MULTI-LOOP CONTROL DESIGN IN MULTIVARIABLE (2X2)

CONTINUOUS STIRRED TANK REACTOR

Abdul Wahid, William Andreas Hambali

Departement of Chemical Engineering, Faculty of Engineering, University of Indonesia

Kampus Baru UI Depok – 16424, Indonesia, Phone No. 7863516

Email: [email protected], [email protected]

Abstract

With this study, the design and tuning of multi-loop for multivariable (2x2) CSTR will be made in order to
achieve optimum CSTR control performance. This study used Bequette model reactor and MATLAB software and is
expected to be able to cope with disturbances in the reactor so that the reactor system is able to stabilize quickly despite
the distractions. In this study, the design will be made using multi-loop approach, along with PI controller as the next
step. Then, BLT and auto-tune tuning method will be used in PI controller and given disturbances to both of tuning
method. The controller performances are then compared. Results of the study are then analyzed for discussions and
conclusions. Results from this study have shown that in terms of disturbance rejection, BLT is better than auto-tune
based on comparison between both of controller performances. For IAE for the case of temperature, BLT is 30% better
than auto-tune, but it is almost the same for the case of concentration. For settling time for the case of concentration,
BLT is 30% better than auto-tune, and for the case of temperature, BLT is 18% better than auto-tune. For rise time for
the case of concentration and temperature, BLT is 30% better than auto-tune.

Keywords : Multi-loop approach, Bequette reactor, Controller performances.

1. Introduction measured controlled variable and adjust only one

manipulated variable, the actions of the controllers are
CSTR reactor is used in big industries that require relatively easy to monitor (Marlin, 2000).
processes that can’t be stopped because the production This study will use a CSTR model with a cooling jacket
has to remain continuous such as oil refineries and so (Bequette, 1998). The method used was based on multi-
on. Usually in this type of reactor, the reaction occurs loop approach with a variety of disorders provision
always not linear and has very high complexity. This replaces variations of set point changes. The main
kind of reaction is difficult to control by conventional interest here is how one can pair the right combination
method because it has multiple steady states in this for each of manipulated and controlled variables,
equipment. (Vaneshani, 2011; Rahmat, 2011). Also because the right pair of it will led to a better
reactor of this kind involves a multivariable system performances for a controller. The used reaction is
that’s why even proportional-Integral (PI) control exothermic irreversible first-order reaction with the
system may not give good results for multivariable multi-input multi-output (MIMO) 2x2 system. The
systems. This kind of condition also led to frequent success of this study will be tested by calculating the
interruptions to the CSTR reactor that would make the Integral Absolute Error (IAE) from the resulting BLT
resulting product does not fit with what is desired. control design with PI control that is made then
That’s also why it’s necessary to have the control compared it with the value of the IAE from model with
system that can work well in nonlinear and multi-loop control design PI control system on
multivariable system. The approach will be used here is simulation program and comparison of the presence or
the multi-loop approach. Although it has been decades, absence of criterion performances for both of control
there are already many successful multi-loop strategies systems.
have been used and proven to be a good approach thus it
continue to be used. Because of its use of simple Objectives of this study are design and tuning the
algorithms, it’s ease to be understanding by plant multi-loop control using PI controller for multivariable
operating personnel, which is the result of its simple (2x2) CSTR in order to achieve optimum CSTR control
control structure. Since each controller will use only one performance. This study is expected to be able to cope
with disturbances in the reactor so that the reactor Here's a derivation model for the Bequette
system is able to stabilize quickly despite the reactor. Equilibrium material in the reactor is shown by
distractions. equation 1.

2. Experimental (1)

The study began by making the material and energy With (amount of substances in the reactor
balance for the system which is Bequette CSTR reactor remains constant), then equation 3.1 can be simplified to
then making the state space eq. in order to make the equation 2.
pairing between inputs and outputs of the system. Then
design a PI controller system based on the pairing and (2)
tuning it using BLT PI tuning method, and calculate its Density of the substance is also assumed to be constant,
controller performance. Compare the controller so it can be gotten the equation 3.
performance of PI with PI tuning from tuning software.
Flow diagram of study in general can be seen in Figure dan (3)
1. Because the reactor volume is constant, then the moles
balance of component A can be written according to
This study uses a model of a non-isothermal CSTR equation 4.
reactor developed by Bequette, as the system to be
controlled. (Bequette, 1998) and using a simulation (4)
software called MATLAB. This study will used multi- The reaction rate is shown by equation 5.
loop approach for PI tuning design.
(5)
For the energy balance equation is shown in equation 6.

