Nice Journal of Emerging Technologies Vol.:6 No.

:1 June:2011

COMPARITIVE ANALYSIS OF MODEL BASED CONTROLLER

DESIGN FOR A CHEMICAL PROCESS

G.Glandevadhas,S.Pushpakumar

and IMC is presented. The approach is used for

Abstract—In the chemical or biochemical design of a robust PID and IMC for the CSTR.
industry most processes are modeled by The reactor has three uncertain parameters: the
nonlinear equations. It is of a great significance reaction enthalpy, the reaction rate constant and
to design high-performance nonlinear the overall heat transfer coefficient. The control
controllers for efficient control of these input is volumetric flow rate of the coolant and
nonlinear processes to achieve closed-loop the controlled output is the temperature of the
system’s stability and high performance. reacting mixture.
However, there are many difficulties which
hinder the design of such controllers due PID controller and linear model predictive
mainly to the process nonlinearity. In this controller are the two most popular control
work, comprehensive design procedures based schemes that have been widely implemented
on robust control have been proposed to throughout the chemical process industries for the
efficiently deal with the design of controllers past two decades. However, control of nonlinear
for nonlinear systems. Since all the design system using above linear control schemes
procedures proposed in this work rely strongly doesn’t give satisfactory performance at all
on the process model, the first difficulty operating points. The introduction of powerful
addressed in this work is the identification of a nonlinear control schemes[7,10] and neural
relatively simple model of the nonlinear adaptive control, the proposed control strategy,
processes under study. The nonlinearity of the because it offers advantages such as simple design
processes makes it often difficult to obtain a and low computational complexity. The main
first-principles model which can be used for contributions of this paper are as follows; firstly,
analysis and design of the controller the nonlinear system is represented as a family of
local linear models. Secondly, local PID
I. INTRODUCTION controllers has been used to control the nonlinear
process, and finally a nonlinear model predictive
Chemical reactors are one of the most control[1,2] scheme using the family of local
important plants in chemical industry. Their linear models has been proposed to control
operation, however, is corrupted with various nonlinear process.
uncertainties. Some of them arise from varying or
not exactly known parameters, as e.g. reaction
rate constants, heat transfer coefficients. In other II. PROCESS DESCRIPTION
cases, operating points of reactors vary or reactor Chemical reactions in a reactor [9] are either
dynamics is affected by various changes of exothermic (release energy) or endothermic
parameters or even instability of closed loop (require energy input) and therefore require that
control systems. Application of robust energy either be removed or added to the reactor
control approach can be one of ways overcoming for a constant temperature to be maintained.
all these problems. In this project, a simple
method for design of robust PID controllers[9]

63
Nice Journal of Emerging Technologies Vol.:6 No.:1 June:2011

Figure 1 shows the schematic of the CSTR H = 5.3kJ/kg

process. In the CSTR process model under Q = 224.1kJ/sec
discussion, an irreversible exothermic reaction R = 0.00831kJ/mol K
takes place. The heat of the reaction is removed
by a coolant medium that flows through a jacket VdcA / dt FcAin FcA Vke( E / RT )cA
around the reactor. A fluid stream A is fed to the VdcA / dt f 1( F , cAin) f 2( F , cA) f 3(cA, T )
reactor. A catalyst is placed inside the reactor.
The fluid inside the reactor is perfectly mixed and (3)
sent out through the exit valve. The jacket Now be used in the general model differential
surrounding the reactor also has feed and exit equation as,
dx / dt f ( x) f ( x0) (df / dx) x
streams. The jacket is assumed to be perfectly
mixed and at a lower temperature than the reactor. (4)
Using eqn (3), eqn (4) can be linearized,
Vd ( cA) / dt ( f 1 / F ) F ( f 1 / cAin) cAin ( f 2 / F) F
( f 2 / cA) cA ( f 3 / cA) cA ( f 3 / T ) T

This can be written as,

Vd ( cA) / dt cAin F F 0 cAin cA0 F F 0 cA
Vke( E / RT ) cA Vke( E / RT 0)cA0( E / RTO 2) T
Vke ( E / RTo)CAo E / RTo ^2 T
(5)
Rearranging terms and introducing the Laplace
operator results in:

Fig1: CSTR process CA ( K1 F /( cs 1)) ( K 2 CAin /( cs 1)) ( K 3 T /( cs 1))

