You are on page 1of 6

Copyright IFAC New Trends in Design of Control Systems,

Smolenice, Slovak Republic, 1997

ROBUST PID CONTROL: APPLICATION TO A CONTINUOUS-STIRRED


TANK REACTOR

Roman Prokop, Petr Dostal, Zdenka Prokopova

Department ofAutomatic Control. Faculty of Technology Zlin.


Nom. TGM 275. 762 72 Zlin. Czech Republic
tel.,fax: ++42067 7211521
E-mail: prokop@Zlin.vutbr.cz

Abstract: A tuning method for a class of Pill-like controllers and a MatIab-Simulink


toolbox were developed for robust control SISO continuous-time systems. Controllers are
obtained via solutions of diophantine equations in the ring of proper and stable rational
functions . Uncertainty and robustness are studied through the infinity norm lL,. A scalar
parameter m > 0 was proposed as a tuning knob for controller design with respect to
robustness and uncertainty evaluation. An application to a nonlinear chemical reactor is
simulated and studied.

Keywords: Robustness, uncertainty, Pill controllers, norms, algebraic approaches.

I . INTRODUCTION analyzed in Prokop and Dostal (1995), Prokop and


Meszaros (1996), Prokop and Corriou (1997). The
Robust controllers and plant uncertainty became a algebraic approach gives nontraditional Pill
useful and popular discipline in control theory structures proposed in e.g. Astrom et al. (1991),
during the last decade. The necessity of robust Morari and Zafiriou, (1989). Moreover, for SI SO
control was naturally developed by the situation controllers a scalar parameter m > 0 can be defined
when the nominal plant (used in control design) through the appropriate ring. This parameter is
differs from the real (perturbed) one. A suitable tool utilized as a tuning knob for a class of proposed
for parameter uncertainty is the infinity norm H00. SI SO continuous-time controllers and it strongly
Hence, a polynomial description of transfer functions influences the dynamics of the feedback system as
had to be replaced by another one. A convenient well as the sensitivity of proposed controllers.
description adopted from Vidyasagar (1995), Kucera
(1993), Doyle et al (1992) is a factorization approach
where transfer functions are expressed as a ratio of 2. SYSTEM DESCRIPTION AND UNCERTAINTY
two Hurwitz stable and proper rational functions
(Rs) . Then conditions of robust stability can be easily The plant uncertainty can be expressed conveniently
formulated in algebraic parlance and all controllers in terms of the fractional description. It means that
are obtained via linear diophantine equations in an linear transfer functions are no more represented as
appropriate ring. a ratio of two polynomial but two elements of an
another ring. For the purposes of robust control in
A class of Pill-like controllers is obtained by this the sense of this contribution any transfer function
way for first and second order SISO systems. The H(s) of a (continuous-time) linear system can be
methodology is proposed and more precisely expressed as a ratio of two elements:

81
H(s) = b(s) = B(s) class of nonlinear lumped parameter systems and
a(s) A(s) their mathematical models are described by
nonlinear differential equations. It is well known
A(s) = a(s) ; B(s) = b(s) (1) that traditional control methods are often
m(s) m(s)
unsatisfactory for a CSTR control. Some control
m(s) =(s +m)n: n = max{dega ; degb} problems are due to the nonlinearity and high
sensitivity of the output variables to the input
where a, b are polynomials. Elements A , B, ... changes. Moreover, the dynamic characteristics of a
constitute a subring Rm(s) of the ring of all Hurwitz CSTR may exhibit non-minimum phase behavior,
stable and proper rational functions Rs(s). A class of the time delay and sometimes also a changing sign
PID like controllers is generated by first and second of the gain in various operating points. The basic
order systems and all controlled plants will be properties of the reactor depend always on the
approximated by such systems. Systems with time reaction type taking place inside.
delays will be approximated by the following
fractional representation in Rm(s) : Here, a perfectly mixed CSTR with a cooling jacket
is considered, in which an exothermic consecutive
. A kl k2
82 reactIOn ~ B ~ C takes place. The
boe- es bo (1-8s+T) product is the component B . The reactant volume as
H (s) =--2--C:.---- ~ 2
well as the reactant volumetric flow rate are assumed
a 2s +a] s +a o a 2s +a]s+a o
2 to be constant. The reactant and coolant densities as
b2 s +b]s+b o (2) well as specific heat capacities are also constant. A
(s+m)2 B(s) perfectly mixed cooling jacket is assumed with
--2~--':"'-- -
a2s +a]s+ao A(s) constant coolant volume. The dynamics of the metal
jacket wall is neglected since the time constant of the
(s+m) 2
wall is very small in comparison with both reactant
and coolant time constants.
Systems of the first order are approximated in
similar way and systems without time delays are Under the above assumptions, the CSTR model can
simply obtained by putting 8 = O. For m = 0 subring be described by four differential equations of the
Rm(s) expands to the traditional ring Rs(s) (see form
Vidyasagar, 1985). An infinity norm (convenient for
uncertainty) in both rings is defined by

