Professional Documents
Culture Documents
Chemical Engineering Communications
Chemical Engineering Communications
To cite this article: HSIAO-PING HUANG , I-LUNG CHIEN , YUEH-CHUNG LEE & GOW-BIN WANG (1998): A SIMPLE METHOD FOR
TUNING CASCADE CONTROL SYSTEMS, Chemical Engineering Communications, 165:1, 89-121
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to
anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should
be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims,
proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in
connection with or arising out of the use of this material.
Chm. &. Comm..1998. Vol. 165, pp. 89-121 Q 1998 OPA (OK- Publirhm Asmiation)
Reprinu availabk directly fmm the publisher Amsterdam B.V. Fublirhcd under lknr
Photompying permitted by l i m w only under thc Oordon and B m h Scicna
Publishers imprint.
Rink4 in India.
A simple method for tuning controllers in a cascade system is presented. In this method, all the
relations that facilitate the tuning procedures are well prepared in terms of figures or simple
equations. Using thew figures and equations, the controller tuning for different configurations
of cascade systems becomes easy and straightfonvard when process models are available. On
the other hand. when process models are not available, a simple method that w s one single run
of step input experiment to develop such models is proposed. Based on these developments in
the controller tuning and process models, an autotuning system that uses relay feedback is
presented. Unlike the existing autotuning systems, this proposed system conducts identification
and controller tuning in a decoupled manner. As a result, no excessive trial- and-error efforts for
modeling.and tuning are required. Simulation results show the potential usage of such a
method. It is interested to see that the resulting systems have almost compatible responses to
those systems which have been designed optimally in one way or another as reported in the
literature. It is not, however, the purpose of this article to emphasize on obtaining superior
performance to all other existing methods, but to emphasize on its effectiveness and simplicity
for application.
Keywordr: Cascade control; primary open-loop; controller tuning; autotuning; relay feedback
1. INTRODUCTION
In a cascade control system, the output of the "primary" (also called master
or outer) controller is used to manipulate the setpoint of the "secondary"
'Corresponding author.
t ~ c p a r t m e n t of Chemical Engineering, National Taiwan University of Science and
T hnology, Taipei 106, Taiwan. R.O.C.
?Department of Chemical Engineering, Chang Gung University, Kweishan, Taoyuan 333,
Taiwan, R.O.C.
90 H.-P. HUANG er a/.
(also called slave or inner) controller, while the secondary controller is the
only one which has an output to the process. These two controllers are
arranged in a way that the secondary loop is nested inside of the primary
one. Each controller has its own process measurement but only the primary
controller will keep its process output follow an independent setpoint
command from outside, while the secondary controller with its intermediate
process output measurement will follow the command from the primary
controller output.
The benefits of a cascade control system are discussed in many process
control books (cf. Smith and Corripio, 1985; Seborg et a/., 1989; Luyben,
1990; Ogunnaike and Ray, 1994; Marlin, 1995) and also in a number of
Downloaded by [Universite Laval] at 00:06 21 June 2013
papers (Franks and Worley, 1956; Schork and Deshpande, 1978; McMillan,
1982; Krishnaswamy el al., 1990; Verhaegen, 1991). The principal
advantages of cascade control are perhaps best described by the book of
Shinskey (1988) as:
I. Disturbances arising within the secondary loop are corrected by the
secondary controller, often before they can influence the primary
variable.
2. Phase lag existing in the secondary part of the process is reduced
measurably by the secondary loop. This improves the speed of response
of the primary loop.
3. Gain variations in the secondary part of the process are overcome within
its own loop.
4. The secondary loop permits an exact manipulation of the flow of mass or
energy by the primary controller.
