Refinements of the Ziegler-Nichols tuning formula

C.C.Fang
K.J. Astrom
W.K. Ho

Indexing terms: Control systems, Control theory

Abstract: The accuracy of the Ziegler-Nichols

tuning formula is reviewed in the context of PID
and PI autotuning. For PID autotuning, it will be
shown that, for excessive overshoot in the setpoint response, set-point weighting can reduce the
overshoot to specified values, and the original
Ziegler-Nichols tuning formula can be retained. It
will also be shown that set-point weighting is
superior to the conventional solution of reducing
large overshoot by gain detuning or set-point filtering. However, for excessive set-point undershoot, the tuning formula will have to be modified.
For PI autotuning, it will be shown that the
Ziegler-Nichols tuning formula is inadequate and
has to be completely revised. The application of
set-point weighting and the modification of the
tuning formula can be based simply on the knowledge of the normalised gain or normalised deadtime of the process. These heuristic refinements,
when incorporated, will give appreciable improvement in the performance of autotuners.
1

Introduction

The Ziegler-Nichols ultimate-cycle or closed-loop tuning

[l, 21 has been widely known as a fairly accurate heuristic method to determine good settings of PID and PI
controllers, for a wide range of common industrial processes. However, this manual tuning method is not often
applied in practice because it is laborious and timeconsuming, particularly for a process with large time constant. It also requires the close attention of the
instrument/control engineer and the operator, because
the process has to be operated near instability to measure
the ultimate gain and period.
Several commercial products for automatic tuning and
adaptive control [3-81 have been available since 1981.
Recently, a simple method of autotuning has been developed. It is based on the automatic measurement of the
ultimate gain and period, from which the optimal PID
and PI controller parameters can be computed using the
Ziegler-Nichols tuning formula. Various techniques, such
as the relay feedback [ S , 91, approximate system identification [lo] and crosscorrelation [ l l , 123, have been
developed to determine the ultimate gain and ultimate
Paper 7639D (CS), received 6th April 1990
C.C. Hang and W.K. Ho are with the Department of Electrical Engineering, National University of Singapore, 10 Kent Ridge Crescent,
Singapore 051 1
K.J. Astrom is with the Department of Automatic Control, Lund Institute of Technology, Sweden

IEE PROCEEDINGS-D, Vol. 138, NO. 2, M A R C H I991

period without the need to drive the system to the verge

of instability.
The accuracy of the Ziegler-Nichols tuning formula
has been found to be quite adequate in manual tuning,
because it can be supplemented by fine-tuning based on
experience. With the automation of the controller-tuning
procedure, it becomes attractive to explore the possibility
of modifying or augmenting the tuning formula by incorporating heuristic knowledge to replace manual finetuning. Heuristics are naturally limited to certain classes
of problem. The heuristic rules presented in this paper
are applicable to the same class of process considered in
the work of Ziegler and Nichols [l]. Processes with integrators or resonant poles are not considered in this
study.
An extensive simulation study has indicated that a
simple means of developing heuristic rules for assessing
the performance of a Ziegler-Nichols tuned PID controller is to characterise the process dynamics by the normalised process gain or the normalised dead-time [13]. This
paper will go a step further by using these two parameters to refine the Ziegler-Nichols tuning formula.
The idea of using normalised dead-time to improve
controller tuning has been around for a long time. One of
the early proposals was made by Cohen and Coon [14].
The earlier tuning formulas based on normalised deadtime are however not very good. A reason is probably
that they were developed using process models that were
too simple. Furthermore, the formula tends to produce
very oscillatory set-point response because it was derived
to give quarter damping for load-disturbance response.

