You are on page 1of 6

Computers them. Engng Vol. 19, Suppl., pp. S409-S414.

1995
Copyright @ 1995 Elsevier Science Ltd
Pergamon 0098-1354(95)00050-x Printed in Great Britain. All rights reserved
0098-1354/95 $9.50 + 0.00

PERFORMANCE OF DIFFERENT REGULATORS


FOR PLANTS WITH RECYCLE

C. SCALI and R. ANTONELLI


Department of Chemical Engineering, University of Pisa
Via Diotisalvi, 2 - 56100 Pisa (I)
r) Tel: +39-50-511241, Faz: $39-50-511266, e-mail: sQicnucevm.cnuce.cnr.it

ABSTRACT

Performance of simple PI and model based (IMC) regulators for plants with recycle, are compared
in control schemes with and without a recycle compensator. By considering different dynamic
characteristic of the process, it is analysed in which cases the adoption of a compensator is necessary
to eliminate completely the effect of recycle streams, both for set point tracking and for disturbance
suppression. The robustness of the compensator in the presence of parametric uncertainty in the
recycle model is also examined.

KEYWORDS

Process Control; Recycle Systems; Recycle Compensator; Internal Model Control

INTRODUCTION

In chemical plants recycle streams are very common; for instance, in the presence of a reactor where
total conversion is not reached, there is a need of separating unconverted reagents and to recycle
them upstream. The design of the control system for plants including units with recycle must be
done taking them into account simultaneously; it turns out to be much more difficult than the case
of units operating in cascade, where it can be done sequentially. The installation of large surge
tanks between different units allows to lead the recycle structure back to cascade units, but it can
be uneconomic and unconvenient for safety and environmental reasons.
The dynamics of recycle processes can be far different from that of the forward process. Denn
and Lavie (1982), pointed out that the recycle is equivalent to a positive feedback and studied
the effect of a delay in the recycle path. Luyben (1993a), analyses a very simple system in open
loop and in closed loop conditions to show how the dynamics can vary with recycle parameters:
neglecting the presence of recycle in the design of the control system may lead to deterioration of
performance and to instability conditions. In other related papers Luyben (1993b, 1993c, 1994)
compare alternative process design for a typical reactor/stripper configuration in order to optimize
the steady state economics and the controllability of the plant. As a generic rule to minimize the
effect of the recycle and to avoid amplification of disturbances (snow-bull effect) is proposed to fix
the flow rate in the recycle loop.
The design of a recycle compensator is proposed by Taiwo (1986) and its performance for set point
tracking is analysed in a more recent paper (Taiwo, 1993).
This paper analyses possible improvements in closed loop performance which can be obtained by
including some knowledge of the recycle process in the controller design. This will be done in
different ways: by a proper tuning of a controller having a fixed structure (PI); by designing a
controller with advanced algorithm (Internal Model Control, Morari and Zafiriou, 1989); by adding
a compensator to a PI or IMC controller.
The complexity of the control structure increases, therefore it will be seen when the different
characteristics of the recycle process and the control objectives (set point tracking or disturbance
suppression) justify the most complex one. The model of the recycle process has to be considered
to include some uncertainty, for this reason the robustness of different control schemes will be
analysed in the presence of parameter errors.

S4OY
s410 EuropeanSymposiumon ComputerAided Process Engineering-5

RECYCLE PROCESSES AND COMPENSATORS

Different recycle processes have been studied in literature and different block diagram schemes
have been proposed to analyse their dynamics. A block diagram containing essential elements of
the recycle process is given in Fig. 1. The process consisting of two chemical reactors in series,
described by Ray (1981) and the chemical reactor plus distillation column illustrated by Luyben
(1993a), can be reduced in a straightforward way to the scheme of Fig. 1. P, Pd, P, account for
the dynamics of process, disturbance and recycle, while K represent the recycle compensator. The
effect of manipulated variable (u) and disturbance (d), on the controlled output (y), in open loop
conditions and without compensator, is given respectively by:

y/d = PdR; y/u = PR; R = (1 - P,)- (1)


The closed loop system response can be expressed as:

PRC PciR
Y(T, d) = (2)
l+PRCr+ ltPRCd
Recycle dynamics appears in a highly nonlinear way in the system response, thus its effect on
stability and performance is not easy to predict a priori. Therefore, the importance of accounting
for the recycle in the design of the control system becomes evident.
The effect of the recycle can be eliminated by adopting a compensator K, which allows to base the
controller design on the main process; see Taiwo (1985,1993). The effects of manipulated variable
u and disturbance d on the output y in the absence of recycle and in the presence of recycle plus
compensator If, are respectively (open loop conditions):

y(u, d) = Pu t Pdd; y(u, 4 = 1 u + I+ PdR


+pRRK PRKd
By imposing that y(u, d) must not change in the two cases, the compensator, designed referring to
the model of the process (P) and of the recycle (fi), becomes:

