1638 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 48, NO.

9, SEPTEMBER 2003

[5] O. M. E. El-Ghezawi, A. S. I. Zinober, and S. A. Billings, “Analysis and with time delay. Traditionally, this model is obtained by applying a step
design of variable structure systems using a geometric approach,” Int. J. input to the plant and measuring at the output the following three pa-
Control, vol. 38, pp. 657–671, 1983. rameters: the steady-state gain, the time constant, and the time delay.
[6] M. Zohdy, M. S. Fadali, and J. Liu, “Variable structure control using
system decomposition,” IEEE Trans. Automat. Contr., vol. 37, pp. Although many of these tuning techniques work in practice, not much
1514–1517, Oct. 1992. is known about the robustness or stability of these algorithms beyond
[7] J.-J. E. Slotine and W. Li, Applied Nonlinear Control. Upper Saddle what has been observed in empirical studies. Perhaps, the only excep-
River, NJ: Prentice-Hall, 1991. tion is the internal model control (IMC) algorithm where the stability
[8] W.-C. Su, S. V. Drakunov, and U. Ozguner, “Constructing discontinuity
surfaces for variable structure systems: A Lyapunov approach,” Auto-
constraint is built into the PID design method. Recent results on PID
matica, vol. 32, pp. 925–928, 1996. stabilization obtained in [6], however, make it possible to revisit these
[9] H. H. Choi, “A new method for variable structure control system de- classical tuning rules and to justify them in terms of stability and ro-
sign: A linear matrix inequality approach,” Automatica, vol. 33, pp. bustness. The main objective of this note is to do precisely that.
2089–2092, 1997. In this note, we will analyze several PID tuning techniques that are
[10] , “Variable structure control of dynamical systems with mismatched
norm-bounded uncertainties: An LMI approach,” Int. J. Control, vol. 74, based on first-order models with time delay. This analysis will attempt
pp. 1324–1334, 2001. to describe when each tuning technique is appropriate in the sense of
[11] , “Variable structure output feedback control design for a class of providing PID controller parameters that are robust in the space of the
uncertain dynamic systems,” Automatica, vol. 38, pp. 335–341, 2002.
[12] R. H. C. Takahashi and P. L. D. Peres, “ H guaranteed cost-switching
controller coefficients. A controller for which the closed-loop system is
destabilized by small perturbations in the controller coefficients is said
surface design for sliding modes with nonmatching disturbances,” IEEE
Trans. Automat. Contr., vol. 44, pp. 2214–2281, Nov. 1999. to be fragile [4]. Any controller that is to be practically implemented
[13] K.-S. Kim, Y. Park, and S.-H. Oh, “Designing robust sliding hyperplanes must necessarily be nonfragile or controller robust (terminology sug-
for parametric uncertain systems: A Riccati approach,” Automatica, vol. gested by W. M. Wonham) [7] so that: 1) round-off errors during im-
36, pp. 1041–1048, 2000.
plementation do not destabilize the closed-loop; and 2) tuning of the
[14] S. Boyd, L. El. Ghaoui, E. Feron, and V. Balakrishnan, Linear Matrix
Inequalities in System and Control Theory. Philadelphia, PA: SIAM, parameters about the nominal design values is allowed.
1994. Four tuning techniques will be discussed: 1) the classical
Ziegler–Nichols step response method; 2) the CHR method; (3) the
Cohen–Coon method; and 4) the IMC design technique. The analysis
starts by ascertaining if the proposed proportional gain value lies
inside the allowable range determined in [6]. We will then examine for
this fixed proportional gain, the location of the integral and derivative
On the Stability and Controller Robustness of Some gain values inside the stability region described in [6]. This procedure
Popular PID Tuning Rules will allow us to determine conditions under which each tuning
technique provides a good l2 parametric stability margin in the space
Guillermo J. Silva, Aniruddha Datta, and S. P. Bhattacharyya of the controller coefficients. In this way, we will avoid undesirable
scenarios such as PID controller parameters that are dangerously close
to instability.
Abstract—In this note, we study the stability and controller robustness The note is organized as follows. In Section II, we recall some recent
of some popular proportional-integral-derivative (PID) tuning techniques
that are based on first-order models with time delays. Using the character- results on PID stabilization of first-order plants with time-delay [6].
ization of all stabilizing PID controllers derived in a previous paper, each These results are used in Section III to analyze the Ziegler–Nichols
tuning rule is analyzed to first determine if the proportional gain value dic- step response method. Section IV summarizes the results of similar
tated by that rule, lies inside the range of admissible proportional gains. analyzes for the other three methods. Finally, Section V contains some
Then, the integral and derivative gain values are examined to determine
concluding remarks.
conditions under which the tuning rule exhibits robustness with respect to
controller parameter perturbations.
Index Terms—Controller robustness, proportional-integral-derivative II. PRELIMINARY RESULTS
(PID) controllers, stability, tuning rules. The tuning techniques analyzed in this note are based on character-
izing the plant to be controlled by the following transfer function

