Journal of Process Control 15 (2005) 81–88

www.elsevier.com/locate/jprocont

A simple robust PI/PID controller design via numerical

optimization approach
R. Toscano *

Laboratoire de Tribologie et de Dynamique des Syst
emes, CNRS UMR5513 ECL/ENISE, 58 rue Jean Parot 42023, Saint-Etienne Cedex 2, France
Received 7 March 2003; received in revised form 4 March 2004; accepted 31 March 2004

Abstract
This paper presents a simple but effective method for designing robust PI or PID controller. The robust PI/PID controller design
problem is solved by the maximization, on a finite interval, of the shortest distance from the Nyquist curve of the open loop transfer
function to the critical point-1. Simulation studies are used to demonstrate the effectiveness of the proposed method.

2004 Elsevier Ltd. All rights reserved.

Keywords: PID controller; Robustness; Numerical optimization; Time-delay

1. Introduction margin is related to the damping of the system, and can

The proportional-integral (PI) and proportional-
integral-derivative (PID) controllers are widely used in
many industrial control systems for several decades
since Ziegler and Nichols proposed their first PID tun-
ing method. This is because the PID controller structure
is simple and its principle is easier to understand than
most other advanced controllers. On the other hand, the
general performance of PID controller is satisfactory in
many applications. For these reasons, the majority of
the controllers used in industry are of PI/PID type.
Most of real plant operate in a wide range of oper-
ating conditions, the robustness is then an important
feature of the closed loop system. When this is the case,
the controller has to be able to stabilize the system for
all operating conditions. To this end, it is possible to
employ an internal-model-based PID tuning method
[12,4]. However, this method gives very slow response to
load disturbance for lag-dominant processes because of
the pole-zero cancellations inherent in the design meth-
odology [2]. Another popular approach with similar
emphasis is the tuning of PI or PID controller by the
gain and phase margin specifications [9,1]. Gain margin
and phase margin have always served as important
measures of robustness. It is well known that phase
therefore also serve as a performance measure [6]. In this
way, numerous progress has been made to improve the
performances of the PI/PID control [8]. In particular,
tuning methods based on optimization approach have
recently received more attention in the literature, with
the aim of ensuring good stability robustness of the
controlled system [10,7]. However, these new methods
are not easy to use for the operating engineer who is the
main user of the PI/PID controller.
The objective of this paper is to propose a novel ro-
bust PI or PID controller which is simple and easy to
use.
The paper is organized as follows. Section 2 presents
the process models used for the synthesis of the PI or
PID controller. In Section 3, the robust PI/PID con-
troller design problem is formulated and solved by the
maximization, on a finite interval, of the shortest dis-
tance from the Nyquist curve of the open loop transfer
function to the critical point-1. Simulation studies are
conducted in Section 4 and comparison with the works
of other authors are given. Section 5 concludes this
paper.
2. Process models

*
Tel.: +33-477-43-84-84; fax: +33-477-43-84-99. The industrial processes are of an extreme variety.
E-mail address: toscano@enise.fr (R. Toscano). Nevertheless, a very broad class is characterized by
0959-1524/$ - see front matter
2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jprocont.2004.03.005
82 R. Toscano / Journal of Process Control 15 (2005) 81–88

aperiodic response. This important category of indus- D(s)

trial systems can be represented by a first-order plus
+
dead time model, as follows: R(s) + E(s) U(s) + Y(s)
K(s) G(s)
ke
t0 s -
GðsÞ ¼ ð1Þ
1 þ ss
Fig. 1. Block diagram of the PID feedback control system.
Note that the above process model is only used for
the purpose of simplified analysis. The actual process
may have multiple lags, non-minimum phase zero, etc.