I? = (fi - 1)(Piq- = PJ- (4)

Under the hypotheses that the structure of the compensator given by (4) is physically realizable
(K = I?) and a perfect knowledge of the process and of the recycle is possible (nominal case
P = 9, R = I?), the effect of the recycle is completely eliminated. The closed loop response is given
by:

y(r+,d)= &y t~ pd
I-tPCd
and the controller C can be designed on the basis of P and Pd, neglecting the recycle. In general,
in the presence of modelling errors (P # P, R # a), th e c 1osed loop response is given by (with
K = I?, or K # k whether or not expression (4) is physically realizable):

y(T,d)= &T +&d; 4 = (1 +ppRRK,;


Pdk
= (1 +itK) (6)
The controller is still designed on the basis of P and Pd, neglecting the recycle; hopefully the effect
of the compensator will still be positive in improving performance, also in this case.

Fig. 1: Scheme of a recycle process with compensator

CONTROL SYSTEM DESIGN FOR A CASE STUDY

The process recently proposed by Luyben (1993a) will be analyzed. This is a very simple process,
but very effective in showing some relevant aspects of the problem previously stated.
European Symposiumon ComputerAided Process Engineering-S s411

Referring to Fig. 1 transfer functions are the following:

P = Pd = KP/(rPs + l)2; P, = PK,/(r,s + 1) (7)

For this process Luyben (1993a) shows that a PI controller, tuned to give good performance (as
a compromise between set point tracking and disturbance rejection) for the process P without
recycle, can give very unsatisfactory performance and even instability, in the presence of recycle.
This performance, which is obtainable by the PI controller in the presence of recycle and perfect
compensator, will be considered optimal and taken as a reference for comparison with the other
controllers.
In the sequel, the reference case (R 1 ) will be compared with the 3 control systems:
- R2: IMC controller designed for a process without recycle P, augmented by a compensator K,
- Rs: PI controller tuned for a process with recycle P = PR, without compensator K,
- Ra: IMC controller designed for a process with recycle P = PR, without compensator K.
The scope is to show when the higher complexity of the control structure is justified by the im-
provement of performance, and to separate improvements coming from the controller algorithm
from those coming from the control structure.
The compensator in this very simple case is physically realizable and has a transfer function:

K = ii = j?:/i, = rl-,/(i,s + 1) (8)

The design of the PI controller: Cpz = K,(l f l/ris) is straightforward; for the controller which
takes into account the recycle process, the tuning is accomplished to obtain the same speed of
response in the y(r) response, as for the reference case.
The feedback controller C, designed according to the IMC design procedure (Morari and Zafiriou,
1989), is obtained as:
CIjuC = q/(1 - Pq); q = jF (9)
The nominal controller $ depends on I, the model of the overall process, as seen by the controller
in the forward path (p = p with compensator, F = pfi without compensator). In general, i
depends also on the input to the process, that is its algorithm is specific for each input.
The filter F, for type-l and type-2 inputs, takes the form respectively:

Fl = l/(Xs + 1); F2 = (nXs + l)/(Xs + l)n (IO)

where: the exponent n is chosen to ensure physical realizability of the equivalent feedback controller
and the filter time constant X to give robustness. By increasing the value of X, the controller action
is damped, thus slowing down the closed loop response.
Being based on the model of the process, the structure of the resulting controller is usually higher
than a standard (PID). In the special case of a minimum phase system, the same controller, optimal
for every input, is given by: 4 = l/p. In this case the output response in the nominal case becomes
for the IMC controller, with and without a perfect compensator k:

y(r, d) = FT + Pd(l - F)d (11)


Y(T, d) = FT + P,#( 1 - F)d (12)
It can be noted that by adopting an IMC controller based on P = PR (without compensator),
the effect of recycle can be completely eliminated for set point tracking, while it is not possible for
disturbance suppression. The relevance of this fact will be evident from the cases illustrated below.

SIMULATION RESULTS

Referring to Luyben (1993a), the process P (7) has these values of the parameters: K = ~~ = 1.
Parameters of the recycle transfer function have been varied to analyse situations wit*K dynamics
of the recycle faster/slower than the process (r, = 0.1~ 10) and with negative/positive recycle gain
K, < 0 + K, > 0. The overall process P = PR, as seen by the controller C in the absence of the
recycle compensator, becomes:

=PR= (S +1)2;s+fll)
_ K,

It is characterized by a negative zero and three poles, which can have different location in the
complex plane (positive/negative, real/imaginary), according to the values of the parameters of
the recycle transfer function (K,, TV).
S412 European Symposium on Computer Aided Process Engineering-5

Four significant processes originated by the recycle parameters are now illustrated. Results can be
considered representative of possible situations generated by variations of recycle parameter for the
process given by (7).
The IMC controllers with (Rs) and without compensator (Rd), become from (9):

R2=_ F = (s t l& 04)


P(l - F)
F = (3 t 1)2(v t 1) - K, F
R4= __ (15)
PR(l- F) 7,s t 1 1-F

In the presence of a compensator the same speed of response of the reference PI controller RI, for
K, = 1, 7; = 1.778 (Luyben, 1993a) can be obtained by the IMC controller Rz, for X = 0.5, n = 2.

