IMC based Robust PID design: Tuning guidelines

and automatic tuning

R. Vilanova
*
Telecomunication and System Engineering Department, ETSE, Universitat Autonoma de Barcelona, 08193 Bellaterra, Barcelona, Spain
Received 18 July 2006; received in revised form 19 April 2007; accepted 16 May 2007
Abstract
This communication addresses the problem of tuning a PID controller for step response. The tuning is based upon a First Order Plus
Time Delay (FOPTD) model and aims to achieve a step response specication while taking into account robustness considerations. The
industrial ISA-PID formulation is chosen. A tuning rule is derived rst where the four parameters of the ISA-PID are determined by
means of two new parameters: one parameter is related to the desired closed-loop time constant and the other one to the robustness
level. On a second step, these two parameters are set to a xed value in order to get a simple and automatic rule that directly gives
the controller parameters in terms of the process model parameters. The proposed automatic tuning rule is compared with other known
tunings.
2007 Elsevier Ltd. All rights reserved.
Keywords: PID tuning; IMC; Minmax optimization; Robustness; FOPTD models
1. Introduction
Proportional-Integrative-Derivative (PID) controllers
are without doubt the most extensive option that can be
found on industrial control applications. Their success is
mainly due to their simple structure and meaning of the
corresponding three parameters. This fact makes the PID
control easier to understand by the control engineer than
most other advanced control techniques. In addition, the
performance of a PID controller is usually satisfactory in
many situations.
Because of the widespread use of PID controllers, it is
interesting to have simple but ecient methods for tuning
the controller. In fact, since ZieglerNichols proposed their
rst tuning rules [1], an intensive research has been done.
From modications of the original tuning rules [24] to a
variety of new techniques: analytical tuning [5,6]; optimiza-
tion methods [7,8]; gain and phase margin optimization
[7,9], just to mention a few.
Recently, tuning methods based on optimization
approaches with the aim of ensuring good stability robust-
ness have received attention in the literature [10,11]. Also,
great advances on optimal methods based on stabilizing
PID solutions have been achieved [12,13]. However, these
methods, although eective, rely on somewhat complex
numerical optimization procedures and do not provide
tuning rules. Instead, the tuning of the controller is dened
as the solution of the optimization problem. The work pre-
sented in this paper which is also based on solving an opti-
mization problem provides tuning rules along with the
controller structure; instead of just a numerical solution.
The purpose of this paper is to obtain PID tuning rules
based on a combination of a simple model description;
First Order plus Time Delay (FOPTD); and closed-loop
specications with robustness considerations. The tuning
rules are given in two forms: rstly parameterized in terms
of desired time constant and robustness level and, sec-
ondly, a completely automatic tuning determined by the
0959-1524/$ - see front matter 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jprocont.2007.05.004
*
Tel.: +34 93 5812197; fax: +34 93 5814031.
E-mail address: Ramon.Vilanova@uab.cat
www.elsevier.com/locate/jprocont
Available online at www.sciencedirect.com
Journal of Process Control 18 (2008) 6170
process parameters. The results presented extend previous
formulations presented in [14,15] and give the tuning guide-
lines an interpretation as an extension of the classical IMC
approach of [16] in the following sense: whereas in [16] the
well known IMC lter determines both the desired closed-
loop time constant and the robustness of the controlled sys-
tem, here both roles are separated and can be dealt with
two separate parameters, showing possible advantages of
this alternative formulation. To get the results as close as
possible to the industrial situation, the widely used ISA
structure [7] is chosen for the PID control law.
The paper is organized as follows. Section 2 presents the
problem formulation: process model, PID structure and
the optimization problem based on a minmax formula-
tion. Section 3 solves the minmax optimization problem
and provides the controller structure and expression for
the PID parameters. Starting from the optimal controller
structure, Section 4 introduces the robust stability con-
straint and provides the guideline for a robust tuning.
Starting from the previous guidelines, a simple tuning rule
for automatic tuning is provided in Section 5. Simulation
examples are presented in Section 6 where the automatic
tuning is compared with other known tuning rules. Finally,
in Section 7, conclusions and considerations for further
extensions are conducted.
2. Problem formulation
In this section, the controller equations are presented
along with the assumed process model structure and the
optimization problem that is posed in order to tune the
PID controller.
2.1. PID controller
There exists dierent ways to express the PID control
law [17]. In this paper, we concentrate on the ISA PID con-
trol law [7]
us K
p
brs ys
1
sT
i
rs ys
_

sT
d
1 sT
d
=N
crs ys
_
1
where r(s), y(s) and u(s) are the Laplace transforms of the
reference, process output and control signal, respectively.
K
p
is the PID gain, T
i
and T
d
are the integral and derivative
time constants, nally N is the ratio between T
d
and the
time constant of an additional pole introduced to assure
the properness of the controller. Usually, for this parame-
ter, a value N 10 is found in the literature. However, here
it will be considered as a design parameter that can be cho-
sen freely and does not need to take an integer value.
Parameters b and c are called set-point weights and con-
stitute a simple way to obtain a 2-DOF controller. As their
choice does not aect the feedback properties of the result-
ing controlled system, with no loss of generality here we
will assume b = c = 1. This way, the PID transfer function
we work with can be written as
Ks K
p
1 s T
i