Accuracy of Ziegler-Nichols formula

The Ziegler-Nichols tuning formula [l, 23 is based on

the empirical knowledge of the ultimate gain k, and ultimate period t u ,as shown in Table 1.
Table 1 : Ziealer-Nichols tuning formula
~~

PI D
Proportional gain k , = 0.6k,
Integral time
T, = 0.5fU
Td= 0.125t,
Derivative time

PI
k, = 0.45kU

T, = 0.85t,

The PID controller is usually implemented as follows:

Yf = 1

+ s&/N

(3)
111

The controller output, process output and set-point are

U,, y and y , , respectively.
A PID controller in the form of eqn. 1 avoids 'derivative kick' and provides noise filtering [SI. The noise filtering constant N is usually in the range of 3-10. Without
loss of generality, N = 10 is used throughout this study.
For simplicity, the continuous-time formulation will be
used, as the discretisation for digital implementations is
straightforward [2].
To illustrate the accuracy of the Ziegler-Nichols
tuning formula and to explore appropriate solutions, the
performance of the following process, tuned using the
equations in Table 1, is first analysed:
Process I

Y(s)
e-eds
-U,@)- (1 s y
~

(4)

2. I Ziegler-Nichols tuned PID control

In general, the performance of the Ziegler-Nichols tuned
control system varies signficantly with the process deadtime. For PID control, when the dead-time is small, both
the set-point and load-disturbance responses are well
tuned in terms of speed of response and damping, as
shown in Fig. 1 for e d = 0.4. The first overshoot in the
set-point response is, however, excessive, and it often
needs to be reduced to 10% or 20%, depending on applications. The load disturbance seen in Fig. 1 acts on the
process input. This assumption is made throughout the
paper.

to /? = 0 in eqn. 5), because the responses may then

become very sluggish. The introduction of /? provides a
mean of adjusting the zeros of the closed-loop transfer
function, which in turn affects the overshoot [17]. An
example is given in Fig. 1.
It is found from extensive simulation studies of the
process in eqn. 4 and other common process models that
set-point weighting is superior to set-point filtering
because the speed of response is sacrificed to a far smaller
degree for the same reduction in overshoot. The examples
in Fig. 2a and b use the same process model studied in
Fig. 1 to show the effect of set-point weighting. In
summary, there is no need to modify the Ziegler-Nichols
tuning formula for a process with small dead-time,
because the associated large overshoot can be overcome
by set-point weighting.

O5I
01
0

O5I

I/

10

15

20

15

20

ii

10

Excessive overshoot in the set-point response for PID control

Fig. 1
8, = 0.4
ZN = Ziegler-Nichols tuning: k, = 3.43; 7; = 1.44; Td = 0.36
RZN = refined Ziegler-Nichols tuning: k, = 3.43; 7; = 1.44; Td = 0.36;/?= 0.45

The problem of excessive set-point response overshoot

can be solved in several ways. A simple solution is to
detune the gain. This will, however, decrease the speed of
both the set-point and load-disturbance responses. An
alternative is to filter the set-point. The time constant tf
of the filter is usually set to a fraction of the integral time,
to achieve a reasonable compromise between overshoot
and rise time. The merit of this solution is that the load
disturbance response will not be affected, as the PID controller settings are not changed.
In a study on dominant pole designs [lS], a new
method of reducing overshoot by means of set-point
weighting has been proposed. In this case, the control law
(1) is modified to include a constant weighting factor B on
the set-point in the proportional term. The control law
(1) is thus modified to

dt
Set-point weighting has the same merit as set-point filtering, since the load response is also not affected. It is also
better than the extreme solution of totally removing the
set-point from the proportional term [161 (corresponding
112

Fig. 2
= 0.4

Comparison of methods for reducing overshoot

e,

(i) p = 0.45
(ii) Set-point filtering, t , = 1 . 1
b (i) p = 0.45
(ii) Detuning gain, kc = 0.9
U

The overshoot in the set-point response reduces as the

dead-time increases, because the loop gain reduces. It has
been found, in practice and in simulations, that a new
problem of excessive undershoot would occur in the later
part of the transient response when the dead-time is
large. In an application where very high performance is
required, such as a quality control loop, dead-time compensation using a Smith predictor [2] or other advanced
control technique would be necessary.
In less demanding applications, or when the dynamic
model of the process is poorly known, PID control may
still be used. Therefore it is useful to resolve the problem
of excessive undershoot. One possible solution is to apply
set-point weighting with /? > 1. However, this is found to
be effective only if the undershoot is moderate. When the
dead-time is large and undershoot serious, the load disturbance response is also poor. An example is shown in
Fig. 3 for Od = 2.5. Fine-tuning experience has indicated
that the problem can be overcome by increasing the integral action, that is, by reducing the integral time. This
results in improvements in both the set-point and load
disturbance responses as shown in Fig. 3. The trade-off
then may be a larger overshoot that can be compensated
IEE PROCEEDINGS-D, Vol. 138, N o . 2, MARCH 1991

as shown by the use of set-point weighting. The automatic setting of the set-point weighting and the modification of the tuning formula will be studied in Section 4.