Case #l: K, = -2;r, = 0.1


The overall process P = PR is stable with poles: pr = -10.23, p3,p3 = -0.88 f i1.467.
Without compensator, the PI controller (Rs), with Kc = 1, ri = 0.25, and the IMC (Rd), from
(15), with X = 0.5, n = 2, give the same speed of response as the reference case.
Simulation results are reported in Fig. 2 for set point tracking y(r) and in Fig. 3 for disturbance
rejection y(d). It can be seen that an appropriate tuning of the PI controller (Rs) is able to give
almost the same performance as the reference controller (RI), which uses a compensator. The
IMC controller (Rd) gives better performance, with respect to the PI, owing to its higher order
algorithm.
The compensator seems to be unnecessary, when an IMC controller can be used, as it gives exactly
the same response for y(r), whl 1 e a somewhat slower return to steady state is shown in the y(d)
curve, (Rs vs. Rd). Th is result, as far as set point tracking is concerned, has been confirmed
for ail the remaining cases and, from now on, we will concentrate on the problem of disturbance
suppression.

Fig. 2: Set point response for case #l. Fig. 3: Disturbance rejection for case #l.

Fig. 4: Disturbance rejection for case #2.

Case #2: K, = 4; r, = 1
The overah process P = PR is open loop unstable with poles: pi = 0.587, ps,p3 = -1.79 f i1.37.
Without compensator the PI controller designed for P = PR, according to the values chosen for
the integral action constant r;, can originate stability (for K, > K,+,), conditional stability or may
not be able to stabilize the process. Acceptable responses, with much higher overshoot in y(r), can
be obtained by the PI regulator Rs, with Kc = 4,~; = 1.
European Symposium on Comiuter Aided Process Engineering-S s413

The IMC design can be easily extended to the case of open loop unstable processes, by adopting
some modifications to the standard procedure. At the expense of a slightly more complex controller
structure, the IMC controller (&) allows successfully to overcome problems of open loop instability
and to give good results. In particular the filter in (15) assumes the form reported below, with:
7 = X3p2 + 3X2p + 3X = 5.107, /3 = pl = 0.587 and X = 1.

F = (7s + l)(Xs + 1)-3; 06)


Simulation results are reported in Fig. 4 for disturbance rejection. Again the compensator does
not seem necessary, when an IMC controller can be used. From now on, we will concentrate on the
comparison between the IMC controller with (R2) and without compensator (Rd).

Case #3: K, = 0.8; T, = 1


The overall process P = PR is open loop stable with poles: pl = -0.07, p2,p3 = -1.46 f i0.79.
The IMC controller (Rd), designed for P = PR, without compensator, is computed from (15)
with: X = 0.5, n = 2. From time responses reported in Fig. 5, it can be noted that the performance
of the IMC controller (Rh) is not acceptable: the presence of the small pole causes a very slow
recovery towards the steady state. Improvements can be obtained in suppressing the disturbance
by adopting a smaller value of the filter time constant A, but at the expense of a loss of robustness.
A type-2 filter (Fr) can improve the time response, at equal robustness, as firstly shown by Scali
et al. (1992). Fr is computed from (lo), with: A = 1, n = 3, and the controller (Rz) is still
obtained from (15). The corresponding response (Fig. 5) shows some improvements, but the use of
the compensator is appropriate in this case.

Case #4: K, = -7; 7, = 1


The overall process P = PR is open loop stable with poles: pl = -2.91, p3,p3 = -0.043 f i1.65.
The IMC controller (Ra), designed for PR without compensator (15), has: X = 0.5, n = 2. Again,
in the case of disturbance rejection (Fig. 6), its performance is unsatisfactory. A step disturbance
is converted by the recycle in a high amplitude oscillation, which persists for very long time. This
is a consequence of the dominating imaginary pole in the recycle process, which influences the y(d)
response, as shown by (12).
In this case, no significant improvements in suppressing the effect of the disturbance can be obtained
by adopting the type-2 order filter. Obviously, the PI controller without compensator (Rs) (not
shown for the sake of brevity), gives oscillating responses both in Y(P) and in y(d). Therefore in this
case the adoption of a recycle compensator is absolutely necessary, for any controller, to guarantee
good performance for disturbance suppression.