/lHI! = sup IH(s)1 = sup/H(jm)/


Re s~O OJ EE

This norm is the radius of the smallest circle


containing the Nyquist plot of the transfer function. (7)
Almost all mathematical models differs from
physical systems. Let H(s) = B(s) be a nominal
A (s)
plant and consider a family of perturbed
B'(S)
with initial conditions CA (0) =C~ , CB (0) =C~ ,
systems H'(s) =- - where
Tr (0) =T/ and Te (0) =Te' .
A ' (S)

In Eqs. (5)-(8) CA and CB are component A and B


IIA - A f: ] liB - B'
I 11 $; 11 $; f: 2 concentrations, Tr and Te are reactant and coolant
(4)
or I A- A ' B - B' 11 $; f:
temperatures, respectively, V are volumes, Q are
volumetric flow rates, p are densities, cp are specific
heat capacities, Ah is the area for heat transfer and U
is the overall heat transfer coefficient. The subscripts
3. A CONTINUOUS-STIRRED TANK REACTOR
are r for the reactant and C for coolant. The feed
MODEL
concentration CA! , and feed temperatures Tif, Tcf are
assumed to be known and constant, the feed
Continuous stirred tank reactors (CSTRs) are units
concentration CB! is zero. The specific reaction rates
frequently used in chemical industry. From the
system engineering point of view they belong to a are given by the Arrhenius equation

82
For the operating point 1, the output y time
j = 1, 2 (9) responses to various step changes of input u are
shown in Fig. 2. The noniinearity of the controlled
system is clearly documented.
where E;. are activation energies and R is the gas
constant. The heat of reaction is expressed as
0.8 r-----======:::Jci
':'~
where h are reaction enthalpies. -0
e
For the next calculation and control simulation the c
'" 0.2
component B concentration takes place of the .0.4
controlled output and the coolant volumetric flow
0.6 ...........--'--~-'--...........-~~..........--'-~--'---'
rate is used as the control input, both as deviations o 20 40 60 80 100 120 140
I (mm )
from their steady-state values so that
Fig. 2. The output y time responses to input u step
yet) =GB (t)-c1 (11) changes: u = 1) -0.015, 2) 0.015, 3) 0.030,
u(t) = Qc(t) - Q: (12) 4) 0.045,5) 0.060 [m3 min- l ] .

All parameter values, feed values and initial


conditions of the model are shown in Table 1. 4. CONTROL DESIGN

A control system depicted in Fig. 3. is considered for


Table 1 Parameter values, feed values and initial design of a class of PID - like controllers.
conditions

v = 1.2 m3 Pr = 985 kg m- 3
3
Vc = 0.64 m Pc = 998 kg m- 3
Ah = 5.5 m2 Gpr = 4.05 kJ kg-
I

