ISA

TRANSACTIONS®
ISA Transactions 43 共2004兲 297–304

Tuning PI controllers for stable processes with specifications

on gain and phase margins
Ibrahim Kaya*
Inonu University, Engineering Fac., Dept. of Electrical & Electronics Eng., 44069 Malatya, Turkey
共Received 28 February 2003; accepted 17 August 2003兲

Abstract
In industrial practice, controller designs are performed based on an approximate model of the actual process. It is
essential to design a control system which will exhibit a robust performance because the physical systems can vary with
operating conditions and time. Gain and phase margins are well known parameters for evaluating the robustness of a
control system. This paper presents a tuning algorithm to design and tune PI controllers for stable processes with a
small dead time while meeting specified gain and phase margins. Simulation examples are given to demonstrate that the
proposed design method can result, in a closed-loop system, in better performances than existing design methods which
are also based on user-specified gain and phase margins. © 2004 ISA—The Instrumentation, Systems, and Automa-
tion Society.

Keywords: IMC design; PI controller; Gain margin; Phase margin; Time delay

1. Introduction ally, vary with operating conditions and time.

Hence robustness of a control system has always
Proportional-integral-derivative 共PID兲 control- been an important issue. Gain and phase margins
lers are widely used in industrial systems despite are two well-known measures for maintaining the
the significant developments of recent years in robustness of the control system. Recently, there
control theory and technology. This is because has been a renewed interest in designing a control
they perform well for a wide class of processes. system to satisfy the specified gain and phase mar-
PID controllers give robust performance over a gins 关1– 4兴.
wide range of operating conditions. Furthermore, This paper presents a controller design, with
they are easy to implement using analog or digital specifications on gain and phase margins, for con-
hardware and familiar to engineers. trolling stable processes with small time delays.
In the practice, the model used to analyze or To accomplish this, the classical single input
design control systems is only an approximation single output 共SISO兲 feedback control system is
of the actual plant transfer function. The most represented as its equivalent internal model con-
common models used for stable plant transfer trol 共IMC兲 关5,6兴. This representation provides the
functions are a first-order plus dead time 共FOPDT兲 parameters of PID-type controllers used in the
or second-order plus dead time 共SOPDT兲 model. SISO system to be defined in terms of the desired
Also, the parameters of the physical systems, usu- closed-loop time constant, which can be adjusted
by the operator, and the parameters of the process
*Tel: ⫹90 422 3410010, ext. 4499; fax: ⫹90 422 model. This means that only one parameter,
3410046. E-mail address: [email protected] namely the desired closed-loop time constant, is

0019-0578/2004/$ - see front matter © 2004 ISA—The Instrumentation, Systems, and Automation Society.
298 Ibrahim Kaya / ISA Transactions 43 (2004) 297–304

