Journal of Chemical Engineering of Japan, Vol. 40, No. 6, pp.

000000, 2007

Research Paper

An Enhanced Performance PID Filter Controller for First Order Time

Delay Processes
Mohammad SHAMSUZZOHA and Moonyong LEE
School of Chemical Engineering and Technology,
Yeungnam University, Dae-dong 214-1, Kyongsan 712-749, Korea

Keywords: PID Controller Tuning, First Order Plus Dead Time Process, Disturbance Rejection, First Order
Lead/Lag Filter, Two-Degree-of-Freedom Controller
An analytical tuning method for a PID controller cascaded with a lead/lag filter is proposed for FOPDT
processes based on the IMC design principle. The controller is designed for the rejection of disturbances
and a two-degree-of-freedom control structure is used to slacken the overshoot in the set-point response.
The simulation study shows that the proposed design method provides better disturbance rejection than
the conventional PID design methods when the controllers are tuned to have the same degrees of robustness. A guideline of a single tuning parameter of closed-loop time constant ( ) is provided for several
different robustness levels.

Introduction
Proportional integral derivative (PID) controllers
have been the most popular and widely used controllers in the process industries because of their simplicity, robustness and wide ranges of applicability with
near-optimal performance. However, it has been noticed that many PID controllers are often poorly tuned
and a certain amount of effort has been made to systematically resolve this problem.
The effectiveness of the internal model control
(IMC) design principle has made it attractive in the
process industries, where many attempts have been
made to exploit the IMC principle to design PID controllers for both stable and unstable processes (Morari
and Zafiriou, 1989). The IMC-PID tuning rules have
the advantage of using only a single tuning parameter
to achieve a clear trade-off between the closed-loop
performance and robustness. The PID tuning methods
proposed by Rivera et al. (1986), Morari and Zafiriou
(1989), Horn et al. (1996), and Lee et al. (1998) are
typical examples of the IMC-PID tuning method. The
direct synthesis (DS) method proposed by Smith et al.
(1975) and the direct synthesis for the disturbance (DSd) method proposed by Chen and Seborg (2002) can
also be categorized into the same class as the IMCPID methods, in that they obtain the PID controller
parameters by computing the ideal feedback controller which gives a predefined desired closed-loop response. Although the ideal controller is often more
complicated than the PID controller for time delayed
Received on November 27, 2006; accepted on February 15,
2007. Correspondence concerning this article should be addressed to M. Lee (E-mail address: [email protected]).

Copyright 2007 The Society of Chemical Engineers, Japan

processes, the controller form can be reduced to that

of either a PID controller or a PID controller cascaded
with a low order filter by performing appropriate approximations of the dead time in the process model.
The control performance can be significantly enhanced by cascading the PID controller with a lead/
lag filter, as given by Eq. (1).

1
1 + as
Gc = K c 1 +
+ D s
1 + bs
Is

(1)

where Kc, I and D are the proportional gain, integral

time constant, and derivative time constant of the PID
controller, respectively, and a and b are the filter parameters.
The structure of the PID controller cascaded with
a filter was also suggested by Rivera et al. (1986),
Morari and Zafiriou (1989), Horn et al. (1996), Lee et
al. (1998) and Dwyer (2003). The PID filter controller
in Eq. (1) can easily be implemented in modern control hardware.
It is essential to emphasize that the PID controller designed according to the IMC principle provides
excellent set-point tracking, but has a sluggish disturbance response, especially for processes with a small
time-delay/time-constant ratio (Morari and Zafiriou,
1989; Chien and Fruehauf, 1990; Horn et al., 1996;
Lee et al., 1998; Chen and Seborg, 2002; Skogestad,
2003). Since disturbance rejection is much more important than set-point tracking for many process control applications, a controller design that emphasizes
the former rather than the latter is an important design
goal that has recently been the focus of renewed research.
1

where pm is the portion of the model inverted by the

controller, pA is the portion of the model not inverted
by the controller and pA(0) = 1. The noninvertible part
usually includes the dead time and/or right half plane
zeros and is chosen to be all-pass.
To obtain a good response for processes with poles
near zero, the IMC controller q should be designed to
satisfy the following conditions.
1. If the process GP has poles near zero at z1, z 2, ...,
zm, then q should have zeros at z 1, z2, ..., z m.
2. If the process GD has poles near zero, z d1, z d2, ...,
zdm, then (1 GPq) should have zeros at zd1, z d2, ...,
zdm.
Since the IMC controller q is designed as q =
p m1f, the first condition is satisfied automatically. The
second condition can be fulfilled by designing the IMC
filter f as

