You are on page 1of 19

Analysis of Hybrid Temperature

Control for Nonlinear Continuous


Stirred Tank Reactor
Om Prakash Verma, Sonu Kumar and Gaurav Manik

Abstract Classical controllers usually require a prior knowledge of mathematical


modeling of the process. The inaccuracy of mathematical modeling degrades the
control performance of the continuous stirred tank reactor (CSTR), which shows
nonlinearity to some extent. It is very necessary to attain desired temperature within
a specied period of time to avoid overshoot and absolute error, with better temperature tracking capability, else the process is disturbed in the nonlinear CSTR
system. This paper studies the output (temperature) tracking and disturbance
rejection problem of nonlinear CSTR control systems with uncertainties via classical control PID, cascade control, and hybrid intelligent controller that includes
FLC, adaptive control, and adaptive neuro-fuzzy inference system (ANFIS). This
paper evaluates change in an adaptive controller response with varying adaptive
gain. It has been observed that OLTF of CSTR is stable, and adaptive controller is
best suitable for temperature control for ISE, and also has much better temperature
tracking capability. Adaptive controller and ANFIS both have observed zero
overshoot.

Keywords CSTR Nonlinearity PID controller


controller MIT rule FLC ANFIS

 Cascade controller  Adaptive

Nomenclature

V
F
CA
rA

Density of the material in the system lb/ft3


Total volume of the system ft3
Volumetric flow rate of the system ft3/h
Molar concentration (moles/volume) of component A in the system
Reaction rate per unit volume component A in the system

O. Prakash Verma  G. Manik (&)


Department of Polymer and Process Engineering, Indian Institute
of Technology Roorkee, Saharanpur Campus, Saharanpur, India
e-mail: manikfpt@iitr.ac.in
S. Kumar
Department of Electrical and Electronics Engineering, Graphic Era University,
Dehradun, India
Springer India 2015
K.N. Das et al. (eds.), Proceedings of Fourth International Conference on Soft
Computing for Problem Solving, Advances in Intelligent Systems and Computing 336,
DOI 10.1007/978-81-322-2220-0_9

103

104

Q
U
Tst, Tj
AH
DH

O. Prakash Verma et al.

Amount of heat exchanged between the system and its surrounding per
unit time
Over all heat transfer coefcient
Temperature of the steam and jacket, respectively
Total area of heat transfer
Heat of reaction at temperature T

1 Introduction
Temperature control is one of the most critically limiting factors in the production
operations of process industries, for example, production of propylene glycol form
hydrolysis of propylene oxide is an exothermic reaction [1] requiring careful
temperature control, due to its high nonlinear dynamics, of continuous stirred tank
reactor (CSTR), controlling its temperature has been a challenging task, for process
automation engineers. The overshoot and undershoot are undesired which badly
affect the nal product, if the temperature is out of the given range. Therefore, it is
necessary for a control engineer to reach steady state at given desired set point
temperature range quickly and also to minimize the overshoot. Since, the CSTR
possesses nonlinearity and its response changes in an unpredictable manner, it is not
easy to control the parameters accurately.
A design methodology is proposed for the analysis and synthesis of robust linear
controllers for a nonlinear CSTR [2]. The achievement of robust stability and robust
performances is guaranteed when the operating regions are dened in phase plane.
To overcome the difculty of achieving zero overshoot, zero absolute error, and
better temperature tracking capability, the intelligent controllers using FLC and
Adaptive Neuro-fuzzy Inference System (ANFIS) have been proposed for water
bath temperature control system [3]. As compared to conventional controller, FLC
and ANFIS produce a stable control signal. It has been found to exhibit better
temperature tracking capability with almost zero overshoot and minimum absolute
error. Model reference adaptive control (MRAC) [47] has been designed previously for second-order system using MIT rule scheme or gradient method to
overcome the variation in process dynamics due to its nonlinear nature and changes
in environmental condition with disturbances variation.
One of the major disadvantages of non-adaptive control is it cannot cope the
variation in the parameters of the process. When the plant undergoes transitions or
exhibits nonlinear behavior and when the structure of the plant is not known, then
adaptive scheme has been found to show efcient control [8]. Adaptive tracking
control is considered for general nonlinear systems using multilayer neural network
(MNN) [9] that is used to realize feedback linearization. This technique ensures the
stability of closed-loop system, and the system output tracks a set point signal while
tracking error converges to neighborhood of zero.