01

20

40

60

Excessive undershoot in the set-point response, and poor load

Fig. 3
disturbance response for PID control
or = 2.5
(i) Ziegler-Nichols tuning: k, = 0.93; = 4.25; = 1.1
(ii) Improvement using 7; = 2.92
(iii) 7; = 2.92;j3 = 0.79

2.2 Ziegler-Nichols tuned PI control

The Ziegler-Nichols tuned PI controller for a process
with small dead-time produces a set-point response with
both large overshoots and undershoots. This is unlike the
Ziegler-Nichols tuned PID controller for a process with
small dead-time, where the undershoot is acceptable and
only the first overshoot is excessive. For example, using
the same plant studied in Fig. 1 for PID control, the
Ziegler-Nichols tuned PI controller as shown in Fig. 4
produces a set-point response that is more oscillatory. As
set-point weighting would only affect the first overshoot
but not the damping, it cannot be used to solve the
underdamped response of the Ziegler-Nichols tuned PI
controller.
For large dead-time, the Ziegler-Nichols tuned PI
controller produces sluggish set-point and loaddisturbance responses. An example is given in Fig. 5.

1 5 r ZN

Fig. 4
Poor damping and excessive overshoot in the set-point response
for PI control
0, = 0.4
ZN = Ziegler-Nichols tuning: k, = 2.57; 7; = 2.45
RZN = refined Ziegler-Nichols tuning: k, = 0.89; 7; = 1.45

Integral action has to be increased to overcome this

problem.
It has been found necessary to revise completely the
Ziegler-Nichols tuning formula for PI control. The
purpose of the modification is to improve the damping of
the set-point response for a process with small dead-time,
and to speed up the set-point response as well as loaddisturbance response for a process with large dead-time.
The improved responses are also shown in Figs. 4 and 5.
In Fig. 4, the larger load-disturbance response can be
expected, as the system is now tuned to give a greater
stability margin.
3

Process characterisation

The recent interests in artificial intelligence [4, 61 and

expert control [lS] have motivated the study of heuristics used by expert human operators and designers. One
such heuristic is the broad characterisation of process
dynamics by means of normalised process gain and normalised dead-time [131, from which an order-of-magnitude prediction of the achievable controller performance
can be made. An extensive simulation study has shown
that the accuracy of the Ziegler-Nichols tuning formula,
as discussed in the previous section, is strongly correlated
with the normalised gain and normalised dead-time of
the process.
The normalised process gain K is defined as the
product of the ultimate gain k , and the process steadystate gain k , . Formally
K = kpk,
(6)
The steady-state gain of the process can be determined
easily in closed-loop by introducing a small set-point
change. It is then computed simply as the ratio of the
steady-state change in the process output to the steadystate change in the control variable. Hence, the normalised process gain is a parameter that is easily determined
once the ultimate gain is known.
The normalised dead-time 0 is defined as the ratio of
the dead-time e,, or apparent dead-time 0, to the apparent time constant T, of the open-loop step response of the
process. Formally,

It can be simply determined using a pretuning pulse test

[l, 41, as shown in Fig. 6a, which is well defined for a
stable and damped process. The same definition can also
be used to characterise a process with inverse dynamics
caused by nonminimum-phase zeros, as shown in Fig. 6b.
The physical pulse testing to determine 0 is not needed
for the relay-feedback autotuner, if the available measurements are further employed to estimate a low-order plus
dead-time model [19]. It can also be avoided for the

'
5
0511

0
0

0
50

100

IEE PROCEEDINGS-D, Vol. 138, NO. 2, M A R C H 1991

150

Fig. 5 Sluggish set-point and load-disturbance responses for PI

control
0, = 2.5
ZN = Ziegler-Nichols tuning: k, = 0.7; 7; = 7.22
RZN = refined Ziegler-Nichols tuning: k, = 0.48; 7; = 2.4

t
a

Fig. 6
0

Step responses

O$TD = ajk,,

a Stable and damped process

h Stable and nonminimum-phase process

113

correlation-based autotuner [ l l , 121, because the normalised dead-time can be readily computed from the
impulse-response estimates generated automatically by
the correlator.
Besides the process given by eqn. 4, two other processes that exhibit different dynamics, namely, a multiplelag process and a nonminimum-phase process, are also
used in the correlation exercise:
Y(S)
1
Process2 -U,(S) - (1 s)"
~