PI : KðsÞ ¼ kp þ ksi
Another important class of industrial processes is ð5Þ
characterized by non-aperiodic response. This category PID : KðsÞ ¼ kp þ ksi þ kd s
of processes can be represented by a second-order plus For this control system, the sensitivity function SðsÞ and
dead-time model, as follows: complementary sensitivity function T ðsÞ which is the
ke
t0 s transfer function of the closed loop system, are respec-
GðsÞ ¼ ð2Þ tively, defined by
s2 þ a1 s þ a0
1 1
Many identification techniques can be used to obtain SðsÞ ¼ ¼ ð6Þ
1 þ KðsÞGðsÞ 1 þ LðsÞ
first-order plus dead-time or second-order plus dead-
time model for PI/PID control [5,3]. A simple method is where LðsÞ ¼ KðsÞGðsÞ is the open-loop transfer func-
based on the analysis of the open-loop step response. tion, and
The first-order plus dead-time model (1) is obtained as
follows: LðsÞ
T ðsÞ ¼ 1
SðsÞ ¼ ð7Þ
8 1 þ LðsÞ
< k ¼ y1
t ¼ 2:8t1
1:8t2 ð3Þ The quantity jT ðjxÞj represents the input-output gain at
: 0 the frequency 2p=x, for a PI/PID controller this gain is
s ¼ 5:5ðt2
t1 Þ
equal to one in the low frequency domain, that is the
where y1 is the final value of the step response of the steady-state error is equal to zero. The quantity Mp ¼
process, t1 (respectively, t2 ) is the time where the output maxx j T ðjxÞj is the peak magnitude of the frequency
attains 28% (respectively 40%) of its final value. For the response of the closed-loop system. It is well known that
second-order plus dead-time model (2), the parameters Mp is related to the overshoot for the step response of
are obtained as follows: the closed-loop system. In order to impose good tran-
8 sient response it is necessary to have
< k ¼ y1
>
t0 : is the apparent time delay Mp 6 Mpþ ð8Þ
ð4Þ
> 2
: a1 ¼ 2j lnðD1 Þj ; a0 ¼ p þlnðD 1Þ
2

ptp p
2
p t 2 where Mpþ > 1 is the upper bound of the maximum of
the complementary sensitivity function. In an equivalent
where D1 is the first overshoot for the unit step response manner the following constraint is required:
of the process and tp is the corresponding time. Alter-
D1 6 Dþ
1 ð9Þ
natively, these models can be derived from relay feed-
back method [2,1]. This method can be extended to where D1 is the first overshoot of the step response and
open-loop unstable processes [13,11]. Dþ1 is the upper bound value of this overshoot. It is then
possible to introduce a lower bound pseudo-damping
factor fm , which is related to the upper bound of the first
overshoot by the relation:
3. Robust PI/PID controller design
j lnðDþ 1 Þj
fm ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð10Þ
In this section, the robust PI/PID controller design 2
p2 þ lnðDþ 1Þ
problem is formulated and solved via numerical opti-
mization method. the relation between Mpþ and the lower bound pseudo-
damping factor fm , is given by [14]
3.1. Problem statement 1
Mpþ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð11Þ
Consider the PID feedback control system shown in 2fm 1
ðfm Þ2
Fig. 1, in which GðsÞ represents the transfer function of For a good transient response it is then required that
the process model (1) or (2) and KðsÞ is the transfer
function of the standard PI/PID controller f P fm ð12Þ
 R. Toscano / Journal of Process Control 15 (2005) 81–88 83

where f is the pseudo-damping factor of the closed-loop qðsÞ ¼ ðs þ aÞðs2 þ 2fx0 s þ x20 Þ
system. The quantity 1=jSðjxÞj represents the distance
between the Nyquist curve of the open-loop transfer with
function LðsÞ and the critical point-1 at the frequency 8
>
> a ¼ t0t0þs
s

