You are on page 1of 6

0263±8762/98/$10.00+0.

00
q Institution of Chemical Engineers
Trans IChemE, Vol 76, Part A, May 1998

DOUBLE-CONTROLLER SCHEME FOR SEPARATING


LOAD REJECTION FROM SET-POINT TRACKING
YU-CHU TIAN and FURONG GAO
Department of Chemical Engineering, The Hong Kong University of Science and Technology, Hong Kong

I
n the design of a conventional feedback control system, a dif® cult compromise has to be
made between the set-point tracking performance and load rejection performance. A
novel double-controller scheme is proposed in this paper to alleviate this compromise.
This control scheme has two controllers, a set-point controller and a load controller, which
result in the separation of the load response from the set-point response in a closed-loop
system. These two controllers can be designed independently to achieve good system
performance for both set-point tracking and load rejection. Also, good robustness of the
double-controller scheme to variations in system dynamics can be obtained by properly
designing the load controller. The proposed scheme is applied to two dif® cult problems, set-
point tracking in the presence of process uncertainty and control of a process with dominant
time delay. The results show that the proposed double-controller scheme is superior to a
conventional single controller system.
Keywords: process control; double controller; set-point tracking; load rejection

INTRODUCTION process is described by:


The design of a control system has two main objectives: set- 4
point tracking and load rejection, which are referred to P(s) = e- 3.5s (1)
7s + 1
respectively as the servo problem and the regulator problem.
In a classical feedback control system with a simple It is assumed that a PID controller is employed to control the
structure, shown in Figure 1, a single controller is used to above process. The ITAE PID controller settings are
simultaneously solve both the servo problem and the Kc = 0.435, Ti = 9.69 and Td = 1.13 for set-point tracking
regulator problem. The symbols R and L represent set- and Kc = 0.654, Ti = 4.98 and Td = 1.34 for load rejection,
point and load respectively. When the set-point changes, the where Kc , Ti and Td are respectively controller gain, integral
controller sends control signals to make the process track time and derivative time. These two sets of controller
the set-point. In the case of a load change, the same parameters are clearly different. To illustrate the effects of
controller regulates the process at a constant set-point to these two sets of controller parameters on system perfor-
reject load disturbance. mance, a unit step change in set-point at t = 0 and a
In the classical control system, this single controller is negative unit step change in load at t = 100 are respectively
expected to perform well for both the servo problem and the introduced as excitation signals to the system. Figure 2
regulator problems, i.e., for set-point tracking and for load shows the responses of the closed-loop system with these
rejection. This requirement is often con¯ icting. Optimal set- two sets of `optimal’ controller parameters. The solid line
point tracking and optimal load rejection cannot be obtained corresponds to the controller settings for set-point and the
simultaneously with a single controller. This con¯ ict can be dashed line to the controller settings for load. It is clear that
demonstrated by the ITAE (integral of the time-weighted the controller settings for set-point are more conservative
absolute error) PID (proportional-integral-derivative) con- than those for load. The settings for set-point result in a
troller design for a ® rst-order plus delay process1 . The sluggish load response, while the settings for load lead to
resulting optimal PID controller settings are different signi® cant overshoots to set-point changes. In reality, set-
depending on whether the set-point tracking or the load point changes and load disturbances are both likely to occur,
rejection are considered. Unless a pure regulatory problem a compromise in the controller settings, therefore, has to be
is considered, a compromise between set-point tracking and employed. There is no clear and systematic methodology to
load rejection has to be performed, which is a dif® cult task perform or eliminate such a compromise.
because there is no clear and simple compromise method. The objective of this paper is to propose a new control
Moreover, a compromise means that `optimal’ performance structure, which will be referred to as a double-controller
can be obtained for neither set-point tracking nor load scheme, to separate load response from set-point response
rejection. in a closed-loop system. The scheme contains two con-
To illustrate this inherent drawback of the single trollers, a set-point controller and a load controller. With this
controller system, consider an example discussed by double-controller scheme, both optimal set-point response
Seborg, Edgar and Mellichamp in their textbook2 . The performance and optimal load response performance can be
445
446 TIAN and GAO