Y(s)
1 - as
Process3 -=
(9)
VAS) (1 + SI3
For convenience, the processes of eqns. 4, 8 and 9 will be
referred to as process 1, 2 and 3, respectively. The normalised process gain K and normalised dead-time @ will
be varied in process 1 by changing the dead-time e,, in
process 2, the order n, and in process 3, the numerator
coefficient a.
~

-1

The setpoint overshoot and undershoot are correlated

to the normalised process gain and the normalised deadtime, in Fig. 7, for PID control, and in Fig. 8, for PI
control. The correlation results provide a fairly good
quantitative measure of the adequacy of the ZieglerNichols tuning formula. For PID control, when the normalised process gain is large, in the range of K > 4, or
equivalently when the normalised dead-time is small, in
the range of 0 < 0.3, the overshoot is about four times
the undershoot. This is better than the well known
quarter damping ratio [l]. The overshoot, however, is
clearly excessive. In the next range of 2 < K < 4 or
0.3 < 0 < 0.6, the damping ratio reduces. For K < 2 or
0 > 0.6, the undershoot increases and exceeds the overshoot.
For PI control, Fig. 8 shows that when K > 4 or
0 < 0.3, the damping is clearly poor, as the magnitude of
the undershoot approaches that of the overshoot. Furthermore, at very large K or very small @, the overshoot
and undershoot even exceed loo%, which means that the

0-

-2 0

-l O ! J L - - - -20

io

15

7Bo[
0

-20 I
0

02

04

06

OB

12

Fig. 7 Percentage overshoot and undershoot for PID control

a Against K
b Against 8
% undershoot
% overshoot

~~~~

x x x process

114

I;

+ + + process 2; 000process 3

Fig. 8 Percentage overshoot and undershoot for PI control

LI Against K
b Against 8
% overshoot
% undershoot
x x x process 1; + + + process 2; 000 process 3

~~~~

IEE PROCEEDINGS-D, Vol. 138, NO. 2, MARCH 1991

system is unstable. In the next range of 2 < K < 4 or

0.3 < 0 < 0.6, the undershoot exceeds the overshoot, and
the load-disturbance response becomes sluggish. Finally,
for K < 2 or 0 > 0.6, the overshoot becomes negative,
which means that the response is overdamped and very
sluggish. An example of this type of response is shown in
Fig. 5. It is thus evident that, for PI control, the existing
Ziegler-Nichols tuning formula has to be completely
revised.
The normalised process gain K and the normalised
dead-time 0 are closely related. Fig. 9 shows the relation
between K and 0 for the three processes. It can be
approximated by the following equation:
110

.=2(

4.1 Large normalised process gain or small

normalised dead-time (2.25 < K < 15;
0.16 < 0 < 0.57)

In this section, we will establish a simple correlation

formula for the set-point weighting factor 8, in terms of
the normalised process gain K , for 2.25 < K < 15, or
0.16 < 0 < 0.57, to obtain an acceptable overshoot in
the set-point response. Depending on applications, a 10%
or 20% overshoot may be a suitable criterion. An extensive simulation study has been carried out using processes 1,2 and 3. The results are shown in Fig. 10.
lr

0.91

+ 13

370-4)

Fig. 10 Set-point Weightingfactors for PID control

x x x process 1; + + + process 2; 000 process 3

- - -- - _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
........................................

The following empirical formulas

0
~

.......
~

05

Relation between
_ process
_
1~

Fig. 9

15

and 0

process 2
process 3
approximation

In this paper, we shall correlate to give the refined

Ziegler-Nichols formula in terms of the normalised
process gain IC. The formula also can be expressed in
terms of 0,by using eqn. 10 to substitute for K . Alternatively, in an earlier study [20], we have correlated and
have given the refined Ziegler-Nichols directly in terms
of 0.