left for tuning, assuming that the model param-

eters have been obtained from a relay autotuning
关7,8兴. The details of the identification method are
not given here and interested readers can refer to
the cited references. However, for the sake of easi-
ness, equations used for parameter estimation of
the FOPDT plant transfer function are given in the
Appendix. The value of the closed-loop time con-
stant is obtained from the specified gain and phase
margins. The proposed design method is compared Fig. 1. IMC control strategy.
with existing design methods based on the speci-
fied gain and phase margins and it is shown by This control structure is referred to as internal
examples that the proposed design method gives model control 共IMC兲 since the plant model Ĝ ( s )
better closed-loop performance. appears in the control structure. Here, G ( s ) and
The next section gives a brief review of the IMC Ĝ ( s ) are the actual process and process model
design, since the tuning rules to tune/design PI transfer functions, respectively. When G ( s )
controllers are derived using IMC principles. ⫽Ĝ ( s ) , that is perfect modeling, and d⫽0, the
Simple tuning rules for obtaining parameters of a system is basically open loop. This provides the
PI controller based on the user-specified gain and
open-loop advantages. When G ( s ) ⫽Ĝ ( s ) or d
phase margins are derived in Section 3. Section 4
⫽0 the system is a closed-loop system. Thus the
gives simulation examples which show that the
IMC control strategy has the advantages of both
design method given in the paper results in a bet-
the open-loop and closed-loop structures.
ter closed-loop system performance than existing
The first step in the IMC controller design is to
design methods which is designed with specifica-
factor the process model,
tions on gain and phase margins as well. Conclu-
sions are given in Section 5. Ĝ 共 s 兲 ⫽Ĝ ⫹ 共 s 兲 Ĝ ⫺ 共 s 兲 , 共1兲
where Ĝ ⫹ ( s ) contains all the time delays and
right-half plane zeros.
2. Internal model control „IMC… The second step is to define the IMC controller
as
A control system design is expected to provide a
fast and accurate set-point tracking, that is, the G imc 共 s 兲 ⫽Ĝ ⫺1
⫺ 共 s 兲F共 s 兲, 共2兲
output of the system should follow the input signal
where F ( s ) is a low pass filter with a steady-state
as close as possible. In addition, any external dis-
gain of 1. The filter is introduced for physical re-
turbances must be corrected by the control system
alizability of the IMC controller G imc ( s ) . The
as efficiently as possible. The first requirement is
simplest filter has the following form 关5,6兴:
achieved by an open-loop control system. With an
open-loop control scheme, the stability of the sys- 1
tem is guaranteed provided that both the plant and F共 s 兲⫽ . 共3兲
controller transfer functions are stable. Also, the 共 ␭s⫹1 兲 n
design of the controller in an open-loop control From the block diagram of the IMC structure
scheme may simply be chosen as G c ( s ) shown in Fig. 1, the closed-loop transfer is given
⫽G ⫺1 ( s ) , where G c ( s ) and G ( s ) are, respec- by
tively, the controller and plant transfer functions.
The drawback of an open-loop control system is G 共 s 兲 G imc 共 s 兲
T r共 s 兲 ⫽ . 共4兲
the sensitivity to modeling errors and inability to 1⫹ 关 G 共 s 兲 ⫺Ĝ 共 s 兲兴 G imc 共 s 兲
deal with external disturbances entering the sys-
tem. In this case, a closed-loop system can be used The disturbance transfer function of the IMC
to deal with disturbances and modeling errors. structure is
Rivera et al. 关5兴 proposed the control structure 1⫺Ĝ 共 s 兲 G imc 共 s 兲
given in Fig. 1. T d共 s 兲 ⫽ . 共5兲
1⫹ 关 G 共 s 兲 ⫺Ĝ 共 s 兲兴 G imc 共 s 兲
Ibrahim Kaya / ISA Transactions 43 (2004) 297–304 299

Substituting Eq. 共2兲 into Eqs. 共4兲 and 共5兲 gives

G共 s 兲F共 s 兲
T r共 s 兲 ⫽ 共6兲
Ĝ ⫺ 共 s 兲 ⫹ 关 G 共 s 兲 ⫺Ĝ 共 s 兲兴 F 共 s 兲
and
Ĝ ⫺ 共 s 兲关 1⫺F 共 s 兲兴
T d共 s 兲 ⫽ . 共7兲
Ĝ ⫺ 共 s 兲 ⫹ 关 G 共 s 兲 ⫺Ĝ 共 s 兲兴 F 共 s 兲 Fig. 2. IMC representation of a SISO control system.

