not constant because of several reasons, such as deterioration and prior information about the process) but the input-output
of components as time elapse or change in parameters and signal level is changed according to sensor and actuator
environment. A satisfactory system must have the ability to
voltages limits.
adaptation. Adaptation implies the ability to self-adjustment
The response in the two cases (level and thermal) is
or modification in accordance with an unpredictable change
acceptable. So this approach can he generalized in DCS
in the condition of the environment or structure. In many
system to control different loops (variables).
cases, the physical measurements of the pertinent quantities
are very difficult and expensive. These difficulties lead to 2- NORMALIZED FUZZY CONTROLLER AND
explore the use of "Artificial Intelligence" (AI) [l-4] as a
SUPERVISORY CONTROLLES
way of obtaining models based on the experimental
measurements. One of the superior capabilities of fuzzy The first stage of the fuzzy controller operation is the
system, as an AI technique, is that it can use the information fuzzification and the last one is the defuuification. On both
expressed in linguistic pattem [I-81. Though most fuzzy stages, the membership functions that describe the different
system have been formed to emulate human decision making values of the linguistic variables are applied. To choose
behavior, the linguistic information stated by an expert may membership function, first of all one needs to consider the
not be precise, or it may be difficult for the expert to universe of discourse for all the linguistic variables, applied
articulate the accumulated knowledge to encompass all to the rules formation. To specify the universe of discourse,
circumstances. Hence, it is essential to provide a tuning one must frstly determines the applicable range foe the
capability [9] for fuzzy system to generate or modify the characteristics variable in the context of the system designed.
controller parameters on line in real system and it is an It is usually desirable to scale, or normalize, the universe of
important issue in intelligent control. Thus, the human discourses of an input/output variable, where a reasonable
operator (supervisor) is often required to provide on line choice of one case can be unsuitable for another case of the
adjustment, which makes the process performance greatly system operation. Normalization means applying the

0-7803-7729-X/03/$17.00 02003 IEEE. 263

 performance not specific performance both supervisory
controllers introduced could achieve the performance
required hence the simplest one is preferred.