Refined Ziegler-Nichols tuning for PID control

The fact that the normalised process gain and the normalised dead-time correlate well with the set-point overshoot and undershoot produced by Ziegler-Nichols
tuning immediately suggests that these two parameters
can also be used to refine the tuning formula. Based on
the above empirical observations, the following heuristic
criterion for refining the P I D tuning formula is recommended. When K > 2.25 or 0 < 0.57 (using eqn. 10 for
conversion), retain the Ziegler-Nichols tuning formula
and apply set-point weighting in the proportional term,
as shown in eqn. 5, to reduce the excessive overshoot.
When K < 2.25 or 0 > 0.57, modify the Ziegler-Nichols
tuning formula by a suitable reduction of the integral
time to improve both the set-point and load disturbance
responses.
IEE PROCEEDINGS-D, Vol. 138, N O . 2, MARCH 1991

15

- K

b=- 15 + K 2.25 < K < 15

and

approximate the simulation results for

6 and 20%
overshoot, respectively. The graphs of /? based on the
above formulas are shown as solid lines in Fig. 10.
Response R Z N in Fig. 1 is an example of the improvement in performance that is obtained using this formula.
Notice that when K is large, or 0 small, the application of set-point weighting reduces not only the overshoot but also the undershoot, hence achieving faster
settling time. The effectiveness of set-point weighting may
reach its limit when K > 15 or 0 < 0.16. This is acceptable because these ranges of K and 0 signify a low-order
process, where other design methods may be more appropriate than heuristics. Otherwise, it is straightforward to
use a combination of set-point weighting and set-point
filtering to satisfy a given overshoot criterion.
4.2 Small normalised process gain or large
normalised dead-time ( 1.5 < K < 2.25;
0.57 < 0 < 0.96)

It has been established, in Section 2, that, for large deadtime, the Ziegler-Nichols tuning formula needs to be
modified to reduce the undershoot in the set-point
response and to improve the load-disturbance response.
As in set-point weighting, an extensive simulation study
of processes 1, 2 and 3 is used to obtain the appropriate
modifications.
115

It is found that a suitable integral time can indeed be

determined without changing the proportional gain and
derivative time. However, the improvement is accompanied by an excessive overshoot, and hence the application of set-point weighting with /3 < 1 is required. The
new integral time is given as

the above formulas has already been shown in Fig. 3.

Notice that, especially at small normalised process gain,
the results are not very accurate for process 3. However,
the computed settings still give good responses. This is
demonstrated by the extreme case example shown in Fig.
15a.

(13)
where p is defined as the ratio of the modified integral
time to the Ziegler-Nichols integral time.
As the control performance in this region of K and 0
cannot be expected to be very tight, the criterion for
adjusting the integral time and set-point weighting is set
at set-point response of 20% overshoot and 10% undershoot. The empirical results for all three processes are
shown in Fig. 11. The following formulas can be used to
approximate the simulation results :
= OSpt,

p = 4.

1.5 < K < 2.25

+ 1)

/3 = &($K

1.5 < K < 2.25

The graphs of p and 8, based on the formulas, are plotted

in Fig. 11 as solid lines. An example of the accuracy of
k"

0 45

0 30 54 t

03t

O6I

0 55

:\

0051

051

16

18

22

, k

O0

2L

10

15k

10

15

a
0

'r

0
0

0 751
14

16

18

, k
22

01
0

24

Fig. 11

Optimal values for PID control

a Values ofp
b Values ofg
x x x

116

~k

process I ,

+ + + process 2; 000 process 3

Fig. 12

Refined Ziegler-Nichols tuning for PI control

a kJk, against K
b TJt,against K
x x x

process 1;

+ + + process 2; 000process 3
IEE PROCEEDINGS-D, Vol. 138, No. 2, M A R C H 1991

k,/k, and TJt, , respectively. The following formulas

k, - _5
_
-

formula was derived to give quarter damping for the load

disturbance response for a first order plus dead-time
process.