For perfect modeling, G ( s ) ⫽Ĝ ( s ) , and nonmini-

mum phase systems, Ĝ ⫺ ( s ) ⫽Ĝ ( s ) , Eqs. 共6兲 and To find the tuning parameters for the controller
共7兲 can be further simplified to give G c ( s ) in Fig. 2, stable first-order plus dead time
共FOPDT兲 plant transfer function is considered. It
T r 共 s 兲 ⫽F 共 s 兲 共8兲 should be pointed out that the FOPDT model is
only used for simplifying calculations and that the
and
actual process may be a higher-order process or a
T d 共 s 兲 ⫽1⫺F 共 s 兲 . 共9兲 process with complex poles, etc. In order to obtain
the IMC controller, the process FOPDT model,
Eqs. 共8兲 and 共9兲 show that the performance of a
Ĝ ( s ) ⫽Ke ⫺ ␪ s / ( Ts⫹1 ) , must be factored as in
closed-loop system designed based on the IMC
Eq. 共1兲:
design method is determined solely by the filter
dynamics. For a filter with the form given by Eq. Ĝ ⫹ 共 s 兲 ⫽e ⫺ ␪ s , 共14兲
共3兲 and for t→⬁, Eqs. 共8兲 and 共9兲 give c r ( t ) →1
and c d ( t ) →0. K
Ĝ ⫺ 共 s 兲 ⫽ . 共15兲
Ts⫹1
3. Controller design
The IMC controller can be obtained from Eq. 共2兲,
The closed-loop transfer function of a classical assuming a filter with n⫽1, as
SISO feedback system and IMC design, for per-
fect matching, are, respectively, given by Ts⫹1
G imc 共 s 兲 ⫽ . 共16兲
K 共 ␭s⫹1 兲
G c共 s 兲 G 共 s 兲
T siso 共 s 兲 ⫽ 共10兲 Using a first-order Taylor series expansion for the
1⫹G c 共 s 兲 G 共 s 兲
time-delay approximation, the classic controller
and G c ( s ) can be obtained from Eq. 共13兲,
T imc 共 s 兲 ⫽G imc 共 s 兲 G 共 s 兲 . 共11兲 Ts⫹1
G c共 s 兲 ⫽ . 共17兲
In order to have the same output for both configu- K 共 ␭⫹ ␪ 兲 s
rations, it is easy to illustrate, by comparing Eqs.
Eq. 共17兲 is rearranged as an ideal PI controller,
共10兲 and 共11兲, that the IMC controller G imc ( s ) is
which has the following controller parameters:
related to the classic controller G c ( s ) through the
transformation T
K p⫽ , 共18兲
G c共 s 兲 K 共 ␭⫹ ␪ 兲
G imc 共 s 兲 ⫽ 共12兲
1⫹G c 共 s 兲 G 共 s 兲 T i ⫽T. 共19兲
or The only unknown in the last two equations is the
G imc 共 s 兲 filter time constant ␭, since it is assumed that the
G c共 s 兲 ⫽ . 共13兲 plant transfer function model is obtained from the
1⫺G imc 共 s 兲 G 共 s 兲
exact relay feedback identification method given
Therefore from Eq. 共12兲 a classical SISO feedback in Refs. 关7,8兴. Thus if a proper value of ␭ is
system can be put into the IMC structure as shown achieved, then the design procedure will be com-
in Fig. 2. pleted. Here, the value of ␭ is obtained from gain
300 Ibrahim Kaya / ISA Transactions 43 (2004) 297–304

and phase margin specifications, which are well-

known robustness parameters.
The characteristic equation of a SISO control
system is given by 1⫹G c ( s ) G ( s ) . Hence the
open-loop transfer function of the SISO control
system, with G c ( s ) given by Eq. 共17兲, is
e ⫺␪s
G c共 s 兲 G 共 s 兲 ⫽ . 共20兲
共 ␭⫹ ␪ 兲 s Fig. 3. The block diagram for computing the controller,
G c (s), parameters.
Therefore from the basic definitions of the gain
and phase margins the following equations can be The recommended ranges for the gain and phase
obtained: margins are between 2 and 5 and 30° and 60°,
arg兵 G c 共 j ␻ p 兲 G 共 j ␻ p 兲 其 ⫽⫺ ␲ , 共21兲 respectively 关9兴. Choosing A m ⫽3, then ␾ m
⫽60°. Therefore the closed-loop time constant ␭
A m 兩 G c 共 j ␻ p 兲 G 共 j ␻ p 兲 兩 ⫽1, 共22兲 is obtained, by rearranging Eqs. 共25兲 and 共26兲, as

兩 G c 共 j ␻ g 兲 G 共 j ␻ g 兲 兩 ⫽1, 共23兲 ␭⫽ ␪ 冉 2A m
␲ 冊
⫺1 ⫽0.91␪ . 共31兲
␾ m ⫽ ␲ ⫹arg兵 G c 共 j ␻ p 兲 G 共 j ␻ p 兲 其 , 共24兲
Hence the PI controller parameters are given by
where the gain margin is given by Eqs. 共21兲 and Eqs. 共18兲 and 共19兲 with ␭ given by Eq. 共31兲.
共22兲, and the phase margin by Eqs. 共23兲 and 共24兲. Remark: If a second-order plus dead time
The frequency ␻ p is known as phase crossover 共SOPDT兲 plant transfer function model, Ĝ ( s )
frequency, where the Nyquist curve has a phase ⫽Ke ⫺ ␪ s / ( T 1 s⫹1 )( T 2 s⫹1 ) , instead of the
lag of ⫺␲, and the frequency ␻ g is known as the FOPDT model, is assumed, then the controller
gain crossover frequency, where the Nyquist curve G c ( s ) will be a PID controller. However, it has
has an amplitude of 1. been observed during extensive simulations that
Substituting Eq. 共20兲 into Eqs. 共21兲–共24兲, results using a PID controller provides a little improve-
in the following set of equations: ment in the closed-loop performance of the sys-
tem. Hence simulation results only for the PI con-
␲
␻ p␪ ⫽ , 共25兲 troller are given.
2
3.1. Tuning procedure
A m ⫽ ␻ p 共 ␭⫹ ␪ 兲 , 共26兲