2fx0
2p=x. The minimum of this distance represents then a <
ðx0 þ2afÞx0 t0 s
1
good measure of the stability margin. kp ¼ k ð17Þ
>
>
Consider an additive error model of the open loop : ax20 t0 s
ki ¼ k
transfer function DLðsÞ, the influence of this error on the
closed-loop transfer function can be deduced from the The closed-loop stability impose a > 0 which is verified
first order Taylor series expansion: if
@T ðsÞ t0 þ s
T ðLðsÞ þ DLðsÞÞ ¼ T ðLðsÞÞ þ DLðsÞ ð13Þ >2 ð18Þ
@LðsÞ fx0 t0 s
The above inequality is satisfied for
which gives the well known result
t0 þ s
DT ðsÞ DLðsÞ ¼b
¼ SðsÞ ð14Þ fx0 t0 s
T ðsÞ LðsÞ
with b > 2. Taking into account the first constraint of
The quantity maxx jSðjxÞj represents then a good eval- (16) one can choose f ¼ fm which gives
uation of the robustness in the face of model uncer- (
x0 ¼ bft0mþst0 s
tainties. The sensitivity function SðsÞ appears also in the ð19Þ
transfer function of the input disturbance DðsÞ to the a ¼ t0t0þs
s

2fm x0
output Y ðsÞ
The optimization problem is then written as follows
Y ðsÞ ¼ GðsÞSðsÞDðsÞ ð15Þ 8
>
> maxb>2 f minx j1 þ Lðjx; bÞjg
>
>
The quantity maxx jSðjxÞj represents then also a good >
> kð1þkp Ti sÞe
t0 s
LðsÞ ¼ Ti sð1þssÞ
>
>
evaluation of the performance rejection of the load >
>
>
< x0 ¼ bft0 þst s
disturbance. Finally, in order to achieve good transient m 0
t0 þs ð20Þ
response, good stability margin, good robustness in >
> a ¼
2fm x0
>
>
t0 s
the face of model uncertainties and good rejection of >
>
>
> kp ¼ ðx0 þ2afmkÞx0 t0 s
1
the load disturbance, it is necessary to determine the >
>
>
: ax2 t s
parameters kp , ki and kd such that ki ¼ k0 0
(
f P fm n  o which is numerically easy to solve. The PI controller is
  ð16Þ
maxkp ;ki ;kd minx  Sðjx;k1p ;ki ;kd Þ  sufficient when the process dynamics is essentially first-
order. For higher-order processes the PI controller is not
There is not a known analytical solution of this performing well, in this case the PID controller will be
optimization problem. A way to solve this problem is used [9].
the numerical optimization. In the following, specific
numerical optimization methods are proposed for the 3.1.2. Numerical optimization of the PID controller with
PI/PID controller for the plant models (1), (2). the first-order plus dead-time process model
The dynamic performance obtained with the PI
3.1.1. Numerical optimization of the PI controller with the controller can be improved by the use of a PID con-
first-order plus dead-time process model troller. Consider the standard PID controller (4.1) and
Consider the standard PI controller (4.1) and the the process model (2), the open-loop transfer function is
process model (1), the open-loop transfer function is then given by
given by kð1 þ kp Ti s þ kd Ti s2 Þe
t0 s
LðsÞ ¼
kð1 þ kp Ti sÞe
t0 s Ti sð1 þ ssÞ
LðsÞ ¼
Ti sð1 þ ssÞ with Ti ¼ 1=ki . Using now the approximation
with Ti ¼ 1=ki . Using the approximation e
t0 s
 e
t0 s  1=ð1 þ t20 sÞ2 , the polynomial characteristic of the
1=ð1 þ t0 sÞ, the polynomial characteristic of the closed- closed-loop system is given by
loop system is given by
t0 þ 4s 3 4ðt0 þ s þ kd kÞ 2 4ð1 þ kp kÞ
t0 þ s 2 1 þ kp k k qðsÞ ¼ s4 þ s þ s þ s
qðsÞ ¼ s þ3
s þ sþ t0 s t02 s t02 s
t0 s t0 s Ti t 0 s 4k
þ 2
which is of the form Ti t0 s
84 R. Toscano / Journal of Process Control 15 (2005) 81–88

in order to obtain an entirely adjustable polynomial with

characteristic, qðsÞ can be put in the form 8
>
> a ¼ 12 a1 þ 2t10
fx0
2 >
>
qðsÞ ¼ ðs þ aÞ ðs2 þ 2fx0 s þ x20 Þ < kp ¼ 2ðafþx0 Þax 0 t0
a0
k
2
a x t 2 ð25Þ
with >
> ki ¼ k 0 o
8 >
>
: ða2 þ4afx0 þx20
a0 Þt0
a1
>
> a ¼ 4sþt
2t0 s
0

fx0 kd ¼ k
>
>
< kp ¼
ðafþx0 Þax0 t02 s
2
2k The closed-loop stability impose a > 0 which is verified
a2 x20 t02 s ð21Þ
>
> i
k ¼ if
>
> 4k
 