12+ K
1.2<~<15
6 15+ 1 4 ~

k,

cc

correspond to the solid curves in the figures. Examples of

the excellent performance that can be expected of the formulas have already been given in Figs. 4 and 5.
6

Comparative studies

In this section, the performance of the refined ZieglerNichols formula will be compared with the performance
of the Ziegler-Nichols and the Cohen-Coon formulas
[14]. A popular model in process control, the first order
plus dead-time model

10

15

where 8 = 0.2, 2.0 will be used in the comparative studies.

To extend to other process dynamics, the performance of
the tuning formulas on a nonminimum phase process
Y(s)
1 - us
U,(s) - (1 $3
where a = 1.4, will also be compared.
The result of the comparison for the first order plus
dead-time model with PID control is shown in Fig. 13.
The superiority of the refined formula is obvious. The
result of the comparison for the first order plus dead-time
model with PI control is shown in Fig. 14. The larger
load-disturbance response produced by the refined
formula for the process with Od = 0.2 is to be expected, as
the system is now tuned to give a greater stability
margin. It is also not surprising that the Cohen-Coon
formula produces very oscillatory set-point response. The

10

20

30

40

Fig. 14 Comparison of tuning formulas for PI control

a ed = 0.2
RZN = refined Ziegler-Nichols: k, = 1.08; 7; = 0.48
ZN = Ziegler-Nichols: k, = 3.83; 7; = 0.63
CC = Cohen-Coon: k, = 4.58; 7; = 0.47
b 8,=2
RZN = refined Ziegler-Nichols: k, = 0.47; 7; = 1.55
ZN = Ziegler-Nichols: k, = 0.68; 7; = 4.68
CC = Cohen-Coon: k, = 0.52; 7; = 1.47

I lccc

01 I
0

10

30

20

10

40

a
n .RZN

01

10

20

30

40

b
Fig. 13 Comparison oftuning formulas for PID control
a 0, = 0.2
RZN = refined Ziegler-Nichols: k, = 5.1; T, = 0.37; = 0.09;j3 = 0.28
ZN = Ziegler-Nichols: k, = 5.1; P i = 0.37; 5 = 0.09
CC = Cohen-Coon: k, = 6.92; 7; = 0.45; 5 = 0.07
b 8,=2
RZN = refined Ziegler-Nichols: k, = 0.91; 7; = 1.86; 5 = 0.68; j3 = 0.79
ZN = Ziegler-Nichols: k, = 0.91; 7; = 2.75; 5 = 0.68
CC = Cohen-Coon: k, = 0.91; 7; = 3.03; Td= 0.53

I E E PROCEEDINGS-D, Vol. 138, No. 2, MARCH 1991

20

40

60

Comparison of tuning formulas, where a = 1.4

Fig. 15
a PIDcontrol
RZN = refined Ziegler-Nichols: k, = 0.92; 7; = 2.33; = 0.85;j3 = 0.79
ZN = Ziegler-Nichols: k, = 0.92; 7; = 3.41; 5 = 0.85
CC = Cohen-Coon: k, = 1.24; 7; = 4.09; 5 = 0.71
b PI control
RZN = refined Ziegler-Nichols: k, = 0.48; 7; = 1.93
ZN = Ziegler-Nichols: k, = 0.69; T, = 5.81
CC = Cohen-Coon: k, = 0.75; 7; = 2.3
117

The comparisons show that the refined ZieglerNichols formula works well for the first order plus deadtime process, even though it was derived using
higher-order process models. The performance of the
refined Ziegler-Nichols formula for the second order
model

commonly used to approximate industrial processes, can

be expected to be sandwiched in between process 1 and
the first order plus dead-time process.
The result of the comparison for the nonminimum
phase process is given in Fig. 15a for PID control and in
Fig. 1% for PI control. The dead-time and time constant
parameters for the Cohen-Coon formula are obtained by
using least square to fit the step response of a first order
plus dead-time model to that of the nonminimum phase
process. The response of Cohen-Coon tuning is highly
oscillatory, as the first order plus dead-time model does
not model the nonminimum phase process adequately.
The good performance of the refined formula is again
obvious.
7

Conclusions

From the normalised gain or normalised dead-time of the

process, an accurate prediction of the performance, and
hence refinements, of the Ziegler-Nichols tuning formula
can be made. With the proposed refinements of the
Ziegler-Nichols tuning formula, PID and PI autotuning
will be accurate over a much larger range of process
dynamics. Consequently, the need for manual fine-tuning
and human expertise will be largely eliminated.
The refinements have several merits. Firstly, they
require only a small set-point change in closed-loop to
determine the normalised process gain, or a simple pulse
test for the measurement of the normalised dead-time.
The pulse test can even be avoided for the relay-feedback
autotuner and the correlation-based autotuner, because
the normalised dead-time can be calculated from available measurements. Secondly, the computations of the
tuning factors are also very simple. Finally, through setpoint weighting, the PID controller is tuned to give both
good set-point response as well as load-disturbance
response.