(a)

s
i

f =
(b)
Fig. 1

Block diagram of IMC and classical feedback control systems: (a) The IMC structure; (b) Feedback
control structure

In the present study, a simple and efficient method

is proposed for the design of a PID filter controller
with enhanced performance. A closed-loop time constant () guideline is recommended for a wide range
of time-delay/time-constant ratios. A simulation study
was performed to illustrate the superiority of the proposed method for both nominal and perturbed processes.

+1

( 4)

i =1

(s + 1)r

where is an adjustable parameter which controls the

tradeoff between the performance and robustness; r is
selected to be large enough to make the IMC controller (semi-)proper; i are determined by Eq. (5) to cancel the poles near zero in GD.

1 GP q s= z

d 1 ,L, zdm

IMC Controller Design Procedure

y = GP qr + 1 G P q GD d

(2 )

According to the IMC parameterization (Morari and

Zafiriou, 1989), the process model G P is factored into
two parts:
G P = pm pA
2

(3)

=0

(5)

s = zd 1 ,L, zdm

q = pm1

Figures 1(a) and (b) show the block diagrams of

the IMC control and equivalent classical feedback control structures, respectively, where GP is the process,
G P the process model, q the IMC controller, fR the setpoint filter, and Gc the equivalent feedback controller.
For the nominal case (i.e., GP = G P ), the set-point
and disturbance responses in the IMC control structure can be simplified as:

(s + 1)r

Then, the IMC controller comes to be

1.

= 1

m
pA i s i + 1

i =1

+1

(6 )

i =1

(s + 1)r

Thus, the closed-loop response is

y=

m
pA i s i + 1

i =1

(s + 1)r

m
i
p
A
i s + 1

i =1
GD d
r + 1
(s + 1)r

(7)

From the above design procedure, one can achieve

a stable closed-loop response by using the IMC controller.
2.

PID filter Controller Design for FOPDT

Process

The ideal feedback controller that is equivalent to

the IMC controller can be expressed in terms of the
internal model G P and the IMC controller q:
JOURNAL OF CHEMICAL ENGINEERING OF JAPAN

Gc =

q
1 G P q

(8)

Gc =
1

+1

1 i =1
m

(s + 1)r

( 9)

pA i s i + 1
i =1

(s + 1)r

Let us consider the first order plus dead time

(FOPDT) process, which is most widely utilized in the
chemical process industries, as a representative model.
GP = GD =

s 2 s 2
+
( s + 1)
2
12
Gc =

s 2 s 2
s 2 s 2
2
K (s + 1) 1 +
+
+
( s + 1)1

2
12
2
12

(15)

Substituting Eqs. (3) and (6) into Eq. (8) gives the ideal
feedback controller:

results in GC as

Ke s
s + 1

( s + 1)1 +

It is important to note that the 2/2 Pade approximation

is precise enough to convert the ideal feedback controller into a finite dimensional feedback controller
with barely any loss of accuracy. Expanding and rearranging Eq. (15) gives
s 2 s 2
Gc = 1 +
+
( s + 1)( s + 1)

2
12

2
2 + +
K (2 + )1 +
s
2 +

(10)

2 2 2
2 2
+
+
12

3
6
2 s2 +
12
+

2 +

2 +

where K is the gain, the time constant, and the time

delay. The IMC filter structure is

s + 1
f =
(s + 1)2

(11)

It is noticed that the IMC filter form in Eq. (11)

was also utilized by Lee et al. (1998) and Horn et al.
(1996). The resulting IMC controller becomes
q=

( s + 1)(s + 1)
K (s + 1)

(12)

Therefore, the ideal feedback controller is obtained as

Gc =

( s + 1)(s + 1)

K (s + 1) e s ( s + 1)
2

(13)