Because the two loops of a cascade system are nested together, how to
properly tune these two controllers becomes very important or else the
potential benefits claimed above will be significantly lost. There are quite a
few papers in the literature targeting on tuning such cascade control
systems. Some of these papers aim at minimizing integral criteria, e.g.,
ITAE, ISE, IAE, etc. (Krishnaswamy et a/., 1990, Krishnaswamy and
Rangaiah, 1992; Wang et a/., 1995; Huang et al., 1996a); some use Ziegler-
Nichols formulae (Schork and Deshpande, 1978; Krishnaswamy and
Rangaiah, 1987). Other approaches include: Hang et al. (1994) using two
relay feedback tests to obtain the tuning parameters of cascade controllers;
and Eker and Johnson (1996) using a controller derived from LQG or H,
approach in a form other than PID. Although the controller parameters are
determined differently, the basic approach is the same: the secondary loop is
tuned first and then the primary one follows.
TUNING CASCADE CONTROL SYSTEMS 91
In all the tuning methods having been proposed in literature, there are
common disadvantages or difficulties in either technical or practicing
aspects. In practicing aspect, during the tuning stage, besides modeling for
the primary process G,, and secondary process Gp2,one has to find ultimate
information (such as: ultimate frequency and ultimate gain) or develop a
parametric model for the primary open-loop. Since this primary open-loop
includes so many parameters, general relations for modeling are not
available. Thus, identification procedures have to be used from time to time
during the design or tuning stage. This would cause inconvenient and would
take a lot of efforts. In technical aspect, difficulties are associated with the
way those tuning rules are established. First, the tuning results are usually
Downloaded by [Universite Laval] at 00:06 21 June 2013
which have been designed optimally in one way or another as reported in the
literature. ~evertheltss,it is really not our purpose to emphasize on
obtaining superior performance to all other existing methods, but to
emphasize on the simplicity and readiness of this method for application.
functions, respectively. Similarly, G,1 and GpI are those for the primary
loop. The dynamics of valve and measurements are neglected for simplicity
purpose. The process models (i.e., G ~ ; , I and cP2) are assumed to be
approximated as first-order plus dead-time (FOPDT) forms.
A conventional approach to tune the controllers in a cascade control
system is to start with tuning the secondary loop first. Tuning this secondary
loop is hardly different from tuning a conventional single loop, except that
the controller is usually confined to P or PI form. Ater this step, the well
tuned secondary loop is combined with Gpl, the primary process, to become
the primary open-loop. A proper model to fit this primary open-loop is then
derived and is used to tune the primary controller. Although it seems simple
Downloaded by [Universite Laval] at 00:06 21 June 2013
then we have
where S(r) designates the step response of G(s) and k is the steady-state
gain of G(s). Thus, if any two transfer functions, say Hl(s) and H2(s), are
94 H.-P. HUANC er al
We should also mention the fit-3 method of Smith and Corripio (1985) to
find a FOPDT model to fit an overdamped step response of a given process.
Assume that S(t) is the given overdamped step response recorded from an
open-loop step test on the process. The corresponding dead time 0 and the
time constant T can be calculated from the following formulae
Downloaded by [Universite Laval] at 00:06 21 June 2013
where 10.~8and 10.63 are the times that S(t) reaches the values of 0.28 S(t,)
and 0.63 S(t,), respectively.
Using first-order Taylor approximation for the dead time, the resulting
IMC-PI settings are given by
TUNING CASCADE CONTROL SYSTEMS 95
where 7 f 2 is the only tuning parameter of this loop. A model for this
secondary loop would be an underdamped second-order process that
approximates the complementary sensitivity function of this loop, i.e.,
where k, 7, <, and e are the steady-state gain, the time constant, the
damping factor, and the dead time of the secondary loop, respectively.