1 The block diagram for obtaining the PI control-

␻ g⫽ , 共27兲 ler, G c ( s ) , parameters, based on the specified gain
␭⫹ ␪ and phase margins, is shown in Fig. 3. The tuning
procedure can be carried out as follows:
␲
␾ m⫽ ⫺ ␻ g␪ . 共28兲 1. When the controller needs to be tuned,
2 switch from the controller mode to relay mode.
2. Measure the limit cycle parameters and esti-
From Eqs. 共26兲 and 共27兲 one can obtain
mate the parameters of the FOPDT plant transfer
A m␻ g⫽ ␻ p . 共29兲 function using the relay feedback method. Equa-
tions to compute the parameters of the FOPDT
Multiplying both sides of Eq. 共29兲 with ␪ and then plant transfer function are provided in the Appen-
substituting values of ␻ g ␪ and ␻ p ␪ from Eqs. 共25兲 dix.
and 共28兲, the relation between gain and phase mar- 3. Find the PI controller parameters using Eqs.
gin can be obtained as 共18兲 and 共19兲, in conjunction with Eq. 共31兲.

冉 冊
4. Switch from the relay mode to the controller
␲ 1 mode with calculated tuning parameters for the
␾ m⫽ 1⫺ . 共30兲
2 Am control of the process.
Ibrahim Kaya / ISA Transactions 43 (2004) 297–304 301

Fig. 5. Control signals for example 1.

Fig. 4. Step responses for example 1.

that with the proposed design method less effort is

4. Simulation examples required for the control action, for both design
methods.
Several examples are presented to illustrate the Example 2: In this example, a real industrial
use of the proposed method. Since the presented HVAC system used in Wang et al. 关3兴 with trans-
design method is model based, the identification fer function of G ( s ) ⫽e ⫺2s / ( 0.12s 2 ⫹1.33s
method given by Kaya 关7兴 or Kaya and Atherton ⫹1.24) is considered. The FOPDT model was ob-
关8兴 has been used to find the FOPDT model. The tained as G ( s ) ⫽0.81e ⫺2.78s / ( 1.011s⫹1 ) , using
identification method has been used for all transfer the estimation method given in Refs. 关7,8兴. The PI
functions in the examples but since it gives essen- controller parameters are K p ⫽0.235 and T i
tially exact results on simulation data the esti- ⫽1.011, when Eqs. 共18兲 and 共19兲 are used in con-
mated plant transfer functions are only given for junction with Eq. 共31兲. The PID controller param-
original plants of higher order. In all the examples, eters used by Wang et al. 关3兴 are K p ⫽0.611, T i
controllers, for both the proposed design method ⫽1.441, and T d ⫽0.564. The response of the
and design methods that are used for comparison, closed-loop system with designed controllers for
are designed for a gain and phase margin of 3 and both design methods to a unity step set-point
60°, respectively. change together with load disturbance introduced
Example 1: Consider a second-order plus dead at time 30 s are given in Fig. 6. The assumed load
time plant transfer function of G ( s ) ⫽e ⫺1.0s / ( s disturbance magnitude was ⫺0.5. In terms of
⫹1 )( 0.5s⫹1 ) , which was given in Ref. 关1兴. The
identification method given in Refs. 关7,8兴 was
used to obtain the FOPDT model as G ( s )
⫽e ⫺1.34s / ( 1.44s⫹1 ) . Therefore the PI controller
parameters are obtained from Eqs. 共18兲 and 共19兲,
in conjunction with Eq. 共31兲, to be K p ⫽0.563 and
T i ⫽1.44. The controller parameters for the design
method proposed by Ho et al. 关1兴 are K p ⫽0.52,
T i ⫽1.00, and T d ⫽0.50. The closed-loop re-
sponses for both design methods are given in Fig.
4 for a unity step set-point change and a distur-
bance with magnitude of ⫺0.5 introduced at time
30 s. As is seen from the figure, the proposed de-
sign method results in a better performance for
both the set-point response and disturbance rejec-
tion. Fig. 5 illustrates control signals, which show Fig. 6. Step responses for example 2.
302 Ibrahim Kaya / ISA Transactions 43 (2004) 297–304

Fig. 8. Step responses for example 3.