Since the ideal feedback controller in Eq. (13) does

not have the PID filter controller form, the remaining
issue is how to design the PID filter controller that
approximates the ideal feedback controller most
closely.
Approximating the dead time e s with a 2/2 Pade
expansion

s s
+
2
12
=
s 2 s 2
1+
+
2
12
2 2

e s

VOL. 40 NO. 6 2007

(14)

(16)
As seen in Eq. (16), the resulting controller has the
form of the PID controller cascaded with a high order
filter. The analytical PID formula can be obtained as
Kc =

, I = , D =
2 K (2 + )
2
6

(17)

The value of the extra degree of freedom is selected

so that it cancels out the open-loop pole at s = 1/
that causes a sluggish response to load disturbances.
From Eq. (5), this requires [1 ( s + 1)e s/( s +
1) 2]| s=1/ = 0. Thus, the value of is obtained as

= 1 1

2

e

(18)

Furthermore, it is obvious from Eq. (5) that the

remaining part of the denominator in Eq. (16) contains
the factor (s + 1). Therefore, the filter parameter b in
Eq. (1) can be obtained by taking the first derivative
of Eq. (19) below
cs 2 + bs + 1
2
2 2
2 2


+
+
+ + 2

3
2
12
2
6
2 s +
12
1+

s
s +
2 +

2 +
2 +
=
s + 1

(19)

Since = G p q = G p p m1 f for the IMC controller, the

resulting Eq. (24) becomes:

and substituting s = 0 as

+ + 2
2
b=

2 +

(20)

The filter parameter in Eq. (1) can be easily obtained from Eq. (16) as

(21)

a=

Since the high order cs2 term has little impact on

the overall control performance in the control relevant
frequency range, the remaining part of the fraction in
Eq. (16) can be successfully approximated to a simple
first order lead/lag filter as (1 + as)/(1 + bs). Our simulation result (although not shown in this paper) also
confirms the validity of this model reduction.
The lead term (s + 1) in the closed-loop transfer
function of Eq. (7) causes excessive overshoot in the
set-point response, which can be eradicated by adding
the set-point filter f R as:

s + 1
fR =
s + 1

(22)

where 0 1. The extreme case with = 0 has no

lead term in the set-point filter which would cause a
slow servo response. On the other hand, = 1 means
that there is no set-point filter. can be adjusted online
to obtain the desired speed of the set-point response.
The proposed study is also applicable to the process
with negligible dead time while it is mainly focused
on the first order time delay process.
3.

The well-known robust stability theorem can be

utilized to analyze the robust stability of the proposed
controller.
Robust Stability Theorem (Morari and Zafirou,
1989): Let us assume that all plants GP in the family

(23)

have the same number of RHP poles and that a particular controller Gc stabilizes the nominal plant G p .
Then, the system is robustly stable with the controller
Gc if and only if the complementary sensitivity function for the nominal plant G p satisfies the following bound:

lm
4

= sup lm ( ) < 1

Thus, the above theorem can be interpreted as | lm | <

1/| | = 1/| G p p m1 f|, which guarantees robust stability
when the multiplicative model error is bounded by
|m(s)| lm .

(s) m (s)

(26)

<1

where m(s) defines the process multiplicative uncertainty bound. i.e., m(s) = (Gp G p )/ G p . This uncertainty bound can be utilized to represent the model reduction error, process input actuator uncertainty, and
process output sensor uncertainty, etc., which are very
frequent in the actual process plants.
For the FOPDT process, the complementary sensitivity function (s) can be obtained as

(s) =

(s + 1)e s
(s + 1)2

(27)

Substituting Eq. (27) and into Eq. (26) yields

the robust stability constraint required for tuning the
adjustable parameters
2


1 1
e s + 1

(s + 1)2

<

1
m (s)

(28)

Substituting s = i into Eq. (28) results in

Robust Stability

Gp (i ) G p (i )
= Gp :
< lm ( )
G p (i )

(25)

G p p m1 flm ( ) < 1

(24)

2

2
2 1 1
e +1

2 2
+1

<

1
m ( )

(29)

It is possible for uncertainty to occur in any of the

three process parameters i.e., , , and K. Consequently,
we have to consider the uncertainty in the different
parameters separately. Let us consider the FOPDT
process having the uncertainty in all three parameters
as