Analysis of Hybrid Temperature Control for Nonlinear Continuous

105

This paper demonstrates a comparative analysis of controller efciency among


conventional controller (PID), advanced controller (Cascade), and intelligent controllers (Fuzzy, MRAC, and ANFIS) for controlling the temperature of nonlinear
CSTR and discusses each.

2 Mathematical Modeling
Chemical reactors often have signicant heat effects [10], so it is important to be
able to add or remove heat from them. In a jacketed CSTR, the heat is added or
removed by virtue of the temperature difference between a jacket fluid and the
reactor fluid. An irreversible exothermic reaction A ! B DH occurs in a CSTR.
The heat of reaction is removed by the coolant medium that flows through a jacket
around the reactor as shown in Fig. 1. Consequently, when the CSTR is at steady
state, the heat produced by the reaction should be equal to the heat removed by the
coolant. The objective is to keep the effluent temperature T at a desired value Ts.
The operation of CSTR is disturbed by external factor, such as changes in the feed
flow rate and inlet temperature (Fi and Ti). At steady state, there is no need to
supervise and to control the system. But, this is not practically true, since Fi and Ti
are frequently changing. And, therefore, some form of control action [11] will be
needed to alleviate the impact of the changing of the disturbance and keep the
temperature T at the desired value. During the process, (i) mixing in the reactor is
perfect; liquid density (q) and the heat capacity Cp are constant (ii) To remove the
exothermic heat, the coolant is introduced into jacket; the reactor is assumed to be
perfectly insulated (iii) Coolant is perfectly mixed in the jacket (iv) volume of
reactor and jacket is to constant.
Total mass balance across reactor gives,
dV
Fi  F
dt

Reactant (

Coolant (

Product (

Fig. 1 Conguration of a perfectly mixed CSTR with a cooling jacket and the process parameters

106

O. Prakash Verma et al.

Balance on component A yields,


CA

dV
dCA
V
Fi CAi  rA V  FCA
dt
dt

Using Eq. (1),


dCA Fi
CAi  CA  rA
dt
V

Since, rA = rate of disappearance of A and


rA K0 eE=RT CA

Applying energy balance across the reactor,


dVqCp T
Fi qCp Ti  FqCp T  Q DH rA V
dt

dT
dV
Q
V
T
Fi Ti  FT 
DH rA
dt
dt
qCp
qCp

From Eq. (1), we get


dT Fi
Q
1
Ti  T 
DH rA
dt
qVCp
qCp
V

Now, using Eq. (4),


dT Fi
Q
K0 eE=RT CA
Ti  T 
DH
dt
qVCp
V
qCp

The state variable form that includes the effect of cooling jacket on CSTR
response is given as,

 Fi
dCA
f1 CA ; T; Tj CAi  CA  K0 eE=RT CA
dt
V

 Fi
UA T  Tj
dT
K0 eE=RT CA
f1 CA ; T; Tj Ti  T 
DH
dt
VqCp
V
qCp

9
10

Analysis of Hybrid Temperature Control for Nonlinear Continuous

107

Similarly, the equation of jacket temperature is given as



 Fj 
 UAT  Tj
dTj
f3 CA ; T; Tj
Tj in  Tj
Vj qj Cpj
dt
Vj

11

Q UAT  Tj

12

where,

Since, the above sets of Eqs. (9)(10) are nonlinear, on linearizing [12],
dCA 1
CAi  CA  K0 eE=RT CA ;
s
dt