(I 2 6 I I LO II I4 16 II m

Fig. 5: Disturbance rejection for case #3. Fig. 6: Disturbance rejection for case #4.

Robustness of the recycle compensator


Simulation results shown above have been obtained in the nominal case, that is by assuming
a perfect knowledge of the main process P and of the recycle R. Every controller is designed
by referring to a model (the nominal process), but, qualitatively speaking, controllers which are
model-based may be more sensitive to process uncertainty than standard PID. This happens also
for the compensator which is able to give a perfect suppression of the effect of the recycle in the
nominal case, but not when uncertainty is present.
To account in a direct way for the robustness of the compensator, the simple case of uncertainty in
the two parameters of the recycle transfer function has been considered (K, = k, f 6; 7, = ?, f 6).
Assuming the same amount of percentage uncertainty b in K, and TV,,four possible extreme cases
are generated, corresponding to possible combinations of f6, in the parameters.
Results for disturbance rejection in the case study #3 (recycle process with slow real poles), for
which the adoption of the compensator have been found appropriate to give good performance, are
reported in Fig. 7 and 8.
s414 European Symposium on Computer Aided Process Engineering-S

It can be seen that the control structure including the recycle compensator plus the IMC controller
shows good robustness up to high uncertainty, 6 = 40%. All the curves correspondin to the
four combination of parametric uncertainty remain very close to the nominal response f Fig. 7).
The control structure including the recycle compensator plus the PI controller (Rr is reported for
comparison with the IMC plus compensator (Rs) for the worst cuse of uncertainty 2Fig. 8). A good
robustness is shown also by the structure including the PI controller. Even if in processes having
more complex dynamics, results may not be so excellent (Antonelli, 1995), the control structure
making use of a recycle compensator shows good robustness.

Fig. 7: Robustness of the compensator plus IMC structure Rz: nominal (-), uncertain (- -).
Fig. 8: Compensator robustness for the two control structures; IMC: nominal (Rz), uncertain (Rl);
PI: nominal (RI), uncertain (R;).

CONCLUSIONS

Some conclusions of general validity can be drawn from the cases examined above.
A specific design of the controller, which takes into account the dynamic characteristics of the
recycle process can be successful in many cases, to counteract negative effects of the recycle. A
PI controller shows some limitations deriving from its simple algorithm in the case of recycle
process with unstable open loop poles, for which may not be able to stabilize the process. An
IMC controller does not suffer of limitations in stabilizing the system and makes the use of the
compensator unnecessary for set point tracking problems. Instead, in the suppression of step
disturbances, for the case of a recycle process with slow imaginary poles, a slow return to steady
state conditions, with persisting oscillations, cannot be avoided.
The adoption of a recycle compensator is necessary in this last case to allow better performance
with respect to the specific controller. Also, a good robustness to uncertainty in the recycle process
parameters is shown by the structure including the recycle compensator.

REFERENCES

Antonelli, R. (1995). Tesi di laurea (in italian). Chem. Eng. Dept., University of Pisa, (I).
Denn, M.M. and R. Lavie (1982). Dynamics of plants with recycle. Chem. Eng. J.; 24, pg.55
Luyben, W. L. (1993a). Dynamics and control of recycle systems. 1. Simple open-loop and
closed-loop systems. Ind. Eng. Chem. Res.; 32, pg.46
Luyben, W. L. (1993b). Dynamics and control of recycle systems. 2. Comparison of Alternative
Process Design. Ind. Eng. Chem. Res.; 32, pg.476
Luyben, W. L. (1993c). Dynamics and control of recycle systems. 3. Alternative process design in
a ternary system. Ind. Eng. Chem. Res.; 32, pg.1142
Luyben, W. L. (1994). Snowball effects in reactor/separator processes with recycle. Ind. Eng.
Chem. Res.; 33, pn.299
Morari, M. and E: -Z&iriou. (1989). Robust Process Control. Prentice Hall Publ.; Englewood
Cliffs. NJ. (USA)
Ray, W: H.-(l&i). Ad vanced Process Control. McGraw-Hill Publ.; New York, N.Y. (USA)
Scali, C., D. Semino, and M. Morari. (1992). Comparison between Internal Model Control and
Linear Quadratic Optimal Control for SISO systems. Ind. Eng. Chem. Res.; 31, pg.1920
Taiwo, 0. (1993). Controlling plants with recycle. Proc. 2nd European Control Conference:
ECCg3; 1, pg.7
Taiwo, 0. (1986). The design of robust control systems for plants with recycle. Int. J. Control. ;
43, pg.671

You might also like