* Corresponding author. Tel.

: #82-42-869-3920; fax: #82-42-869-

3910.
E-mail address: swpark@mail.kaist.ac.kr (S. Park)
Chemical Engineering Science 55 (2000) 3481}3493
PID controller tuning for integrating and unstable processes
with time delay
Yongho Lee, Jeongseok Lee, Sunwon Park*
Department of Chemical Engineering, Korea Advanced Institute of Science and Technology, 373-1 Kusong-dong, Yusung-gu, Taejon 305-701, South Korea
Received 4 January 1999; accepted 12 October 1999
Abstract
A new method for PID controller tuning based on process models for integrating and unstable processes with time delay is
proposed in this paper. The proposed method is an extension of the PID controller tuning method by Lee, Lee, Park and Brosilow
(1998) to general unstable and integrating processes with time delay. The proposed method is simpler to use because the proposed
method is an explicit tuning method compared with other frequency response methods. Closed-loop responses tuned by the proposed
method are compared with those of existing methods. The results show that the proposed tuning method gives better closed loop
performance than the existing methods. 2000 Elsevier Science Ltd. All rights reserved.
1. Introduction
There still exist many unstable processes in the chem-
ical plants, even though most chemical processes are
open-loop stable. The most common example is the
batch chemical reactor, which has a strong nonlinearity
due to the heat generation term in the energy balance. In
general, two types of time delayed unstable processes are
the "rst-order delayed unstable process (FODUP) and
the second-order delayed unstable process (SODUP)
(Huang & Chen, 1997).
FODUP: G(s)"
KeFQ
ts!1
,
SODUP: G(s)"
KeFQ
(ts!1)(as#1)
.
Many researchers investigated the tuning methods of
PID controller for unstable processes. However, most of
them are in the category of graphical techniques, which
are often di$cult for the operators to understand and use.
De Paor and O'Malley (1989) proposed the tuning
method of P, PI, and PID controllers for the FODUP.
The method is based on the gain and phase margins
criterion. But in their unit step tests, PI and PID control-
ler gave greater than 200% overshoot by the gain and
phase margin design.
Quinn and Sanathanan (1989) developed a method for
the design of controllers for integrating processes and the
unstable delayed processes based on model matching in
the frequency domain. They achieved low-order control-
lers in a unity-output-feedback con"guration based on
the Guillemin}Truxal synthesis procedure.
Rotstein and Lewin (1991,1992) proposed explicit PID
tuning rules based on IMC for the "rst- and second-order
unstable process models without time delay. PID tuning
rule using frequency domain method for the FODUP is
also proposed by Rotstein and Lewin (1991).
Venkatashankar and Chidambaram (1994) derived ap-
proximate analytical formulae of work by De Paor and
O'Malley (1989) for a P and PI controllers tuning for
the FODUP. The two design methods for P and PI
controllers by De Paor and O'Malley (1989) require
a numerical technique such as iteration method.
Sha"ei and Shenton (1994) proposed a graphical tech-
nique for tuning PID-type controllers based on the
method of D-partition for stable and unstable SISO
processes.
Huang and Lin (1995) developed PID tuning equa-
tions for the FODUP and SODUP using optimization
techniques. In their research, they obtained optimum
PID tuning parameters based on the minimum IAE
criterion and using the least-squares method.
0009-2509/00/$- see front matter 2000 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 0 9 - 2 5 0 9 ( 0 0 ) 0 0 0 0 5 - 1
Fig. 1. A classical feedback diagram.
Poulin and Pomerleau (1996) suggested a graphical
tuning method of PI and PID controllers for integrating
and unstable processes (FODUP and SODUP). Their
method is based on the analysis of the open-loop fre-
quency response of the process on the Nichols chart, and
the controller parameters are determined to satisfy the
speci"cation on the maximum peak resonance of the
closed-loop system according to the ITAE criterion.
Huang and Chen (1997) suggested a three-element
structure, which is equivalent to a two-degrees-of-free-
dom control scheme, for controlling the open-loop un-
stable processes. However, PID controller tuned by their
method still gave about 100% overshoots to a setpoint
change.
In open literature, there is no explicit PID controller
tuning method based on IMC for the FODUP and
SODUP. In this paper, the newly developed PID tuning
method by Lee, Lee, Park and Brosilow (1998) is ex-
tended to integrating and unstable processes with time
delay. Explicit PID controller tuning rules based on IMC
are proposed for the unstable processes with one RHP
pole (FODUP and SODUP), delayed unstable processes
with two RHP poles and integrating processes with time
delay. Six unstable process examples are provided to
demonstrate the proposed method and to compare re-
sults with existing tuning methods. Using a setpoint "lter,
the controller tuned by the proposed method does not
give overshoots in servo problems.
2. Theory
A classical feedback diagram is shown in Fig. 1.
The closed-loop transfer functions for designing the
feedback controller, G
!
are:
C
R
"
G
!
G
1#G
!
G
, (1)
C
d
"
G
"
1#G
!
G
. (2)
The controller G
!
should be designed to insure the
stability of these two transfer functions.
Eqs. (1) and (2) can be reformulated as the IMC struc-
ture (Morari & Za"riou, 1989) by
C
R
"
Gq
1#q(G!GI)
, (3)
C
d
"
(1!GIq)G
"
1#q(G!GI)
. (4)
If G"GI, then
C
R
"Gq, (5)
C
d
"(1!GIq)G
"
, (6)
where q is the IMC controller.
Let us consider an unstable process model
G(s)"P
+
(s)P

