IFAC Conference on Advances in PID Control

PID'12
Brescia (Italy), March 28-30, 2012

FrA2.3

Tuning Method of PI Controller with Desired Damping Coefficient

for a First-order Lag Plus Deadtime System
Yuji Yamakawa*. Yohei Okada**
Takanori Yamazaki***. Shigeru Kurosu****

*Univ. of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo, Japan

(Tel: +81-3-5841-6937; e-mail: Yuji_Yamakawa@ipc.i.u-tokyo.ac.jp)
**SINTOKOGIO LTD., Aichi, Japan
(e-mail: yohei_japan67@hotmail.com)
***Oyama National College of Technology, 771 Nakakuki, Oyama, Tochigi, Japan
(e-mail: yama@oyama-ct.ac.jp)
**** Research Inst. Crotech, 61-88 Ozakata, Chikusei, Ibaraki, Japan
(e-mail: kurosu_shi@yahoo.co.jp)
Abstract: This paper presents an entirely new tuning method of PI controller with a desired damping
coefficient. We consider a plant model described by a first-order lag plus deadtime system in the process
control. The ultimate sensitivity method presented by Ziegler and Nichols has been still widely used, but it
has a disadvantage that gives an oscillatory response. In recent years a less oscillatory response has been
judged to be more appropriate for process controls. In tuning PI controller it is often convenient to
determine the damping coefficient to obtain the desired control performance. The deadtime element can be
approximated by the Pad equation to determine the damping coefficient for the equivalent second-order
lag system. In this paper the integral time is normally chosen equal to the time constant of a plant to
degenerate the order of the closed-loop transfer function so that pole-zero is cancelled. As a result, the
relations among the gain constant, the time constant, the deadtime of the plant and the proportional gain of
the controller are clarified when the desired damping coefficient is provided.
Keywords: process control, PID control, damping coefficient.

1. INTRODUCTION
In most process controls, tuning PID controllers can be
achieved on the assumption that a plant model is
approximated by a first-order lag plus deadtime system. The
ultimate sensitivity method developed experimentally by
Ziegler and Nichols in 1942 has been by far the most
common controller tuning method. However, since the decay
ratio, which is called quarter damping ratio (i.e. the damping
coefficient = 0.22), has been chosen traditionally, this leads
to poor attenuation, so that the closed-loop response can be
rapidly oscillatory due to huntings.
In typical specifications on a control system it is desirable to
have a critically damped response with no overshoot. To do
this, we need a simple, easy-to-use, intuitive tuning method
that gives moderate damping effect. In this study a pair of
conjugate complex poles to form an oscillatory mode for the
closed-loop transfer function are primary concern for PI
control used when the deadtime element of the plant can be
approximated by the Pad equation. The PI controller tuning
method with a desired damping coefficient for the closedloop system is developed.
2. CONTROLLED PLANT AND OVERALL CONTROL
SYSTEM
Consider a first-order lag plus deadtime system with the
transfer function:

k p 1

T
is

K
L2 s 2 6Ls 12

Ts 1 L2 s 2 6Ls 12

Fig. 1 PI control system

GP (s)

K
eLs ,
Ts 1

(1)

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

deadtime of the controlled plant. Since the pure deadtime
element cannot be described by a finite-order model, it can be
practically approximated by the second-order Pad equation as:

GP (s)

K L2 s 2 6Ls 12

.
Ts 1 L2 s 2 6Ls 12

(1)

Consider the simple feedback loop, shown in Fig. 1, and

composed of a plant and a PID controller. In Fig. 1, r is the
setpoint, e is the control error, u is the control input, and x is
the control output, kp is the proportional gain and Ti is the
integral time. For such a control system the transfer function
from the setpoint r to the control output x is given by

k p Ti s 1 K L2 s 2 6Ls 12

Ti s Ts 1 L2 s 2 6Ls 12
.
Gcl (s)
k T s 1 K L2 s 2 6Ls 12
1 p i
2 2

Ti s Ts 1 L s 6Ls 12

(2)