Gp =

( K + K )e ( + ) s
( s + 1)( s + 1)

(30)

It is most common practice that the FOPDT model

approximated from the high order process in the real
JOURNAL OF CHEMICAL ENGINEERING OF JAPAN

process plant. Due to this for the time constant uncertainty it is assumed that the small time constant is
neglected/missing in developing the nominal model as
considered in Eq. (30) (Seborg et al., 2004). Then the
process multiplicative uncertainty bound becomes
1 + K e s

K
m (s) =
1
s + 1

(31)

Substituting the above result into Eq. (29), we

obtain the robust stability constraint as follows:
2

2

2
2 1 1
e +1

<1
2 2 + 1

> 0

1 + K e i

K
1,
i + 1

(32)

The above robust stability constraint is very useful to adjust where there is uncertainty in the process parameters. The robust stability constraint in Eq.
(32) can also be used to determine the maximum allowable values of uncertainty in K, and or
various combinations of them for which robust stability can be guaranteed. For example, a plot of
| () lm ()| vs. can be constructed for a small value
of any parametric uncertainty and/or combination of
different uncertainties.
4.

Simulation Study

This section deals with the simulation study conducted for three representative FOPDT processes: the
lag time dominant process, the equal dead time and
lag time process, and the dead time dominant process.
To evaluate the robustness of a control system,
the maximum sensitivity, Ms, which is defined by
Ms = max |1/[1 + GpG c(i )]|, is used. Since the Ms is
the inverse of the shortest distance from the Nyquist
curve of the loop transfer function to the critical point
(1, 0), a small Ms value indicates that the stability
margin of the control system is large. The Ms is a wellknown robustness measure and is used by many researchers (Skogestad and Postlethwaite, 1996; strm
et al., 1998; Chen and Seborg, 2002; Skogestad, 2003).
Typical values of Ms are in the range of 1.22.0
(strm et al., 1998; Seborg et al., 2004). To ensure a
fair comparison, it is widely accepted for the modelbased controllers (DS-d, DS, and IMC) to tune by adjusting so that the Ms values become the same values. Therefore, throughout all our simulation examples, all of the controllers compared were designed to
have the same robustness level in terms of the maximum sensitivity, Ms.

VOL. 40 NO. 6 2007

To evaluate the closed-loop performance, two performance indices were considered in the case of both a
step set-point change and a step load disturbance, viz.,
the integral of the time-weighted absolute error (ITAE)
defined by ITAE = 0 t|e(t)|dt, and the overshoot which
acts as a measure of how much the response exceeds
the ultimate value following a step change in the setpoint and/or disturbance.
In this paper, the simulation study has been conducted using the PID controller in the form of Eq. (1).
However, for real implementation, the parallel form
of the PID controller, G(s) = Kc {1 + 1/( Is) + Ds/
[(0.1Ds) + 1]}(1 + as)/(1 + bs), which is widely used
in the real processes, can be applied to approximately
the same performance.
To evaluate the usage of manipulated input values, we compute TV of the input u(t), which is the sum
of all of its movement of up and down. If we discretize
the input signal as a sequence [u1, u2, u3, ..., ui, ...],
then TV = i=1|ui+1 ui| should be as small as possible.
TV is a good measure of the smoothness of a signal
(Skogestad and Postlethwaite, 1996; Chen and Seborg,
2002; Skogestad, 2003).
4.1 Example 1: Lag time dominant process ( / =
0.01)
Consider the following FOPDT process (Chen and
Seborg, 2002; Seborg et al., 2004):
Gp = GD =

100e 1s
100 s + 1

(33)

The proposed PID filter controller is compared

with other controllers based on existing methods, such
as the DS-d method, and those proposed by Rivera et
al. (1986), Horn et al. (1996), Lee et al. (1998) and
Lee et al. (1998) with a conventional filter. The controller parameters, including the performance and robustness matrix, are listed in Table 1. In order to ensure a fair comparison, all of the controllers compared
are tuned to have Ms = 1.94 by adjusting . Figure 2
compares the set-point and load responses obtained
using the proposed method, the DS-d method, and the
methods proposed by Lee et al. (1998) and Horn et al.
(1996). The 2DOF controller using the set-point filter
was used in the DS-d method and the methods proposed by Lee et al. (1998) and Horn et al. (1996) to
obtain an enhanced set-point response. It is important
to note that the set-point filter used for the set-point
response has a clear benefit when the process is lag
time dominant. In this case, it is observed that 0.4
0 gives smooth and robust control performances. In
the proposed controller, in the set-point filter is selected as = 0.45. The closed-loop response for both
the set-point tracking and disturbance rejection signifies that the proposed method provides a superior response for the same robustness.