(s), (7)
where P
+
(s) contains the invertible portion of the model
and P

(s) contains all the noninvertible portion. The

invertible portions are the part of the model with stable
poles and unstable poles. The noninvertible portions are
the portion of model with right-half-plane zeros and time
delays. The following two conditions should be satis"ed
to stabilize the closed-loop response.
1. If the process model G has unstable poles up

,
2
, up
I
,
q should have zeros at up

,
2
, up
I
.
2. If the process model G
"
has unstable poles
dup

,
2
, dup
K
, (1!GIq) should have zeros at
dup

,
2
, dup
K
.
If these two conditions are satis"ed, the closed-loop res-
ponses for both a setpoint change and a load change
become stable.
The IMC controller is set as q"P
+
(s) f. Here, q has
zeros at up

,
2
, up
I
because P
+
(s) is the inverse of the
model portion with the unstable poles. Thus, the "rst
condition is satis"ed. Then, through the "lter design,
the second condition should be satis"ed. The "lter
for the IMC controller can be designed by two parts; f
Q
is
the portion to make the controller proper, and f
B
is the
portion to cancel the unstable poles or stable poles near
zero of G
"
.
f"f
Q
f
B
, (8)
f
Q
"
1
(zs#1)L
, (9a)
f
B
"
K
G
:
G
sG#1
(zs#1)K
, (9b)
Ces=3201=KCT=VVC=BG
3482 Y. Lee et al. / Chemical Engineering Science 55 (2000) 3481}3493
where n is chosen to make the controller realizable, :
G
are
determined to cancel the unstable poles of G
"
and m is
the number of unstable poles. Eq. (8) functions as a "lter
with adjustable time constant z.
1!GIq
QBSNG
2
BSNK
"0 (10)
where dup
G
O0.
Thus, the IMC controller is
q"
P
+
(s)
(zs#1)L
;
(K
G
:
G
sG#1)
(zs#1)K
. (11)
Then, we get
C
R
"Gq"
P

(s)
(zs#1)L
;
(K
G
:
G
sG#1)
(zs#1)K
, (12)
C
d
"(1!Gq)G
"
"

1!
P

(s)
(zs#1)L
(K
G
:
G
sG#1)
(zs#1)K
G
"
.
(13)
The lead term (K
G
:
G
sG#1) in Eq. (12) causes an
overshoot in the closed-loop response to a setpoint
change. This problemcan be resolved if we add a setpoint
"lter
f
0
"
1

K
G
:
G
sG#1

. (14)
So far, we have designed an IMC controller to obtain
stable closed-loop responses. Using the IMC controller,
we can get the closed-loop responses by Eqs. (12) and
(13). Next, the classical feedback controller we need is
obtained as
G
!
"
q
1!Gq
(15)
Thus, the controller G
!
is
G
!
"
P
+
(s)
(zs#1)L
;
(K
G
:
G
sG#1)
(zs#1)K
1!
P

(s)
(zs#1)L
(K
G
:
G
sG#1)
(zs#1)K
. (16)
The order n is chosen so that the controller G
!
is realiz-
able. Then, we convert the controller in Eq. (16) to the
standard PID form. Recently, Lee et al. (1998) developed
this conversion procedure for general process models as
follows. The controller G
!
can be approximated to a PID
controller by "rst noting that it can be expressed as
G
!
,f (s)/s. (17)
Expanding G
!
(s) in a Maclaurin series in s gives
G
!
(s)"
1
s
f (0)#f ' (0)s#
f " (0)
2!
s`#2

. (18)
The "rst three terms of the above expansion can be
interpreted as the standard PID controller given by
G
!
(s)"K
!