6- REFERENCE

output [I] L. H. Tsoukvlas and R. E: Uhring, "Fuzzy and Neural

Scale Approaches in Engineering", John Wiley Sons, inc.,
factor 1997
[2] H. Hargras, "Applications on Fuzzy Logic Control,
Genetic Algorithms in Intelligent process Control", M.
Sc. Thesis, Alex. University, 1996.
[3] C. J. Harris, C. G. Moore, and M. Brown, Intelligent
Control Aspects of Fuzzy Logic and Neural Nets, World
Scientific, 1993.
[4] Leonid Remik, "Fuzzy Controllers", Newness, 1997.
[5] L. A. Zadeh, Fuzzy sets, Information Control, vol. 8, pp.
338-353, 1965
[6] T. W. Vaneck, "Fuzzy Guidance Controller for an
2 Autonomous Boat",' IEEE Transaction on control
1.5
system, pp 43-51, April 1997.
[7] T. Terono, K. asai, and M. Sugeno, "Applied Fuzzy
1 Systems", USA, 1997.
0.5 [8] J. M. Mendel, "Fuzzy Logic Systems for Engineering: A
Input tutorial", Proc. IEEE, vol. 83, pp. 345-377, 1995.
0
scale [9] Marco Russo Lakhumi C. Join, Fuzzy Learning and
factor 0.5 Applications, CRC Press, 2001
-1
[IO] Han-Xiong Li and Shouping Guan," Hybrid Intelligent
Control Strategy", IEEE Control System Magazine, pp
-1.5 36-48, June 2001
-7 [ 1 I ] L. X. Wang, Adaptive Fuzzy System & Control design
& Stability Analysis, Prentice-Hall, 1994.
[I21 Junhang Nie and Derek. A. Linkens, Fuzzy-Neural
Control Principle, algorithms and application, Prentice-
Time (sec) Hall Europe, 1995.
[I31 Plam R. "Scaling of Fuzzy Controllers Using the Cross-
Fig. 14 output and input scale factor of correction", IEEE Transaction on Fuzzy Systems, vol. 3
response Fig.13 nol,pp 116-130, 1995.
[I41 Yager R. R. and Filer D. P. Essentials of Fuzzy
5- CONCULOSION Modeling and Control, John Wiley, 1994.
In this paper, a fuzzy controller with self adapted scale [IS] Renmik L. and Stoica A., "Some Tricks in Fuzzy
factors using another supervisory fuzzy controller is Controller design", In: Proceedings of Australia and new
implemented, and its robustness is checked when disturbance Zealand Conference on Intelligent information systems,
occur. It is shown that, this type of .controller can be ANZIIS-93, IEEE, Perth, Westem Australia, pp. 60-64,
generalized with different inputsioutputs ranges for different 1993
loops and different variables in DCS system, which needs a [I61 M. A. El-Geliel, "Computer Based control of Thermal
simple and fast controller. Process Using Fuzzy Logic and PID controller", MSc.
Using a scale factor tuning limits the FLC parameters, Thesis, Arab Academy for Science and Technology,
this needs to tune, such as membership-functions, rule base, June 2000
fuzzy inference, etc. The priority of supervising output scale [I71 Kal Johan Astroum and Bjom Wittenmark, "Adaptive
factor is selected duo to its direct impact on stability and control", Addison-weslly, 1995
oscillation tendency. But input scale factor has the most
influence on basic sensitivity of the controller. Supervisory
controller can tune both input and output scale factors
according to the required performance. Using model
reference with supervisory fuzzy controller forces the
process to follow the desired specifications hence it is
preferred in case of the specified performance required. If the
required from the process is to achieve some accepted

264
standard range of [-I, + I ] for the universe of discourses of
the inputs and the outputs. The structure of the fuzzy
controller is shown in Fig. 1. controller

- - - - - - - - - _ - _ - .._ .. !
I, . .. . . .. .. .. .. .. . . .. . . .. ..

I Controller I 1
71 ~

\
Normalized
Output
fuly
controller scale
factor
I

Crisp Crisp
input Process or object output
under control
Fig. 2 Over-all Block of the process and supervisory fuvy
controller with the same inout Siena1 ofthe main controllers
Fig, 1 Block diagram of normalized fuuy controller

Model

In the case of the normalized universe of discourse, an Supemisow

appropriate choice of specific operating areas requires
controller
scaling factors. An input scale factor transforms a crisp input
into a normalized input in order to keep its value within the
universe. An output-scaling factor provides a transformation
of the defuzzified crisp output from the normalized universe
of the controller output into an actual physical output [4]. 1L
Some priority list of scale factor choice is recommended in
[13]. Similarity between coeficient Ki and Kp of the PI
controller and the scaling factors of the normalized fuzzy
controller is analyzed in [14]. Selection of scale factors by
trail and error is suggested and recommended in [4]. In this
study, firstly tuning scaling factor of input and output is
achieved using trail and error based on the human
experience. After that the tuning is achieved using another
fuzzy controller (supervisor). In the two cases the procedure
oftuning the factors is related to the research results in [IS].
The main objectives of supervisory controller are tuning
on lines the scale factors of the normalized fuzzy controllers. Fig 3 over-all supervisory fuzz). controller using model
The tuning procedure depends on trail an error (human reference technique
supervision) or adaptation algorithms.
The output gains of the supervised fuzzy controllers 3- THERMAL PROCESS AND LEVEL PROCESS
depend on the error signal between the reference input to the HARD-WARE CONFIGURATION
process and the actual output as shown in Fig.2. In this case
The thermal process is a continuous process. The
the system response can perform accepted specification but if
apparatus consists of an electrically heated model process,
it is desired to force the over-all system to achieve a desired
conditioning unit and digital controller. The model is an
specification such as overshoot, rise time etc ... the output
electrically heated aluminum process block surrounded by a
gains should be related to that specifications. Model
reference adaptive technique is one of the adaptation water jacket; into which is inserted a platinum resistance
methods used o force the system performance to achieve a thermometer is inserted. The model is deTigned so that it is,
specified specification [3, 171. Fig. 3 shows the over-all in effect, a speed up version of an industrial process with
block diagram of the model reference supervisory controller time constant shortened, to make the experiment of a suitable
where, the input signals to the supervisory controller are duration for laboratory work. The control problem
error (between the desired model and actual output of the investigated is that to maintain the process temperature under
variation of heat losses (by variable cooling water flow rate).
process) and its rate of emor.