Table 1

Tuning methods

Proposed method
Lee et al. (1998)b
DS-dc
Horn et al. (1996)d
Rivera et al. (1986)e
Lee et al. (1998)f
a

Gc = K c 1 +

fR =
fR =

Is

PID controller parameters and performance matrix for example 1 (/ = 0.01)

Kc

1.131
0.124
0.50
1.330
0.806
3.947
1.202
0.826
4.059
1.689 15.038 100.50
0.408
0.714 100.50
0.248
0.805 100.41

+ D s

Set-point

0.167
0.3068
0.353
0.497
0.4975
0.399

1 + as

1 + bs , where a = 3.222, b = 0.139;

Disturbance

ITAE

Overshoot

TV

ITAE

Overshoot

TV

1.96
8.46
3.13
12.45
3.86
3.15

0.007
0.0
0.015
0.0
0.025
0.018

3.17
0.542
0.87
0.430
1.47
1.136

14.55
19.77
20.43
31.18
3785.0
3354.0

1.206
1.314
1.273
1.478
1.411
1.273

1.96
1.79
1.88
1.69
1.35
1.53

fR =

1.45s + 1
3.22 s + 1

1
3.66 s + 1
2.03s + 1

1.43s + 4.06 + 1
1
1 + as

1
d Gc = K c 1 +
+ D s
f =
Is
1 + bs + cs 2 , where a = 4.311, b = 100.2, c = 21.34; R 4.31s + 1
1
1

e Gc = K c 1 +
+ D s
, where b = 0.145
1

s
+
bs

I
f
The Lee et al. (1998) method based on the conventional IMC filter form of f = 1/( s + 1)
*1DOF controller is used only for the methods of Rivera et al. (1986)e and Lee et al. (1998)f
2

(a)

Process variable

1
0.8
0.6

Proposed method
Lee et al. (1998)
DS-d
Horn et al. (1996)

0.4
0.2
0

0
1.5

Time [min]

12

Process variable

(b)

15

Proposed method
Lee et al. (1998)
DS-d
Horn et al. (1996)

0.5

12

15

Time [min]

Fig. 2

Simulation results for example 1

The robust performance is evaluated by inserting

a perturbation uncertainty of 20% in all three parameters in the worst direction simultaneously and finding the actual process as Gp = GD = 120e 1.2s/(80s + 1).
The simulation results for the model mismatch for various methods are given in Table 2. The performance
and robustness indices obviously demonstrate that the
proposed method has more robust performance than
the others.

4.2 Example 2: Equal lag time and dead time

process ( / = 1)
Consider the process model described by Chen and
Seborg (2002) as follows

JOURNAL OF CHEMICAL ENGINEERING OF JAPAN

Gp = GD =

1e 1s
1s + 1

(34)

PID controller parameters and performance matrix for example 2 ( / = 1)

Table 3

Tuning methods

Proposed method
Lee et al. (1998)b
DS-dc
Horn et al. (1996)d
Rivera et al. (1986)e
Lee et al. (1998)f
a

0.499
0.596
0.771
0.73
0.503
0.309

a = 0.907, b = 0.102; fR =

fR =
fR =

Kc

0.458
1.042
1.055
1.010
0.998
1.055

0.5
1.304
1.444
1.50
1.50
1.382

0.166
0.270
0.313
0.333
0.333
0.289

Set-point

Disturbance

ITAE

Overshoot

TV

ITAE

Overshoot

TV

2.28
2.83
2.71
3.59
2.33
1.95

0.0034
0.0025
0.0006
0.0002
0.1481
0.1273

2.835
1.050
1.375
1.059
2.124
1.981

2.66
3.31
3.85
4.12
4.26
3.52

0.626
0.622
0.633
0.672
0.670
0.634

2.837
1.368
1.419
1.163
1.151
1.394

1
0.908s + 1

1
0.94 s + 1
0.722 s + 1
0.452 s + 1.44 s + 1
2

a = 0.975, b = 1.179, c = 0.179; fR =

1
0.976 s + 1

b = 0.167
*1DOF controller is used only for the methods of Rivera et al. (1986)e and Lee et al. (1998)f