Figure 1. A conventional single controller feedback system.

obtained simultaneously by independently designing the set- Figure 3. The double-controller scheme.
point controller and the load controller. Performance
compromise is no longer necessary. The effectiveness of
this proposed double-controller scheme will be demonstrated denote the controller 1 (set-point controller) and the
by applying to two dif® cult control problems: (1) set-point controller 2 (load controller) respectively, as shown in
tracking in the presence of process uncertainty; and (2) Figure 3. The process and its model are represented by P and
control of a process with dominant time delay. P* . Then:
Y (s) Gc1 (s)P(s) 1 + Gc2 (s)P* (s)
DOUBLE-CONTROLLER SCHEME R(s) = Hr (s) =
1 + Gc1 (s)P* (s) 1 + Gc2 (s)P(s)
The main feature of this scheme is the employment of two (2)
separate controllers to independently control set-point
Y (s) P(s)
L(s) = l = 1 + Gc2 (s)P(s)
response and load response. The controller for set-point is H (s) (3)
referred to as the set-point controller, and the controller for
load is called the load controller. To assure the separation of Equation (3) clearly indicates that the load response of the
these two responses, the control scheme should have the closed-loop system is determined only by the load controller
following features in its structure: Gc2 and has been separated from the set-point response.
(1) The load rejection loop does not contain the set-point Therefore, the load controller can be designed indepen-
controller; dently to achieve a good load rejection performance. This is
(2) The set-point controller does not use the process output an important feature of the double-controller scheme.
or any variables affected by load changes in the load From equation (2), with a good process model, i.e.,
rejection loop as its feedback signal; and P* (s) < P(s), the set-point response becomes:
(3) In contrast to the set-point controller as stated in (2), the Gc1 (s)P(s)
load controller uses the process output as its feedback Hr (s) = (4)
1 + Gc1 (s)P* (s)
signal.
Taking into account these three requirements, the ideal The above equation implies that the set-point response is
feedback measurement for the set-point controller is the independent of the load controller and determined only by
load-free process output, which is unfortunately unavail- the set-point controller. As a result, the set-point controller
able. This ideal feedback signal, however, can be produced can be also designed independently. This is another
by a process model. The process model input is the set-point important feature of the double-controller scheme. The
controller output. Figure 3 gives the structural diagram of design objective of the set-point controller is to obtain a
the double-controller scheme, where Y and Y * are process good set-point tracking performance.
output and model output respectively. It is seen from this For a properly designed set-point controller and load
® gure that the process model output is also fed to the load controller, the following relations hold:
controller as its set-point to eliminate the system set-point lim Hr (s) = 1, lim Hl (s) = 0 (5)
effect on the load rejection loop and to compensate for the s! 0 s! 0

inevitable model mismatch. These relations imply that the process output will track the
To analyse the double-controller scheme quantitatively, set-point at steady state without any error for set-point and
the overall transfer functions relating the system output to load changes. For example, with the PID control law in both
the set-point and the load are derived below. Let Gc1 and Gc2 the set-point controller and the load controller, the above
relations are true for most chemical processes if the closed-
loop system is stable. In addition to the PID controller, other
more sophisticated control algorithms can also be used in
the double-controller scheme to get improved system
performances.
The effectiveness of the double-controller will be shown
through applications to two dif® cult control problems in the
following sections.

APPLICATION TO SET-POINT TRACK ING WITH


PROCESS UNCERT AINTY
Figure 2. A comparison of PID controllers designed using ITAE criteria for Set-point tracking with process uncertainty is an
set-point (solid line) and for load (dashed line). important problem in the process industries. For example,