Let k= l and by normalizing time with respect to e2, we have
Downloaded by [Universite Laval] at 00:06 21 June 2013
<
It is found that ?/02 vs. rf2/02 and vs. v2/& have a linear relationship
as shown in Figure 2. The results are obtained from simulating the step
responses of both sides of Eq. 10 and followed by employing proper
identification methods (for example, Sundaresan et al. (1978) for over-
damped SOPDT model; Chen (1989) for underdamped SOPDT model) to
match the resulting responses. On the other hand, the apparent dead time
of this equivalent open-loop process can be taken as
FIGURE 2 The relations o f ? and vs. its tuning factor TD for a secondary PI-loop.
where
Simulation results to match the step responses of both sides of Eq. 13 give
the following relations as shown in Figure 3. The quantified equations for
TUNING CASCADE CONTROL SYSTEMS
Downloaded by [Universite Laval] at 00:06 21 June 2013
FIGURE 3 The relations of r and < vs. its tuning factor kn for a secondary P-loop.
this figure are given in Appendix A. Similarly, the equivalent deadtime
for this secondary loop is given as
Thus, once the tuning parameter 7/2 or kc2 is assigned to a given secondary
process, the equivalent second-order process to this secondary loop can be
found.
where k*, T * , C*, and 8' are the steady-state gain, the time constant, the
damping factor, and the dead time of the SOPDT model for the primary
Downloaded by [Universite Laval] at 00:06 21 June 2013
If the right hand side of Eq. 21 gives a negative value, AB' is assumed to
be zero.
Although the closed-loop transfer function of the secondary loop is
usually specified as an underdamped system for fast disturbance
rejection purpose, the primary open-loop can vary from an under-
damped to an overdamped system depending on how slow the primary
process G,,, is. By using the dimensionless groups, matching the time
domain step responses of each side of Eq. 18 gives the results as shown
in Figure 4.
According to the same simulation results, it is very interesting to find
<
that, for = 0.25 0.7, r*and t have the following relation (Lee, 1996)
N
TUNING CASCADE CONTROL SYSTEMS
Downloaded by [Universite Laval] at 00:06 21 June 2013
On the right hand side of the above equation, the first three terms are
known from the process models. Meanwhile, 7/02 can be read from
Figure 2 and C* can be read from Figure 4. Thus, if 712 has been
assigned, the value of '8 can be easily found.
where ko, T O , and Bo are the steady-state gain, the time constant, and the
dead time of the FOPDT model for the primary open-loop, respectively,
and ko = & I . It is interested to find that TO/? can be formulated in terms
of T I / ? as shown in Figure 5 (Lee, 1996). In other words, the equivalent
time constant r0is a function of r 1and 7 only when the secondary loop is
obviously overdamped. The value of Bo for such cases can be estimated as
Downloaded by [Universite Laval] at 00:06 21 June 2013
If GpIand Gp2 are known at the moment of tuning, the procedure would be
most straightforward. In this case, we don't need to perform any further
open-loop test but simply follo.wing the conventional tuning approach, i.e.,
from inner loop to the outer, by making uses of those relations presented in
the previous section and the charts for performance specifications to be
shown in the following.
To tune the secondary loop, the relations of damping factor and the
tuning factor rf2 have been given in Figure 2. The relating performance
indices such as maximum peak gain M, rising time (T,-B2)/B2, and peak
time (Tp-e2)/B2to each choice of rf2are given in Figure 6. If the secondary
loop is to emphasize its fast response to correct for the intermediate load
TUNING CASCADE CONTROL SYSTEMS
Downloaded by [Universite Laval] at 00:06 21 June 2013
FIGURE 6 The performance charts for a PI-loop with open-loop FOPDT process.
can then be read from Figure 4 or Figure 5 together with Eqs. 22 and 23 or
Eq. 25, depending on whether it is highly damped or not. The tuning
parameters based on IMC can be calculated for an underdamped process as
Downloaded by [Universite Laval] at 00:06 21 June 2013
FIGURE 7a The relations of tuning factor ~,,/ff VS. @'/refor the SOPDT primary open-
loop (0.4 < C' < 0.9).