T
d
N
_ _
s
2
T
i
T
d
N
N 1
sT
i
1 s
T
d
N
_ _ 2
2.2. Process model
An important category of industrial processes can be
represented by a First Order Plus Dead Time (FOPDT)
model as
G
n
s
Ke
Ls
1 Ts
3
where K is the process gain, T the time constant and L is
the time delay. These kind of models are easy to nd by
means of a simple step response experiment to get the pro-
cess reaction curve. In order to deal with the delay term it is
usual to use a rational approximation. Here, we will work
with the following simple rst order Taylor expansion of
the e
Ls
term
e
Ls
1 Ls: 4
Another usual approximation for the time delay is the rst
order Pade approximation. However, as it will be clear
when solving the optimization problem, at this stage we
need to consider this approximation if we want to retain
the PID as the resulting structure of the optimizing control-
ler. Other situations are to be considered as extensions of
this initial situation.
2.3. Optimization problem
In order to take into account robustness considerations,
the design problem must be posed accordingly. One, rather
usual, approach is to use frequency dependent uncertainty
descriptions and to include them into the design problem
[18]. Assume the real process G(s) belongs to a family of
plants F dened by
F fGs G
n
s1 D
m
s : jD
m
jwj
< jW
m
jwjg 5
where G
n
(s) is the nominal model that obeys (3), D
m
(s) is
the uncertainty description based on the multiplicative
model error dened as
D
m
s
:
Gs G
n
s
G
n
s
6
and W
m
(s) is a frequency dependent weight that denes the
family of plants by bounding the modelling error.
It is well known that a controller, K(s), that stabilizes the
control system on the nominal system, also stabilizes all the
control systems built up around the family F of plants
such that
kW
m
sTsk
1
< 1 7
62 R. Vilanova / Journal of Process Control 18 (2008) 6170
where T(s) is the nominal Complementary Sensitivity
transfer function:
Ts
KsG
n
s
1 G
n
sKs
8
However, if one uses the Internal Model Control paradigm
(IMC) that can be found in [18] it turns out that T(s) has a
very simple expression in terms of the so called IMC con-
troller, or Youla parameter, for stable plants:
Ts G
n
sCs 9
The IMC synthesis gets C(s) on a rst step and recovers
K(s) on a second step from:
Ks
Cs
1 G
n
sCs
10
The main feature of the IMC method is that the desired
closed-loop time constant is provided as a tuning parame-
ter, commonly known as the IMC lter. Robustness is
dealt through the reduction of this desired closed loop
bandwidth. A detailed description of the IMC synthesis
method can be found in [18,19] (see Fig. 1).
Here, we will make use of the IMC formulation just to
set up the minmax problem on which we will base the
design. This way we directly design the C(s) transfer func-
tion as the solution to the following problem:
min
Cs
kW sMs CsG
n
sk
1
11
where M(s) is a Reference Model for the closed-loop sys-
tem response and W(s) is a weighting function. In this com-
munication, we will use the Reference Model to specify the
desired closed-loop time constant, T
M
. Therefore it will
take the form:
Ms
n
M
s
d
M
s

1
1 T
M
s
12
With respect to the weighting function, W(s), in order to
automatically include integral action and keep it as simple
as possible, we will assume the following form:
Ws
n
W
s
d
W
s
c
1 zs
s
13
In order to include robustness considerations, the solution
to this minimization problem must be constrained to (7).
3. Solution to the minmax optimization problem
This section will present a solution to the optimization
problem (11). Several approaches exists to solve this H
1
problem. See [20,21] among others. Here, we will follow
a particularization of the solution presented in [22] where
a polynomial approach was taken. This has the advantage
of providing the structure of the optimal controller. There-
fore, as we will do here, the problem statement can be con-
strained in order to provide a solution that leads to a PID
controller.
The problem at hand is, in fact, a minmax approxi-
mation problem: given two transfer functions M(s),
Ns 2 RH
1
, nd Cs 2 RH
1
such that the following
cost function in the 1-norm is minimized
J
1
kEsk
1
kW sMs NsCsk
1
14
where N(s), M(s) and W(s) are factored as
Ns
n
N
s
d
N
s
Ms
n
M
s
d
M
s
W s
n
W
s
d
W
s
The solution to the minimization of the cost function (14)
lies in optimal interpolation theory. First, factorize the
plant numerator n
N
(s) as
n
N
s n