13

dT 1
Q
Ti  T 
SK0 eE=RT CA
dt
s
qVCp

14

where, s V=Fi

DH
: The nonlinear term present in the modeling equation, eE=RT CA ;
where, S  qC
p

is linearized close to the operating point, T0 and CA0 as


" 
#
@ eE=RT CA
e
CA e
CA0
T  T0
@T
T
;C
0 A0
" 
#
@ eE=RT CA
15

CA  CA0
@CA
T0 ;CA0


E E=RT0
E=RT0
eE=RT CA eE=RT0 CA0
e
C
CA  CA0
A 0 T  T0 e
RT20
E=RT

E=RT0

16
The linearized model,
dCA 1
E E=RT0
CAi  CA  K0 eE=RT0 CA0
e
CA0 T  T0
s
dt
RT20
eE=RT0 CA  CA0 

17

dT 1
Q
E E=RT0
Ti  T 
SK0 eE=RT0 CA0
e
T  T0
dt s
VqCp
RT20
eE=RT0 CA  CA0 

18

108

O. Prakash Verma et al.

At steady state condition,


dCA
1
0 CAi0  CA0  K0 eE=RT0 CA0
s
dt

19

dT0
1
Q0
0 Ti0  T0 
SK0 eE=RT0 CA0
s
dt
VqCp

20

Equations (19)(20) represent the modeling equation. Linearized model (in


terms of deviation variables for concentration) yields,
d
1
E
CA  CA0 CAi  CAi0  CA  CA0   K0 2 eE=RT0 CA0 T  T0
dt
s
RT0
 K0 eE=RT0 CA  CA0

21

i
d 0
1h
E
CA CA0 i  CA0  K0 2 eE=RT0 CA0 T 0  K0 eE=RT0 CA0
dt
s
RT0

22

where the deviation variables are given as,


CA0 i CAi  CAi0 ;

CA0 CA  CA0 ;

T 0 T  T0

Likewise, expressing the linearized model (in terms of deviation variable for
temperature) as
d
1
Q  Q0
T  T0 Ti  Tio  T  T0  
dt
s
VqCp


E E=RT0
E=RT0
SK0
e
C

T

T

C

C

A0
0
A
A0
RT20



dT 0 1  0
Q
E E=RT0
0
E=RT0 0
Ti  T 0 
SK0
e
C
T

e
C
A0
A
s
VqCp
dt
RT20

23
24

Therefore, steady-state solution is given as



 dCA
Fi
f1 CA ; T; Tj
0 CAi  CAi  K0 eE=RT CA
dt
V

25


 dT
UAT  Tj
Fi
K0 eE=RT CA
0 Ti  T 
DH
f2 CA ; T; Tj
dt
VqCp
V
qCp
26

Analysis of Hybrid Temperature Control for Nonlinear Continuous

109


 dTj
UAT  Tj
Fi
0 Tj in  Tj 
f3 CA ; T; Tj
Vj qj Cpj
dt
V

27

The above equations are solved and the matrices, A, B, C, and D of state
equation are found. The state and input variables represented in deviation variable
form are as follows:
2

X
2
6
A4
"
B

X1
X2

CA  CAS
;
T  Ts

Y T  Ts ;

U1

E=RT0

E=RT0

 FVi

 K0 e

DH
 qC
p

K0 e
#

0
UAH ;

K0 E

RT20

C 0 1;

E=RT0

Tj  Tjs

7
6U 7 6T  T
is
7
6 27 6 i
U6 76
7
4 U3 5 4 CAi  CA is 5
U4

 FVi

CA0

DHK0 E E=RT0
e
CA0
RT20 qCp

F  Fs

7
5;
UA
H

VqCp

D 0

VqCp

From reactor parameters values used previously [1316] and shown in Table 1,
the state matrices determined are given as:

A


7:991 0:0137
;
2922:9 4:5564

Table 1 Parameters values


of proposed design of CSTR
system


B


0
;
1:4582

C 0

1 ;

D 0

Reactor parameters

Parameter values

E (Btu/lb mol)
Ko (h1)
DH (Btu/lb mol)
UAH (Btu/h F)
qcp (Btu/ft3 F)
R (Btu/lb mol F)
Fi/V (h1)
CAi (lb mol/ft3)
Ti (F)
CAo (lb mol/ft3)
To (F or R)
Tji (F)
Tji (F)
jCpj (Btu/ft3 F)

32,400
16.96 1012
39,000
6,600
53.25
1.987
4
0.132
60
0.066
101.1 or 560.77
80
0
55.5

110

O. Prakash Verma et al.

The transfer function Gp s of the system based on above matrices obtained is


found as
G p s

T
1:458s 11:65

Tj s2 3:434s 3:557

28

The evaluated Eigen values are as follows:


Y 1:7173 i0:8273

and

 1:7173  i0:8273

Complex Eigen value with negative real part ensures the stability of CSTR.