IFAC Conference on Advances in PID Control

PID'12
Brescia (Italy), March 28-30, 2012

FrA2.3

3. CHARACTERISTICS OF POLES FOR PI CONTROL

SYSTEM
Expanding Eq. (2), Gcl(s) can be expressed as a ratio of
polynomials, and the orders of the numerator polynomial and
the denominator one are 3 and 4, respectively. In pole
assignment design it is attempted to assign a pair of conjugate
complex poles of the closed-loop system equal to the
dominant poles, and the PI parameters should be chosen so
that the damping coefficient of the dominant poles leads to
better response. In this current analysis it is desirable that the
order of the denominator polynomial should be 3 to obtain
satisfactory control performance.
3.1 Close-loop transfer function with integral time fixed
Considering the closed-loop transfer function given by Eq.
(2) and letting the integral time Ti be the time constant T, the
order of the denominator polynomial can be easily reduced
through the pole-zero cancellation. As a result the closedloop transfer function Gcl(s) is given by

k p K L2 s 2 6Ls 12
2 2
Gcl (s) Ts L s2 2 6Ls 12
k K L s 6Ls 12
1 p 2 2
Ts L s 6Ls 12

Gcl (s)

k p K A (Ls)2 6(Ls) 12
(Ls)3 6 k p K A (Ls)2

Im.
(, j)

Re.

(p, 0)

(a) 0 < 90

Re.

(p, 0)

(b) 90 < 180

Fig. 2 Angle between a line connecting real pole and

oscillatory poles, and real axis

1
a
Q R p

L
3

(5)

3 Q R
1 Q R a

j
s2,3
j
2
L
2
3

s1

In this process the following interim parameters have been

introduced:

M
(3)

12 6 k p K A (Ls) 12 k p K A

3b a 2
2a 3 9ab 27c
,N
,P M3 N2
54
9
3

P N, R

P N .

According to Eq. (5) both real and oscillatory roots can be

given by the function of 1/L. Writing the real and imaginary
parts of oscillatory roots as and we get the damping
coefficient through straightforward calculations as

where A = T/L and Ti = T.

In this case we find that the closed-loop zeros of Eq. (3) are
practically kept constant, and cannot be adjustable by the
controller. On the contrary, it is useful that the closed-loop
poles should be assigned in accordance with the required
design criteria. Eq. (3) indicates major great advantage on the
PI controller that the proportional gain kp becomes the only
one adjustable parameter because the integral time Ti is kept
fixed. Therefore, the pole-assignment for the PI controller
gives the simple problem solving the four parameters, namely
kp, K, T, and L.
3.2 Relations of poles to damping coefficient
To get some insight into the relations between poles and
damping coefficient, let us investigate the following
simplified characteristic equation given in the denominator
polynomial of Eq. (3):

L3s 3 aL2 s 2 bLs c 0

Im.
(, j)

(4)

where a =6 kpK/A, b =12 6 kpK/A, and c =12 kpK/A.

The closed-loop poles are the roots of Eq. (4) and the polezero configuration may be varied significantly. Many simple
feedback loops, however, will be given by a real pole s1 and
complex pair of poles s2 and s3 in case that P2 0 (discussed
later), which can be computed by

2
2

(Q R) a

2
3
2
Q R a 3(Q R)

2
2
3

(6)

Notice that can be determined by a, b, and c only,

regardless of L. Since Q and R are the functions of a, b, and c,
all of these parameters are the functions of kpK/A, so that
can be easily obtained as the function of kpK/A.
3.3 Effects of the real pole on the oscillatory poles
The damping coefficient can be determined by a set of the
real pole and the oscillatory poles. Thus, if the damping
coefficient of the closed-loop response is assumed to be equal
to that of oscillatory poles, the real pole may be neglected by
assigning far from the oscillatory poles of concern.
To examine the effects of the real pole on the oscillatory
poles, we define the angle between the real axis and the
connecting line of real pole and the oscillatory poles, as
shown in Fig. 2.
From Eq. (5), the angle is given by,

IFAC Conference on Advances in PID Control

PID'12
Brescia (Italy), March 28-30, 2012

tan 1

3 (Q R )
,
tan 1

p
3(Q R )

FrA2.3

(7)

where normally varies within the range from 0 to 90 (0

< 90). In case that is gets over the range (90 < 180),
is given by,

180 .