The relations of Eqs. 31 and 32 are plotted for the SOPDT primary open-
loop as shown in Figures 8a and 8b. Consequently, we can tune this primary
loop by adjusting r f l to give desired performance and robustness. On the
other hand, a primary open-loop of FOPDT form would result in a PI
controller for this loop. Thus, to choose a tuning factor for such a PI
controller, we can refer to Figure 2 and Figure 6 by replacing
rn/B2 with rfl/BO.From these figures, we can choose T f 1 in terms of the
ratio of T,-~/B~ to meet the requirements for system performance and
H.-P. HUANG e t a / ,
Downloaded by [Universite Laval] at 00:06 21 June 2013
FIGURE 7b The relations of tuning factor rfl/B'vs. B'/T' for the SOPDT primary open-loop
(1.0 < C' < 6.0).
robustness. In general, rf1 can be taken any value between 0.45 Bo and 0.8 Bo
depends on performance and robustness requirements. Thus, when tuning
this primary loop, it can be guided by what kind of model for this primary
open-loop is identified and what requirements of the system have been
specified.
Although it seems straightforward to tune a cascade control system
according to the procedure described above, choice of the damping factor in
the secondary loop would need some precautions. A default value of 0.25 for
TUNING CASCADE CONTROL SYSTEMS
Downloaded by [Universite Laval] at 00:06 21 June 2013
FIGURE 8a The performance charts for a PID-loop with open-loop SOPDT process
(0.4 < C'
< 0.9).
FIGURE 8b The performance charts for a PID-loop with open-loop SOPDT process
(1.0 < C' < 6.0).
Illustration
In the following, we shall use the tuning procedure described above to find
the controller settings for the selected cascade control system. It is the
purpose to show that using this simple procedure, the resulting system can
have almost compatible performance compared with those designed
optimally.
TUNING CASCADE CONTROL SYSTEMS
y+TL given 7
Ye. 4
~ q . .44-45 Find
1- OPl OP2
4
'
F1O. orng.l Find: , E ~ S76
. 8
Find: G,
Downloaded by [Universite Laval] at 00:06 21 June 2013
or ~ C Z
I
Flp.2 6 Eq.12
Find:?, e,if,
Fig.4 6 Eqs.22 21
FIg.6 6 Eg.25
or
r-lFind: y1
D-@
Find: G,,
Then, from Figure 2 together with Eq. 12, it is found that a second-order
model of the following can be obtained for the secondary loop
r = 2.123 and f = 0.25
Since the ratio of r1/t is large ( 13), according to Figure 4, we shall have
Downloaded by [Universite Laval] at 00:06 21 June 2013
0.8
YI 0.0
0.4
1 0 20 40 60 80 100 110 140 160'
0.2 ~ ~
lims
::;
4.8
1
.l.Z
0 20 40 MI 80 100 120 140 160
Time
FIGURE 10 Response curves of example 1: (a) for setpoint change, (b) for load 1 change,
(c) for load 2 change. (dash line - Wang's method, solid line - proposed method).
TUNING CASCADE CONTROL SYSTEMS 109
To tune a cascade control system when Gpl and Gp2 are unknown, the usual
practice is to carry out in the following manner:
1. A step input is introduced to identify a model for the secondary process
when both loops are opened.
2. The secondary controller is tuned according to the model obtained in the
previous step.
3. A further step test is conducted to identify a model for the primary open-
loop when the secondary loop is closed.
4. The primary controller is tuned using the model of the primary loop.
In this conventional tuning procedure, it needs to carry out open-loop
step test twice. With the help of the data presented previously in this article,
a simple tuning procedure that needs only one step test is presented in the
following.
Just like the first step in the conventional procedure, a step input is
introduced when both two loops are opened. From the primary and
secondary measurements, we have two responses. These two responses can
be fitted into two FOPDT models, i.e.,
where GA(s)and dB@)are the models for Gpl Gp2(s) and Gp2(s),respectively.
Thus, we have
110 H.-P. HUANG er a1
Rough estimations for the dead time and time constant of the primary
process model which are expressed as 81 and ?I, respectively, can be
calculated by
Downloaded by [Universite Laval] at 00:06 21 June 2013
The result in Eq. 36 is derived from matching the first derivatives at s=O
of GP1Gp2(s)and that of G"(s).
The above values of 81 and are then refined according to the following
equations defined as
A
91= m a x { i ~%}
, and ql%nin{?l, 6) (37)
and let
Therefore, we have
Notice that 681 is to compensate for the apparent dead time which is
contributed by the two time constants (Smith and Corripio, 1985).