Qr = 0.08 m 3 min-1 G pc = 4.18 kJ kg-


I

U = 43.5 kJ m- 2K lmin- 1
l
k OI = 5.616e16 min- E21R = 15290 K
I
k02 = 1.128e18 min- hi = 4.8e4 kJ kmorl
EI/R = 13477 K h2 = 2.2e4 kJ kmorl Fig. 3. General control system.
GAjS = 2.85 kmol m-
3
3
Tr/ = 323 K
Q/ = 0.03 m min- 1
Tc/ = 293 K Note that for Q = R a traditional feedback regulator
3
GAS = 0.165 kmol m- T/ = 350.2 K operating on the tracking error is obtained. Transfer
CB = 0.944 kmol m-
s 3
T/ = 330.5 K . G", Gy
fu nctlOns w = - and v =- represent the
F", Fy
The dependence of the output variable upon the
reference and disturbance signal, respectively. The
control input in the steady-state is shown in Fig. 1.
control design in Rm(s) is proposed in Prokop and
Two operating points were chosen for estimation,
Corriou (1997), Prokop and Meszaros (1996),
approximation and control of the CSTR model. Both
Prokop and Dostal (~995) . For the given nominal
are depicted in Fig. 1. and they represent steady-state
plant H(s) in the form of (I) it consists of the
values of CB s = 0.9435 and GBs = 1.3413 ,
following steps:
respectively.

~: All stabilizing (feedback) controllers are


1.8

1.6
' . ... given by all solutions of the equation:

1.4
AP + BQ = i (13)
~
OpcnlUlI pOlntl
1.2
-0
E 1.0
in the parameter form expressed by the fraction :
..,'"
~
0.8
OpuatUlI pomt I