N
sn

N
s
where the polynomial n

N
s only has stable roots and n

N
s
is the remaining part. In order to obtain a unique factoriza-
tion, the polynomial n

N
s is assumed to be monic. Let
m = deg(n

N
s) and {z
1
, z
2
, . . . , z
m
} be the distinct zeros of
n

N
s. From Eq. (14) it results that the error function
E(s) is subjected to the following interpolation constraints:
Ez
i
W z
i
Mz
i
i 1 . . . m 15
If z
i
is a zero with multiplicity m
i
, then additional dierential
interpolation constraints should be imposed.
A well established theory [23,24], [20] that solves this
problem exists and a closed form solution can be obtained
from the following lemma [20]:
Lemma 3.1. The optimal E
o
(s) which minimizes kE(s)k
1
is
of an all-pass form
E
o
s
q
qs

qs
if m P1
0 if m 0
_
16
K (s)
C(s) G(s)
G(s)
G
n
(s)
u
u
y
y
d
d
r
r
a
b
Fig. 1. Feedback control system: (a) Conventional conguration and (b)
Internal Model Control conguration.
R. Vilanova / Journal of Process Control 18 (2008) 6170 63
where q(s) = 1 + q
1
s + q
2
s
2
+ + q
m1
s
m1
is a strictly
Hurwitz polynomial and q
*
(s) = q(s).
Furthermore, the constants q and fq
i
g
m1
i1
are real and are
uniquely determined by the interpolation constraints (15).
Besides, the minimum cost of (14) is given by
J
o
1
min kEsk
1
kE
o
sk
1
jqj
Now we will proceed with the application of this lemma
in order to compute the optimal C(s), C
o
(s). Note rst that
in our case m = 1 and z
1
= 1/L. Therefore, the interpolation
constraints give the following value for the optimal cost q:
q W 1=LM1=L cL
L z
LT
M

17
Application of the above lemma gives the following equa-
tion for the optimal parameter C
o
(s)
WsMs W sNsC
o
s q
q

s
qs
then
C
o
s W sNs
1
WsMs q
q

s
qs
_ _

d
W
sd
N
s
n
W
sn

N
sn

N
s

n
W
sn
M
sqs qq

sd
W
sd
M
s
d
W
sd
M
sqs
_ _
In order for C
o
(s) to be a stable transfer function, n

N
s
must be a factor of the numerator. That is to say, there
must exist a polynomial v(s) such that
n

N
svs n
W
sn
M
sqs qq

sd
W
sd
M
s 18
It follows that, to determine the optimal controller C
o
(s),
the v(s) polynomial must be known. In any case, the opti-
mal C
o
(s) will obey the following structure:
C
o
s
d
N
svs
n
W
sn

N
sd
M
sqs
19
Moreover, as m = 1, it follows from the previous lemma
that q(s) = q
*
(s) = 1. Also, d
N
(s) = (1 + Ts), n

N
s K
and d
M
(s) = (1 + T
M
s). Therefore, C
o
(s) further simplies
to
C
o
s
1
K
1 Tsvs
1 T
M
s1 zs
20
With respect to v(s), it must obey to (18) so, if
v(s) Gv
0
+ v
1
s, then:
1 Lsv
0
v
1
s 1 zs
qs
c
1 T
M
s 21
It is easily seen that
v
0
1 v
1
z L
q
c
22
Therefore, the solution for the optimal C
o
(s) is
C
o
s
1
K
1 Ts1 v
1
s
1 T
M
s1 zs
23
and the resulting optimal feedback K
o
(s) = C
o
(s)/
(1 G
n
(s)C
o
(s)) becomes
K
o
s
1
K
1 Ts1 v
1
s
s
q
c
T
M
_ _
T
M
q
c
z
_ _
s
_ _

1
K
q
c
T
M

1 Ts1 v
1
s
s 1 T
M
q
c
z
q
c
T
M
s
_ _ 24
Thus, identifying (2) and (24), the following tuning rela-
tions arise:
K
p

T
i
Kq=c T
M

T
i
T v
1
T
M
q=c z
q=c T
M

T
d
T
M
q=c z
q=c T
M

N
N
T
T
i
q=c
L
q=c T
M

q=c z
1
25
Note that these relations provide all the PID parameters,
including the derivative lter, N. The benets of providing
a tuning of this parameter have been reported in [25,26].
Although the tuning relations (26) look somewhat com-
plicated, note that they are directly expressed in terms of
the process model (K, L and T) and the denition of the
optimization problem (c, z and T
M
). Moreover, note that
c always appears as q/c. Therefore, because of (17) it results
that q/c is independent of c. Without loss of generality, we
can assume c = 1 and the previous relations simplify to:
K
p

