Control Engineering Practice 27 (2014) 1122

Contents lists available at ScienceDirect

Control Engineering Practice

journal homepage: www.elsevier.com/locate/conengprac

An automatic tuning methodology for a unied

dead-time compensator
Julio E. Normey-Rico a,n, Rafael Sartori a, Massimiliano Veronesi b, Antonio Visioli c
a

Dep. de Automao e Sistemas, Universidade Federal de Santa Catarina, 88040-900 Florianpolis, SC, Brazil
Yokogawa Italia srl, Milan, Italy
c
Dipartimento di Ingegneria Meccanica e Industriale, University of Brescia, Italy
b

art ic l e i nf o

a b s t r a c t

Article history:
Received 6 August 2013
Accepted 1 February 2014
Available online 7 March 2014

In this paper, an automatic tuning methodology for a modied Smith predictor control scheme is
proposed. The main feature of the procedure is that it is applied in closed-loop (by either evaluating a
set-point or a load disturbance step response) and it is suitable for self-regulating, integral and unstable
processes. Further, the process parameter estimation technique is based on the evaluation of the integral
of signals, thus making it inherently robust to measurement noise. Simulation and experimental results
demonstrate the effectiveness of the methodology.
& 2014 Elsevier Ltd. All rights reserved.

Keywords:
Smith predictor
Dead-time compensators
Robustness
Automatic tuning
Performance assessment

1. Introduction
Dead time compensator control schemes have been widely
investigated in the last 50 years because of the need of obtaining a
higher and higher performance even with processes with a large
dead time, which are well known to be difcult to control by
means of standard feedback control systems. In fact, processes
with a signicant dead time are frequently encountered in
industry. The time delay can be due to the transportation of
material, energy, information, or can be introduced by the sensor,
or can appear as a result of the series of a large number of loworder systems (Normey-Rico & Camacho, 2007; Palmor, 1996).
Many modications have been proposed starting from the
Smith predictor scheme (Smith, 1957), by trying to overcome its
drawbacks (in particular, its poor robustness and its incapability to
deal with processes that are not self-regulating). See for example
the works (strm, Hang, & Lim, 1994; Matauek & Mici, 1996,
1999; Normey-Rico & Camacho, 2002; Zhong & Normey-Rico,
2002) based on simple modications of the SP for stable and
integrative models or the ones in Kwak, Whan, and Lee (2001),
Kaya (2003), Chien, Peng, and Liu (2002), Hang, Wang, and Yang
(2003), Tan, Marquez, and Chen (2005), Liu, Cai, Gu, and Zhang
(2005), Liu, Zhang, and Gu (2005), Lu, Yang, Wang, and Zheng
(2005), and Rao and Chidambaram (2005) for the unstable case.
n

Corresponding author.
E-mail addresses: julio.normey@ufsc.br (J.E. Normey-Rico),
sartori.rafael@grad.ufsc.br (R. Sartori),
max.veronesi@it.yokogawa.com (M. Veronesi), antonio.visioli@unibs.it (A. Visioli).
http://dx.doi.org/10.1016/j.conengprac.2014.02.001
0967-0661 & 2014 Elsevier Ltd. All rights reserved.

A review of these techniques can be found in Normey-Rico and

Camacho (2007). A relevant methodology in this context has been
proposed in Normey-Rico and Camacho (2009), which consists in
using a modied structure of the Smith predictor that allows the
decoupling of the disturbance rejection and set-point following
tasks. In this way, the controller can be tuned in order to achieve a
compromise between performance and robustness. It has to be
stressed that the same approach can be used for self-regulating,
integral and unstable processes. All these features, together with
its overall simplicity, make this methodology very suitable to be
applied in industrial settings.
However, it has also to be recognized that, for its widespread
use in industry, an industrial controller should be equipped with
the automatic tuning functionality, which allows the user to put
the controller in place without the need of (possibly time consuming) trial-and-error procedures. Actually, the presence of wellestablished automatic tuning techniques is one of the reasons of
the great success of Proportional-Integral-Derivative (PID) controllers (strm & Hgglund, 2006; Visioli, 2006). For this reason,
different methods have been proposed for the automatic tuning of
Smith predictor based control schemes (Kaya, 2003, 2004; Majhi &
Atherton, 2000; Matausek & Kvascev, 2003; Tan, Chua, Zhao, Yang,
& Tham, 2009). However, all of them have been applied to control
schemes which address particular issues of the dead time compensators. Further, they all need a special experiment in order to
determine the process parameters upon which the controller
is tuned.
In this paper we propose an automatic tuning procedure which
can be applied to the unied design of the dead-time compensator

12

J.E. Normey-Rico et al. / Control Engineering Practice 27 (2014) 1122

Fig. 2. The equivalent ltered Smith predictor scheme.