1#
1
t
'
s
#t
"
s

, (19)
where
K
!
"f ' (0) (20a)
t
'
"f ' (0)/f (0), (20b)
t
"
"f " (0)/2f ' (0), (20c)
t
'
*0; t
"
*0. (20d)
In this way, using Eqs. (20a)}(20d) we can convert the
ideal controller given by Eq. (16) to a standard PID
controller. When the process model has one or more
dominant lead time constants, the derivative and/or inte-
gral time constants computed from Eqs. (20b) and (20c)
can be negative independent of the choice of "lter time
constant. In this case, a PID controller cascaded with
a "rst or second-order lag of the form 1/([s#1) or
1/([
`
s`#[

s#1) is recommended (Lee et al., 1998).

2.1. PID controller settings
2.1.1. FODUP model
The process model is
G(s)"G
"
(s)"
Ke'FQ'
ts!1
(21)
and f
Q
"1/(zs#1), f
B
"(:s#1)/(zs#1), therefore the
IMC controller becomes q"(:s#1)(ts!1)/K(zs#1)`.
Then, we get
G
!
"
(ts!1)(:s#1)
K[(zs#1)`!(:s#1)exp(!0s)]
. (22)
Expanding G
!
(s) in a Maclaurin series in s gives
K
A
"
t
'
!K(2z#0!:)
, (23a)
t
'
"!t#:!
z`#:0!0`/2
2z#0!:
, (23b)
t
"
"
!t:!(0`/6!:0`/2)/(2z#0!:)
t
'
!
z`#:0!0`/2
2z#0!:
, (23c)
Ces=3201=KCT=VVC=BG
Y. Lee et al. / Chemical Engineering Science 55 (2000) 3481}3493 3483
where : value is calculated by solving 1!(:s#1)e'FQ'/
(zs#1)`
QO
"0.
That is
:"t[(z/t#1)`eFO!1]. (23d)
With this controller, the transfer function of the set-
point change is given by
C
R
"
(:s#1)eFQ
(zs#1)`
(24)
Therefore, if only PID controller is used, the closed-
loop response is at best Eq. (24), and the lead term
(:s#1) causes an overshoot. To eliminate the overshoot,
we propose a novel idea of adding a setpoint "lter
f
0
"1/(:s#1). Then the resulting closed-loop transfer
function is
C
R'
"
eFQ
(zs#1)`
, (25)
R'"
1
:s#1
R. (26)
2.1.2. Integrating process model with time delay
The process model is
G(s)"G
"
(s)"
Ke'FQ'
s
. (27)
We cannot directly use Eq. (16) for integrating pro-
cesses as the terms including : disappear at s"0. How-
ever, we can use Eq. (16) for integrating processes by
approximating the integrator as an unstable pole near
zero. Usually, the controller based on the model with an
unstable pole near zero can give more robust closed-loop
responses than that based on the model with an integ-
rator. Generally, we can design PID controllers with the
FODUP model for the integrating processes with time
delay. For example, the Eq. (27) is approximated with
a FODUP model as follows:
G(s)"G
"
(s)"
Ke'FQ'
s!0.01
"
100Ke'FQ'
100s!1
. (28)
Then, the PID controller is designed by (23a)}(23d).
2.1.3. SODUP model
The process model is
G(s)"G
"
(s)"
Ke'FQ'
(ts!1)(as#1)
(29)
and f
Q
"1/(zs#1), f
B
"(:s#1)/(zs#1), therefore the
IMC controller becomes q"(:s#1)(ts!1)(as#1)/
K(zs#1)`. Then, we get
G
!
"
(ts!1)(:s#1)(as#1)
K[(zs#1)`!(:s#1)exp(!0s)]
(30)
Expanding G
!
(s) in a Maclaurin series in s gives
K
A
"
t
'
!K(2z#0!:)
, (31a)
t
'
"!t#a#:!
z`#:0!0`/2
2z#0!:
, (31b)
t
"
"
!t:#a:!at!(0`/6!:0`/2)/(2z#0!:)
t
'
!
z`#:0!0`/2
2z#0!:
, (31c)
where : value is calculated by solving 1!(:s#1)e'FQ'/
(zs#1)`
QO
"0. That is
:"t[(z/t#1)`eFO!1]. (31d)
With this controller, the transfer function of the set-
point change is given by
C
R
"
(:s#1)eFQ
(zs#1)`
. (32)
Therefore, if only a PID controller is used, the closed-
loop response is at best Eq. (32), and the lead term
(:s#1) causes an overshoot. Adding a setpoint "lter
f
0
"1/(:s#1) results the closed-loop transfer function
C/R'"eFQ/(zs#1)`.
2.1.4. Unstable process model with two unstable poles and
time delay
The process models are
G(s)"G
"
(s)"
Ke'FQ'
(t