(8)

According to Eq. (7), (8) is the function of kpK/A as well as

. In the previous paper 3), if is less than 50 ( 50) on
third-order system consisted of real pole and oscillatory poles,
it has been clarified that the effects of real pole can be
neglected. Thus, when specifying of oscillatory poles, it is
necessary to examine if the condition for of less than 50 is
satisfied. As a result, the value of affects the response (over
damped or oscillation) of the control system.
4. PI PARAMETERS FOR DESIRED DAMPING
COEFFICIENT
4.1 kp for desired coefficient
To derive kp for the desired damping coefficient , Eq. (6) can
be varied with kpK/A. The relation between and kpK/A is
shown as in Fig. 3. It can be seen that there exist three regions
as kpK/A increases. The first region means the region where the
oscillatory poles join the real axis at the junction point as kpK/A
increases. The Second region means the region where all three
poles become real. The third region means the region where
the oscillatory poles moves from the separation point to the
stability limit at the imaginary axis. The behaviors for these
three regions are depicted as shown in Fig. 4.

Next let us consider how Eq. (7) and (8) are varied with kpK/A.
The relation of to kpK/A can be obtained as shown in Fig. 5.
When looking at Fig. 5 corresponding to Fig. 3, is less than
50 in the third region and the damping coefficient of
oscillatory poles can be desired as the damping coefficient of
a step response in this region. Therefore, kp for the desired
damping coefficient can be calculated by using the region 3
(0.368 kpK/A < 1.58). In region 3 in Fig. 3, kpK/A can be determined uniquely when is desired. Thus, Eq. (9) can be
obtained by replacing kpK/A by
k p K / A ( )

(9)

where is the function of . for desired is shown in Table 1.

From Eq. (10), kp for the desired damping coefficient can be
given by
k p ( ) A / K

(10)

From Eq. (10) and the results shown in Table 1, kp for the
desired damping coefficient can be found to be a completely
linear function of A/K, as shown in Fig. 6.

IFAC Conference on Advances in PID Control

PID'12
Brescia (Italy), March 28-30, 2012

4.2 Validation for the pure deadtime system

In section 4.1, we proposed the PI tuning method based on the
damping coefficient of oscillatory poles. So that it is response
of = 0.5 is required for this plant, the proportional gain kp =
6.61 can be obtained from Fig. 6. Then the integral time Ti =
10 can be found as Ti = T. The step responses for the desired
= 0.5 and 1.0 are shown in Fig. 7(a), (c) respectively. Because
the damping coefficient of step responses for the desired =
1.0 cannot be obtained from the amplitude ratio, the damping
coefficient is assumed to be = 0.5.
An enlarged figure of the step responses desired as = 0.5 is
shown in Fig. 7(b). In this figure, the dashed line is the step
responses for first order lag plus deadtime system approximated by the second-order Pad equation. In the calculation of
the damping coefficient of the step responses from amplitude
ratio, it is found that the step response corresponds approximately to the damping coefficient desired by oscillatory poles.
In this figure, the solid lines show the step responses for firstorder lag plus deadtime system with pure deadtime, and it can
be found that the step response (solid line) almost corresponds
to the step response (dashed line). Thus, it can be concluded
that this parameter method is also effective for pure deadtime.
As additional consideration, the plant which the deadtime is