(6)
MATHEMATICAL MODELLING With
The component balance for the reactor can be
given as, c V /( Fo Vke ( E / RTo ))
VdcA/ dt F cAin cA Vke( E / RT )cA K1 ((CAino CAo ) /( Fo Vke ( E / RTo )))
(1) K 2 Fo /( Fo Vke ( E / RTo ))
The energy balance by, K 3 (Vke ( E / RTo )CAo ( E / RTo ^ 2)) /( Fo Vke ( E / RTo )))
PVcpdT / dt FPcp(Tin T ) Vke( E / RT )cA H Q
(2) After substitution of the steady state values gains
of eqn.(3.6)we get into the time constant and
PARAMETERS USED process
V = 5m3
CA = 200.13kg/m3 c 250s, K1 3 *10^ 4, K 2 3.174 0.25, K 3
CAin = 800kg/m3 (7)
F = 0.005m3/s The second equation (energy balance) of the
K = 18.75s-1 reactor model can be rewritten as:
E = 30kJ/mol
T = 413K PVCpdT / dt FPCp (Tin T ) Vke ( E / RTCA H
Tin = 353K Q)
P = 800kg/m3 FPCpTin FpCpT Vke ( E / RTCA H Q)
Cp = 1.0kJ/kg k f 1( F , Tin ) f 2( F , T ) f 3(CA, T ) f 4 * (Q)

64
Nice Journal of Emerging Technologies Vol.:6 No.:1 June:2011

Using eqn. (3), eqn. (4) can be written as: Eqn. (10) while setting changes in CAin and Tin to
zero:
PVCpd ( T ) / dt f 1/ F F f 1/ Tin Tin f 2/ F F f 2/ T F T (K 4 F /( Ts 1)) (K 6 /( Ts 1))[(K1 F /( cs 1)) (K 3 T /( cs 1))]
f 3 / T CA f 3/ T T Q
(15)
(8) This equation can be rearranged to:
This can be rewritten as: T / F (( K 4 K 6 K1) /(1 K 6 K 3)) *
VPCpd ( T ) / dt PCp (Tin T ) F FPCp Tin
{( K 4 c /( K 4 k 6 K1)) s 1} /{( T c ?(1 K 3K 6)) s ^ 2
FPCp T Vke ( E / RT ) H CA VkE _(E / RT )
(( T c) /(1 K 3K 6)) s 1}
CA H E / RTo ^ 2 T Q
(16)
After substituting values for the time constants
(9) and gains, Eqn. (3.13) and Eqn.(3.14) can be
Rearranging terms and introducing the Laplace written as:
transform operator results in,
T ( K 4 F /( Ts 1)) ( K 5 Tin /( Ts 1)) CA / F (120 .02 s 0.262 ) /( s ^ 2 0.0049 s 3.921e 6) .
( K 6 CA /( Ts 1)) T / F ( 20.86 s 0.0456 ) /( s ^2 0.0049 s 3.921e 6) .
(10) The above two Equations are the transfer function
With of concentration and temperature of the CSTR
models.
T VPCp /( FoPCp Vke ( E / RTo)CAo H ( E / RTo ^ 2))
III. MODEL PREDICTIVE CONTROLLER
K 4 PCp (Tino To) /( FoPCp (Vke ( E / RTo)CAo H ( E / RTo ^ 2))
K 5 FoPCp (Vke ( E / RTo)CAo H ( E / RTo ^ 2))
The basic idea of this controller is predict
K 6 (Vke ( E / RTo) H ) /( FoPCp Vke ( E / RTo)CAo H ( E / RTo ^ 2)) the output. An important property of a model is its
predictive power that is it gives as the possibility
to predict future values of interesting variables. A
(11) natural use of a model for control design is to
Substitution of the steady state values in the time calculate expected future values of the controller
constant and process gains of eqn. (11) variables as a function of possible control actions
Results in: with this knowledge; it is possible to choose a
control action which is the best one according to
T 1091 .8s, K 4 1.31 *10 ^ 4, K 5 1.09, K 6 0.022 some criterion. More formally we can proceed as
(12) follows.
The response of the change in reactor outlet At time t compute or predict a number of
concentration CA to a change in reactor future outputs y (t k / t ), k 1,............, M . They
throughput F can now be obtained by combining will in general depend on future inputs
Eqns. (6) and Eqn. (10) while setting changes in u (t j ), j 0,1,.............N .
CAin and Tin to zero:
1. Apply u (t ) to the physical plant. Wait for the
CA (K1 F /( cs 1)) (K 3 /( cs 1)) (K 4 /( Ts 1)) (K 6 CA /( Ts 1)) . next sampling instant t 1 and go to 1 .
(13) This is a very general and flexible method. It is of
This equation can be rearranged to: great value that it is easy to include in the
CA / F (( K1 K 3K 4) /(1 K 3K 6)) *
criterion realistic constraints on the magnitude
[( K1 T /( K1 K 3K 4)) s 1] /[( c T /(1 K 3K 6)) s ^ 2
and rate of change of the control variables. The
[( c T ) /(1 K 3K 6)) s 1] method is called Model Predictive Controller.
(14) MPC is a control strategy that uses an optimizer to
Then, the response of the change in reactor outlet solve for the control trajectory over a future time
temperature T to a change in reactor throughput F
can now be obtained by combining Eqn. (.6) and