G(s) = 1 +kTs e0Ls

I. INTRODUCTION
(1)
Over the last forty years, numerous methods have been developed
for setting the parameters of a proportional-integral-derivative (PID)
controller [1]. Some of these methods are based on characterizing the where k is the steady-state gain, L is the apparent time delay, and T
dynamic response of the plant to be controlled with a first-order model is the apparent time constant. We will consider the feedback control
system shown in Fig. 1 where r is the command signal, y is the output
of the plant, G(s) given by (1) is the plant to be controlled, and C (s)
Manuscript received July 25, 2002; revised April 25, 2003. Recommended by is the controller. We focus on the case when the controller is of the PID
Associate Editor A. Bemporad. This work was supported in part by the National type, i.e., the controller has a proportional term, an integral term and a
Science Foundation under Grant ECS-9903488, in part by the Texas Advanced derivative term. There are different ways of representing the PID con-
Technology Program under Grant 000512-0099-1999, and in part by the Na-
trol algorithm [1]. In our case, we will use the following representation:
tional Cancer Institute under Grant CA 90301.

C (s) = kp + ksi + kd s
G. J. Silva is with the IBM Server Group, Austin, TX 78758, USA (e-mail:
[email protected]).
A. Datta and S. P. Bhattacharyya are with the Department of Electrical En-

where kp is the proportional gain, ki is the integral gain and kd is the

gineering, Texas A & M University, College Station, TX 77843-3128 USA
(e-mail: [email protected]; [email protected]).
Digital Object Identifier 10.1109/TAC.2003.817008 derivative gain.

0018-9286/03$17.00 © 2003 IEEE

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 48, NO. 9, SEPTEMBER 2003 1639

Fig. 1. Feedback control system.

The following theorem [6] provides an analytical characterization of

( )
the set of controller parameters kp ; ki ; kd for which the closed-loop
system in Fig. 1 is stable.
Theorem 2.1: The range of kp values for which a given open-loop
()
stable plant, with transfer function G s as in (1), continues to have
closed-loop stability with a PID controller in the loop is given by

0 k1 < kp < k1 TL 1 sin(1 ) 0 cos(1 ) (2)

where 1 is the solution of the equation Fig. 2. Stabilizing region of (k ; k ) for: (a) 0(1
=k ) < k < (1=k );
(b) k = (1=k ); and (c) (1=k ) < k < k .
tan() = 0 T +T L (3)