: ða2 þ4afx0 þx20 Þt02 s
4ðt0 þsÞ
kd ¼ 1 1 1
4k a1 þ >1 ð26Þ
fx0 2 2t0
The closed-loop stability impose a > 0 which is verified
if The above inequality is satisfied for
 
4s þ t0 1 1 1
>1 ð22Þ a1 þ ¼b
2fx0 t0 s fx0 2 2t0
The above inequality is satisfied for with b > 1. Taking into account the first constraint of
4s þ t0 (16) one can choose f ¼ fm which gives
¼b (  
2fx0 t0 s
x0 ¼ fm1 b 12 a1 þ 2t10
with b > 1. Taking into account the first constraint of ð27Þ
a ¼ 12 a1 þ 2t10
fm x0
(16) one can choose f ¼ fm which gives
(
4sþt0
x0 ¼ 2bf The optimization problem is then written as follows:
m t0 s
ð23Þ 8
a ¼ 4sþt 0

fm x 0 >
> maxb>1 f minx j1 þ Lðjx; bÞjg
2t0 s >
> kð1þkp Ti sþkd Ti s2 Þe
t0 s
>
> LðsÞ ¼ Ti sðs
The optimization problem is then written as follows: >
>  2 þa1 sþa0Þ
>
>
8 > 1 1
> x0 ¼ fm b 2 a1 þ 2t0
1
> maxb>1 f minx j1 þ Lðjx; bÞjg <
>
> a ¼ 12 a1 þ 2t10
fm x0 ð28Þ
> LðsÞ ¼ kð1þkp Ti sþkd Ti s2 Þe
t0 s
> >
>
> Ti sð1þssÞ >
> 2ðafm þx0 Þax0 t0
a0
>
> 4sþt0 >
> kp ¼
>
> x ¼ >
>
k
>
<
0 2bfm t0 s > 2 2
> ki ¼ a xk0 t0
a ¼ 4sþt 0

fm x 0 >
>
2t0 s ð24Þ >
: ða2 þ4afm x0 þx20
a0 Þt0
a1
>
> ðaf þx 2
0 Þax0 t0 s
2 kd ¼
>
> kp ¼ m k
>
> 2k
>
> a2 x20 t02 s
>
> k ¼
>
>
i 4k
: ða2 þ4afm x0 þx20 Þt02 s
4ðt0 þsÞ
kd ¼ 4k 4. Results

3.1.3. Numerical optimization of the PID controller with In this section various examples are presented to
the second-order plus dead-time process model illustrate the proposed robust PI/PID controller design
The methods presented above can be extended for the method.
second-order plus dead-time process model, the open-
loop transfer function is then given by 4.1. Example 1
kð1 þ kp Ti s þ kd Ti s2 Þe
t0 s e
s
LðsÞ ¼ Consider the first-order plus dead-time model: sþ1 .
Ti sðs2 þ a1 s þ a0 Þ The proposed tuning method gives the following PI/PID
with Ti ¼ 1=ki . Using the approximation e
t0 s  controllers parameters ðfm ¼ 0:7Þ:
1=ð1 þ t0 sÞ, the polynomial characteristic of the closed-

PI : kp ¼ 0:646 ki ¼ 0:5712
loop system is now given by
    PID : kp ¼ 0:846 ki ¼ 0:7007 kd ¼ 0:2501
1 3 a1 þ k d k 2
qðsÞ ¼ s4 þ a1 þ s þ a0 þ s For comparison, simulation results are presented for the
t0 t0
PI controller tuned by the Gain and Phase Margin
a0 þ k p k k (GPM) method [9]. The GPM-PI controller parameters
þ sþ
t0 Ti t 0 are: kp ¼ 0:52, ki ¼ 0:52 (Am ¼ 3, Um ¼ 60).
which can be put in the form Comparison results are shown in Fig. 2 for unit step
response and load–disturbance response, respectively. It
2
qðsÞ ¼ ðs þ aÞ ðs2 þ 2fx0 s þ x20 Þ is observed that the performance of the proposed PI/
 R. Toscano / Journal of Process Control 15 (2005) 81–88 85