65
Nice Journal of Emerging Technologies Vol.:6 No.:1 June:2011

horizon based on a dynamic model of the process. problem at each time step. The current values of
Predictive control based on linear models is the process states are used for calculating the
acceptable when the process operates at a single control action and the optimization process is
set point and the primary use of the controller is repeated at the next time horizon. A small time
the rejection of the disturbances. Linear MPC interval and a long prediction horizon are required
frequently results[6,7] in poor control to maintain closed-loop robustness. However, the
performances. In order to properly control these smaller the time interval is, the heavier the
plants, a nonlinear predictive control technology computational load. The model predictive control
is needed. MPC is an advanced method of generally formulates the optimization problem
process control that has been in use in the process over a finite prediction horizon to decrease the
industries such as chemical plants and oil computational load. In general, the nonlinear
refineries. It has a long history in the field of model-predictive control leads to a non-convex
control engineering. Three major aspects of model optimal control problem is a local optimum.
predictive controller are However, the model predictive control can
1. Design formulation. achieve a global optimum by using suboptimal
2. The ability of method to handle both soft model predictive control.
constraints and hard constraints in a The model predictive control has been
multivariable control system. used widely in the process industries because of
3. The ability to perform process online its many appealing features such as handling
optimization. multivariable systems with time delays. In
addition, the constraints on manipulated inputs,
states and output variables are explicitly handled
in the formulation of the optimization problem.
However, the closed-loop stability and feasibility
are major concerns in the model predictive
control. The local optimization[8] in a finite
horizon does not guarantee closed-loop stability.
Thus, the model predictive control formulates the
optimization problem with special constraints or
penalty terms based on a Lyapunov function to
Fig.2. Block diagram of MPC ensure the closed-loop stability. The stability is
guaranteed by imposing an equality or inequality
Model predictive control (MPC) refers to a wide constraint on the final state in the prediction
class of control algorithms that use an explicit horizon, adding a weight on the final state in the
process model to predict the behavior of a plant. objective function ,or using an infinite prediction
The most significant feature that distinguishes horizon with a finite control horizon. However,
MPC from other controllers is its long range the measurable process states, perfect process-
prediction concept. This concept enables MPC to model and high computational load to determine
perform current computations to account the the attraction domain boundaries of the linear
future dynamics, thus facilitating it to overcome controller are required.
the limitations of process dead time, non- Furthermore, a large number of tunable
minimum phase behavior and slow dynamics. In parameters are needed when the optimization
addition, MPC exhibits superior performance by problem of the terminal state constraint is
systematically handling constraints violation. complex. To avoid the high computational load,
Model Predictive Control is a discrete- the contractive constraint method introduces a
time[3] control in which control action is the stabilizing state constraint and requires the
solution to an open-loop constrained optimization