T
i
Kq T
M

T
i
T v
1
T
M
q z
q T
M

T
d
N
T
M
q z
q T
M

N 1
T
T
i
q
L
q T
M

q z
26
4. Tuning for robustness
The previous section has provided both the IMC con-
troller structure corresponding to the optimal solution to
(11) and the associated PID tuning relations (26). That
problem does not take into account robustness consider-
ations, so a further step is needed in order to deal with
model uncertainty. Now we will start from the optimal
solution and introduce the robust stability constraint.
It is worth to stress the point that our primary interest is
to provide a simple way of dealing with uncertainty. From
the optimal solution obtained in the previous section, two
64 R. Vilanova / Journal of Process Control 18 (2008) 6170
parameters dene the optimization problem (T
M
and z).
We will now see what is the role of these parameters.
Assuming we have the set of plants dened in terms of
an uncertainty description weight, W
m
(s), the robust stabil-
ity constraint takes the form:
kG
n
sCsW
m
sk
1
< 1
) K
p
1 Ls
1 Ts
1
K
p
1 Ts1 v
1
s
1 T
M
s1 zs
W
m
s
_
_
_
_
_
_
_
_
1
< 1
)
1 Ls1 v
1
s
1 T
M
s1 zs
W
m
s
_
_
_
_
_
_
_
_
1
< 1
)
1 Ljw1 v
1
jw
1 T
M
jw1 zjw

<
1
W
m
jw

8 w 27
Now, as L is given by the process model and v
1
by the opti-
mization problem, the two parameters left to be chosen are
T
M
and z. As 1/W
m
(s) usually has a low pass form, both
parameters can contribute to provide the necessary roll-
o in order to satisfy the constraint.
Moreover, we can consider T
M
as a nominal specica-
tion for the closed-loop step response. Therefore, at this
step it would be desirable not to change it and use z to
try to meet the robustness specication. This way, even z
was dened from the weighting function of the nominal
optimization problem (13), we see that it can be interpreted
as a robustness parameter. The choice of z will obey the
satisfaction of a constraint of the form:
1 v
1
jw
1 zjw

<
1 T
M
jw
1 Ljw
1
W
m
jw

8 w 28
The choice of z becomes easy if we take into account that z
is also a pole of the nominal closed-loop characteristic
equation and the fact that jW
m
(jw)j
1
usually imposes a
low pass characteristic to C(s)G
n
(s). The choice of z will
need to provide such a shape to the left-hand side of
(28). One way of assuring this is by choosing z > T
M
(this
will assure z > v
1
). However, lower values of z may be pos-
sible if T
M
> L as this will help in making the right-hand
side of (28) less restrictive. This situation will be exempli-
ed in example 1.
This way, the optimal value for z will be the smallest one
that satises (28). Of course, if we state a highly stringent
T
M
for the nominal situation, it may be possible that the
value for z is such that (28) is satised and is excessively
high, or even it may not exist. In such case the nominal
response can deteriorate and it may be benecial to slightly
increase T
M
and get lower values for z. This is the usual
tradeo between the desired nominal performance and sat-
isfaction of the robust stability constraint.
This approach could be easily seen as extension to the
IMC method [16] because of the role played by the second
parameter, z. Within the IMC approach, we have the well
known IMC lter time constant, that turns out to be the
desired closed-loop time constant (therefore it plays the
same role as T
M
) and also determines the robustness of
the closed-loop. What is suggested here is to provide
robustness by using z instead of relaxing the desired
closed-loop time constant. This does not mean that we
can achieve any desired closed-loop time constant even in
the presence of uncertainty.
4.1. Simulation example
This section presents a simulation example that shows
the application of the outlined design method and how to
incorporate frequency domain uncertainty considerations
into the design procedure. The purpose of the example is
to show that frequency shaping considerations can be eas-
ily incorporated into the design phase.
As it has been stated, the nominal process model the
design is based upon is a FOPTD model where the time
delay has been approximated by using (4). Therefore, con-
sider the following nominal model:
Gs K
n
e
Lns
1 T
n
s
! G
n
s K
n
1 L
n
s
1 T
n
s
with K
n
= L
n
= T
n
= 1. In addition, an uncertainty of 30%
associated to each parameter is considered. We choose
T
M
= 1 as the desired closed-loop time constant. The other
parameter z is chosen such that (27) is satised. In order to
guide the selection of z, the inverse of the multiplicative
uncertainty, D
m
(jw), of the members of the plant family
1
is plotted against jC(jw)G
n
(jw)j (see Fig. 2).
It is found that the value of z can be reduced to z 1.35.
The corresponding ISA-PID parameters that result from
the application of the Tuning rules (26) are:
K
p
0:4 T
i
1:31 T
d
0:0015 N 0:0013
10
-2
10
-1
10
0
10
1
10
-1
10
0
10
1
10
2
Freq. (rad/sec)
M
a
g
n
i
t
u
d
e