Table 2

Robustness analysis for example 1

Tuning methods

Proposed methoda
Lee et al. (1998)b
DS-dc
Horn et al. (1996)d
Rivera et al. (1986)e
Lee et al. (1998)f

Set-point

Disturbance

ITAE

Overshoot

ITAE

Overshoot

13.36
14.40
40.41
16.32
28.73
21.21

0.3418
0.0387
0.2817
0.0091
0.5354
0.5797

33.36
76.26
98.25
62.27
3766.0
3338.0

1.8798
1.7399
1.7011
2.2096
2.1152
1.9450

The proposed PID filter controller is compared

with the DS-d controller and the controllers designed
by Lee et al. (1998), Horn et al. (1996), Rivera et al.
(1986) and Lee et al. (1998) with a conventional filter.
The controller parameter values are listed in Table 3
along with the performance matrix, where Ms = 1.84
is selected for all controller designs. Unit step changes
are introduced both in the set-point and in the disturbance for the simulation. The simulation results in Figure 3 indicate that both the disturbance and the setpoint responses are faster in the proposed controller.
The 2DOF controller structure is used for each design
method except Rivera et al. (1986), and Lee et al.
(1998) with a conventional filter. = 0 is selected for
the proposed controller. It is clear from Figure 3 and
Table 3 that the proposed controller exhibits better
performance for both the set-point and disturbance response.
4.3 Example 3: Dead time dominant process ( / =
5)
Consider the process with a long dead time studied by Luyben (2001) and Chen and Seborg (2002)
VOL. 40 NO. 6 2007

Gp = GD =

1e 5 s
1s + 1

(35)

The proposed and aforementioned design methods are compared. The controller settings with the performance matrices are given in Table 4. All of the controllers are designed to have Ms = 1.74. Since in the
case of a dead time dominant process, the 1DOF controller is sufficient to achieve satisfactory control performance, no set-point filter is used for any design
method.
The set-point and load responses are shown in
Figure 4. From this figure, it is apparent that the proposed controller and the one designed by Lee et al.
(1998) with the conventional filter provide similar responses, while the DS-d and Horn et al. (1996) methods exhibit sluggish responses and take a long time to
settle the response.
The proposed controller has excellent performance
when the lag time dominates, but its performance becomes similar to that of the methods based on the conventional filter when the dead time dominates. When
/ >> 1, the filter time constant should be chosen as
>> for the sake of closed-loop stability. Therefore, the process pole at 1/ is not a dominant pole in
the closed-loop system. Instead, the pole at 1/ determines the overall dynamics. Thus, introducing the
lead term (s + 1) into the IMC filter to compensate
the process pole at 1/ has little impact on the disturbance response.
Furthermore, the lead term usually increases the
complexity of the IMC controller, which in turn degrades the performance of the resulting PID controller
7

PID controller parameters and performance matrix for example 3 ( / = 5)

Table 4

Tuning methods

Proposed method
Lee et al. (1998)b
DS-dc
Horn et al. (1996)d
Rivera et al. (1986)e
Lee et al. (1998)f

Kc

1.408
1.423
2.706
2.648
3.117
1.798

0.366
0.408
0.316
0.430
0.431
0.417

2.5
2.799
2.555
3.5
3.5
2.838

0.833
0.721
0.053
0.714
0.714
0.759

Set-point

Disturbance

ITAE

Overshoot

TV

ITAE

Overshoot

TV

29.95
30.79
37.57
41.76
40.86
30.66

0.0451
0.0641
0.0115
0.0502
0.0436
0.0642

1.268
1.044
0.766
1.022
0.996
1.115

68.34
67.67
85.88
87.01
87.06
66.77

0.989
0.992
0.984
0.993
0.992
0.990

1.312
1.116
1.052
1.061
1.048
1.066

a = 0.998, b = 0.689
a = 2.164, b = 3.155, c = 2.156
e
b = 0.96
*1DOF controller is used for all of the methods
d