FrA2.3

larger than the time constant, K = 1, A = 0.1 (T << L) is

considered. For this plant, the step responses for the desired =
0.5 and 1.0 are shown in Fig. 8(a), (c) respectively, and
enlarged figure of Fig. 8 (a) is shown in Fig. 8(b). In case that
T << L, the step response for the desired damping coefficient
can be obtained successfully as well as T >> L.
4.3 Comparison with PMM (Partial Model Matching)
Table 2 shows PI parameters obtained by the proposed method
and PMM (Partial Model Matching method, proposed by
Kitamori, 1985) for =1.0.
As you can see Table 2, both PI parameters are about the same.
This result is significantly interesting.
Since the zero of the controller compensates the pole of the
plant in the proposed method, it can be considered that the
pole-zero cancellation also occurs in the PMM. In the future,
we will discuss about the other damping coefficients.
Fig. 9 shows pole-zero maps for the closed-loop system (the
characteristics of the plant is A = 1.) using PI controller. In Fig.
9(a), the PI parameters are decided by using the proposed
method, and the PI parameters are determined by using the
PMM in Fig. 9(b). Fig. 10 shows the results of the step
response for both control systems.

IFAC Conference on Advances in PID Control

PID'12
Brescia (Italy), March 28-30, 2012

FrA2.3

Table 2: Comparison of PI parameters

of the response depends on the value of the deadtime L.

Plant
T
1
2
3
5
10
15
10
20
30

L
1
2
3
1
2
3
1
2
3

( K=1)

Proposed
Ti
kp
T/L
1 0.368
1
2
1 0.368
3
1 0.368
5 1.84
5
5 1.84
10
5 1.84
15

PMM
kp
Ti
0.329 0.935
0.329 1.87
0.329 2.80
1.85 4.99
1.85 9.98
1.85 15.0

10
10

3.68
3.68

10
20

3.73

9.99

3.73

10

3.68

30

3.73

20.0
30.0

(=1.0)

It can be seen from Fig. 9 that the control system with the
PMM cannot achieve pole-zero cancellation for the plant (A =
1) that the deadtime L is the same as the time constant T.
However, in this situation, the non-oscillation response can be
obtained. And, both responses are the almost same by
comparing the response using the proposed method with that
using the PMM.
In addition, Fig. 11 shows pole-zero maps for the closed-loop
system (the characteristics of the plant is A = 10.) using PI
controller. In Fig. 11(a), the PI parameters are decided by using
the proposed method, and the PI parameters are determined by
using the PMM in Fig. 11(b). Fig. 12 shows the results of the
step response for both control systems. In this case, the both
characteristics (the pole-zero map and the step response) are
the about perfectly same. Moreover, although the both
response between Fig. 10 and Fig. 12 are the same, the speed

5. CONCLUSIONS
We proposed PI controller tuning method for first-order lag
plus deadtime system. On the assumption that the integral
time Ti is chosen equal to the time constant T of a plant to
degenerate the order of the closed-loop transfer function, we
obtained the following conclusions.
1) The step response desired by of oscillatory poles can be
approximated the equivalent response of second-order
system since real pole can be almost neglected.
2) A proportional gain kp for the desired damping coefficient
is an exactly linear function of A/K (A = T/L).
3) We demonstrated an applicability of the tuning method of
PI controller for the given damping coefficient.
REFERENCES
J.G. Ziegler and N.B. Nichols (1942). Optimum Setting for
Automatic Controllers, Trans. ASME, Vol. 64, pp.759-768.
R. Kuwata (1987). An Improved Ultimate Sensitivity Method
and PID; I-PD Control Characteristics, Transaction of
SICE, Vol. 23, No. 4, pp.232-239.
Y. Okada, Y. Yamakawa, T. Yamazaki, S. Kurosu (2006).
Tuning Method of PI Controller for Given Damping
Coefficient, SICE-ICASE International Joint Conference,
pp. 284, SP07-4.
T. Kitamori (1985). A Method of Control System Design Based
upon Partial Knowledge about Controlled Processes,
Transaction of SICE, Vol. 15, No. 4, pp. 549-555.