(
d
B
)
|C(jw)G
n
(jw)|
|1/
m
(jw)|
Fig. 2. Robust stability constraint for example 1.
1
It is worth to notice that the uncertainty has been computed with
respect to Gs K
e
Ls
1Ts
. This way the eect of the delay approximation (4)
is also taken into account.
R. Vilanova / Journal of Process Control 18 (2008) 6170 65
The step response of this PID with respect to the plant fam-
ily is shown in Fig. 3. It can be seen that the step response
performance over all the family set is highly acceptable and
of similar dynamics as in the nominal situation.
5. Proposed automatic robust tuning rule
Tuning relations (26) allows to choose z on the basis of
the uncertainty levels available in each situation. In some
cases, however, one may just think of having an automatic
tuning. Either because there is no uncertainty description
available or because of the simplicity of an automatic tun-
ing rule where just the parameters of the FOPTD model are
needed.
In order to achieve this, values for z and T
M
are needed.
The approach we will follow here is to propose a selection
of these parameters on the basis of the robust stability con-
straint (7). The usual shape for multiplicative uncertainty,
imposes T(s) a low-pass shape. This way, we will look
how to assure that
1 v
1
jw1 Ljw
1 zjw1 T
M
jw

< 1 8 w 29
and to behave as a low-pass lter. This constraint may
seem too conservative because it looks like W
m
(s) was cho-
sen such that jW
m
(jw)j = 1. However, mid and high fre-
quency uncertainty greater than 1 will be allowed; as it
will be clear from the discussion below; because of the
way the dierent parameters are selected. With the previ-
ous constraint we are imposing, in some sense, that the
closed-loop transfer function does not have spikes and
behaves as a low pass lter. This is to be accomplished
by a suitable pole-zero pattern. The following ordering is
needed: 1/z < 1/v
1
and 1/T
M
< 1/L. In addition, because
of (22), in order to have 1/z < 1/v
1
we must have z > T
M
.
Note that z > T
M
also implies 1/v
1
< 1/T
M
. These inequal-
ities suggests that z and T
M
could be chosen on the basis of:
T
M
= bL and z = aT
M
with b > 1 and a > 1. As a and b
grow the system becomes more robust.
Our choice here is to take a b

2
p
. This way, 1/T
M
,
is located on the geometric center between 1/L and 1/z.
Also, with this choice, we have x
c
L 0.6, where x
c
is
the gain crossover frequency. We have from [27] that
x
c
L < p/2 = 1.57 provides a phase margin of at least 45
and x
c
L < 0.7 for max
w
jS(jw)j < 2, being S(s) the Sensitiv-
ity function. This way, the simple choice produces reason-
able classical robustness margins and by assuring (29) is
below 1 for all frequencies, we also have tolerance to mid
and high frequency non-modelled dynamics.
Application of the relations T
M
bL

2
p
L and z = a
T
M
= 2L to tuning rule (26) provides the following simple
tuning rule:
K
p

T
i
KL2:65
T
i
T 0:03L
T
d
N
1:72L
N 1
T
T
i
30
6. Evaluation and comparison of the proposed tuning
In this section, we will evaluate, by means of representa-
tive examples, the proposed simple automatic tuning rule.
In order to cover possible dierent situations, the examples
consider one lag-dominant process, one lead-dominant
process and one with balanced lag and delay. The examples
are taken from [4]. Each one of the processes is approxi-
mated by a FOPTD model of the form (3). The three pro-
cesses and corresponding FOPTD models are:
0 2 4 6 8 10 12 14 16 18 20
0
0.2
0.4
0.6
0.8
1
1.2
1.4
time