Process variable

(a)

0.8
0.6
0.4

Proposed method
Lee et al. (1998)
DS-d
Horn et al. (1996)

0.2
0

Time [min]

8
Proposed method
Lee et al. (1998)
DS-d
Horn et al. (1996)

(b)

0.7

Process variable

0.5
0.3
0.1
-0.1
0

Fig. 3

3e 10 s
100 s + 1

Simulation results for example 2

by causing a larger discrepancy between the ideal feedback controller and thus the PID controller.
It is also important to note that as the order of the
filter increases, the power of the denominator term
( s + 1) also increases, which can cause an unnecessarily slow output response. As a result, in the case of
a dead time dominant process, the PID controller based
on the IMC filter that includes no lead term offers better performance.
4.4 Example 4: Polymerization process
An important viscosity loop in a polymerization
process was identified by Chien et al. (2002) as follows:

Gp = GD =

Time [min]

(36)

The above-mentioned process has a large openloop time constant of 100 min and a dead time of 10
min, which is also quite noteworthy. Chien et al. (2002)
designed the PI controller with the modified Smith
Predictor (SP) by approximating the above process in
the form of an integrating model with a long dead time.
Figure 5 compares the nominal responses by the proposed PID filter controller and that by the modified
SP. In the proposed controller, = 8.0 is selected and
the resulting tuning parameters are obtained as Kc =
0.6446, I = 5.0, D = 1.6667, a = 23.4146 and b =
0.9781. The simulation was conducted by inserting the
step set-point change at t = 0 followed by a load step
change of 1.0 at t = 90.
The proposed controller uses a simple feedback
control structure without any dead time compensator.
Nevertheless, the proposed PID filter controller provides a superior performance, as shown in Figure 5.
JOURNAL OF CHEMICAL ENGINEERING OF JAPAN

(a)

Process variable

1
0.8
0.6

Proposed method
DS-d
Horn et al. (1996)
Lee et al. (1998)R

0.4
0.2
0

Process variable

10

15

Time [min]

20

25

30

(b)
Proposed method
DS-d
Horn et al. (1996)
Lee et al. (1998)R

0.8
0.6
0.4
0.2
0
-0.2

10

15

20

25

30

Time [min]

Fig. 4

Simulation results for example 3

1.5
Proposed method
Chien et al. (2002)

0.9
0.8

Process variable

Process variable

0.7
0.6
0.5
0.4
0.3

0.5

0.2
Proposed method

0.1
0

Chien et al. (2002)

0

30

60

90

120

150

180

Time [min]

Fig. 5

Simulation results of the polymerization process

The disturbance rejection afforded by the proposed

controller has a smaller settling time, whereas the
modified SP controller described by Chien et al. (2002)
shows a sluggish and required long settling time.
As regards the set-point response, the modified
SP controller has an initially fast response, because of
the elimination of the dead time, but afterwards it becomes slow. On the other hand, the speed of the response for the proposed controller is uniform and the
settling time is similar to that by the modified SP.
It is important to note that the SP control configuration has a clear advantage of eliminating the time
delay from the characteristic equation, which is very
effective to set-point tracking performance. However,
this advantage is lost if the process model is inaccurate. In order to evaluate the robustness against model
uncertainty, a simulation study was conducted for the
VOL. 40 NO. 6 2007

100

200

300

400

500

600

Time [min]

Fig. 6

Simulation results of the polymerization process

with model mismatch

worst case of model mismatch by assuming that the

process has a 20% mismatch in the three process parameters in the worst direction, as follows
Gp = GD =

3.6e 12 s
80 s + 1

(37)

The closed-loop responses are presented in Figure 6. Notice that the proposed method and the modified SP method described by Chien et al. (2002) have
similar disturbance rejection responses for the model
mismatch case. However, the set-point response afforded by the modified SP controller shows severe oscillation, while the proposed controller gives a more
robust response.
9

compared with the more sophisticated controller, such

as the modified Smith Predictor, in the case of the viscosity loop in a polymerization process. The result
shows that the proposed controller gives satisfactory
performance without the external dead time compensator. A guideline of closed-loop time constant was
also proposed for a wide range of / ratio.