(0 )
in the interval ; . For kp values outside this range, there are no
stabilizing PID controllers. The complete stabilizing region is given by
(see Fig. 2) the following.
( (1 ) (1 ))
1) For each kp 2 0 =k ; =k , the cross section of the stabi-
( )
lizing region in the ki ; kd space is the trapezoid T.
= (1 )
2) For kp =k , the cross-section of the stabilizing region in the
( ) 1
ki ; kd space is the triangle .
((1 )
3) For each kp 2 := (1 )[( ) sin( )
=k ; kupp =k T=L 1 1 0
cos( )])
1 , the cross-section of the stabilizing region in the
( )
ki ; kd space is the quadrilateral Q.
=12
In Fig. 2, the parameters mj ; bj , and wj , for j ; , are defined as
follows:

mj = 1 L2
zj2
1 0 L sin(z ) + T z cos(z )
bj =
Fig. 3. Comparison of k given by the Ziegler-Nichols Method and the upper
j
L j j bound k .
kzj
wj =1 zj sin(z ) + T z (cos(z ) + 1) now define the parameter as the ratio of the apparent time delay to
j
L j j (4)
kL the apparent time constant of the plant, i.e.,
where z1 ; z2 ; z2 > z1 are the solutions of
= TL :
kkp + cos(z ) 0 T z sin(z ) = 0
L First, we consider the proportional gain value given in (5) and rewrite
it as a function of
in the interval (0; ).
kp = 1:2 : (6)
III. ZIEGLER–NICHOLS STEP RESPONSE METHOD
k
A simple way to determine the parameters of a PID controller based
0 0
Since k > and > (the plant is open-loop stable), then kp > . 0
From Theorem 2.1, we can rewrite the upper bound on kp as a function
on step response data was developed by Ziegler and Nichols in 1942
of the parameter
[8]. This method first characterizes the plant by the parameters L and
a, where the parameter a is defined as kupp = 1 1 1 sin(1 ) 0 cos(1 ) (7)
k
a = kL: where 1 is now the solution of the equation
T
Once these parameters are determined, the PID controller parameters tan() = 0 1 +1
are then given in terms of L and a by the following formulas:
(0 )
in the interval ; . We now compare kp and kupp by plotting kp k
kp = 1:2 ki = 0:6 kd = 0:6L : (5) and kupp k as functions of the parameter . As can be seen from Fig. 3,
a aL a the proportional gain value given by the Ziegler-Nichols step response
This tuning rule was developed by empirical simulations of many dif- method is always less than the upper bound kupp . Thus, this tuning
ferent systems and is only applicable to open-loop stable plants. We technique always provides a feasible proportional gain value kp . We
1640 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 48, NO. 9, SEPTEMBER 2003

Fig. 4. Location of the parameters (k ; k ) when 1:2. Fig. 5. Comparison of 0:2(kL =T )x ; 0:8(kL =T )x , and (kL =T )k for
1:2.

now set kp = kp and consider two cases, requiring different treatment

according to the results presented in Section II. Moreover, for clarity
of presentation, let us rewrite the parameters ki and kd in (5) as

0:6T
ki = (8)
kL2
0:6T
kd = : (9)
k
Case 1: 1:2. In this case, we have 0 < kp (1=k). Then
the stabilizing set is given either by Fig. 2(a) or by Fig. 2(b). Notice
from (9) that the parameter kd is always less than (T=k) as illustrated
in Fig. 4. The derivative gain value provided by the Ziegler–Nichols
method is robust in the sense that it is not close to the stability boundary
(T=k ). Following the same principle, we would like to guarantee that
the integral gain value is also far away from the stability boundary. Let
x1 be the ki -coordinate of the point where the line kd = kd intersects Fig. 6. Location of the parameters (k ; k ) when 0 < < 1:2.
the line kd = m1 ki + b1 . From Fig. 4, we now find the conditions under
which the parameter ki lies in the range (0:2x1 ; 0:8x1 ). Following the
same derivation used in [6], x1 can be expressed as follows where z2 > z1 > 0 is the solution of