3 Control Topologies
The controllers are classied as classical and intelligent. P, PI, PID, and Otto-Smith
are examples of classical controllers that require a prior and accurate knowledge of
mathematical model of the process in order to design an efcient controller. Any
inaccuracy in mathematical modeling results in inaccurate calculation of control
parameters and hence poor controlling action. Intelligent controllers, overcome this
disadvantage, by using a new approach to the controller design that does not require
knowledge of mathematical model of the process. FLC, MRAC, Articial Neural
Network (ANN), ANFIS, and Model Predictive Control (MPC) are some examples
of intelligent controllers.

3.1 PID Controller


A PID controller calculates an error value as the difference between a measured
process variable and a desired set point. The controller, as shown in Fig. 2, attempts
to minimize the error by adjusting the process control inputs. Control signal of PID
controller is given as follows.

Fig. 2 Conguration of a block diagram of a conventional PID controller

Analysis of Hybrid Temperature Control for Nonlinear Continuous

111

This may also be written as


Zt
ut Kp et Ki

esds Kd

det
dt

29

The transfer function, Gc s; of PID controller can be represented as




1
Gc s Kp 1
sd s
si s

30

In discrete form, transfer function of PID controller may be represented as given


below



1
ek  ek  1
Uk K ek ek ek  1    DT Td
Ti
DT
31
The Eq. (29) is used for the realization in FLC.
The selection of a controller type and its parameters Kp ; Ki ; Kd is related to the
model of process to be controlled. The adjustment and selection of controller
parameters to achieve satisfactory control is essentially an optimization problem.
Ziegler and Nichols (Z-N) developed a closed-loop method for parameters tuning
[17]. In this paper, the parameters of PID have been tuned by Z-N tuning methods
and are determined as, Kp 0:1; Ki 0:116 and Kd 0:020:

3.2 Cascade Controller


The response of simple feedback control has been improved by the changes in the
coolant temperature [18] by measuring Tj and taking control action before its effect
can be felt by the reacting mixture. If Tj rises, the flow rate of coolant is increased to
remove the same amount of heat and vice versa. Therefore, two control loops have
been involved using two different measurements, T and Tj , that share a common
variable, Fj : This is illustrated in Fig. 3.
The state space model matrices for cascaded controller [19] reduce to
2

7:991 0:0137
A 4 2922:9 4:5564
0 
4:7482
0
D
0

3
0
1:4582 5;
5:8977

3
0
5;
B 40
3:2558

0
C
0

1
0


0
;
1

112

O. Prakash Verma et al.

TC 1

TC 2

Primary controller

Jacket
Dynamics

Secondary
Controller

CSTR
dynamics

TM 2

TM 1
Fig. 3 Generalized block diagram of a cascade controller

The transfer function for primary controller, Gp1 s; and secondary controller,
Gp2 s; are found as
Gp1 s

T
4:747s  37:937

Tj s3 9:332s2 16:97s  33:9

32

Gp2 s

Tj
3:256s2  11:18s  11:83
3
Fj s 9:332s2 16:97s  33:9

33

The primary and secondary controllers have also been tuned using Z-N methods
and parameters determined as: Kp1 655:007; Ki1 4; 265:269; Kd1 20:197; for
primary loop, and Kp2 0:092; Ki2 0:022; Kd2 0:006; for secondary loop.

3.3 Fuzzy Logic Controller


Fuzzy inference system (FIS) is proposed as a highly efcient method for controlling the settling time, delay time, and overshoot parameters for nonlinear CSTR
systems. Fuzzy logic is a way to make machines more intelligent, enabling them to
reason in a fuzzy manner like humans [20].
The FLC makes a nonlinear mapping between the input and the output using
membership functions and linguistic rules (normally in the form If_Then_). The
CSTR system uses two input variables error (e), change in error (de), and one output
variable (Y). The computational structure of fuzzy logic control scheme is composed
of fuzzication, inference engine, and de-fuzzication as shown in Fig. 4. Fuzzication converts a crisp value (real value) into a member of fuzzy sets. FIS formulating the mapping from a given input to an output using fuzzy logic, while defuzzication converts the fuzzy output determined by the inference system to a crisp