O
u
t
p
u
t

Response for
e
s
s +1
_
Fig. 3. Step responses of the ISA-PID with respect to the plant family.
66 R. Vilanova / Journal of Process Control 18 (2008) 6170
G
1
s
1
1s10:1s10:01s10:001s

e
0:073s
1:073s 1
G
2
s
e
s
10:05s
2

e
s
0:093s 1
G
3
s
1
1s
4

e
1:42s
2:9s 1
31
In addition a comparison with other tuning rules is pro-
vided. A complete comparison with multiple tuning rules
will not be provided because (i) the number of existing
tuning rules in the literature is really large and (ii) the
approach presented here is in the spirit of simple tuning
rules. Therefore, we will concentrate on two existing
approaches that also provide very simple and eective
tunings:
S-IMC tuning rule: This is a really simple and eective
analytic tuning rule proposed in [6]. This method is
based on the so-called half-rule for process model
reduction and approximation. It is worth noting that
for a FOPTD, the S-IMC rule gives a PI controller.
Therefore, we are comparing a S-IMC PI with respect
to a PID.
AMIGO tuning rule: Tuning rule proposed in [4] is along
the lines of the classical ZieglerNichols rules. Under
this approach the design is carried out by means of the
optimization of load disturbance rejection with con-
straints on robustness to model uncertainties.
The evaluation and corresponding comparison will be
done according to the criteria aimed to represent both
robustness and performance. Following [6], the following
measures will be used:
Robustness: As a rather usual measure for robustness,
will use the Sensitivity and Complementary Sensitivity
peaks, M
s
and M
t
, respectively, providing M
t
that is a
measure of the allowable multiplicative uncertainty
bound.
Output performance: The Integrated Absolute Error
(IAE) of the error e = r y will be computed. This
value should be as small aspossible
IAE
_
1
0
jetjdt
Input performance: To evaluate the manipulated input
usage, the total variation (TV) of the control signal,
u(t) is computed. This value is dened, for a discrete sig-
nal as the sum of the size of its increments:
TV

1
i1
ju
i1
u
i
j
This quantity should be as small as possible and provide
a measure of the smoothness of the control signal. In or-
der to dene it properly for a continuous signal (that is
the case in our examples), a sampled version of the con-
trol signal has to be used.
Even the presented tuning approach is formulated on
the basis of a step response specication and no distur-
bance attenuation considerations have been taken into
account when deriving the automatic tuning rule, both per-
formance measures, IAE and TV, will be computed for a
step change as well as a load disturbance. This will provide
a more global and complete comparison framework. Table
1 shows the presented performance measures applied to the
considered processes (31) when the proposed, S-IMC and
AMIGO tuning rules are used. Note for the disturbance
performance also the peak deviation, e
p
, following the load
disturbance is shown.
As it can be seen in Figs. 46, a feature of the proposed
tuning rule is that it generates quite smooth responses. This
is veried by the achieved closed-loop bandwidth; see
Fig. 7; and the low values for the manipulated input vari-
able usage. In fact, the proposed tuning provides the lower
TV values for both step change and load disturbance. The
eect this low input usage has on the output performance is
dierent if one looks at the setpoint or the disturbance per-
formance. Except for process G
2
(s); where the AMIGO
tuning exhibits a clear superior performance; the proposed
tuning provides lower setpoint IAE with less input usage.
With respect to the robustness margins, note the M
t
= 1.0
for all the designs. This is a direct consequence of constraint
(27). Moreover, the value of 1.0 corresponds to x = 0. In
Table 1
Performance evaluation for the processes G
1
(s), G
2
(s) and G
3
(s) and Proposed, S-IMC and AMIGO tuning rules
System Tuning M
s
M
t
Setpoint Disturbance
IAE TV IAE TV e
p
G
1
(s) Proposed 1.41 1.0 0.23 1.16 0.1 0.11 0.07
S-IMC 1.63 1.11 0.24 1.72 0.04 0.13 0.06
AMIGO 1.44 1.19 0.30 1.52 0.03 0.13 0.05
G
2
(s) Proposed 1.41 1.0 2.66 0.009 1.33 0.009 0.5
S-IMC 1.59 1.0 2.18 0.01 1.09 0.01 0.5
AMIGO 1.41 1.0 1.94 0.014 0.97 0.013 0.5
G
3
(s) Proposed 1.41 1.0 3.97 0.482 1.89 0.31 0.30
S-IMC 1.59 1.0 4.08 0.776 1.54 0.38 0.28
AMIGO 1.40 1.06 3.98 0.723 1.26 0.35 0.25
R. Vilanova / Journal of Process Control 18 (2008) 6170 67
addition, this low-pass shape for the complementary sensi-
tivity transfer function generates sensitivity peaks of 1.41.
Therefore, showing the clear robustness margins of the
resulting design and major tolerance to unstructured uncer-
tainty. This robustness is in accordance with the smoothness
of the corresponding control and output signals.
0 5 10 15
0
0.2
0.4
0.6
0.8
1
1.2
1.4
System G2 : Proposed (solid) S-IMC (dashed) AMIGO (dash-dot)