T 1:2 + cos(z ) 0 z sin(z) = 0

x1 = z [ sin(z1 ) + z1 (cos(z1 ) + 0:6)] (10)
kL2 1 in the interval (0; ). By sweeping in the range (0; 1:2), it can be
where z1 is the solution of shown that kd < b2 . Fig. 6 shows the location of kd with respect to the
stabilizing set in the space of (ki ; kd ).
T
kkp + cos(z ) 0 z sin(z ) = 0
As in the previous case, we will now analyze for which values of
L , the parameter ki lies inside the range (0:2x1 ; 0:8x1 ). As in Case 1,
, 1:2 + cos(z) 0 z sin(z) = 0 [using (6)] we can plot the terms 0:2(kL2 =T )x1 ; 0:8(kL2 =T )x1 , and (kL2 =T )ki
versus . This graph is shown in Fig. 7 for 0 < < 1:2. From this
in the interval (0; ). From (8) and (10), we can plot the terms graph we see that ki lies in the range (0:2x1 ; 0:8x1 ) for 0 < < 1:07.
2 2
(kL =T )x1 , and (kL =T )ki versus . This graph is shown in Fig. 5 For the relaxed condition where ki lies in the range (0:1x1 ; 0:8x1 ), we
for 1:2. As can be seen from this graph, ki does not lie in the have 0 < < 1:2.
range (0:2x1 ; 0:8x1 ) for any value of . If we relax our robustness From the previous analysis, we conclude that the Ziegler–Nichols
condition and now make ki lie inside the range (0:1x1 ; 0:8x1 ), we step response method gives a controller-robust PID controller for 0 <
see from Fig. 5 that this occurs for 1:2 < 3. In this way, for < 1:07. Controller robustness is here understood as good parametric
1:2 < 3; ki will be located 10% of x1 away from the kd -axis stability margin in the space of (ki ; kd ).
which corresponds to a good l2 parametric stability margin. Remark 3.1: It has been determined empirically [1] that the
Case 2: 0 < < 1:2. In this case, we have (1=k) < kp < kupp . Ziegler–Nichols rule is applicable if 0:1 < < 0:6. In this range, the
The stabilizing set is given by Fig. 2(c). We now show that the param- derivative action often gives significant improvement of performance.
eter kd is less than b2 for all < 1:2. From (4), b2 can be rewritten as Comparing this range with the one previously obtained for controller
follows: robustness, we see that the former is included in the latter. Thus, for
0:1 < < 0:6, the Ziegler–Nichols step response method not only
T sin(z2 )
b2 = 0
gives good performance but also is robust with respect to controller
+ cos(z2 )
k z2 parameter perturbations.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 48, NO. 9, SEPTEMBER 2003 1641

REFERENCES
[1] K. Astrom and T. Hagglund, PID Controllers: Theory, Design, and
Tuning. Research Triangle Park, NC: Instrum. Soc. Amer., 1995.
[2] K. L. Chien, J. A. Hrones, and J. B. Reswick, “On the automatic control
of generalized passive systems,” Trans. Amer. Soc. Mech. Eng., vol. 74,
pp. 175–185, 1952.
[3] G. H. Cohen and G. A. Coon, “Theoretical consideration of retarded
control,” Trans. Amer. Soc. Mech. Eng., vol. 76, pp. 827–834, 1953.
[4] L. H. Keel and S. P. Bhattacharyya, “Robust, fragile or optimal?,” IEEE
Trans. Automat. Contr., vol. 42, pp. 1098–1105, Aug. 1997.
[5] M. Morari and E. Zafiriou, Robust Process Control. Upper Saddle
River, NJ: Prentice-Hall, 1989.
[6] G. J. Silva, A. Datta, and S. P. Bhattacharyya, “New results on the syn-
thesis of PID controllers,” IEEE Trans. Automat. Contr., vol. 47, pp.
241–252, Feb. 2002.
[7] W. M. Wonham, private communication, 2001.
[8] J. G. Ziegler and N. B. Nichols, “Optimum settings for automatic con-
trollers,” Trans. Amer. Soc. Mech. Eng., vol. 64, pp. 759–768, 1942.

Fig. 7. Comparison of 0:2(kL =T )x ; 0:8(kL =T )x , and (kL =T )k for

0 < < 1:2.