Fig. 7. Control signals for example 2.

illustrates that with the proposed method less ef-

fort is necessary for the control action.
maximum overshoot, the proposed design method Example 4: A high-order oscillating plant trans-
gives better performance than the design method fer function of G ( s ) ⫽e ⫺s / ( s 2 ⫹s⫹1 )( s⫹3 ) ,
proposed by Wang et al. 关3兴 for the set-point re- which was used by Wang and Shao 关4兴, is consid-
sponse. In terms of settling time, both design ered. Again, the parameter estimation method
methods give similar responses. The load distur- given in Refs. 关7,8兴 was employed to generate the
bance rejection of the design method suggested by FOPDT model as G ( s ) ⫽0.333e ⫺3.1s / ( 0.075s
Wang et al. 关3兴 is slightly faster than the proposed ⫹1 ) . Once a proper model is found, the PI con-
one. This is expected, since in the proposed design troller tuning parameters were calculated to be
method pole zero cancellations are used and this in K p ⫽0.038 and T i ⫽0.075. Wang and Shao sug-
some cases may lead to a sluggish load distur- gested a PID controller with settings of K p
bance rejection. However, when comparing the ⫽1.298, T i ⫽1.034, and T d ⫽1.017. With these
control signals, shown in Fig. 7, it is seen that the calculated controller settings, the step response of
proposed design method requires less attempt for the closed-loop system to a unity step set-point
the control action. change and a disturbance of magnitude of ⫺0.5
Example 3: A high-order plant transfer function introduced at time 50 s is shown in Fig. 10. Again,
of G ( s ) ⫽1/( s⫹1 ) 8 , which was given in Wang the proposed design method results in a better per-
and Shao 关4兴, is considered in this example. The formance, especially for set point response. In Fig.
identification method given in Refs. 关7,8兴 was 11, control signals for both design methods are
used to find the FOPDT model, G ( s )
⫽e ⫺5.10s / ( 4.35s⫹1 ) . Using Eqs. 共18兲 and 共19兲
together with Eq. 共31兲, the PI controller settings
were found to be K p ⫽0.447 and T i ⫽4.340. The
PID controller parameters suggested by Wang and
Shao 关4兴 are K p ⫽0.677, T i ⫽4.340, and T d
⫽1.649. Fig. 8 illustrates responses for both de-
sign methods to a unity set-point change together
with load disturbance introduced at time 80 s. The
load disturbance magnitude was again assumed to
be ⫺0.5. The proposed design method, as is seen
from Fig. 8, results in a better closed-loop system
performance in terms of maximum overshoot,
while in terms of the settling time both designs
give quite similar performances. Control signals
for this example are given in Fig. 9, which again Fig. 9. Control signals for example 3.
Ibrahim Kaya / ISA Transactions 43 (2004) 297–304 303

controller may lead to sluggish load disturbance

rejection, again due to pole zero cancellation used
in the design procedure. However, in the examples
given in this paper, it is seen that the load distur-
bance rejection of the proposed design method is
also satisfactory.

Appendix: Model identification

In this section, equations used to identify the

unknown parameters of the FOPDT plant transfer
function are given. The identification method
Fig. 10. Step responses for example 4. makes use of the relay autotuning. The given
equations will result in exact parameter estima-
tions, assuming no measurement errors. The de-
given. Again, it is seen that the process can be tails can be found in Refs. 关7,8兴.
controlled with less effort by the proposed design Two equations for the limit cycle frequency ␻
method. and the pulse duration ⌬t 1 can be obtained and are
given by

冉 冊
5. Conclusions
⫺ ␻ ⌬t 1 ␲ 共 e ⌬t 1 /T ⫺1 兲 e ␪ /T
Simple tuning rules for a PI controller for con- K ⫹
2 共 e 2 ␲ /␭ ⫺1 兲
trolling stable process with small time delays have
been derived using specified gain and phase mar-
gin specifications. The design method presented in
this paper is model based. Therefore first a
⫽ 冉 ␲
h 1 ⫺h 2 冊冉 R⫺⌬⫺
G 共 0 兲关 h 1 ⌬t 1 ⫹h 2 ⌬t 2 兴
P 冊
FOPDT plant transfer function model was ob- 共A1兲
tained from a single relay feedback test with exact and

冉 冊
limit cycle analysis. Once the model was found,
simple tuning rules provided in the paper were ␻ ⌬t 1 ⫺2 ␲ ␲ 共 e 共 ⫺ ␻ ⌬t 1 ⫹2 ␲ 兲 /␭ ⫺1 兲 e ␪ /T
used to control the process. Since the proposed K ⫹
design method incorporates IMC design prin-
2 共 e 2 ␲ /␭ ⫺1 兲