66
Nice Journal of Emerging Technologies Vol.:6 No.:1 June:2011

process states at the end f the prediction horizon consequence of its assumption about the basic
to be norm-contracted with respect to the process plant model. Although GPC is capable of
states at the beginning of the prediction horizon. controlling such systems, the control performance
The stability of the closed-loop system in the of GPC needs to be ascertained if the process
model predictive control can be tested by constraints are to be encountered in nonlinear
employing Lyapunov functions or a sequence of processes. Camacho (1993) proposed a
monotonic objective functions only when the constrained generalized predictive controller
prediction horizon is infinite or when a terminal (CGPC) for linear systems with constrained input
state constraint is applied. and output signals. By this strategy, the optimum
a)Principles of MPC: values of the future control signals are obtained
Fig 3 shows the block diagram of MPC. It by transforming the quadratic optimization
is a multi variable control algorithm that uses: problem into a linear complementarily problem.
1. An internal dynamic model of the Camacho demonstrated the results of the CGPC
process. strategy by carrying out a simulation study on a
2. A history of past control moves. linear system with pure delay. Clarke et al. (1987)
3. An optimization cost function J over the have applied the GPC to open-loop stable
receding prediction horizon. unconstrained linear systems. Camacho applied
the CGPC to constrained open-loop stable linear
The optimization cost function J is given by, system. However, most of the real processes are
nonlinear and some processes change behavior
(17) over a period of time. Exploring the application of
Where, GPC to nonlinear process control will be more
xi = ith controlled variable. interesting. In this study, a constrained
ri = ith reference variable. generalized predictive control (CGPC) strategy is
ui = ith manipulated variable. presented and applied for the control of highly
wxi = weighting coefficient reflecting the nonlinear and open-loop unstable processes with
relative importance of xi. multiple steady states. Model parameters are
wui = weighting coefficient penalizing big updated at each sampling time by an adaptive
changes in ui. mechanism.

b)Generalized predictive control (GPC) IV. GAIN SCHEDULING CONTROLLER

Gain scheduling controller is an approach to
control of nonlinear systems that uses a family of
linear controllers, each of which provides
satisfactory control for a different operating point
of the system. One or more observable variables,
called the scheduling variables are used to
determine what operating region the system is
Fig.3.Block diagram of GPC control law currently in and enable the appropriate linear
controller. It is one of the simplest and most
The generalized predictive control (GPC) intuitive forms of adaptive control. It consists of a
is a general purpose multi-step predictive control family of controllers and scheduler selects the
algorithm (Clarke et al., 1987) for stable control controller depending on the operating region.
of processes with variable parameters, variable
dead time and a model order which changes
instantaneously. Fig3 shows the block diagram of
GPC, adopts an integrator as a natural

67
Nice Journal of Emerging Technologies Vol.:6 No.:1 June:2011

Different gain scheduling approaches can be V. DISCUSSION AND RESULTS

distinguished, which may be classified in different
ways. To start with gain scheduling methods can
be divided into methods decomposing non-linear
design problems into linear sub-problems[4] and
methods decomposing non-linear design problems
into simpler non-linear sub-problems of affine
sub-problems. Furthermore, gain scheduling
methods decompose into continuous gain-
scheduling methods and discrete, i.e, hybrid or
switched gain scheduling methods. Discrete or
hybrid in this sense involves the switching of a
system or controller between dynamical regimes.
Fig.5. Input concentration MPC
.

Fig.4.Block diagram of Fuzzy gain schedule

controller

Procedure for Gain scheduling controller:

The designer selects several operating Fig.6. Output concentration response of MPC
points[4,5] which span the range of operation of
the process. At each of these operating points, the
designer constructs a linear time invariant
approximation of the plant and designs a linear
compensator for the linearised plant model.
In between operating points, the parameters or
gains of the compensators are then interpolated, or
scheduled, thus resulting in a global compensator
applicable to the whole window of operation.