Fig. 1. The considered ltered Smith predictor scheme.

presented in Normey-Rico and Camacho (2009). The devised

methodology suitably extends the process estimation parameters
technique that has already been applied to the automatic tuning of
PID controllers (Veronesi & Visioli, 2009, 2010a,b) and of cascade
control systems (Veronesi & Visioli, 2011). In particular, the
process parameters are obtained by evaluating routine operating
data with the dead-time compensator control structure (possibly
roughly tuned) already in place. In this context, both set-point and
(measurable) load disturbance step responses can be evaluated.
Integral of signals are employed so that the technique is inherently
robust to measurement noise. Then, if the performance is not
satisfactory (in this context, performance assessment indexes have
been suitably devised), the dead-time compensator can be suitably
retuned by considering the trade-off between performance and
robustness.
The paper is organized as follows. In Section 2 the unied
approach for the robust dead time compensator design is reviewed
by highlighting the three different cases related to self-regulating,
integral, and unstable processes. Then, in Section 3 the parameters
estimation procedure is described for the considered kinds of
processes and for both the set-point and load disturbance step
responses. Performance assessment indexes and tuning rules are
proposed in Section 4. Simulation results are then shown in
Section 5 and experimental results are presented in Section 6.
Conclusions are drawn in Section 7.

From (Eqs. (1) and 2) and a proper tuning, C(s) and F(s) can be
used to dene the set-point response while Fr(s) can dene the
characteristics of the disturbance response in such a way that Hr(s)
and Hd(s) can have different closed-loop poles, therefore decoupling the set-point and disturbance rejection responses. (Details
on the design of the control scheme are presented in Normey-Rico
& Camacho, 2009.) Hereafter, we briey review the most relevant
cases which will be exploited in the automatic tuning procedure
that is the subject of this paper, that is, processes that can be
modelled by stable, integrative or unstable rst order plus deadtime systems. It is in any case worth noting that the control
structure shown in Fig. 1 is used for analysis while, in order to cope
with an internal instability when integral or unstable processes
are to be controlled, alternative implementation should be applied
(Palmor, 1996; Watanabe & Ito, 1981). In particular, as dead-time
compensator structures are only implemented in digital devices, a
simple implementation of the FSP in the discrete time domain can
be exploited considering the scheme in Fig. 2, where the discretetime representation of Ceq(s) is a simple rational function in z
(Normey-Rico & Camacho, 2009).

2.2. Self-regulating processes

Self-regulating (asymptotically stable) processes can be modelled, as it is industrial practice, as a rst-order-plus-dead-time
(FOPDT) transfer function, that is
P n s

Km
e  Lm s :
T m s 1

2. The modied Smith predictor

In this case, a convenient strategy is to select C(s) as a PI controller

2.1. Control scheme

Cs K p

The control scheme considered in this paper is the ltered Smith

predictor (FSP) originally proposed in Normey-Rico, Bordons, and
Camacho (1997) and shown in Fig. 1, where P(s) is the true process
transfer function, P n s Gn se  Lm s is the process model (i.e., Gn(s) is
the delay-free model and Lm is the estimated value of the dead time)
and C(s) is the feedback controller. Note that, with respect to the
classical Smith predictor, two additional lters are employed. In
particular, the set-point lter F(s) is used to improve the set-point
response and the predictor lter Fr(s) is used to improve the predictor
properties in such a way that eliminates all the drawbacks of the
original Smith predictor when controlling lag-dominant or unstable
plants.1
In fact, in case of a perfect process model, that is, P n s Ps,
the nominal closed-loop transfer functions are

with Ti Tm and K p T=K m T r where Tr denes the nominal

closed-loop pole of H r s e  Lm s =1 sT r (in this case F(s)1).
To decouple the set-point and disturbance responses,

H r s
and
H d s

Ys FsCsPs

Rs 1 CsGn s


Ys
F r sCsPs
Ps 1 
:
Q s
1 CsGn s

F r s

T is 1
T is

T r s 11 s 1
T 0 s 12

is used, where 1 T m 1  1  T 0 =T m 2 e  Lm =T m , giving H d s P n

s1  1 s 1 =1 sT 0 2 e  Lm s  that has not a pole at s  1=T m
(Normey-Rico & Camacho, 2009). It has to be noted that Tr
determines the performance with respect to the set-point step
response while T0 handles the trade-off between performance (in
the load disturbance rejection task) and robustness of the control
system (this point is analyzed with more details in Section 4.2).