Trans IChemE, Vol 76, Part A, May 1998


DOUBLE-CONTROLLER SCHEME FOR SEPARATING LOAD REJECTION FROM SET-POINT TRACKING 447

set-points are adjusted to keep the process in optimum. This


may be due to the changes in the catalyst activation. In
injection molding, injection velocity set-point is pro® led to
be harmonious with the product shape. In cold rolled steel
strip annealing, the set-points of annealing furnace tem-
peratures are required to be continuously changed according
to the measurements of steel grip temperatures. In all these
examples, the closed-loop systems are required to track the
set-points quickly to ensure the production ef® ciency and/or
product quality.
Set-point tracking with process uncertainty is also a
dif® cult problem in the conventional single controller Figure 5. Set-point tracking responses of the double-controller scheme
system, since a good set-point tracking performance (solid lines) and the single controller system (dashed lines) for the delay
cannot be simultaneously met with a good load rejection time deviating from the nominal value of 3.5 to (a) 4.2 and (b) 2.8.
requirement. This constraint can, however, be easily
removed using the double-controller scheme due to its same load rejection performance. Therefore, the load
separation of the load rejection response from the set-point performances of the two systems will not be shown in the
tracking response. One may obtain good set-point tracking following simulations. It can be seen from Figures 4 and 2
performance by designing the set-point controller, while the that the double-controller scheme is clearly superior to the
load rejection performance can also be optimized indepen- single controller system in having `optimal’ performance in
dently. It will be shown through simulations that even in both set-point tracking and load rejection.
the presence of the process uncertainty, the set-point The effects of the process uncertainty on the closed-loop
tracking performance of the double-controller scheme is system performance are now studied with the controller
still superior to that of the conventional single controller parameters kept unchanged. The variation of the process
system. delay time is considered ® rst. The delay is changed 6 20%
Let us reconsider the example of equation (1), where the from its nominal value of 3.5. Corresponding to the positive
process was controlled by a conventional single controller. and negative delay changes respectively, Figures 5(a) and
Using the double-controller scheme, the ITAE PID set-point (b) give the set-point responses of the double-controller
controller settings are Kc1 = 0.435, Ti1 = 9.69 and Td1 = scheme (solid lines) and the single controller system
1.13, while the ITAE PID load controller settings are (dashed lines). It can be seen that the double-controller
Kc2 = 0.654, Ti2 = 4.98 and Td2 = 1.34. These two con- scheme has smaller oscillations and faster set-point
trollers are designed independently and no compromise tracking.
between the set-point tracking and the load rejection is The variation in the process gain strongly affects the
required. The results are to be compared to that of the single closed-loop system performance. The deviation of the
controller system. To compare the control performances process gain from its nominal value of 4 is considered.
under the same standard, the conventional single controller The set-point responses of the double-controller scheme
system has to be designed to behave with the same set-point (solid lines) and the single controller systems (dashed
response or load response as the double-controller scheme. lines) are shown in Figures 6(a) and (b) for the process
In this section, the set-point responses of the two control gain deviating to 5 and 2 respectively. Better set-point
schemes are compared with each other under the same load tracking performance with clearly shorter settling time and
rejection. smaller overshoot is obtained with the double-controller
Figure 4 gives the simulation results for double-controller scheme.
(solid line) and single controller (dashed line) systems when The variation of the process time constant also affects the
the process model is perfect. The excitation signal is the closed-loop system performance. Suppose the process time
same as that discussed in the Introduction, i.e., a unit step constant deviates to 6 and 10 respectively from its nominal
change in set-point at t = 0 and a negative step change in value of 7. The corresponding set-point responses are shown
load at t = 100. As designed, the load responses of the in Figures 7(a) and (b) respectively, where the solid lines
double-controller scheme and the single controller system
overlap each other, indicating these two systems have the

Figure 4. A comparison of the set-point tracking performances of the Figure 6. Set-point tracking responses of the double-controller scheme
double-controller scheme (solid line) and the single controller system (solid lines) and the single controller system (dashed lines) for the process
(dashed line). gain deviating from the nominal value of 4 to (a) 5 and (b) 2.

Trans IChemE, Vol 76, Part A, May 1998


448 TIAN and GAO

Figure 8. An alternative structure of the double-controller scheme for


dominant delay processes.