s!1)(t
`
s!1)
(33)
and from Eqs. (9a) and (9b) f
Q
"1/(zs#1)`, f
B
"
(:
`
s`#:

s#1)/(zs#1)`. The IMC controller becomes

q"(t

s!1)(t
`
s!1)(:
`
s`#:

s#1)/K(zs#1)`. Then,
we get
G
!
"
(t

s!1)(t
`
s!1)(:
`
s`#:

s#1)
K[(zs#1)"!eFQ(:
`
s`#:

s#1)]
. (34)
Expanding G
!
(s) in a Maclaurin series in s gives
K
A
"
t
'
K(4z#0!:

)
, (35a)
t
'
"!t

!t
`
#:

!
6z`!:
`
#:

0!0`/2
4z#0!:

, (35b)
Ces=3201=KCT=VVC=BG
3484 Y. Lee et al. / Chemical Engineering Science 55 (2000) 3481}3493
t
"
"
:
`
#t

t
`
!(t

#t
`
):

!(4z`#0:
`
#0`/6!:

0`/2)/(4z#0!:

)
t
'
!
6z`!:
`
#:

0!0`/2
4z#0!:

, (35c)
where :
`
, :

values are calculated by solving

1!
(:
`
s`#:

s#1)e'FQ'
(zs#1)"
QO O`
"0. (35d)
With this controller, the transfer function of the setpoint
change is given by
C
R
"
(:
`
s`#:

s#1)eFQ
(zs#1)"
. (36)
Therefore, if only a PID controller is used, the closed-
loop response is at best (36), and the lead term
(:
`
s`#:

s#1) causes an overshoot. Adding a setpoint

"lter f
0
"1/(:
`
s`#:

s#1) results the closed-loop

transfer function C/R'"eQ/(zs#1)".
The resulting tuning rules for the FODUP model, the
SODUP model with one unstable pole and two unstable
poles are summarized in Table 1.
2.2. Limitation of PID controllers for unstable processes
with time delay
For some model equations, the ideal controller in Eq.
(16) can give the desired closed-loop response but the
standard PIDcontroller in Eqs. (20a)}(20c) does not exist
that can give the desired closed-loop response. This can
be explained as follows: Let us consider a FODUP
model. In Eq. (1), the controller G
!
is substituted with
a PID controller. Thus,
C
R
"
G
!
G
1#G
!
G
"
K
!
(1#1/(t
'
s)#t
"
s) KeFQ/(ts!1)
1#K
!
(1#1/(t
'
s)#t
"
s)KeFQ/(ts!1)
. (37)
The Routh (1905) stability criterion can be used to
analyze the stability of the closed-loop transfer function.
To use this criterion, eFQ is replaced by a 2/2 PadeH
approximation. We can obtain a necessary condition for
stability as follows:
a
"
"
1
12
0`t
'
(t
"
#
t
KK
!
)'0, (38a)
a
`
"
0t
'
12
(0!6t
"
#
6t!0
KK
!
)'0, (38b)
a
`
"t
'
(t
"
!
0
2
#
1
KK
!
(t!
0
2
))#
1
12
0`'0, (38c)
a

"t
'
(1!
1
KK
!
)!
0
2
'0 (38d)
a
"
"1 (38e)
and a necessary and su$cient condition for stability is as
follows:
a
"
'0, a
`
'0, b