0.6

0.4
Q = Qo - AT
(14)
~~~~~~~~~~~~~

0,01 0.02 0 .03 0,04 O.O~ 0,06 0,07 0 ,08 0.09 0.10 0.11 P Po +BT
Qc' (m' min") where Po, Qo E Rm(s) are particular solutions and T
Fig. 1. Steady-state input-output behavior of CSTR. is arbitrary in Rm(s) .

83
Step 2: Feedback controllers (if necessary) are given which yield a class of generalized Pill-like control
in a similar way by all solutions of structures. The program system enables design and
simulation of a wide spectrum of robust control
(15) problems.

Step 3: Compute P E Rm(s) such that The simulation of the perturbed plant with the
regulator computed for the nominal transfer function
Fw divides P for asymptotic tracking problems is carried out in standard Simulink environment.
Fv divides P for disturbance rejection problems Simulation parameters as simulation horizon,
(16) reference, load disturbance and disturbances, input
constraints, ... can also be defined by the user. All
Step 4: For perturbed plants (4) choose such P, Q in simulation variables can be stored and transferred
(13), (14) which fulfill the conditions out of the Matlab workspace and IT AE, IAE or IE
can be calculated as a tool of comparison and quality
evaluation of the control behavior.
(17.a)

(17.b) The proposed methodology is illustrated by the


example with the first order system and a time delay.
Suppose a nominal plant in Rm(s) given by
Steps 1 - 3 represents the first and simplest case of
robust design. All solutions of diophantine equations b}s+b o
in Rm(s) are expressed as functions of the parameter s+m
,m~O
m > O. The value of this parameter strongly s+a o
influences dynamics as well as robustness of the s+m
proposed control system. For a deeper insight into (20)
robustness the notion of the sensitivity function :
The control goal is to design in a robust way:
y I
E= - = = A(Po + BT) (18)
v l+HCb a) an asymptotic tracking feedback controller (Q=R)
for a stepwise reference signal
can be utilized as the second possibility. Similarly, b) the same as in a) but with feedback and
for SISO systems E is also a function of a scalar feedforward parts (Fig. 3.)
parameter m > 0 and it can be minimized by a
simple scalar optimization method. In this way the The denominator of the reference generator in Rm(s)
"most robust" controller for a given nominal plant is in the form Fw =_s_ .
obtained. The third case covers situations where s+m
perturbed plants are known and the norms in (4) can a) All solutions of AP+BQ = 1 in the parameter
be evaluated. Then inequalities (17) yield generally form are:
nonlinear inequalities for the scalar parameter m > 0
which can be solved numerically. b}s+b o s+a
P= P + T' Q=q - - - T; (21)
o s+m' 0 s+m

5. ROBUST AND SENSITIVITY TUNING bo -b)m m-ao


where Po = ; qo = and T is
bo -aOb) ' bo -aOb)
A. MA TLAB-like toolbox with simulation support in
SIMULINK was developed for nominal plants: arbitrary in Rm(s) . The simplest but not robust
solution is given for T= O. Robust controllers are
given for P divisible by Fw which is obtained for
F)(s)=~ T= -~. bo - b)m Then the feedback (only)
s +ao
bo bo - aOb) .
F2 (s) = bo e -es ~ bo (1- 8s) controller is given by
s+ao s+ a o
bo p, _ s bo - mb) .
F3 (s) = -,?:-----"--- (19) ) - s + m b0 Po,
s - +a)s+a o
2 m aopo (22)
bo ( 1 - 8s + 8
2
S2 J Q1-
_
(qo + - po)s+m(qo + -- )
bo bo
s+m

84
and the control law is generated by the differential order system (with and without time delay) in the
equation: form :

es
(23) H(s) = 20e- ~ _B_(s_)
400s 2 +40s+1 A (s)
2
b 2s +bls+bo
_ b o -mb l m (27)
where P I = b Po; q l =qo +bPo; B(s) (s+m) 2
, m'?
0
o o 2
_ a A (s) 400s +40s+ 1
o po)m.
qo = (qo +- (s+m) 2
ho
Naturally, equation (23) represents a PI regulator.
where the time delay is either 8 = 0 or 8 = 5.
b) All feedforward parts are given by solutions of Coefficients b j are computed according to (2). The
FS + BR = I: approximation was obtained by a four-parameter
method adopted from Astrom and Hagglund (1995)
m s applied to step responses in Fig. 2. Because the
R=----Z (24) CSTR model is non-linear, the approximation is
bo s +m
rather rough and it depends on a given operating
point. In spite of this situation, it is desirable to find
where Z is again arbitrary in Rm(s). For the simplest a PID-like controller which is " sufficiently
z=o we get the control law: acceptable" for both operation points. More
precisely, the control goal is to design (in a robust
2
u= ~ [~w+~fwdt-ql fydt-qoY] (25)
way) a FB and FBFW controllers according to above
PI bo bo mentioned methodology for nominal plants (27) and
choose such values of m > 0 which give control
which represents a generalized PI controller responses without overshooting. From the
proposed e.g. by Astrom et al. (1991) in the form: technological point of view, aperiodic responses after
step changes of the set point are inherent and natural
k conditions for chemical reactors control.
u=k(pw- y ) + - f (w - y )ch (26)
T; The "most robust controller" was obtained by scalar