Analysis of Hybrid Temperature Control for Nonlinear Continuous


Knowledge
base
Inputs

Scaling factors,
normalization

113

Rule base

Fuzzifification

De-fuzzifification,
denormalization

Inference

Output-scaling factors,
normalization

CSTR

Sensors

Fig. 4 Block diagram of fuzzy logic control (FLC)

value. In FLC, the membership functions are utilized to nd the degree of membership of the element in a given set. The fuzzy control rules have the form [21]:
R1 :
R2 :
..
.
Rn :

If X is A1
If X is A2
..
.

and
and
..
.

If X is An

and

Y is B1 ;
Y is B2 ;
..
.
Y is Bn ;

then
then
..
.

Z is C1
Z is C2
..
.

then

Z is Cn

where X, Y, and Z are linguistic variables that represent two process state variables
and one control variable, respectively;
Ai ; Bi and Ci are linguistic values of the linguistic variable X, Y, and Z in the
universe of discourse U, V, and W, respectively, with i = 1, 2, 3 n; AND
operator, MAMDANI-type FIS and 11 rule base has been utilized here and shown
in Fig. 5. Centroid method has been used for controlling the fuzzy inference control
action to real value. The gain for error (e) and derivative of error (de) has been
tuned to 0.5 and 0.5, and the gain for FLC controller taken as 150.

Fig. 5 FIS editor of fuzzy control model for a CSTR system

114

O. Prakash Verma et al.

3.4 Adaptive Control


Adaptive control is a combination of a parameter estimator, which generates
parameter estimates online, with a control law in order to control classes of plants
whose parameters are completely unknown and/or could change with time in an
unpredictable manner [22]. An adaptive control can be used when the plant
undergoes the transitions and show nonlinearity and also plant structure is not
known exactly [23]. This control can adjust the parameter automatically to compensate for variations in the characteristics of the CSTR. An adaptive control for
CSTR consists of two loops: a normal feedback loop and the parameter adjustment
loop as shown in Fig. 6.

3.4.1 Model Reference Adaptive Control (MRAC)


MRAC is also regarded as adaptive servo system [2426] in which, the desired
performance is expressed in terms of a reference model, which gives the desired
response to a command signal. The parameters are changed on the basis of feedback
from the error, e Yp  Ym ; where Yp is the plant output and Ym ; the model output.
The mechanism for adjusting the parameters in a MRAC can be obtained using a
gradient method.

3.4.2 MIT Rule


For a closed-loop system, the controller has one adjustable parameter, h; dependent
on the minimized square of the prediction error, e. The desired closed-loop system
response is specied by a model whose output is Ym :
And,
e Yp  Ym

34

Reference
Model ( )
Adjustment
Mechanism

Controller
Parameters
Command
Signal

CSTR

Controller
Control Signal

Fig. 6 Block diagram of MRAC applied to a CSTR

O/P

Analysis of Hybrid Temperature Control for Nonlinear Continuous

115

Adjusting the parameters in such a way that the loss function, Jh 12 e2 is


minimized. To make J small, the parameter h is changed in the direction of the
negative gradient of J, i.e.,
dh
@J
@e
c
ce ;
dt
@h
@h

35

@e
the sensitivity derivative of the system. It gives
where c is the adaptation gain and @h
an idea about how the error is influenced by the adjustable parameters. The transfer
function of the process is then represented as given below

Yp
K
2
s a1 s a2
u

36

where K, a1 and a2 are the process parameters and control signal given as
u h1 uc  h2 Yp

37

where h1 ; h2 and uc are adjustable parameters and command signal, respectively.