265
 The required process temperature (set point) is adjusted by The two experimental process tested in this work are a
gradual dial on the panel. The error signal is suitably simulation model of two process variable (temperature and
amplified to drive a temperature deviation of 200mv for one level) in actual liquefied petroleum gases recovery process in
degree centigrade of the temperature error change. The petrochemical industries. The system is exposed to a variable
control of the power to the process heater is achieved using condition irregularly so the exist controller (mainly PI) not
thyristor circuit operating in burst firing mode, which suitable.
transforms the control signal of +IO v from the controller to
0-220V to the heater. The schematic diagram of the process
4- EXPREMENTAL WORK
is shown in Fig. 4.
The level process is a continuous process. The apparatus The first parameter of the fuzzy logic controller (FLC)
consists of level tank driving pump and two electric valves, was obtained from a manual tuning of process that gave the
one for feedback water and the other for output water flow most appropriate response [16]. Fig. 6 shows the normalized
rate driven by their positioning units. The control problem membership function of inputs and outputs signals of the
investigated is that to maintain the tank level constant under FLC and Table 1 shows the fuzzy rule base. The process
variation of output water flow rate. The required process response for step input 30"c and 50% flow rate cooling
level (set point) is adjusted in the controller. The actual level waters and using static fuzzy controller described above is
is measured using level sensor. The level signal is suitably shown in Fig. 7. Using supervisory fuzzy controller to tune
amplified to drive 50mv for one percent of the tank capacity. the main fuzzy controller output scale factors, with
The controller output signal is 0-5 volt to operate the membership function is shown in Fig. 8, and the rule base is
feedback valve and the feeding pump is supplied from shown in Table 2. These parameters are selected according to
constant 22OV AC supply. The schematic diagram of the the experience and manual tuning results. The input signals
process is shown in Fig. 5 .
c-"- .. Digital
for both the supervisory controller and the main controller
are the same but the outout is the new scale factor. The block
supply diagram of the overall control system is shown in Fig. 2. The
controller
~ 220 V system response using supervisory controller for step input
5O"c and different cooling water flow rates (disturbance
case); i s shown in Fig. 9 when scale factor for inputs fixed to
IO and 20 respectively. The results show the robustness of
Thermometer
~ ..... .......... ~ ~ ~ ~ . . . ~ . . the supervisory fuzzy controller
TABLE 1
FUZZY ALLOCATION MATRIX OF
NORMALIZED FUZZY CONTROOLER

roportional NB NM Z PM PB
~ Drain valve
I ...... ~.~~~~
................... ~ ...............
. ~ .~ ~ ~ ~~ ~~ . . .
Process
Fig. 4 Schematic diagram ofthermal process
@MA
-I -0.5 0 0.5 I

Fig. 6 Normalized membership function of

innuts and oumut variables

NM PM PM
Z PM PM Z NM NM
PM Z NM NR NB
PM Z NM NB NB NB

I- -7-7 I
the supervised system. The relation between each output
Fig, 5 Schematic diagram of level process from the supervisory controller and its input signal is
specified according to the tule base in Table 2. Each output
membership functions are taken as shown in Fig. 8 for all
outputs. By comparing both results, the system response is