"a
`
!
a
"
a

a
`
'0,
c

"a

!
a
`
a
"
b

'0. (38f)
There exist ranges of model parameters of unstable pro-
cesses that can be controlled using a standard PID con-
troller. An extensive simulation study indicates that there
exist sets of PID parameters that satisfy the stability
criteria, Eqs. (38a)}(38d) and (38f) when the following is
satis"ed:
0)
0
t
(2. (39)
None of the sets of PID parameters satisfy the stability
criteria when 0/t*2.
When a time delay 0 is zero, a necessary and su$cient
condition for stability is as follows:
KK
!
'1; t
'
*0; t
"
*0. (40)
3. Examples
Six examples are used to show the advantage of the
proposed method.
All simulations were performed using MATLAB 4.0
(control system design and simulation software) (Shahian
& Hassul, 1993).
Example 1. First order delayed unstable process (FOD-
UP). Consider an unstable process with process and
disturbance models as follows (Huang & Chen, 1997)
G(s)"G
"
(s)"
e""Q
s!1
. (41)
The closed-loop time constant was chosen as z"0.5.
The PID tuning values used in the simulation are
presented in Table 2. The tuning values of existing
methods in the Table 2 of this paper were obtained from
the Table 2 of the work by Huang and Chen (1997).
Figs. 2a and b show the closed-loop responses of the
unstable process given by Eq. (41) to a unit step change in
Ces=3201=KCT=VVC=BG
Y. Lee et al. / Chemical Engineering Science 55 (2000) 3481}3493 3485
T
a
b
l
e
1
P
I
D
t
u
n
i
n
g
r
u
l
e
s
f
o
r
F
O
D
U
P
a
n
d
S
O
D
U
P
p
r
o
c
e
s
s
e
s

P
r
o
c
e
s
s
K
A
t
'
t
"
S
e
t
p
o
i
n
t
"
l
t
e
r
F
O
D
U
P
G
(
s
)
"
K
e

F
Q
t
s
!
1
K
A
"
t
'
!
K
(
2
z
#
0
!
:
)
t
'
"
!
t
#
:
!
z
`
#
:
0
!
0
`
/
2
2
z
#
0
!
:
t
"
"
!
t
:
!
(
0
`
/
6
!
:
0
`
/
2
)
/
(
2
z
#
0
!
:
)
t
'
!
z
`
#
:
0
!
0
`
/
2
2
z
#
0
!
:
1
/
(
:
s
#
1
)
S
O
D
U
P
a
.
w
i
t
h
o
n
e
u
n
s
t
a
b
l
e
p
o
l
e
G
(
s
)
"
K
e

F
Q
(
t
s
!
1
)
(
a
s
#
1
)
K
A
"
t
'
!
K
(
2
z
#
0
!
:
)
t
'
"
!
t
#
a
#
:
!
z
`
#
:
0
!
0
`
/
2
2
z
#
0
!
:
t
"
"
!
!
t
:
#
a
:
!
a
t
t
'
!
(
0
`
/
6
!
:
0
`
/
2
)
/
(
2
z
#
0
!
:
)
t
'
!
z
`
#
:
0
!
0
`
/
2
2
z
#
0
!
:
1
/
(
:
s
#
1
)
b
.
w
i
t
h
t
w
o
u
n
s
t
a
b
l
e
p
o
l
e
s
G
(
s
)
"
K
e

F
s
(
t

s
!
1
)
(
t
`
s
!
1
)
K
A
"
t
'
K
(
4
z
#
0
!
:

)
t
'
"
!
t

!
t
`
#
:

!
6
z
`
!
:
`
#
:

0
!
0
`
/
2
4
z
#
0
!
:

t
"
"
:
`
#
t

t
`
!
(
t

#
t
`
)
:

!
(
4
z
`
#
0
:
`
#
0
`
/
6
!
:

0
`
/
2
)
/
(
4
z
#
0
!
:

)
t
'
!
6
z
`
!
:
`
#
:

0
!
0
`
/
2
4
z
#
0
!
:

1
/
(
:
`
s
`
#
:

s
#
1
)

:
"
t
[
(
z
/
t
#
1
)
`
e
F

O
!
1
]
,
d
e
s
i
r
e
d
c
l
o
s
e
d
l
o
o
p
r
e
s
p
o
n
s
e
:
C
/
R
"
e

F
Q
/
(
z
s
#
1
)
`
i
n
F
O
D
U
P
a
n
d
S
O
D
U
P
(
a
)
,
:
`
,
:

v
a
l
u
e
s
a
r
e
c
a
l
c
u
l
a
t
e
d
b
y
s
o
l
v
i
n
g
1
!
(
:
`
s
`
#
:

s
#
1
)
e
'