Having these two models at hand, the remaining steps of the procedure
will then be the same as those in the previous section. Usually, the cascade
system tuned in this way is more conservative than the one of which both
GpI and Gp2 are well defined.
Illustration
For the same processes as those in the previous section, it is assumed here
that open-loop process models are unknown at the moment of tuning. In
order to tune the cascade controllers using the procedure depicted
TUNING CASCADE CONTROL SYSTEMS 111
previously, the estimations of FOPDT models using one single run of step
test for the demonstrated example are given in the following.
Example 2
Consider the same process as example 1:
e-9~ e-3s
G P ~ ( s=) and Gp2(s)= -
8s+ 1
"(s) = 1 4 . 7 ~ 1~ - 7.95s
Thus, we have
As we can see, the estimated FOPDT models are very close to the original
GpI and G p 2 Thus, the resulting cascade control system which is tuned
according to these G ~ I ( Sand) G ~ ~ ( Scould
) only have very small
differences from that we have in the previous example.
where
when yf# 0
Then, by shifting the switch to close the primary loop, we can estimate the
FOPDT model for GplGp2(s):
When eP2(s) and dA(s) are known, a FOPDT model for the primary
process can be derived as depicted in the previous section. Figure 9 also
shows the systematic tuning procedure for a cascade system by using the
proposed autotuning test. Two examples for such estimations are given in
the following.
Example 3
Consider again the process of example 1:
e-9~ e-3s
GPl(s) = - and Gp2(s) = -
IOs+ I 8s+ 1
The output and input responses of the proposed relay feedback test are
shown in Figure 12. Table I gives summaries about the estimations. The
resulting process models are
e-16.348s e-3.006s
and dP2(s)=
= 13.651s+ 1 7.987s + 1
Therefore, the final process model of the primary loop is derived as
H.-P. HUANG er al.
FIGURE 12 Testing results of example 3. (G,I (s)= e-9'/(10s + I ) and GP2(s) = e-3'/
(8s + I)). (a) ouput responses for the relay feedback test; (b) input responses for the relay
feedback test.
It is apparent that the estimated FOPDT models are very close to the
original GpI and Gp2. Thus, the resulting cascade control system which
is tuned according to these cP;,l(s)and cP2(s) would have very small
differences from that we have in the previous example.
Example 4
A cascade control system where the primary and secondary loop
processes are in the form of
2e-" 1.5e-0.5S
GPl(s) =
(5s + 1)(2s + 1) and Gp2(s) =
+
(S l)(O.Ss 1) +
is studied next. The output and input responses of the proposed test
are shown in Figure 13. And the estimation results are summarized in
Table I1 from which we obtain that
2.00 le-2.3033s 1,se-0.795s
GI(.) = and (;' s
5.7097s + I P'( ) = I.l95s+ 1
And open-loop step responses of primary loop process Gpl(s) and its
model cPl(s) are also shown in Figure 14. The derived FOPDT model
does have very close step response behavior to that of the process.
Downloaded by [Universite Laval] at 00:06 21 June 2013
FIGURE 13 Testing results of example 4. (a) output response for the relay feedback test;
(b) input responses for the relay feedback test.
TABLE I Estimation results of example 3 using relay feedback test (r=2 and h= 1)
Inner loop: G,(s) = &
Relay: hl = 2 and h2= -1 l,= 120
/ydr= 10.2098 a1=0.6263 / rydr= 377.8791
/u@= 10.2100 a?=-0.3135 /rudr= 265.6401
+ kp2=1.00 P = 11.0681 92 + 4 = 10.9937
+ 92 = 3.006 + 6 = 7.987
+ Model : 6,2(s) =
-.I13
=+ Model : G ~( sI) =
1 16 H.-P. HUANG et al.