266
improved in case of tuning the input and output scale factors
simultaneously, where the overshoot is reduced. Table 2
The output gains of the supervised fuzzy controller, in all FUZZY ALOCCATION MATKIX OF
of the responses obtained, depend on the error signal SUPERVISORY FUZZY
between the reference input to the process and the actual
output. In this case the system response performs accepted
specification but if it is desired to force the over-all system to
achieve a desired specification such as overshoot, rise time
etc ... the output gains should be related to that
specifications. Model reference adaptive technique Fig. 3 is H H
one of the adaptation methods used to force the system
performance to the desired specifications. If the reference
output was taken as a second order system with 2%
overshoot and 150 sec rise time. Fig.1 I shows the forcing of
the system for different operating condition of the process - 100
50% and 80% cooling water flow rates. Fig. 12 shows the - 90
system response for step input 2O"c and different cooling - 80
water flow rates (disturbed case). The results show the - 70 Cooling
robustness of the model reference supervisory fuzzy - Water
60 Flow
controller. By analysis the results, the supervisory controller
improve the system performance, but if it is desired to
- 50 Rate

restrict the performance model reference supervisory

- 40 (04

controller is preferred.
- 30
- 20
30
- 10
25 ' 0
o 400 eo0 1200 1600 2000
l
2. .. ... .i" i:2 ~ .-.>;.J -..*..: ..... ~:&.i
L_. ._.I AI t
Time (sec)

Fig. 9 Output response for distributed system

-
- - -----
Response
Cooling water flow rate

Time (sec)
Fig. 7 Normalized fuzzy controller response (without
supervision)

N Z P N Z P

b) Kate of error

x
a) Error

Time (sec)

Fig. l a1 Response of supervisory controller in case supervising

of output-input scale factors and output only
-Output scale factor supervision
c) Output scale factor
-
-. . ..-. Input-output scale factors supervision

Fig. 8 lnputloutput membership function of

supervisory fizzy controller

267
 Applying the supervisory fuzzy controller based on the
i(55 variation of the output response to the level process in Fig. 5
I
with the same parameter used in the thermal process with
150
i
changing the controller output to 0-5V corresponding to 0%-
output
Temp
I 45
100% opening of the feedback valve and the sensor signal
level 0-5V corresponding to 0%-100% level. The level
(“c) A0 response for step input 60% and disturbed suction valve is
i shown in Fig. 13. The variation of scale factor of input and

i
i
L
35

30
output of the supervisory controller is shown in Fig. 14. It is
no need to apply model reference adaptation technique in this
case where the system achieves the best performance with
I minimum rise time and 0% overshoots.
After testing the supervisory fuzzy controller in thermal
0 200 400 600 800 900
and level process without prior information about the level
I. ._.,~ - -._-~

process it is note that supervisory fuzzy controller can be

standard controller for different loops (process variables) in
Fig. I 1 Model reference supervisory fuzzy controller DCS system and also can be used for any single input and
response for different cooling water flow rate output process control loop hy some changing in the ranges
Flow rate 50% of scales, input and outputs signal limits.
-
--- - --- - Flowrate80%
- -
.- . - . .-Keference model output .

Time (sec) Time (sec)

- _ I - . .T”_~_I_.I.”4.~-
~ .-__-_-.-”---.-__-~
{/ 55 - 100
! - 90
output 50 - 80
Temp.
- 70
(OC) 45
- 60
40 - 50
-
i 35 -
A0
30
i -
1 30 -
20
10
- 1

0 400 800 1200 1600 2000

il._
I . ~ -.-&;.*.>:,:”:.&&
‘ A .,.-.. ::- ..
“Z.U..,.” .&22 :
Time (sec) Time (sec)
Fig. I2 Control signal and output response
for distrihuted system
Fig 13 Control signal and actllai level with disturbed suction

-Desired model output -Actual level

valve

- - - - _- Percentage opening of suction valve (V2)

-- --- Cooling water flow rate
Response

268