1.4 0.8

0.7
1.2
0.6
1
0.5

0.8 0.4

0.6 0.3

0.2
0.4
0.1
0.2
0

0 -0.1
0 5 10 15 0 5 10 15
Time Time

Fig. 2. Unit step response and load–distrbance response. GPM-PI controller (--), proposed PI controller (–) and proposed PID controller (. . .).

1.4 0.6

1.2 0.5

1 0.4

0.8 0.3

0.6 0.2

0.4 0.1

0.2 0

0 -0.1
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
Time Time

Fig. 3. Unit step response and load–distrbance response. IMC-PI controller (. . .), proposed PI controller (–).

1.2 0.8

0.7
1
0.6
0.8 0.5

0.4
0.6
0.3
0.4
0.2

0.2 0.1

0
0
-0.1

-0.2 -0.2
0 5 10 15 20 25 0 5 10 15 20 25
Time Time

Fig. 4. Unit step response and load–distrbance response. GPM-PID controller (--), proposed PID controller (–).

PID controller is better than that of GPM-PID con-
troller.
For comparison with IMC-PI controller, consider the

0:1s
lag dominant first-order plus dead-time model e1þs . The
proposed tuning method gives the following PI con-
trollers parameters ðfm ¼ 0:5Þ: kp ¼ 0:86, ki ¼ 2:66. The
corresponding parameters of the IMC-PI controller are
[2]: kp ¼ 0:5, ki ¼ 2.
Examples of responses to step change in set point and
load–disturbance are shown in Fig. 3. The performance
of the proposed PI controller is superior to that of IMC-
PI controller.
86 R. Toscano / Journal of Process Control 15 (2005) 81–88

Consider now, the non-minimum phase zero process

1
s Parameter Notation Value
model: ðsþ1Þ 3 . The model used for the designing PID
e
1:58s Process flow rate q 100 l/min
controller is: ðsþ1Þ 2 . The proposed method gives the fol-

lowing PID controllers parameters: kp ¼ 0:7983, Feed concentration CAf 1 mol/l

ki ¼ 0:3514, kd ¼ 0:4497 ðfm ¼ 0:75Þ. With the Gain and Feed temperature Tf 350 K
Phase Margin method the parameters are kp ¼ 0:66, Coolant inlet Tcf 350 K
ki ¼ 0:33, kd ¼ 0:33 (Am ¼ 3, Um ¼ 60). temperature
Comparison results are shown in Fig. 4 for unit step Reactor volume V 100 l
response and load–disturbance response, respectively. Heat transfer hA 7 · 105 cal/min/
Globally, the performance of the proposed PID con- coefficient K
troller is superior to that of GPM-PID controller. Reaction rate k0 7.2 · 1010
constant min
1
Activation energy E=R 1 · 104 K
4.2. Example 2 term
Heat of reaction DH )2 · 105 cal/
Consider the model of a stirred tank reactor mol
q Liquid densities q;, qc 1 · 103 g/l
C_ A ¼ ðCAf
CA Þ
k0 CA e
ðE=RT Þ Specific heat Cp ; Cpc 1 cal/g/K
V
q DHk0
T_ ¼ ðTf
T Þ
CA e
ðE=RT Þ ð29Þ
V qCp
q Cpc 
þ c qc 1
e
ðhA =qc Cpc qc Þ ðTcf
T Þ In this example, CA is the measured output, qc is the
qCp V
control variable and CAf is the disturbance. Consider the
whose variables, parameters and nominal values are the stable region CA 2 ½0:06 0:14, the proposed synthesis
same as defined in [7] and reproduced below. method was applied in the worst-case of this domain,