(6)
Final differential equations to calculate the
concentration of the product and the reactor temperature
is shown by equations 7 and 8

(7)

(8)

Reactor parameters used in equations 7 and 8 are shown

by Table1 (Bequette, 1998).
Table 1. Parameter Value Reactor Model
Parameter Value
F/V, hr-1 1
Figure 1. Flowchart k0, hr-1 9703*3600
(-∆H), kcal/kgmol 5960
ΔE, kcal/kgmol 11843
Material used in this study is a model of CSTR ρcp, kcal/(m3K) 500
(Bequette, 1998). The manipulated variables are feed Tf, K 298
concentration and initial temperature of cooling jacket. CAi , kgmol/m 3
10
The controlled variables are the reactor temperature and UA/V, kcal/(m3K
the concentration of the product. The disturbance is 150
hr)
what will happen when we change the manipulated
Tj, K 298
variables. Reaction contained in this reactor is a simple
reaction A  B. The assumptions used are the reaction
A to B is a first order reaction.
Study procedures:
 In order to make the pairing, first material and
energy balance of the system need to be made
because they will be used to find the state (19)
space of the system
 The material and energy balance are from last
eq.
(20)

(9)
(21)

(22)
 For C and D on the state space formula, C is
using the matrix identity because the output is
the state variable and since the C is the state
variable, just use zero matrix for D
(10)
 Find the state space by converting the previous
equation into: (23)

(11) (24)
 Make the pairing between the inputs and the
outputs of the system
o Find the transfer function using
MATLAB (commands can be seen at
appendices)
o By using the [x,y]=ss2tf(a,b,c,d,1),
[x,y]=ss2tf(a,b,c,d,2) command and g
(12)
=tf(x(1,:),y), i =tf(x(2,:),y) the transfer
 Make the derivative equation based on the
function can be found
previous equation, for A and B, according to
Result of the transfer function
the state space formula:

(13)
(25)
(14)
(26)
 The derivatives:

(27)
(15)

(28)
(16) o Using the transfer function, find the
λ11 (pairing constant) using below
equation. To find the value of K
(17) (gain), use MATLAB (commands can
be seen at appendices)

(29)
(18)
 Use the step command to find each gain of the The graph steady at -0.19; means the value of K21 = -
transfer function, by plotting it and measure the value of 0.19.
the amplitude until the graph is steady.

Figure 5 Gain (K22)

Figure 2 Gain (K11) The graph steady at 2.6; means the value of K21 = 2.6.

The graph steady at 0.51; means the value of K11 = 0.51. o Find the RGA by using the table
below, thus the correct pairing for the
system will be known (pick the
combination that the number of it
close to 1).
The calculation will be:

(30)
o The array will be:

Table 2. Array of Pairing

MV1 MV2
CV1 λ11 1- λ11
CV2 1- λ11 λ11
Figure 3 Gain (K12)
Thus the results are:
The graph steady at 4.5; means the value of K12 = 4.5.
Table 3. Result of Pairing
MV1 MV2
CV1 0.392
CV2 0.392
Thus the result is pairing between controlled variable 1
with manipulated variable 1 and controlled variable 2
with manipulated variable 2.
 Make the PI controller.
 Tuning for BLT, using the Kc and from
Ziegler Nichols, we can determine the Kc and
of BLT by dividing the Kc with F and
multiplying with F. the F value for this