O
u
t
p
u
t
time
0 5 10 15
0
0.2
0.4
0.6
0.8
1
1.2
1.4
System G2 : Proposed (solid) S-IMC (dashed) AMIGO (dash-dot)
C
o
n
t
r
o
l
time
0 5 10 15
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
System G2 : Proposed (solid) S-IMC (dashed) AMIGO (dash-dot)
O
u
r
p
u
t
time
0 5 10 15
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
System G2 : Proposed (solid) S-IMC (dashed) AMIGO (dash-dot)
C
o
n
t
r
o
l
time
Fig. 5. Time responses for process G
2
(s) for a step change and load disturbance. performance of the proposed (solid), S-IMC (dashed) and AMIGO (dash-
dot) tuning rules.
0 0.5 1 1.5 2 2.5 3
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
System G1 : Proposed (solid) S-IMC (dashed) AMIGO (dash-dot)

O
u
t
p
u
t
time
0 0.5 1 1.5 2 2.5 3
-2
0
2
4
6
8
System G1 : Proposed (solid) S-IMC (dashed) AMIGO (dash-dot)
C
o
n
t
r
o
l
time
0 1 2 3 4 5
-0.08
-0.06
-0.04
-0.02
0
0.02
System G1 : Proposed (solid) S-IMC (dashed) AMIGO (dash-dot)
O
u
r
p
u
t
time
0 1 2 3 4 5
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
System G1 : Proposed (solid) S-IMC (dashed) AMIGO (dash-dot)
C
o
n
t
r
o
l
time
Fig. 4. Time responses for process G
1
(s) for a step change and load disturbance. performance of the proposed (solid), S-IMC (dashed) and AMIGO (dash-
dot) tuning rules.
68 R. Vilanova / Journal of Process Control 18 (2008) 6170
On the other hand, Table 1 shows that both S-IMC and
AMIGO exhibit clear superior performance for the load
disturbance attenuation. However, the time responses
depicted in Figs. 46 show that, except for process G
1
(s),
time responses are of comparable benets as they have sim-
ilar settling time and the controller reacts quite similarly.
The improvement of the load disturbance performance is
a subject of current study and preliminary results have been
presented in [28] where the design problem also formulates
din terms of the sensitivity function and not only the com-
plementary sensitivity.
7. Conclusions
Tuning relations for PID design have been presented. In
order to get results closer to Industrial applications, the
10
-2
10
-1
10
0
10
1
10
2
10
3
0
0.5
1
1.5
2
System G1 : Proposed (solid) S-IMC (dashed) AMIGO (dash-dot)
10
-2
10
-1
10
0
10
1
10
2
10
3
0
0.5
1
1.5
2
System G2 : Proposed (solid) S-IMC (dashed) AMIGO (dash-dot)
10
-2
10
-1
10
0
10
1
10
2
10
3
0
0.5
1
1.5
2
System G3 : Proposed (solid) S-IMC (dashed) AMIGO (dash-dot)
Fig. 7. Sensitivity and Complementary Sensitivity transfer functions frequency responses of the closed-loop systems achieved by using the proposed
(solid), S-IMC (dashed) and AMIGO (dash-dot) tuning rules.
0 5 10 15 20 25 30
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
System G3 : Proposed (solid) S-IMC (dashed) AMIGO (dash-dot)