0.2 0.5
CA CA
0.18 0.45

0.16 0.4

0.14 0.35

0.12 0.3

0.1 0.25

0.08 0.2

0.06 0.15

0.04 0.1
0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35
Time Time

140 300
qc qc
130 250

120 200

110 150

100 100

90 50

80 0
0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35
Time Time

Fig. 5. Closed-loop responses to successive step changes in set point. LMI-PID controller (--), proposed PID controller (–).
 R. Toscano / Journal of Process Control 15 (2005) 81–88
Fig. 6. Closed-loop responses to disturbance. LMI-PID controller (--), proposed PID controller (–).
that is for CA ¼ 0:14 which gives undamped modes. The
PID controller parameters are obtained as kp ¼ 698:1,
ki ¼ 1126; 6, kd ¼ 367:1. For comparison, simulation
results are presented for the PID controller given by [7],
which use a LMI approach for the synthesis. The LMI-
PID controller parameters are kp ¼ 516:6, ki ¼ 765:5,
kd ¼ 143:8.
Comparison results are shown in Fig. 5 for successive
step change in the effluent concentration CA that varies
between 0.06 and 0.3. It is observed that in the whole
operating regimes ðCA 2 ½0:06 0:14Þ, the performances
of both PID controllers are similar. However for suc-
cessive step in CA between 0.15 and 0.3 (in this domain
the process is unstable), the LMI PID control system
becomes unstable while the proposed PID control sys-
tem still remains stable. Fig. 6 presents the disturbance
response for two operating points. It is observed that the
performance of the proposed PID controller is better
than that of LMI-PID controller.
5. Conclusion
In this paper, a simple robust PI/PID controller
design method was developed via numerical optimiza-
tion approach. Various simulation studies have dem-
onstrated the effectiveness of the proposed approach.
87
References
[1] K.J. Astrom, T. Haggl€ und, Automatic tuning of simple regulators
with specification on phase and amplitude margin, Automatica 20
(1988) 645–651.
[2] K.J. Astrom, T. Haggl€ und, PID Controller, 2nd ed., Instrument
of Society of America, Research Triangle Park, NC, 1995.
[3] K.J. Astrom, T. Haggl€ und, C.C. Hang, W.K. Ho, Automatic
tuning and adaptation for pid controllers-a survey, Contr. Eng.
Practice 1 (1993) 699–714.
[4] I.-L. Chien, P.S. Fruehauf, Consider IMC tuning to improve
controller performance, Chem. Eng. Progr. (86) (1990) 33–41.
[5] P.B. Deshpande, R.H. Ash, Computer Process Control, Instru-
ment Society of America, 1988.
[6] G.F. Franklin, J.D. Powell, A.E. Naeini, Feedback Control of
Dynamic Systems, Addison-Wesley, Reading, MA, 1986.
[7] M. Ge, M.-S. Chiu, Q.-G. Wang, Robust PID controller design
via LMI approach, J. Process Contr. 1 (12) (2002) 3–13.
[8] C.C. Hang, K.J. Astrom, Q.G. Wang, Relay feedback auto-tuning
of process controllers-a tutorial review, J. Process Contr. 1 (12)
(2002) 143–162.
[9] W.K. Ho, C.C. Hang, L.S. Cao, Tuning of pid controller based on
gain and phase margin specification, Automatica 31 (3) (1995)
497–502.
[10] C. Hwang, C.-Y. Hsiao, Solution of a non-convex optimization
arising in pi/pid control design, Automatica 38 (11) (2002) 143–
162.
[11] G. Marchetti, C. Scali, D.R. Lewin, Identification and control of
open-loop unstable processes by relay methods. Automatica
(2001), preprint.
[12] M. Morari, E. Zafirou, Robust Process Control, Prentice-Hall,
Englewood Cliffs, NJ, 1989.
88 R. Toscano / Journal of Process Control 15 (2005) 81–88

[13] C. Scali, G. Marchetti, D. Semini, Relay and additional delay for [14] J.J. Di Stefano, A.R. Stubberud, I.J. Williams, Feedback and
identification and autotuning of completely unknown processes, Control Systems, McGraw-Hill, New York, 1975.
Ind. Eng. Chem. Res. 1 (38) (1999) 1987–1997.