Figure 4 Gain (K21) system is 1.65 and Kc = 8.63 and = 0.4 for
temperature variable whereas Kc = 5.9 and
= 1.515 for concentration variable. We get the
 looking at the graph then see if the criteria is already
Kc = 5.23 and for temperature, and good or not as the controller performance.
for concentration the Kc value is 3.57 and
3. Results
As for the auto-tune, we can use auto-tune from Reactor model used in this study is the MIMO
software. Press the tune button then tune the PI system with size 2x2, and the study variables are 2
controller. independent variables / input variables which are feed
concentration, initial temperature of cooling jacket. And
the 2 dependent variables / output variables which are
the reactor temperature and the concentration of the
product. In this study, the manipulated variable or
independent variable is the factor that will be changed in
an experiment. So the manipulated variables will be
feed concentration and initial temperature of cooling
jacket. And controlled variables are the variables that
are input into the control system which the researcher
holds constant (controls) during an experiment. So the
controlled variables will be reactor temperature and
concentration of the product. Basically, the controlled
variables need to be constant because that means the
reactor is at steady state condition, and the product will
not be interrupted by disturbances. That’s why in order
Figure 6 Auto-tune to control it; the manipulated variables will be changed.
 Do the simulation control to CA and T using Based from the pairing method for multi-loop approach
both controllers; also give disturbance with the it can be determined that the change of input
same amount of it. temperature or input concentration will only affect the
o From this simulation we can get output of them, respectively. This is because the pairing
response graph of CV, and can be constant or λ11 is simplified, so it can be safely assume
calculated the IAE value for each that it is close to 1, means that the interaction between
simulation control using this eq.: the temperature and the concentration can be neglected,
respectively, or there are almost no interaction between
them, so it can be safely assume that the tuning for this
(31) system can be done separately. λ11 itself is a relative gain
o We can see from the graph that, IAE and it can be defined as ratios of open-loop to closed
is absolute area from the difference loop gain. This number is important because it is
between graph areas of set point with determining whether the system can be tuned correctly
graph area of CV response. The or not. Thus, great accuracy is required in calculating
smaller the IAE means that CV is the relative gain. Changes made to the magnitude of Ti,
getting closer to its set point which Tc, and CAi is a reduction of 30%, 50%, and 70% and
means that the controller used is a the addition of 30% of the initial value of them. These
good one. As shown in the example of values have been assumed to represent the entire range
response graph of CV, and the red of those variables. The reactor will be controlled by the
area is the magnitude of IAE (Integral PI controller, and the manipulated variable will be inlet
Absolute Error). concentration and input temperature. Figure 8 is about
Bequette reactor and Figure 9 is about the reactor after
given the controller.

Figure 7. IAE
For others controller criteria, after the graph is
made, we can see from the graph to determine the good Figure 8. Bequette Reactor
criteria for those controller performance by simply
 Table 4. Comparison of BLT PI’s IAE and PI’s (auto-
tune) IAE on concentration and temperature control (Ti
change)
T (Auto- C T C
Disturbance
tune) (Auto-tune) (BLT) (BLT)
(%)
(K) (kgmol/m3) (K) (kgmol/m3)
-30 101.53 0.44 22.77 0.60
-50 163.45 0.49 32.64 0.66
-70 225.37 0.53 42.00 0.71
+30 101.73 0.72 25.08 0.73

Table 5. Comparison of BLT PI’s IAE and PI’s

(software) IAE on concentration and temperature
control (Ca change)
Sp C T C
T (Auto-
Change (Auto-tune) (BLT) (BLT)
tune) (K)
(%) (kgmol/m3) (K) (kgmol/m3)
Figure 9. Multi-loop Reactor control after pairing -30 11.40 104.10 7.85 104.92
-50 13.17 173.42 8.08 174.58
For Figure 10, those are about design of the -70 14.93 242.75 8.30 244.23
+30 11.53 104.04 7.98 104.91
system after given controller in MATLAB software. It
is then given disturbances and change of set point in Table 6. Comparison of BLT PI’s IAE and PI’s
order to be analysed whether the controller is good or (software) IAE on concentration and temperature
not.
control (Tc change)
C
Sp T T C
(Auto-
Change (Auto-tune) (BLT) (BLT)
tune)
(%) (K) (K) (kgmol/m3)
(kgmol/m3)
-30 0.45 0.58
-50 0.45 0.58
-70 0.45 0. 58
+30 Unstable Unstable Unstable Unstable