冉 冊冉 冊
ciples, where pole zero cancellation is used, the ⫺␲ G 共 0 兲关 h 1 ⌬t 1 ⫹h 2 ⌬t 2 兴
closed-loop system with designed PI controller re- ⫽ R⫹⌬⫺ ,
sults in good set point responses. The designed h 1 ⫺h 2 P
共A2兲
where h 1 and h 2 are the relay heights and ⌬ is the
hysteresis. ⌬t 1 and ⌬t 2 are the pulse durations
and P⫽⌬t 1 ⫹⌬t 2 is the period of the oscillation.
R is a constant valued signal entering the system
and ␭⫽ ␻ T.
Two more equations can be obtained for the
maximum and minimum of the plant output wave
form which are given by the following equations:

a min⫽
G 共 0 兲关 h 1 ⌬t 1 ⫹h 2 ⌬t 2 兴
P
⫹ 冉
h 1 ⫺h 2
␲ 冊
Fig. 11. Control signals for example 4.
⫻ 冉 ⫺ ␻ ⌬t 1 ␲ 共 e ⌬t 1 /T ⫺1 兲
2
⫹
共 e 2 ␲ /␭ ⫺1 兲
冊 共A3兲
304 Ibrahim Kaya / ISA Transactions 43 (2004) 297–304

and either Eq. 共A3兲 if a min is measured or Eq. 共A4兲 if

冉 冊
a max is measured. Finally, with K and T known,
G 共 0 兲关 h 1 ⌬t 1 ⫹h 2 ⌬t 2 兴 h 1 ⫺h 2
a max⫽ ⫹ the dead time ␪ can be calculated from either Eq.
P ␲ 共A1兲 or Eq. 共A2兲.

⫻ 冉 ⫺ ␻ ⌬t 1 ␲ e 2 ␲ /␭ 共 1⫺e ⫺⌬t 1 /T 兲
2
⫹
共 e 2 ␲ /␭ ⫺1 兲
. 冊 References
关1兴 Ho, W. K., Hang, C. C., and Cao, L., Tuning of PID
controllers based on gain and phase margin specifica-
共A4兲 tions. Automatica 31, 497–502 共1995兲.
关2兴 Fung, H. W., Wang, Q. G., and Lee, T. H., PI Tuning
Although Eqs. 共A1兲–共A4兲 obtained for a stable in terms of gain and phase margins. Automatica 34,
FOPDT transfer function are sufficient to identify 1145–1149 共1998兲.
关3兴 Wang, Q. G., Fung, H. W., and Zhang, Y., PID tuning
the unknown parameters, K, T, and ␪, initial with exact gain and phase margins. ISA Trans. 38,
guesses are required for the unknowns to solve 243–249 共1999兲.
these nonlinear equations. To reduce the number 关4兴 Wang, Y. G. and Shao, H. H., PID autotuner based on
of unknowns and make the solution easier to find, gain- and phase-margin specifications. Ind. Eng.
Chem. Res. 38, 3007–3012 共1999兲.
Fourier analysis can be used to identify the steady- 关5兴 Rivera, D. E., Morari, M., and Sigurd, S., Internal
state gain K from model control. 4. PID controller design. Ind. Eng.
Chem. Process Des. Dev. 25, 252–265 共1986兲.
冕 c共 t 兲dt
0
P
关6兴 Morari, M. and Zafiriou, E., Robust Process Control.
Prentice-Hall, Englewood Cliffs, NJ, 1989.
K⫽G 共 0 兲 ⫽ 共A5兲 关7兴 Kaya, I., Relay feedback identification and model
冕 y 共 t 兲dt
P
,
based controller design. Ph.D. thesis, University of
0 Sussex, U.K., 1999.
关8兴 Kaya, I. and Atherton, D. P., Parameter estimation
where c ( t ) is the plant output, y ( t ) is the relay from relay autotuning with asymmetric limit cycle
data. J. Process Control 11, 429– 439 共2001兲.
output, and P is the period. 关9兴 Åström, K. J. and Hägglund, T., PID Controllers:
Once the steady-state gain K is obtained from Theory, Design and Tuning. Instrument Society of
Eq. 共A5兲, the time constant T can be obtained from America, Research Triangle Park, NC, 1995.