IV. OTHER TUNING TECHNIQUES

The analysis presented in the previous section can be applied to other Discretization Behaviors of Equivalent Control Based
PID tuning techniques that are based on first-order models with time Sliding-Mode Control Systems
delays. The main criterion is to ensure first that the controller param-
eters kp and kd are inside the stabilizing set of gain values. Then, the Xinghuo Yu and Guanrong Chen
parameter ki is forced to lie inside an interval located 20% of x1 away
from the boundaries of the stabilizing set in the (ki ; kd ) space. Here,
x1 represents the maximum stabilizing integral gain value for the fixed Abstract—In this note, discretization behaviors of the equivalent con-
trol based sliding-mode control (SMC) systems are studied. Some inherent
proportional and derivative gains provided by the particular tuning rule.
dynamical properties of the discretized second-order systems are first ex-
As a result of this criterion, the range of (L=T ) values that ensures con- plored. Upper bounds for the system steady states are established. The
troller robustness can be determined for each tuning technique. These system’s steady-state behaviors are discussed. The analysis for the second-
values are summarized as follows: order systems is then extended to higher order systems. Simulations are
L presented to verify the theoretical results.
0
Ziegler Nichols Step Response Method : 0
T
<
< 1:07
Index Terms—Discretization, dynamical behavior, equivalent control,
L
Cohen0Coon Method [3] : 0 <
sliding-mode control.
< 8:53
T
L
CHR Method [2] : 0:37 < I. INTRODUCTION
T
L Discrete sliding-mode control (DSMC) has been extensively
IMC Design Technique (for =L = 0:25) [5] : 0:37 < :
T studied to address some basic problems associated with the SMC of
From this table, we conclude that the Cohen–Coon method gives discrete-time systems that have relatively low switching frequencies.
resilient PID parameters in the sense of the parametric stability margin Major research efforts in DSMC have been devoted to the develop-
when the plant under study satisfies the property 0 < (L=T ) < 8:53. ment of various controllers using specific guiding principles [1]–[9].
It is interesting to note that for both the CHR method and the IMC However, the study of discretizing a continuous-time SMC for digital
Design Technique the same resilience of the PID parameters is obtained implementation has not been fully explored.
if the ratio (L=T ) is greater than a lower bound, which is 0.37. In the In this note, we study the discretization behaviors of the most
case of the IMC Design Technique, the design variable > 0 should popular SMC systems—the equivalent control based SMC systems.
be selected properly in order to obtain a PID controller with a good Unlike most existing research on DSMC, our present interest is in
compromise between performance and robustness. It is commonly rec- the discretization effect on a continuous-time SMC system.
ommended [8] that =L should be fixed at 0.25, which was the value
used in the above table.

V. CONCLUDING REMARKS Manuscript received February 11, 2002. Recommended by Associate Editor
D. E. Miller. The work of X. Yu was supported by the Center for Chaos Control
In this note, we have presented an analysis of the robustness of some and Synchronization of City University of Hong Kong during his sabbatical
leave in 2001/2002. The work of G. Chen was supported by the Hong Kong
common PID tuning techniques in the space of the controller parame-
Research Grants Council under the CERG Grants CityU 1098/00E, 1018/01E,
ters. This analysis was motivated by the fact that a good PID controller and 1004/02E.
design should exhibit robustness with respect to small perturbations X. Yu is with the School of Electrical and Computer Engineering, Royal
in the controller coefficients. Since the results of [6] yield a complete Melbourne Institute of Technology, Melbourne, VIC 3001, Australia (e-mail:
characterization of all stabilizing PID controllers for a particular class [email protected]).
G. Chen is with the Department of Electronic Engineering, City Uni-
of plants, it is clear that in principle a similar robustness analysis with versity of Hong Kong, Kowloon, Hong Kong SAR, P. R. China (e-mail:
respect to plant parameter perturbations is also possible. The details, [email protected]).
however, remain to be worked out. Digital Object Identifier 10.1109/TAC.2003.816970

0018-9286/03$17.00 © 2003 IEEE