TABLE I1 Estimation results of example 4 using relay feedback test (7'2 and h = I)
Inner loop: Gp2(s) = &
Relay: hl = 2 and h2= - l I,= 20
Jydt=4.245 a l = 1.0449 / rydr =40.4525
J"u+ = 2.830 a 2 = -0.5525 /rudt= 21.3366
e kp2= 1 .SO P = 2.8601 +&+4=1.990
+ $2 = 0.795 + 4 = 1.195
Model : dPl(s) =
"
0 5 10 15 20 25 30 35 40
Time
FIGURE 14 Open-loop step responses for example 4: process Gpl(s) = 2e-'/(5s + I ) @ + 1)
and model Gpl(s) = 2.001e-2.303"/(5.7097s I). +
TUNING CASCADE CONTROL SYSTEMS 117
The responses of this cascade system to the setpoint change and the load
changes are as shown in Figure 15.
Downloaded by [Universite Laval] at 00:06 21 June 2013
:m
0.4
0.2
:imm:!m
0
0 10 20 30 40 60 M) 70 80 0 10 20 30 40 SO 60 70 80
Time Time
0.4
0.2
0 .1.5
0.2
!-q
0 10 20 30 40 50 60 70 80 0 LO 20 30 40 50 60 70 80
Time Time
y2
0.1
0
-0.1 -0.2
-0.2 4.4
4.6
-0.3 -0.8
-0.1
0 10 20 $0 40 50 64 70 80 0 10 20 30 40 60 60 m 80
Time Time
FIGURE 15 Response curves of example 4 using relay feedback test: (a) for setpoint change,
(b) for load I change, (c) for load 2 change.
H.-P. HUANG el a/.
6. CONCLUSIONS
Acknowledgement
Financial support from the National Science Council of the Republic of
China (NSC-86-2214-E002-018) is gratefully acknowledged.
NOMENCLATURE
parameters in Eq. 42
average amplitude of the constant cycles
coefficient constants
primary and secondary loop disturbances
functions
process models for GpIGp2(s)and Gp2(s)
primary and secondary controllers
primary' and secondary processes .
amplitude of relay output
controller gains of the primary and secondary loops
process gains of the primary and secondary loops
maximum peak gain
period of process output at constant cycling
parameters defined as Eq. 37
(unit) step response of the process
final time of the relay feedback test
peak time and rising time
time original for integration
TUNING CASCADE CONTROL SYSTEMS
Greek Symbols
pre-filter constant of a parallel PID controller
parameter calculated by Eq. 44 or Eq. 45
ratio of positive peak to negative peak of relay output
dead time calculated from Eq. 21
Downloaded by [Universite Laval] at 00:06 21 June 2013
Superscripts
secondary loop
primary open-loop
estimated value or process model
roughly estimated value
Subscripts
primary open-loop
References
Chen, C. L. (1989) "A Simple Method for On-Line Identification and Controller Tuning,"
AIChE J., 35, 2037-2039.
Chien, I-L. and Fmehauf, P. S. (1990) "Consider IMC Tuning lo Improve Controller
Performance," Chem. Eng. Prog., Oct., 33-41.
Corripio, A.B. (1990) Tuning of Industrial Control Systems, ISA, Research Triangle Park, NC.
Eker, I. and Johnson, M. A. (1996) "New Aspects of Cascade and Multi-Loop Process
Control," Trans. IChemE, 74, 38-54.
Franks, R. G. and Worley, C. W. (1956) "Quantitative Analysis of Cascade Control," Ind. Eng.
Chemistry, 48, 1074- 1079.
Hang, C. C., Loh, A. P. and Vasnani, V. U. (1994) "Relay Feedback Auto-Tuning of Cascade
Controllers," I E E E Trans. Control Systems Technology, 2, 42-45.
Downloaded by [Universite Laval] at 00:06 21 June 2013
TUNING CASCADE CONTROL SYSTEMS
where
Downloaded by [Universite Laval] at 00:06 21 June 2013
1 1.
with (02/r2)
3. Quantitative equation for Figure 4
(A- 10)
where
a3 = -0.9187 + 1.784
4. Quantitative equation for Figure 5