For Figure from 11 to 12 those are about the

response of the output after given disturbances. In this
case, the disturbance is change in Ti. It can be seen that
Figure 10. Disturbance Using Simulink (change of Ti) the trend of the graph is following the change in Ti, for
example if the change is -30% of Ti, then the
For the change of Ca and Tc, the only different disturbance peak will go up then down a bit, but if the
is the step input is at Ca and Tc before the PI controller. change is +30%, then the peak will go down then up.
Comparator used in this study is the PI controller
between the one that using the multi-loop approach with
BLT method, and the one that using the auto-tune. PI
controller will do the handling of the disturbance, which
will then be compared with each other. The structure of
the PI controller used is a multi-loop control, in other
words when the disturbance is given, there’s a chance
that both of the inputs will be affected as well as the
outputs. The result of it can be seen in the following
tables and figures. For Table 4.3 to 4.5 is about the
comparison of both PI tuning method’s IAEs.

Figure 11. Response of concentration caused by ΔTi = -

30%
Figure 12. Response of temperature caused by ΔTi = - Figure 14. Response of temperature caused by Δ CA = -
30% 30%
It is shown that BLT and auto-tune method has almost
Overall criterion performances of the system suggest the same results in terms of criterion performances.
that BLT approach is a better method than auto-tune. Even though, BLT has a slightly better criterion
For the settling time, BLT suggests a faster response in performance than auto-tune in terms of settling time.
reaching the system’s steady state condition, despite of Settling time is important because the system has to
the overshoot and the decay ratio is there. In summary, reach its steady state as fast as it can be, after receiving
the smaller the number in criterion performances, then disturbances or the change in set-points, in order to
the faster the response of a controller. maintain the product of the reactor.
For Figure 13 to 14, those are about the set point change For Figure 15 to 16, those are about the set point change
in Ca. It can be seen that because there is a change in set in Tc, same with previous set-point change, it can be
point, that causes disturbance in the temperature. It is on seen that because there is a change in set point, that
purpose in order to know the response of the controllers. causes disturbance in the concentration. It is on purpose
in order to know the response of the controllers.