F
Q
'
/
(
z
s
#
1
)
"

O
`
"
0
,
d
e
s
i
r
e
d
c
l
o
s
e
d
-
l
o
o
p
r
e
s
p
o
n
s
e
:
C
/
R
"
e

F
Q
/
(
z
s
#
1
)
"
i
n
S
O
D
U
P
(
b
)
.
Table 2
Tuning values by the proposed and existing methods for Example 1
Method K
!
t
'
t
"
Setpoint "lter
De Paor and O Malley 1.459 2.667 0.250
Rotstein and Lewin 2.250 5.760 0.2
Huang and Chen 2.636 5.673 0.118
Proposed 2.634 2.52 0.154
Proposed
(with setpoint "lter)
2.634 2.52 0.154 1/(2.36s#1)
setpoint and load. Figs. 2c and d show controller outputs
for the responses in Figs. 2a and b. The process responses
to a unit step change in the load by the proposed method
with/without the setpoint "lter are the same. The results
shown in the "gures illustrate the superior performance
of the proposed method. Figs. 3a}c show closed-loop
responses when there exist plant-model mismatches. The
model values of parameters, K, t, 0 are di!erent from the
true values by 20%. In model-plant mismatch, robust
responses are shown in the "gures.
Example 2. Second order delayed unstable process
(SODUP). Consider an unstable process as (Huang
& Chen, 1997)
G(s)"G
"
(s)"
e"`Q
(5s!1)(2s#1)(0.5s#1)
. (42)
Huang and Chen (1997) approximated the model with
the "rst-order lag following SODUP model:
G(s)"
e""`"Q
(5s!1)(2.07s#1)
. (43)
The "lter time constant z was chosen as 1.2 to give the
desired closed-loop response. In Table 3, the PID tuning
values used in simulation are presented. The tuning
values of existing method in the Table 3 of this paper
were obtained from the Example 3 of work by Huang
and Chen (1997). Figs. 4a and b show the closed-loop
responses of the unstable process given by Eq. (43) to
a unit step change in setpoint and load. Figs. 5a}c show
closed-loop responses when there exist plant-model mis-
matches. The values of model parameters, K, t, 0 are
di!erent from the true values by 20%. The "gures show
the controllers tuned by the proposed tuning give robust
responses when there exist model-plant mismatches.
Example 3. Consider an unstable process with two un-
stable poles as follows:
G(s)"G
"
(s)"
2e"`Q
(3s!1)(s!1)
. (44)
The "lter time constant z was chosen as z"0.45.
Ces=3201=KCT=VVC=BG
3486 Y. Lee et al. / Chemical Engineering Science 55 (2000) 3481}3493
Fig. 2. Closed-loop responses by the proposed method with z"0.5 and existing methods when G(s)"G
"
(s)"e""Q/(s!1). (a) Process response to
a setpoint change, (b) process response to a load change, (c) response of controller output to a setpoint change, (d) response of controller output to
a load change.
The tuning parameters are K
!
"2.3153, t
'
"1.7843,
t
"
"1.8859.
Setpoint "lter is f
0
"1/(3.252s`#1.7147s#1). Figs.
6a and b show the closed-loop responses of the unstable
process given by Eq. (44) to a unit step change in setpoint
and load. Figs. 6c and d show controller outputs for the
responses in Figs. 6a and b.
Example 4. Consider an unstable process with a strong
lead time constant and two unstable poles as follows:
G(s)"G
"
(s)"
2(5s#1)e"`Q
(3s!1)(s!1)
. (45)
Using Eq. (16), an ideal controller G
!
is obtained to
give a desired closed-loop responses. The "lter time con-
stant z was chosen as z"0.45. But, the strong lead term
(5s#1) in Eq. (45) causes tuning parameters, t
'
, t
"
, to be
negative values. In such cases, a PID controller cascaded
with a "rst- or second-order lag of the form 1/([s#1) or
1/([
`
s`#[

s#1) is recommended (Lee et al., 1998).