Using Eqs. (36) and (37),
Yp
Kh1

uc s2 a1 s a2 Kh2

38

Ym
Km
2
uc
s A1 s A2

39

where Km ; A1 and A2 are the reference model parameters. And, for a perfect model
(Fig. 7),
s2 a1 s a2 Kh2 s2 A1 s A2 ;

40

and
@e
Kuc

@h1 s2 A1 s A2

and

KYp
@e
 2
@h2
s A1 s A2

41

Therefore,
dh1
@e
Kuc
c
e c 2
e
@h1
dt
s A1 s A2

42

KYp
dh2
@e
c
ec 2
e
@h2
dt
s A1 s A2

43

116

O. Prakash Verma et al.

Fig. 7 Block diagram of a rst-order MRAC

Then,
h1 



c
Kuc
e
s s2 A1 s A2

and

h2



KYp
c
e
s s2 A1 s A2

44

Second-order model may be represented as:


Gm s

s2

x2n
2exn s x2n

45

The transfer function model of the CSTR temperature process from state space
matrix values is determined to be
Gp s

T
1:458s 11:65
k
2
2
Tj s 3:434s 3:557 s 3:434s 3:557

46

For the temperature control, a maximum overshoot (Mp) is 2 % and a settling


time (ts ) less than 3 min are selected (assuming tolerance band (TB) is 2 %) [27].
For this second-order system,

 v
ln Mp =100 u
1
u
n
h
i2
t
p
lnMp =100
1
p

and

xn 4=n ts

47

Analysis of Hybrid Temperature Control for Nonlinear Continuous

117

where n is the damping ratio, and xn is the natural frequency of second-order


system. For optimum performance characteristics, n 0:707 and xn 1:885 rad/s
has been chosen.
The transfer function of the model is given as
Gm s

3:55
:
s2 2:66s 3:55

48

3.5 Adaptive Neuro-fuzzy Inference System (ANFIS)


A neuro-fuzzy (ANFIS) system [28] is a combination of neural network and fuzzy
systems in such a way that neural network is used to determine the parameters of
fuzzy system. This is illustrated in Fig. 8.
The neuro-fuzzy system with the learning capability of neural network and with
the advantages of the rule-based fuzzy system can improve the performance signicantly and can provide a mechanism to incorporate past observations into the
classication process. The fuzzy inference system under consideration has two
inputs x and y and one output z. For a rst-order Sugeno fuzzy model, a rule set is
the followed:
RULE 1: If x is A1 and y is B1, then f1 p1 x q1 y r1 ;
RULE 2: If x is A2 and y is B2, then f2 p2 x q2 y r2 ;
Figure 8 illustrates the reasoning mechanism for this Sugeno model with the
corresponding equivalent ANFIS architecture, where nodes of the same layer have

Fig. 8 Architecture of ANFIS

118

O. Prakash Verma et al.

similar functions. First layer is the premise parameters, calculated by parameterized


membership function. Output of second layer is the product of two incoming signals, third layer output nds the ring strengths of each node, the output of layer 4
is the product of normalized ring weight with node function, and layer 5 overall
output of the ANFIS model can be written as follows:
P2
w1
w2
wi fi
 1 f1 w
 2 f2 Pi1
f
f1
f2 w
2
w1 w2
w1 w2
i1 wi

49

 2 are normalized ring strengths.


 1 and w
where w
The overall output expressed in Eq. (49) is a linear combination of the consequent
parameters when the values of premises parameters at level 1 are xed. Hybrid
learning algorithm combines the gradient descent method and the least square
estimator (LSE) method which used to rene these parameters. In implementing the
training scheme, a sequence of random input signals in ANFIS limited between 0
and 5 is injected directly to the CSTR and the gain for ANFIS controller is 4.95.

4 Results and Discussion


The outputs of PID, Cascade, FLC, ANFIS, and adaptive controller for a step change
in set point are shown in Fig. 9 and their performances are presented in Table 2 for
comparison. The goal is to screen and design a controller that will control the
temperature of CSTR system most efciently and follow a reference prole.
Results show that adaptive and cascade controllers have comparatively smaller
delay time (td ) of 21.4 and 19.0 min, while PID, Fuzzy, and ANFIS show much
higher values of 141, 199.3, and 199.5 min. It means adaptive and cascade controller has faster response capabilities. The faster response is also exhibited through
rise times of cascade control exhibits the low rise time tr of 32.6 and 22.7 min,
respectively, for adaptive and cascade controllers verses other controlling actions.
Adaptive and cascade controlling actions also improve tracking of set point by
showing low-error ISE values of 0.000157 and 0.001686 compared to all other
controllers. Cascade and PID, however, have been found to suffer from high
overshoots of 29 and 87 % versus (which is not desirable for CSTR system) other
controlling actions for which the overshoot was negligible or zero. The tracking of
set point temperature was found to be best with adaptive controller with lowest ISE
of 0.000157 compared to relative higher values of 0.02279 for FLC. The temperature tracking capability of ANFIS was, however, found to be poor with higher
value of ISE *0.23. In view of faster response (rise, delay, and settling times)
though both adaptive and cascade control perform better, however, due to higher
overshoots with cascade control, adaptive controlling action is proposed as the most
efcient controlling actions among all control techniques considered.