10

Ms=1.4
Ms=1.5
Ms=1.6
Ms=1.8
Ms=1.9

0.1

0.01
0.01

Fig. 7

Acknowledgements
This research was supported by the 2006 Energy Resource
and Technology Development Program of Korea.
0.1

10

guideline for the proposed tuning method

In the proposed tuning rule, the closed-loop time

constant controls the tradeoff between the robustness and performance of the control system. As decreases, the closed-loop response becomes faster and
can become unstable. On the other hand, as increases,
the closed-loop response becomes stable but sluggish.
A good tradeoff is obtained by choosing to give an
Ms value in the range of 1.22.0 (strm et al., 1998;
Seborg et al., 2004). The guideline for several robustness levels is plotted in Figure 7.
Conclusions
A simple analytical design method for a PID controller cascaded with a lead/lag filter was proposed
based on the IMC principle in order to improve its disturbance rejection performance. The proposed method
also includes a set-point filter to enhance the set-point
response like the 2DOF controller suggested by Lee et
al. (1998), Horn et al. (1996) and Chen and Seborg
(2002). FOPDT processes with three representative
different / ratios were used for the simulation study.
The proposed PID filter controller consistently provides superior performance over the whole range of
the / ratio, while the other controllers based on the
IMC-PID design methods take their advantage only in
a limited range of the / ratio. In particular, the proposed controller shows excellent performance when the
lag time dominates. The proposed controller was also

10

Literature Cited
strm, K. J., H. Panagopoulos and T. Hgglund; Design of PI
Controllers Based on Non-Convex Optimization, Automatica,
34, 585601 (1998)
Chen, D. and D. E. Seborg; PI/PID Controller Design Based on
Direct Synthesis and Disturbance Rejection, Ind. Eng. Chem.
Res., 41, 48074822 (2002)
Chien, I. L. and P. S. Fruehauf; Consider IMC Tuning to Improve
Controller Performance, Chem. Eng. Prog., 86, 3341 (1990)
Chien, I.-L., S. C. Peng and J. H. Liu; Simple Control Method for
Integrating Processes with Long Deadtime, J. Process Control, 12, 391404 (2002)
Dwyer, A. O.; Handbook of PI and PID Controller Tuning Rules,
Imperial College Press, London, U.K. (2003)
Horn, I. G., J. R. Arulandu, J. G. Christopher, J. G. VanAntwerp
and R. D. Braatz; Improved Filter Design in Internal Model
Control, Ind. Eng. Chem. Res., 35, 34373441 (1996)
Lee, Y., S. Park, M. Lee and C. Brosilow; PID Controller Tuning
for Desired Closed-Loop Responses for SI/SO Systems, AIChE
J., 44, 106115 (1998)
Luyben, W. L.; Effect of Derivative Algorithm and Tuning Selection on the PID Control of Dead Time Processes, Ind. Eng.
Chem. Res., 40, 36053611 (2001)
Morari, M. and E. Zafiriou; Robust Process Control, Prentice-Hall,
Englewood Cliffs, U.S.A. (1989)
Rivera, D. E., M. Morari and S. Skogestad; Internal Model Control. 4. PID Controller Design, Ind. Eng. Chem. Process Des.
Dev., 25, 252265 (1986)
Seborg, D. E., T. F. Edgar and D. A. Mellichamp; Process Dynamics and Control, 2nd ed., John Wiley & Sons, New York, U.S.A.
(2004)
Skogestad, S.; Simple Analytic Rules for Model Reduction and
PID Controller Tuning, J. Process Control, 13, 291309 (2003)
Skogestad, S. and I. Postlethwaite; Multivariable Feedback Control; Analysis and Design, John Wiley & Sons, New York,
U.S.A. (1996)
Smith, C. L., A. B. Corripio and J. Martin; Controller Tuning from
Simple Process Models, Instrum. Technol., 22, 3944 (1975)

JOURNAL OF CHEMICAL ENGINEERING OF JAPAN