O
u
t
p
u
t
time
0 5 10 15 20 25 30
0.8
1
1.2
1.4
1.6
1.8
2
System G3 : Proposed (solid) S-IMC (dashed) AMIGO (dash-dot)
C
o
n
t
r
o
l
time
0 5 10 15 20 25 30 35 40
-0.35
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
System G3 : Proposed (solid) S-IMC (dashed) AMIGO (dash-dot)
O
u
r
p
u
t
time
0 5 10 15 20 25 30 35 40
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
System G3 : Proposed (solid) S-IMC (dashed) AMIGO (dash-dot)
C
o
n
t
r
o
l
time
Fig. 6. Time responses for process G
3
(s) for a step change and load disturbance. performance of the proposed (solid), S-IMC (dashed) and AMIGO (dash-
dot) tuning rules.
R. Vilanova / Journal of Process Control 18 (2008) 6170 69
discussion has concentrated on the ISA formulation. The
design has been done from a minmax optimization prob-
lem stated in terms of a desired time constant for the
closed-loop step response. In order to cope with model
uncertainty, robust stability considerations allow to give
a concise meaning to the tuning parameters. As a result,
the nal PID tuning relations are governed by two param-
eters: T
M
that express the desired time constant for the
nominal situation, and z that allows to deal with model
uncertainty. Fixed values for these parameters are also pro-
posed and a very simple automatic tuning rule is provided.
Simple examples show the performance over the whole
plant family set.
Future work is conducted to introduce a Sensitivity con-
straint in order to improve disturbance attenuation. This
way a mixed sensitivity problem will need to be solved.
Although optimization approaches based on non-convex
numerical methods could be used it would be helpful if
an analytical solution along the lines of the one presented
could be found. Also, considerations to include the set-
point weights and to design the overall ISA-PID controller
are being considered.
Acknowledgement
This research was done under the nancial support of
the Spanish CICYT program grant DPI2004-06393.
References
[1] J. Ziegler, N. Nichols, Optimum settings for automatic controllers,
Trans. ASME (1942) 759768.
[2] I. Chien, J. Hrones, J. Reswick, On the automatic control of
generalized passive systems, Trans. ASME (1952) 175185.
[3] C. Hang, K. Astrom, W. Ho, Renement of the zieglerNichols
formula, IEE Proc. Part D 138 (1991) 111118.
[4] K. Astrom, T. Hagglund, Revisiting the zieglerNichols step response
method for PID control, J. Process Control 14 (2004) 635650.
[5] S. Hwang, H. Chang, Theoretical examination of closed-loop
properties and tuning methods of single loop pi controllers, Chem.
Eng. Sci. 42 (1987) 23952415.
[6] S. Skogestad, Simple analytic rules for model reduction and PID
controller tuning, J. Process Control 13 (2003) 291309.
[7] K. Astrom, T. Hagglund, PID Controller, Instrument of Society of
America, 1995.
[8] K. Astrom, H. Panagopoulos, T. Hagglund, Design of pi control-
lers based on non-convex optimization, Automatica 34 (1998) 585
601.
[9] H. Fung, Q. Wang, T. Lee, Pi tuning in terms of gain and phase
margins, Automatica 34 (1998) 11451149.
[10] M. Ge, M. Chiu, Q. Wang, Robust PID controller design via LMI
approach, J. Process Control 12 (2002) 313.
[11] R. Toscano, A simple PI/PID controller design method via numerical
optimization approach, J. Process Control 15 (2005) 8188.
[12] G. Silva, A. Datta, S. Battacharayya, New results on the synthesis of
PID controllers, IEEE Trans. Automat. Control 47 (2) (2002) 241
252.
[13] M. Ho, C. Lin, PID controller design for robust performance, IEEE
Trans. Automat. Control 48 (8) (2003) 14041409.
[14] R. Vilanova, P. Balaguer, Robust PID tuning relations based on min
max optimisation, ROCOND06, in: Proceedings of the 5th IFAC
Symposium on Robust Control, July 57, Toulouse, France.
[15] R. Vilanova, Pid controller tuning rules for robust step response of
rst-order-plus-dead-time models, ACC-2006, in: Proceedings of the
25th American Control Conference, Minneapolis, USA.
[16] D.E. Rivera, M. Morari, S. Skogestad, Internal model control 4. PID
controller design, Ind. Eng. Chem. Res. 25 (1986) 252265.
[17] P. Cominos, N. Munro, PID controllers: recent tuning methods and
design to specication, IEE Proc. Part D 149 (2002) 4653.
[18] M. Morari, E. Zarou, Robust Process Control, Prentice-Hall,
Englewood Clis, NJ, 1989.
[19] Y. Lee, S. Park, M. Lee, C. Brosilow, Pid controller tuning for desired
closed-loop responses for SI/SO systems, AICHE J. 44 (1998) 106
115.
[20] B. Chen, Controller synthesis of optimal sensitivity: multivariable
case, IEE Proc. Part D 131 (1984) 547551.
[21] B. Francis, A Course in H
1
Control theory, Springer Verlag, 1987.
[22] R. Vilanova, I. Serra, Model reference control in two degree of
freedom control systems: adaptive minmax approach, IEE Proc.
Part D 146 (1999) 273281.
[23] D. Sarason, Generalized interpolation in h
1
, Trans. AMS 127 (1967)
179203.
[24] G. Zames, B. Francis, Feedback, minmax sensitivity and optimal
robustness, IEEE Trans. Automat. Control 28 (1983) 585601.
[25] A. Isaakson, S. Graebe, Derivative lter is an integral part of PID
design, IEE Proc. Part D 149 (2002) 4145.
[26] W. Luyben, Eect of derivative algorithm and tuning selection on the
PID control of dead-time processes, Ind. Eng. Chem. Res. 40 (2001)
36053611.
[27] P. Albertos, A. Sala (Eds.), Iterative Identication and Control,
Springer Verlag, 2001.
[28] R. Vilanova, O. Arrieta, Sensitivity function optimization approach
to PID design for improved disturbance attenuation, IMECS/
ICCA2007, in: Proceedings of the 2007 IAENG International Con-
ference on Control and Automation, Hong Kong, 2123 March.
70 R. Vilanova / Journal of Process Control 18 (2008) 6170