Analysis of Hybrid Temperature Control for Nonlinear Continuous

(a)

(b)
200

200

Temperature

100
50

Set-point
Cascade control response

Temperature

Set-point
PID control response

150

150
100
50
0
0

0
0

200

400

600

800

1000 1200 1400 1600 1800

Time

(c)

Temperature

Temperature

400

600

800

1000 1200 1400 1600 1800

Time

200
Set-point
FLC response

150
100
50
0
0

200

(d)

200

200

400

600

800

(d)

ANFIS response
Set-point

150
100
50
0
0

1000 1200 1400 1600 1800

200

400

600

(e)

Time

800

1000 1200 1400 1600 1800

Time

120

200

100

Set-point
Adaptive control response

Temperature

150

Temperature

119

100
50

=3
=2.7
=2.5

80
60
40

=0.7
=1.0
=1.8
=2.0

20
0

0
0

10

20

30

40

50

60

70

80

90

100

-20
0

10

20

Time

30

40

50

60

70

Time

Fig. 9 Illustration of output temperature response of a PID controller, b cascade controller,


c fuzzy logic controller, d ANFIS controller, and e adaptive controller, f adaptive controller with
varying adaptation gain, c
Table 2 Comparisons of control parameters for different controlling actions
Controlling
action

td (delay time
in min)

tr (rise time
in min)

ts (settling time
in min)

Mp (2 %
TB)

ISE

PID
Cascade
Fuzzy
ANFIS
Adaptive

141
19
199.3
199.5
21.4

175.5
22.7
429
430
32.6

174
22.4
372
373
30.2

87
29
0
0
0

0.168
0.001686
0.02279
0.2262
0.0001568

Adaptation gain ensures the convergence rate and is illustrated in Fig. 9f which
shows the parameter estimates for different values of the adaptation gain, c: The
convergence rate increases with increasing values of c: When c ranges between 1.8
and 2.0, an optimum response is observed, however, for larger values of c [ 2:0;
overshoot appears and below 1.8, the response becomes sluggish.

120

O. Prakash Verma et al.

5 Conclusion
The present study observes that for a nonlinear CSTR system, adaptive controller
gives the best controlling performance among the set of controlling actions studied
(PID, cascade, fuzzy, ANFIS, and adaptive). Adaptive controller exhibits a zero
overshoot, better temperature tracking capabilities (as shown by very low ISE
values), and a much quicker response (very small rise, delay, and settling times)
than other controllers.