Figure 7. Set-point tracking responses of the double-controller scheme


(solid lines) and the single controller system (dashed lines) for the process of the load controller. As a result, the set-point control loop
time constant deviating from the nominal value of 7 to (a) 6 and (b) 10. does not contain any time delay and therefore can be
designed to have a fast set-point tracking speed. This
alternative double-controller scheme is used throughout in
represent the double-controller scheme and the dashed this section. The delay may be included in the set-point
lines represent the single controller system. Clearly, the control loop without clear performance deterioration if the
double-controller scheme gives superior set-point tracking delay is not dominant.
Let P(s) = G(s)e- ds , and consequently P* (s) = G* (s)e- d s ,
*
performance to the single controller system.
The above simulations show that with an `optimal’ load where G and G* do not contain any delay, d * is an estimate
rejection design, the double-controller scheme provides a of the delay d. The overall transfer function of the double-
better set-point tracking performance than the single controller scheme shown in Figure 8 for set-point changes
controller system even without a good process model. In is:
the same token, with the same `optimal’ set-point tracking
Gc1 (s)Gp (s)e- ds 1 + Gc2 (s)G*p (s)e- d s
*
performance, the double-controller scheme provides super- Hr (s) =
ior load rejection to the single controller system in presence 1 + Gc1 (s)G*p (s) 1 + Gc2 (s)Gp (s)e- ds
of system uncertainty. The results, however, will not be
(6)
shown here to reduce the paper length.
The overall transfer functions of Figures 8 and 3 for load
changes are the same as equation (3). Therefore, the
APPLICATION TO DOMINANT DELAY schemes of Figures 8 and 3 have the same load rejection
PROCESSES performance. From equation (6), if the process model is a
Time delay is a common phenomenon in industrial good description of the process, then:
processes. Also, many higher-order delay-free processes Gc1 (s)Gp (s)e- ds
can be approximated by a lower-order process plus a time Hr (s) = (7)
1 + Gc1 (s)G*p (s)
delay. The control of time delay processes is, therefore, an
important subject for the process industry. It is, however, a Compared with equation (4) for Figure 3, equation (7) does
dif® cult problem due to the severe limitations introduced by not contain any delay in its denominator, implying that
time delay, especially dominant time delay, on control better set-point tracking performance could be obtained for
performance and system stability. The Smith predictor is an the alternative double-controller scheme of Figure 8 than
effective control scheme to compensate the dominant delay that for the original double-controller scheme in Figure 3.
with a good process model3 . An inherent drawback of the Suppose the process dynamics are governed by a ® rst
Smith predictor is its performance sensitivity to the process order plus delay process as:
model, especially to the delay time. Deviation of the process Kp*
P* (s) = * e- d d
s
dynamics from the process model, for which the Smith (8)
Tp s + 1
predictor is designed, results in poor or even unstable
closed-loop performance. The double-controller scheme is This assumption is reasonable because many industrial
shown below to be very effective and more robust than the processes can be approximated by this kind of model. As
Smith predictor for processes with dominant time delay. shown below, the controllers designed on the basis of a ® rst-
This is another useful application of the double-control order plus delay model are also effective for higher-order
scheme to dif® cult control problems. processes, including an under-damped process. It has been
As in any closed-loop system, it can be seen from Figure mentioned previously that for a ® rst-order plus delay
3 that if the processes and consequently the process model process, the relations shown in equation (5) obtained for
contain a pure time delay, the set-point tracking speed will Figure 3 still hold for the alternative double-controller
be limited in the set-point controller design, especially in the scheme shown in Figure 8 with a PID type of controller,
presence of a dominant time delay. To further improve set- which indicates that the closed-loop system will track the
point tracking in a closed-loop system, a slight modi® cation set-point at steady state without offset for a stable closed-
is made to the original double-controller scheme shown in loop system. The design of the set-point and load controllers
Figure 3 to yield an alternative structure shown in Figure 8. is discussed below.
It can be observed from Figure 8 that the delay term of the Equation (7) can be viewed as a combination of a pure
process model has been moved to the position between the delay term and a delay-free feedback system with the set-
feedback signal of the set-point controller and the set-point point controller. The set-point controller can then be