Figure 13. Response of concentration caused by Δ CA =

-30% Figure 15. Response of concentration caused by ΔTc =
-30%
 concentration and the temperature is much faster for
BLT design rather than the one using the software.
Although for set-point change Tc for +30% of initial
value, even the BLT couldn’t handle the system when
its set point is change. It means that the PI controller
couldn’t handle the disturbance or already reach its
limit. The auto-tune works by processing feedback
information in the form of error generated. Also auto-
tune works by “fulfilling” all the criterion performances,
like no decay ratio or overshoot. That’s why the
response became too long and it takes longer time to
reach steady state condition after received disturbances.
Whilst for BLT, the detuning factor is adjusted so that
the biggest log modulus, which is a measure of how far
the system is from closed-loop instability, has a
specified value. That’s why this method provides
Figure 16. Response of concentration caused by ΔTc = reasonable preliminary controller settings with
-30% guaranteed closed-loop stability. The detuning factor is
needed because of the Ziegler-Nichols method earlier
Simulations conducted in this study are essentially used is causing the response to be too oscillatory and it
making multiloop control and is used to handle a given causes low robustness. Therefore one needed to detune
disorder in the system. In this study, given the the controller to obtain a more stable response and
disruption to the system in the form of steps. increased robustness. The controller will be less
Disruptions in the system are disturbance (change in Ti) oscillatory and more tolerant to changes in process
and set-point change (Ca and Tc). Then BLT characteristic. So in conclusion, the BLT control design
capabilities that have been made will be compared with is proven to be better than the auto-tune because it’s
the PI in the tuning software capabilities in dealing with considering things that will help the controller to be a
the same disorder. For IAE for the case of temperature, better one, for example the robustness.
BLT is 30% better than auto-tune, but for the case of
concentration, BLT and auto-tune is almost the same. 4. Conclusions
For settling time for the case of concentration, BLT is
30% better than auto-tune, and for the case of Based on the result and discussion, it can be
temperature, BLT is 18% better than auto-tune. For rise conclude that:
time for the case of concentration, BLT is 30% better 1. After doing pairing analysis, design of multi-loop
than auto-tune, and for the case of temperature, BLT is control for multivariable (2x2) CSTR is output
30% better than auto-tune. For decay ratio for the case temperature (CV1) paired with input temperature
of both concentration and temperature, auto-tune is (MV1), and output concentration (CV2) paired with
better than BLT because auto-tune doesn’t have decay input concentration (MV2).
ratio. Auto-tune and BLT doesn’t have offset for the 2. Tuning results for BLT are for temperature loop,
case of both concentration and temperature. For
overshoot for the case of both concentration and Kc = 5.23 and , and for concentration
temperature, auto-tune is better than BLT because auto-
tune doesn’t have decay ratio. It can be seen from the loop, Kc = 3.57 and with the control
criterion performances, for the BLT control design, each performance parameters are better than the auto-
of them has a better value than the PI that using the tuner.
auto-tune, especially from the offset on the temperature 3. For IAE for the case of temperature, BLT is 30%
of PI that using the tuning software. It means that the better than auto-tune, but for the case of
value of it will never reach the set point given in the concentration, BLT and auto-tune is almost the
system. It may control it but it will not give the same same. For settling time for the case of
performance as the BLT control because of it. But the concentration, BLT is 30% better than auto-tune,
offset is so small it can be neglected, it just affected the and for the case of temperature, BLT is 18% better
settling time. It will get longer to settle after rejecting than auto-tune. For rise time for the case of
the disturbances. That is why the IAEs of auto-tune are concentration, BLT is 30% better than auto-tune,
bigger than BLT. The BLT control also has a faster and for the case of temperature, BLT is 30% better
response on controlling the system, proven by the value than auto-tune. For decay ratio for the case of both
of the settling time on the concentration and the concentration and temperature, auto-tune is better
temperature. It is also proven to have a better than BLT because auto-tune doesn’t have decay
performance because of the value of the rise time for the ratio. Auto-tune and BLT doesn’t have offset for
 the case of both concentration and temperature. For
overshoot for the case of both concentration and
temperature, auto-tune is better than BLT because
auto-tune doesn’t have decay ratio.

Notation List
T Temperature (K)
C Concentration (kgmol/m3)
Sp Set-point
CV Controlled Variable
MV Manipulated Variable
PI Propotional-Integral

References
Altmann, W. (2005). Practical Process Control for
Engineers and Technicians. Newnes, Elsevier.
Bequette, B. W. (1998). Process Dynamics : Modeling,
Analysis, and Simulation. New Jersey : Prentice
Hall, Inc.
Chau, P.C., (2001). Chemical Process Control: A First
Course with MATLAB. University of California, San
Diego.
Farouq, S. M., Jayakummar, N. (2009). An Algorithm
for Stabilizing Unstable Steady States for Jacketed
Nonisothermal Continually Stirred Tank Reactors.
Journal of Process Control.
Fogler, S. H. (2006). Elements of Chemical Reaction
Engineering Fourth Edition. Massachusetts :
Prentice Hall International Series.
Ingham, D. B., et al. (2000). The Optimisation of
Reaction Rate Parameters for Chemical Kinetic of
Combustion using Genetic Algorithms. Computer
Methods in Applied Mechanics and Engineering,
190, 1065-1090.
Han, J. (2009). From PID to Active Disturbance
Rejection Control. IEEE Industrial Electronics
Society, 56, 900-906
Marlin, T. (2000). Process Control : Designing
Processes and Control Systems for Dynamic
Performance 2nd Edition. New York : McGraw-
Hill.
Qin, S. J., Badgwell, T.A. (2002). A survey of industrial
model predictive control technology. Control
Engineering Practice, 11, 733-764.
Wu, W. (2000). Nonlinear Bounded Control of a
Nonisothermal CSTR. Journal of Process Control.
Johnson, M. A., Moradi, M. H. (2005). PID Control :
New Identification and Design Methods. London :
Springer.
Anonymous., (n.d.). Multiloop and Multivariable
Control.
www.cc.ntut.edu.tw , Accessed on 20 June 2013