References
1. Zuhwar, Z.F.: The control of non isotherma CSTR using different controller strategies. Iraqi
J. Chem. Pet. Eng. 13(3), 3545 (2012)
2. Doyle III, F.J., Packard, A.K., Morari, M.: Robust controller design for a nonlinear CSTR.
Chem. Eng. Sci. 44(9), 19291947 (1989)
3. Verma, O.P., Singla, R., Kumar, R.: Intelligent temperature controller for water bath system,
WASET. Int. J. Comp. Sci. Eng. 6(9), 12321238 (2012)
4. Jain, P., Nigam, M.J.: Design of MRAC using modied MIT rule for a second order system.
Adv. Electron. Electr. Eng. 3(4), 477484 (2013)
5. Anbu, S., Jaya, N.: Design of adaptive controller based on Lyapunov stability for a CSTR. Int.
J. Electr. Electron. Sci. Eng. 8(1), 183186 (2014)
6. Zhang, T., Guay, M.: Adaptive nonlinear control of continuously stirred tank reactor systems.
In: Proceeding of American Control Conference Arlington, pp. 12741279 (2001)
7. Pankaj, S., Kumar, J.S.: Comparative analysis of MIT rule and Lyapunov rule in model
reference adaptive control scheme. Innovative Syst. Design Eng. 2(4), 154162 (2011)
8. Abatneh, Y., Sahu, O.: Adaptive control design for a Mimo chemical reactor. Autom. Control
Intel. Syst. 1(3), 6470 (2013)
9. Ge, S.S., Hang, C.C., Zhang, T.: Nonlinear adaptive control using neural networks and its
application to CSTR systems. J. Process Control 9, 313323 (1998)
10. Raju, S.S., Siddiqa, M.A., Kiran, T.K.S.R., Viswanath, M.: Control of concentration in CSTR
using DMC and conventional PID based on relay feedback system. Int. J. Eng. Sci. Tech.
IJSET 5(4), 925932 (2013)
11. San, K.Y., Stephanopoulos, G.: Optimal control policy for substrate-inhibited kinetics with
enzyme deactivation in an isothermal CSTR. AIChE J. 29, 417424 (1983)
12. Bequette, B.W.: Nonlinear control of chemical processes: a review. Ind. Eng. Chem. Res. 30,
13911413 (1991)
13. Elisante, E., Rangaiah, G.P., Palanki, S.: Robust controller synthesis for multivariable
nonlinear systems with unmeasured disturbances. Chem. Eng. Sci. 59, 977986 (2004)
14. Perez, M., Albertos, P.: Self-oscillating and chaotic behaviour of a PI-controlled CSTR with
control valve saturation. J. Process Control 14, 5159 (2004)
15. Bequette, B.W.: Process Control: Modeling, Analysis and Simulation, Module 8, 1st edn,
pp. 641657. Prentice Hall, Upper Saddle River (2003)
16. Bequette, B.W.: Process Dynamics: Modeling, Analysis and Simulation. Series in the Physical
and Chemical Engineering Sciences, Module 7, 1st edn, pp. 506524. Prentice Hall, Upper
Saddle River (1998)
17. Coughanowr, D.R.: Process Systems Analysis and Control. Chemical Engineering Series, 2nd
edn, pp. 282287. McGraw-Hill International Editions (1991)

Analysis of Hybrid Temperature Control for Nonlinear Continuous

121

18. Stephanopoulos, G.: Chemical Process Control: A Introduction to Theory and Practice, 1st
edn, pp. 395402. PHI Learning, Upper Saddle River (1984)
19. Vasickaninova, A., Bakosova, M.: Cascade fuzzy control of a chemical reactor. In:
Proceedings of 15th International Conference Process Control, Strbske Pleso, Slovakia,
pp. 175-1175-5 (2005)
20. Galluzo, M., Cosenza, B.: Control of a non-isothermal CSTR by type-2 fuzzy logic
controllers, pp. 295302. Springer, Berlin WILF (2009)
21. Gizi, A.J.H.A., Mustafa, M.W., Alsaedi, M.A., Zreen, N.: Fuzzy control system review. Int.
J. Sci. Eng. Res. 4(1), 18 (2013)
22. Ioannou, P., Fidan, B.: Adaptive Control Tutorial, Chapter 1, pp. 111. Society for Industrial
and Applied Mathematics, Philadelphia (2006)
23. Kumar, N., Khanduja, N.: Mathematical modelling and simulation of CSTR using MIT rule.
In: IEEE 5th India International Conference on IICPE, Delhi, India, pp. 15 (2012)
24. Astrom, K.J., Wittenmark, B.: Adaptive Control, Chapter 5, 2nd edn, pp. 185198. Pearson
Publication (1995)
25. Astrom, K.J.: Adaptive Feedback Control. Proc. IEEE 75(2), 185217 (1987)
26. Dostal, P., Bobal, V., Gazdos, F.: Simulation of nonlinear adaptive control of a continuous
stirred tank reactor. Int. J. Math. Comp. Simul. 5(4), 370377 (2011)
27. Prabhu, K., Bhaskaran, V.M.: Optimization of a control loop using adaptive method. IJEIT 1
(3), 133138 (2012)
28. Jang, J.S.R., Sun, C.T., Mizutani, E.: Neuro-fuzzy and Soft Computing: A Computational
Approach to Learning and Machine Intelligence, Chapter 12, pp. 335340. Prentice Hall,
Upper Saddle River (1997)

You might also like