118

References

6
7

ZIEGLER, J.G., and NICHOLS, N.B.: Optimum settings for automatic controllers, Trans. ASME, 1942,65, pp. 4 3 3 4 4 4
DESHPANDE, P.B., and ASH, R.H.: Computer process control
(ISA Pub., USA, 1981)
HOOPES, H.S., HAWK, W.K., and LEWIS, R.C.: A self-tuning
controller, I S A Transactions, 1983, 22, pp. 49-58
KRAUS, T.W., and MAYRON, T.J.: Self-tuning PID controllers
based on a pattern recognition approach, Control Engineering,
1984, pp. 10&111
ASTRoM, K.J., and HWGGLUND, T.: Automatic tuning of
simple regulators with specifications on phase and amplitude
margins. Automatica. 1984.20. DD. 6 4 5 6 5 1
HICHAM, E.H.: A self-tuning controller based on expert systems
and artificial intelligence. Proc. Control 85, UK, 1985, pp. 1 1 0 - 1 15
HESS, P., RADKE, F., and SCHUMANN, R.: Industrial applications of a PID self-tuner used for system start-up. Proc. IFAC
World Congress, Munich, FRG, 1987,3, pp. 21-26
RADKE, F., and ISERMANN, R.: A parameter-adaptive PID controller with stepwise parameter optimization, Automatica, 1987, 23,
pp. 4 4 9 4 5 7
ASTRoM, K.J. : Ziegler-Nichols auto-tuners. Report TFRT-3167,
Dept. of Automatic Control, Lund Inst. of Tech., Lund, Sweden,
1982

IO HANG, C.C., LEE, T.H., and TAY, T.T.: The use of recursive
parameter estimation as an auto-tuning aid. Proc. ISA Annual
Conf., USA, 1984, pp. 387-396
1 1 HANG, C.C., LIM, C.C., and SOON, S.H.: A new PID auto-tuner
design based on correlation technique. Proc. 2nd Multinational
Instrumentation Conf., China, 1986
12 HANG, C.C., and SIN, K.K.: On-line auto-tuning of PID controllers based on cross correlation. Proc. International Conference on
Industrial Electronics, Singapore, 1988, pp. 4 4 4 6
13 ASTRoM, K.J., HANG, C.C., and PERSSON, P.: Toward intelligent PID control. Proc. IFAC Workshop on AI in real-time
control, 1989, pp. 3 8 4 3
14 COHEN, G.H., and COON, G.A.: Theoretical consideration of
retarded control. Trans. ASME 75,1953, pp. 827-834
15 HWGGLUND, T., and ASTRoM, K.J.: Automatic tuning of PID
controllers based on dominant pole design. Proc. IFAC Workshop
on Adaptive Control of Chemical Processes, Frankfurt, FRG, 1985
16 GREY, J.P.: A comparison of PID control algorithms, Control
Engineering, March 1987, pp. 102-105
17 FRANKLIN, G.F., and POWELL, J.D.: Digital control of
dynamic systems (Addison-Wesley, USA, 1980)
18 ASTRoM, K.J., ANTON, J.J., and ARZEN, K.E.: Expert control,
Automatica, 1986,22, pp. 277-286
19 ASTRoM, K.J., and HWGGLUND, T.: A new auto-tuning design.
Proc. IFAC Symp. on Adaptive Control of Chemical Processes,
Lyngby, Denmark, 1988
20 HANG, C.C., and ASTRoM, K.J.: Refinements of the ZieglerNichols tuning formula for PID auto-tuners. Proc. ISA Conf., USA,
1988, pp. 1021-1030

IEE PROCEEDINGS-D, Vol. 138, N O . 2, MARCH 1991