Trans IChemE, Vol 76, Part A, May 1998


DOUBLE-CONTROLLER SCHEME FOR SEPARATING LOAD REJECTION FROM SET-POINT TRACKING 449

designed as for a process without delay under the


assumption of a perfect model. This is the same as the
Smith predictor. For a ® rst-order process model without
delay, the application of the direct synthesis method is
straightforward for the set-point controller design. Suppose
the desired set-point response of the closed-loop system
is Hre (s) = Gr (s)e- ds . Making equation (7) equal to Hre (s)
gives:
1 Gr (s)
Gc1 (s) = * (9)
Gp (s) 1 - Gr (s) Figure 9. A comparison of the double-controller scheme and the Smith
For the process model described by equation (8), the desired predictor in the case that the process model is perfect.
Gr (s) can be chosen to be:
Gr (s) = 1/ (Te s + 1) (10) (14). Like the previous simulation example, a unit step
change in set-point at t = 0 and a negative step change in
where Te is the desired closed-loop time constant.
load disturbance at t = 100 are introduced as excitation
Substituting equation (10) into (9) leads to:
signals. To compare the performances of the two control
Tp* 1 schemes under the same standard, the two schemes are
Gc1 (s) = 1+ * (11)
Kp* Te Tp s tuned to have the same set-point response at the nominal
case. This implies that the same PI controller settings are
Equation (11) is a PI controller with the controller gain Kc1 adopted in the set-point controller of the double-controller
and integral time Ti1 as: scheme and the Smith predictor controller.
Kc1 =T *
p / (Kp Te ),
*
Ti1 =T *
p (12) Figure 9 shows the responses of the double-controller
scheme (solid line) and the Smith predictor (dashed line)
Selecting 0 < Te < T or Te $ T will result in a lead or lag
*
p
*
p with a perfect process model. The set-point responses of the
control action. When Te = Tp* , the open-loop time constant two systems overlap each other, implying the two systems
is retained in the closed-loop system. have the same set-point tracking performance. The load
The direct synthesis method is improper for the load response of the double-controller scheme is slightly sluggish
controller design because both the numerator and the compared to that of the Smith predictor to accommodate the
denominator of the load response Hl (s) contain delay inevitable model uncertainty.
term. A PI type control law has been selected for the load For all the following simulations, the process model and
controller design for simplicity. This, however, does not consequently all the controller settings, are kept unchanged,
mean that only PI control law should be used for the load while the process dynamics are changed to simulate the
controller design. Conventional PI tuning methods cannot process uncertainty in delay time and dynamic order. To
be used directly for processes with a dominant time delay. It reduce the length of this paper, the variations of process
has been found that Haalman’ s formulae give acceptable time constant and process gain, which affect the perfor-
results4 although it is conservative for dominant delay mance of the closed-lop system but not as strongly as
processes. It has, therefore, been decided to modify process delay and process dynamic order, are not con-
Haalman’ s formulae for tuning the load controller. For a sidered. For delay time variations, simulations show that the
® rst order plus delay process model (8), Haalman’ s PI Smith predictor becomes unstable for d > 5.2 or d < 2.7,
controller parameters are: while the double controller scheme has a much larger stable
2Tp* range of 0 < d < 9.9. Figure 10 gives the responses of the
Kc = , Ti = Tp
*
(13) double-controller scheme (solid line) and the Smith
3Kp* d *
predictor (dashed line) for d = 3. The double-controller
In the case of dominant time delay, a common modi® cation scheme has fast set-point trackings and good load rejection
factor a is introduced into the Haalman’ s controller gain and without obvious oscillations, indicating that the double-
integral time to speed up the load response with: controller scheme is superior to the Smith predictor.
Kc2 = a Kc , Ti2 = a Ti , a $ 1, (14) If process dynamics change from the original ® rst-order
where a = 1 corresponds to the original Haalman’ s
formulae (13). The selection of a should guarantee a
suf® ciently large gain margin and phase margin.
As an example, the process approximated by the ® rst-
order plus delay model (8) is considered with Kp* = 2,
Tp* = 0.8 and d * = 4. The double-controller scheme of
Figure 8 is applied to this process. To tune the PI set-point
controller, the desired closed-loop time constant Te in
equation (10) is set to be the same as the open-loop time
constant Tp* = 0.8, leading to the PI set-point controller
settings of Kc1 = 0.5 and Ti1 = 0.8 from equation (12). The
modi® cation factor a is chosen to be 2.5 for the load Figure 10. A comparison of the double-controller scheme and the Smith
controller design. The corresponding PI load controller predictor for the process delay time deviating from the nominal value of 4
settings are Kc2 = 1/ 6 and Ti2 = 2 from equations (13) and to 3.