Fig.7. Input temperature of MPC

68
Nice Journal of Emerging Technologies Vol.:6 No.:1 June:2011

Fig .9. Output responses: concentration

and temperature using Gain Schedule controller

Table 4.2 Performance index for MPC

Concentratio Temperature
n
Rise 0.0800 0.0800
Time
Settling 0.0980 0.0980
Time
Settling 2.2284℮-005 1.3010℮-004
Fig.8. Output temperature response of MPC Min
Settling 2.2609℮-005 1.2823℮-004
Table 4.1 Performance index for Gain Scheduling Max
Controller Oversho 3.7385℮005 3.7381℮005
ot
Undersh 0 0
Rise Time oot
Concentrati Temperat Peak 0.0833 0.4795
on ure
Peak 0 0
Settling Time 2.2367 0.0352 Time
Settling Min 45.8326 0.0625
Settling Max 0.6076 0.9005
Overshoot 1.6412 1.0000 VI .CONCLUSION
Undershoot 64.1230 0.0301 The nonlinear chemical process systems have now
Peak 0 0 been developed in various controlling techniques,
Peak Time 1.6412 1.0000 such as MIMO model of Gain Scheduling
5.9990 25.000 Controller and MIMO model of Model Predictive
Controller. A nonlinear gain scheduling controller
has the characteristic of a nonlinear controller
with time varying PID gains. While it is easy to
design for the linear or nonlinear time invariant
systems. The usefulness and effectiveness were
verified through the computer simulations for
CSTR systems. A robust output tracking and
disturbance rejection scheme for nonlinear
process by using a controlling techniques. The
Gain Scheduling Controller ,Internal Model
Controller and Model Predictive Controllers are
designed by using an simulation techniques. The
simulated responses are compared by using an
performance indexes and the better performing
controllers were mentioned. Then the compared
results a Model Predictive Controller is the highly

69
Nice Journal of Emerging Technologies Vol.:6 No.:1 June:2011

controlling the uncertainties and disturbances in

the nonlinear process. Also the MPC techniques 9.Shahin Salehi and Mohammad Shahrokhi,
are an advanced controlled devices compared with (2009) ‘ Adaptive fuzzy backstepping approach
the conventional controllers for temperature control of continuous stirred tank
REFERENCES reactors ‘, Fuzzy sets and systems vol. 160, pp.
1.Prakash J, and Srinivasan K, (2009) ‘Design of 1804 – 1818.
nonlinear PID controller and nonlinear model
predictive controller for a continuous stirred tank 10.Milan S, Matijevic, and Milic R. Stojie. (2006)
reactor ’, ISA Trans vol. 48, pp. 273 – 282. ‘The fuzzy controller design for processes
with dead times ’, Mechanic, Automatic
2.Prakash J, and Senthil R, (2008) ‘Design of control and Robotics vol. 17, pp.131 – 144.
observer based nonlinear model predictive
controller for a continuous stirred tank reactor ‘, ABOUT THE AUTHORS
Journal of process control vol. 18, pp. 504 – 514.
G.Glan Devadhas1 got his B.E.degree from
3.Garcia-Sandoval J.P, Gonzalez-Alvarez V. University of Madras and M.E.in Process Control
Castillo-Toledo B. and Pelayo Ortiz C, (2008) and Instrumentation from Annamalai University,
‘Robust discrete control of nonlinear processes: and pursuing his doctoral degree from PRIST
Application to chemical reactors ‘, Computers and University Thanjavur. He has more than ten years
chemical Engineering vol. 32, pp. 3246 – 3253. of experience and presently he is working as
. Asst.Professor in Noorul Islam Centre for Higher
4.Wang F.Y. Bahri P. Lee P.L. and Cameron I.T. Education. His areas of Interest are Controller
(2007) ‘A multiple model, state feedback strategy design, System modeling and Intelligent
for robust control of nonlinear processes ’, Journal controller design.His email is
of process control vol. 31, pp. 410 – 418. [email protected] and contact no is
9894896257
5.Juergen Hahn, Martin monnigmann, and
Wolfgang marquardt, (2008) ‘ On the use of Dr.S.Pushpa Kumar2 got his doctoral degree in
bifurcation analysis for robust controller tuning power systems from Roorkee University. He has
for nonlinear systems ‘, Journal of process control more than thirty four years of teaching
vol. 18, pp. 408 – 420. experience. He published several research papers
in national and International conferences and
6.Juing-Shian Chiou, Chi-Jo wang, Chan-Ming journals He is former principal of Government
Cheng and Chih-Chieh Wang, (2010) ‘Analysis college of Engineering Kannur and presently he is
and synthesis of switched nonlinear systems using working as principal in Heera College of
the T-S Fuzzy model ‘, Applied mathematical Engineering and Technology.He is presently
modelling vol. 34, pp.1467 – 1481. guiding three Ph.d students in PRIST University,
Thanjavur. His areas of interest are Power
7.Antonelli R. and Astolfi A. (2003) ‘Continuous Systems and Power Electronics. His email id is
stirred tank reactors: easy to stabilize ‘, [email protected]
Automation vol. 39, pp.1817 – 1827.

8.Yu D.L. Chang T.K. and Yu D.W. (2007) ‘ A

stable self-learning PID control for multivariable
time varying systems ‘, Control Engineering
practice vol. 15, pp.1577 -1587.

70