A PID lag controller is obtained as follows:
To get a PID with a "rst-order lag, we write G
!
(s) as
G
!
(s),f (s)/s"
1
s
[f (s)h(s)]
h(s)
, (46)
where h(s),1#[s
Ces=3201=KCT=VVC=BG
Y. Lee et al. / Chemical Engineering Science 55 (2000) 3481}3493 3487
Fig. 3. Closed-loop responses to a setpoint change when there exist model-plant mismatches for example 1. (a) Parameter mismatch in K,
(b) parameter mismatch in t, (c) parameter mismatch in 0.
The expansion of Eq. (46) then becomes
G
!
(s)" f (0)#[f ' (0)#[f (0)]s
#[ f " (0)#2[f ' (0)]s`/2
#[ f ''' (0)#3[f " (0)]s`/6/s([s#1). (47)
Selecting the lag parameter, [, to drop the third-order
term gives
["!f ''' (0)/3f " (0) (48a)
and the PID parameters are
K
!
"f ' (0)#[f (0); t
'
"K
A
/f (0);
t
"
"(f " (0)#2[f '(0))/(2K
A
). (48b)
Using Eqs. (48a) and (48b), a following PID controller
with a "rst-order lag is obtained as
G
!
(s)"2.3012

1#
1
1.7733s
#1.9172s

1
4.989s#1
. (49)
The set point "lter is
f
0
"1/(3.252s`#1.7147s#1).
Figs. 7a and b show the closed-loop responses of the
unstable process given by Eq. (45) to a unit step change in
setpoint and load.
Ces=3201=KCT=VVC=BG
3488 Y. Lee et al. / Chemical Engineering Science 55 (2000) 3481}3493
Fig. 4. Closed-loop responses by the proposed method with z"1.2 and existing methods when G(s)"G
"
(s)"e"`Q/(5s!1)(2s#1)(0.5s#1). (a)
Process response to a setpoint change, (b) process response to a load change.
Table 3
Tuning values by the proposed and existing methods for Example 2
Method K
!
t
'
t
"
Setpoint "lter
Huang and Chen (1997) 6.186 7.170 1.472
Huang and Lin (1995) 3.954 4.958 2.074
Poulin and Pomerleau
(1996)
3.050 7.557 2.070
Proposed 7.144 6.684 1.655
Proposed
(with setpoint "lter)
7.144 6.684 1.655 1/(4.276s#1)
Example 5. Consider an integrating and unstable pro-
cess as follows:
G(s)"G
"
(s)"
e"`Q
s(s!1)
. (50)
To use the proposed method, the model is changed as:
G(s)"G
"
(s)"
e"`Q
(s!0.01)(s!1)
. (51)
The closed-loop time constant was chosen as z"0.8.
The tuning parameters are K
!
"0.8412, t
'
"3.3066,
t
"
"2.8113. The setpoint "lter is f
0
"1/(8.4593s`#
3.3607s#1). Figs. 8a and b show the closed-loop re-
sponses of the unstable process given by Eq. (50) to a unit
step change in setpoint and load.
Example 6. Let us consider a real industrial batch chem-
ical reactor. For a reactor with exothermic reactions, the
process is often modeled with unstable process models.
An experimental study on a batch polymerization reactor
was done by Lee, Lee & Park (1999) The process model is
a FODUP model and the controllers in the application
were tuned by the proposed method. The tuning based
on the proposed method eliminated considerable
overshoots caused by original empirical settings in the
application.
4. Conclusions
A new method for PID controller tuning for unstable
processes was proposed. The tuning rule is based on the
process model and the desired closed-loop response. The
ideal controller is found that can give the desired closed-
loop response. The PID approximation of the ideal
controller is obtained by taking "rst three terms from
Maclaurin series expansion of the ideal controller.
Closed-loop responses of unstable processes tuned by the
proposed method are compared with those of existing.
The results show that the proposed tuning method is
superior to the existing methods. Simulation results show
that the proposed method is robust to model-plant mis-
match.
Ces=3201=KCT=VVC=BG
Y. Lee et al. / Chemical Engineering Science 55 (2000) 3481}3493 3489
Fig. 5. Closed-loop responses to a setpoint change when there exist model-plant mismatches for example 2. (a) Parameter mismatch in K, (b)
parameter mismatch in t, (c) parameter mismatch in 0.
Notation
C controlled variable
dup
I
number of unstable poles of disturbance model
f
B
portion of a controller "lter to cancel the
unstable poles and stable poles near zero
f
Q
portion of a controller "lter to make the con-
troller proper
f
0
setpoint "lter
G process
GI process model
K process gain
K
A
gain, proportional tuning parameter
M manipulated variable
P
+
(s) portion of a process model inverted by the
controller. It must be minimum phase
P