Trans IChemE, Vol 76, Part A, May 1998


450 TIAN and GAO

The following conclusions may be drawn. A simple and


practical double-controller scheme with two alternative
structures has been proposed in this paper to separate load
rejection response from the set-point tracking response in a
closed-loop system. The compromise between these two
responses required in the conventional single controller
system is no longer necessary. The double-controller
scheme contains two controllers, a set-point controller and
a load controller. Set-point tracking performance and load
rejection performance can be optimized separately by
Figure 11. Responses of the double-controller scheme and the Smith independently designing each controller. The proposed
predictor in the case that the process dynamic changes signi® cantly to an double-controller scheme has been applied to two dif® cult
under-damped higher-order plus delay.
problems: (1) set-point tracking in the presence of the
process uncertainty and (2) control of the processes with
plus delay to an under-damped higher-order plus delay dominant delay time. It has been shown that in both cases
with P(s) = 2/ [(0.8s + 1)(s2 + 0.7s + 1)]e- 4s , as shown in the double-controller scheme performs well. The perfor-
Figure 11, the Smith predictor becomes unstable and mance of the double-controller scheme can be improved
therefore fails to work. In contrast, the double controller even further by using sophisticated algorithms for the set-
scheme works well even when the process dynamics have point controller and/or the load controller, but the emphasis
been changed signi® cantly. These simulation results of this paper is to introduce this novel control structure. The
indicate that excellent robustness, fast set-point tracking authors believe that the double control scheme, with its
and good load rejection can be achieved with the proposed simple structure, can be easily applied to many industrial
double-controller scheme. This scheme provides a structure processes.
that allows a separation control of the set-point tracking and
the load rejection. It is, therefore, possible to design a load
controller to accommodate large system uncertainty, while a REFERENCES
fast set-point response is maintained by a separate set-point 1. Lopez, A. M., Murrill, P. W. and Smith, C. L., 1967, Controller tuning
controller. relationships based on integral performance criteria, Instrum Technol,
14(11): 57.
2. Seborg, D. E., Edgar, T. F. and Mellichamp, D. A., 1989, Process
Dynamics and Control (John Wiley & Sons, New York) pp. 285±286.
DISCUSSIONS AND CONCLUSIONS 3. Smith, O. J. M., 1957, Closer control of loops with dead time, Chem
Utilizing multiple-controller in a system has been Eng Prog, 53(5): 217±219.
successful for dif® cult industrial problems. For example, 4. Haalman, A., 1965, Adjusting controllers for deadtime processes,
Control Eng, 12: 71±73.
in a multi-input multi-output system, additional controllers
are used to eliminate the interactions of different control
loops. A cascade control scheme also uses two controllers, a ADDRESS
master controller in the outer loop and a slave controller in Correspondence concerning this paper should be addressed to Furong
the inner loop, to suppress the load effect and to speed up Gao, Department of Chemical Engineering, The Hong Kong University
set-point response. These multiple-controller schemes, of Science and Technology, Clear Water Bay, Kowloon, Hong Kong.
however, do not address the response decoupling issue of (E-mail: KefGao@ust.HK). Yu-Chu Tian is on leave from the Institute
the load rejection and set-point tracking. A system using the of Industrial Process Control, Zhejiang University, Hangzhou 310027,
P.R. China.
novel double-controller structure proposed in this paper
allows a separate optimal design of both the load rejection The manuscript was received 4 November 1997 and accepted for
response and set-point response. publication after revision 26 January 1998.

Trans IChemE, Vol 76, Part A, May 1998

You might also like