Is.n .~n/I.--_ _----,=S:.:c'::.:Dl:.:.:itc;..:IV.:.:itL..!.!fu'-'!D.::.;ct""lo:.::.n_ _- - - ,
minimization of the sensitivity function (18) for the
la M in .n o rm = 1.5 2 2 4 . O ptimal mO = 0 .1 700 nominal plant (27) without time delay (8 = 0). Both
FB and FBFW control schemes gave the optimal
8 value mo = 0.05 . Derivation of controller parameters
6
does not bring any problems but for the second order
the resulting expressions are a bit more complex
than the ones in (21-26).

Fig. 5. to 8. show the results of Simulation 1 to


oL--~--~--~--~-~
o 0. 2 0.4 0 .6 0. 8 I Simulation 4 of proposed controllers, respectively.
mO
The following list describes the conditions of
Fig. 4. Sensitivity function .
control :
Fig. 4. illustrates the sensitivity function for
Simulation 1: Oper~ting point 1, FB controller,
125e - ss
H(s) = . and PI regulator as a function of without time delay (8 = 0), set points w = +0.5; +0.1.
2s+ I Simulation 2: Operating point 1, FBFW controller,
m > 0 with the optimal value mo = 0.17. It is clear with time delay (8 = 5), set points w = +0.5; +0.1.
that all controller parameters are functions of m > O. Simulation 3: Operating point 2, FB controller,
A deeper insight into conditions (17) can be found in without time delay (8 = 0), set points w = O.32.
Prokop and Corriou (1997), Prokop and Meszaros Simulation 4: Operating point 2, FBFW controller,
(1996), Prokop, Bobal and Prokopova (1997). with time delay (8 = 5), set points w = O.32.
Simulation results document that there does not exist
a feedback PID controller which enables the
6. MODEL APPROXIMATION AND aperiodic control responses under given
SIMULATION RESULTS circumstances. Fig. 5. and 7. show control responses
in both operating points with the optimal FB PID
Dynamics of the CSTR described in section 3 in both controller obtained by sensitivity minimization. The
operating points is approximated by the second

85
situation was improved by involving a feedforward
part of the control law (see Fig 6. and 8.). The 0,1 r-~~----------------------~

resulting control law is of a generalized PID-type in ~


, 0,4
the sense of (26) and Astrom et al. (1991). '"
0,2

7. CONCLUSIONS
200 500 100
T im e
A design method based on the fractional
Fig. 5. Simulation 1 - output and reference
representation was developed for SI SO continuous-
time systems generally with time delay. Control laws 0 , 01 , - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - ,
for second order systems yield a class of PID - like
controllers. The proposed methodology enables to 0,04 _~
tune and influence the robustness and control ~ 0 0 2 ~ r,_ _ ____--,
behavior by a single scalar positive parameter. The
0,00
design methodology was successfully applied to a
nonlinear chemical reactor.
T im e

Fig. 5u. Simulation 1 - input


REFERENCES
0 ,1 r-----------------------------,
Astrom, Kl , Hang, e.e. , Persson, P., Ho, WK " " " " ":.;.";,.-.
" - --"
0 ,4
(1991) : Towards Intelligent PID Control. , '

Automatica, 28, pp.I-9.


0,2
Astrom, KJ., Hagglund, T. (1995): PID Controllers:
Theory, Design and Tuning, 343 p. Instrument
0 , 0 0L.L....~1:-::0'-:'0~-:2~0'-:'0~-:J~0"':"0~-:4:-:'0-,-0---::5:-:'0-'-0---::I...J
O0
Society of America.
T im e
Doyle le., Francis B.A. , Tannenbaum A.R.(1992):
Fig. 6. Simulation 2 - output and reference
Feedback Control Theory. Maxwell McMillan.
Kucera, V. (1993) : Diophantine Equations in o, 0 1 5 r-----------------------------,
Control - A Survey, Automatica, 29, No.6,
.. \ .
pp. 1361-75. 0,01 0 b

Kucera, V. (1995) : Algebraic Methods in Control -


Theory and Applications. Prepr. 0 , 005

IFIPIIFA CIIFORS Con! System Modelling and


Optimization, Prague. 0 , 00 0 0"--~1:-::0'-:'0~-:2~0'-:'0~--::J~0"':"0~-:4:-:'0-,-0---::5-:'0-'-0---::1-"00

Morari, M., Zafirou, E. (1989): Robust Process T im e

Control. Prentice Hall, New Jersey. Fig. 6u. Simulation 2 - input


Prokop, R. , Dostcil, P.(1995) : Design of Simple
Robust Controllers. Prepr. 10th Int. 0 ,4
Conference.on Process Control '95, Tatranske .. -~ ...- .. ---
J:. 0.2 b
Matliare, pp. 103-108.
0,0
Prokop, R. , Prokopova Z. (1996) : Design of Robust
0 , 2
Control of SISO Continuous-time Systems,
Prepr. Con! RIP-PC '96, Homi Beeva. ,0 , 4

Prokop, R. , Corriou, lP. (1997) : Design and 100 200


I
500 100
T im e
Analysis of Simple Robust Controllers, Int.
Fig. 7, Simulation 3 - output and reference
Journal of Control, Vol. 66, No. 6, pp. 905-92l.
Prokop, R. , Bobal V. and Prokopova Z. (1997) :
Design of Robust Controllers for Time - delay 0,4
- --.-- - .-----
~
Systems, Prepr. Con! CIS '97, IFA C-IFlP- 0,2

IMA CS, Belfort. 0,0

Venkatashankar, v., Chidambaram, M. (1994) : .0 ,2


Design of P and PI Controllers for Unstable Firs- 0 , 4
" ",, .. .. , "'-- ---1

order Plus time Delay Systems, Int. J.of Control, 1 00 200 JOO 400 500 100
Vol. 60, No.l , p.137-144. T im e

Vidyasagar, M. (1985): Control System Synthesis: A Fig, 8, Simulation 4 - output and reference
Factorization Approach. MIT Press, Cambridge.

86

You might also like