(s) portion of a process model not inverted by

the controller. It is usually non- minimum
phase (i.e. contains dead times and/or right-
half-plane zeros)
q IMC controller
R setpoint
up
I
number of unstable poles
Greek letters
: "lter parameter in the "lter portion to cancel
the unstable poles
Ces=3201=KCT=VVC=BG
3490 Y. Lee et al. / Chemical Engineering Science 55 (2000) 3481}3493
Fig. 6. Comparison of the closed loop responses by the proposed tuning method with and without a setpoint "lter when
G(s)"G
"
(s)"2e"`Q/(3s!1)(s!1), z"0.45. (a) Process response to a setpoint change, (b) process response to a load change, (c) response of
controller output to a setpoint change, (d) response of controller output to a load change.
[ parameter to make PID lag controller
t
'
reset time, integral time constant
t
"
derivative time constant
z desired closed-loop time constant
t time constant
0 dead time
Acronyms
RHP right-half-plane
IMC internal model control
PID proportional}integral-derivative
fodup "rst-order delayed unstable process
sodup tsecond-order delayed unstable process
Acknowledgements
This work was supported in part by the Korea Science
and Engineering Foundation(KOSEF) through the
Automation Research Center at POSTECH.
Ces=3201=KCT=VVC=BG
Y. Lee et al. / Chemical Engineering Science 55 (2000) 3481}3493 3491
Fig. 7. Comparison of the closed loop responses by the proposed tuning method with and without a setpoint "lter when
G(s)"G
"
(s)"2(5s#1)e"`Q/(3s!1)(s!1), z"0.45. (a) Process response to a setpoint change, (b) process response to a load change.
Fig. 8. Comparison of the closed-loop responses by the proposed tuning method with and without a setpoint "lter whenG(s)"G
"
(s)"e"`Q/s(s!1),
z"0.8. (a) Process response to a setpoint change, (b) process response to a load change.
References
De Paor, A. M. (1989). Controllers of Ziegler}Nichols type for unstable
process with time delay. Internatoional Journal of Control, 49, 1273.
Huang, H. P., & Chen, C. C. (1997). Control-system synthesis for
open-loop unstable process with time delay. IEE Process-Control
Theory and Application, 144, 334.
Huang, C. T., & Lin, Y. S. (1995). Tuning PID controller for open-loop
unstable processes with time delay. Chemical Engineering Commun-
ications, 133, 11.
Lee, J., Lee, Y., & Park, S. (1999). Local linearized model identi"cation
and PID tuning for unstable process models in batch polymeriz-
ation reactors. AIChE spring meeting.
Lee, Y., Lee, M., Park, S., & Brosilow, C. (1998). PID controller tuning
for desired closed-loop responses for SI/SO systems. A.I.Ch.E. Jour-
nal, 44(1), 106.
Morari, M., & Za"riou, E. (1989). Robust process control. Englewood
Cli!s, NJ: Prentice Hall.
Poulin, ED ., & Pomerleau, A. (1996). PID tuning for integrating and
unstable processes. IEE Process Control Theory and Application,
143(5), 429.
Ces=3201=KCT=VVC=BG
3492 Y. Lee et al. / Chemical Engineering Science 55 (2000) 3481}3493
Quinn Jr., S. B., & Sanathanan, C. K. (1989). Controller design for
integrating and runaway processes involving time delay. A.I.Ch.E.
Journal, 35(6), 923.
Rotstein, G. E., & Lewin, D. R. (1991). Simple PI and PID tuning for
open-loop unstable systems. Industrial Engineering and Chemical
Research, 30, 1864.
Rotsetin, G. E., & Lewin, D. R. (1992). Control of an Unstable Batch
Chemical Reactor. Computers in Chemical Engineering, 16(1), 27.
Routh, E. J. (1905). Dynamics of a system of rigid bodies, Part II.
London: MacMillan.
Sha"ei, Z., & Shenton, A. T. (1994). Tuning of PID-type controllers for
stable and unstable systems with time delay. Automatica, 30(10),
1609.
Shahian, B., & Hassul, M. (1993). Control system design using matlab.
Englewood Cli!s, NJ: Prentice Hall.
Venkatashankar, V., & Chidambaram, M. (1994). Design of P and PI
controllers for unstable "rst-order plus time delay systems. Interna-
tional Journal of Control, 60(1), 137.
Ces=3201=KCT=VVC=BG
Y. Lee et al. / Chemical Engineering Science 55 (2000) 3481}3493 3493