2.3. Integral processes

1
For stable plants, SP cannot achieve a faster disturbance rejection response
than the open-loop one. For unstable plants, Fr(s) allows to obtain an internally
stable system, which cannot be obtained with the standard SP.

In case of an integral process, the typical model applied for

tuning purposes is the integrator plus dead time (IPDT) transfer
function P n s K m =se  Lm s . Using Cs K p 1=K m T r , Fs 1
and Fr(s) as in (5) with 1 2T 0 Lm the same closed-loop
characteristics (Hr(s) and Hd(s)) as in the stable case are obtained.
Tr and T0 have the same physical meaning of the self-regulating
processes case.

J.E. Normey-Rico et al. / Control Engineering Practice 27 (2014) 1122

2.4. Unstable processes

When the process is unstable, it can be conveniently modelled
as an unstable rst-order-plus-dead-time (UFOPDT) transfer function P n s K m =T m s 1e  Lm s . To obtain the same Hr(s) and Hd(s)
as in the stable case, a PI controller (4) with K p T r 2T m =K m T r
and T i T r 2 T r =T m together with a set-point lter Fs
T r s 1=T i s 1 and a predictor lter F r s T r s 12 1 s 1=
T i s 1T 0 s 12 with 1 T m 1 T 0 =T m 2 eLm =T m  1 are used.
Again, the trade-off between performance and robustness is
handled by Tr and T0 for the set-point following and load
disturbance rejection tasks separately.

should be equal to the amplitude of the measurement noise) and

to rewrite the condition as y4 NB (note that a positive step is
considered without loss of generality).
The process gain K can then be estimated by considering the
following relation (which follows directly from the nal value
theorem) where u is the control variable:
As
lim ut
K t- 1
lim sC eq sEf s
s-0

Kp
lim E s
T i K p K m 2T 0  1  T r Lm s-0 f

Ef s
From the analysis presented in the previous sections some
important conclusions can be derived: (i) the FSP control structure
and tuning gives a unied solution for the control of dead-time
processes modelled by simple transfer functions; (ii) FSP has ve
(four in the integrative case) tuning parameters that can be
grouped in two sets: model parameters K m ; Lm ; T m and specication parameters T r ; T 0 . Thus, the following sections propose some
methodologies to obtain the process model, to asses the closedloop performance and to tune T r ; T 0 for a robustness/performance
trade-off.

1 As
 Ys:
F r s s

Taking into account the primary controller tuning (Ti Tm and

K p T=K m T r ), Eq. (8) is simplied as
K pT r
As

lim E s:
K T m 2T 0  1 Lm s-0 f

10

Therefore
K

As
:
R1
K pT r
0 ef t dt
T m 2T 0  1 Lm

11

Regarding the estimation of the time constant T, the following

variable can be considered:

3. Estimation of model parameters

As already pointed out, in order to design the controller,
parameters K m , Lm , T m need to be estimated. This can be done
by using set-point or load disturbance step response data of
normal plant operation in closed-loop with the FSP. It is considered here that the controller has been tuned but its performance is
not satisfactory, therefore, the process parameters must be estimated rst and then the controller can be conveniently retuned.
The different cases for the process parameters estimation can
be indeed dealt with by suitably extending the methodology
already proposed in Veronesi and Visioli (2009, 2010b). In this
context it is worth noting that the control scheme of Fig. 1 is
equivalent to a standard unity-feedback control system with a setpoint lter as shown in Fig. 2 where
CsF r s
1 CsGn s1  F r se  Lm s

Fs
:
F r s

and
F eq s

where

2.5. Discussion

C eq s

13

Hereafter we will consider the three different kinds of processes

separately and for each case the parameter estimation will be
addressed by evaluating either a set-point step or a load disturbance step response. Null initial conditions are assumed in
all cases.
3.1. Self-regulating processes
Assuming a FOPDT model Ps K=Ts 1e  Ls , in the presence
of a set-point step signal of amplitude As the (apparent) dead time
of the process L can be determined by considering the time
interval from the occurrence of the step signal and the time
instant when the process output attains the 2% of the new setpoint value Ar, namely, when the condition y 40:02Ar occurs.
Actually, from a practical point of view, in order to cope with the
measurement noise, a simple sensible solution is to dene a noise
band NB (strm, Hgglund, Hang, & Ho, 1993) (whose amplitude

eu tKut yt:

12

Then, we can obtain the following expression, which allows the

determination of the sum of the time constant and of the dead
time of the process:
Z t
1
eu v dv lim s Eu s
lim
t-1 0
s-0 s
F eq sC eq sK  Ps As
lim
1 C eq sPs
s-0
s
As T L:
13
Thus, the time constant of the process can be estimated as
Z
1 1
eu t dt  L:
T
As 0

14

A similar procedure can be applied also in the presence of a step

load disturbance when the time instant of its occurrence is known.
In this case the dead time of the process can be estimated in the
same way as for the set-point response case. Then, the amplitude
Ad of the load disturbance can be estimated by considering that
Z t
1
A
yv dv lim s Ys d
lim
t-1 0
s-0 s
s
Ps
Ad
lim
s-01 C eq sPs s
Ad

T i K p K m 2T 0  1  T r Lm
Kp

15

that can be simplied taking into account the primary controller

tuning to obtain
Z t
T m 2T 0  1 Lm
lim
yv dv Ad
16
t-1 0
K pT r
and therefore, we have
Z 1
K pT r
Ad
yt dt
T m 2T 0  1 Lm 0

17

The process gain can be determined by computing the integral of

the variable it ut dt, where d(t) is the constant signal of

14

J.E. Normey-Rico et al. / Control Engineering Practice 27 (2014) 1122

amplitude Ad. In fact, we have

Z

lim

t-1



1
1
Ad Ad T i

iv dv lim s
K m 2T 0  1  T r Lm
s-0 s 1 C eq sPs s
K Kp

18
and therefore


Ti
Ad
K m 2T 0  1  T r Lm
Kp
R1
K
0 it dt

19

Finally, the time constant of the process can be determined by

considering the variable
ei t Kit yt

20

so that we have
Z tZ
1
lim
ei d d lim s 2 Ei s
t- 1 0
s-0 s
0
1 K  Ps Ad
lim
s-0 s 1 C eq sPs s

21

22

3.2. Integral processes

A procedure similar to that presented in the previous subsection can be also applied, with appropriate modications, to
integral processes described by an IPDT transfer function
Ps K=se  Ls . In case a set-point step (of amplitude As) response
is evaluated, the process gain can be estimated by considering that
As
:
K R1
0 ut

23

Then, the dead time can be determined by means of the following

variable:
Z t
uv dv  yt:
24
eiu tK
0

Indeed, if we compute the limit (for t-1) of the integral of this

variable, we obtain
Z t
1 C eq sK sPs Ar
eiu v dv lim s 2
lim
t-1 0
1 C eq sPs s
s-0 s
25

As L
so that, it trivially results
R1
eiu t dt
:
L 0
As

26

If a load disturbance step response is evaluated, the amplitude of

the step signal can be determined by considering the integral of
the process output y(t). In fact, we have
Z t
1
Ps
Ad 1
Ad K m L2m 4Lm T 0 2T 20
yv dv lim s
lim
t-1 0
s-0 s 1 C eq sPs s
2
27
so that
R1
Ad 2

yt dt

K m L2m 4Lm T 0 2T 20

Ad
K m L2m 4Lm T 0 2T 20
2K

28

29

from which it is trivial to obtain

K

Ad K m L2m 4Lm T 0 2T 20
R1 R
2 0 0 i d dt

30

Finally, the dead time of the process is estimated by considering

Rt
the double integral of the variable eii t K 0 i d  yt. In fact,
we have
Z tZ
1 K  sPs Ad
lim
eii d d lim s 3
t-1 0
s-0 s 1 C eq sPs s
0
1
LAd K m L2m 4Lm T 0 2T 20 :
2

A
d T i K m K p 2T 0  1  T r Lm T L
Kp
and therefore
R1 Rt
Kp
0
0 ei d dt
L
T
Ad T i K m K p 2T 0  1  T r Lm

The gain of the process can then be estimated by computing the

double integral of the variable i(t). Indeed, it is
Z tZ
1
1
Ad
lim
i d d lim s 2
t-1 0
s-0 s 1 C eq sPs s
0

Hence, the value of the dead time results to be

R1 Rt
2 0 0 eii d dt
L
Ad K m L2m 4Lm T 0 2T 20

31

32

3.3. Unstable processes

Finally, the case of unstable processes described by the UFOPDT
model K=Ts  1e  Ls is addressed.
Regarding the set-point response, the dead time value L can be
determined as in Section 3.1. Then, the process gain can be
estimated by applying the same procedure as in (8)(11), thus
obtaining
As
 lim ut  lim sC eq sEf s
t- 1
K
s-0
Kp
lim Ef s

K p K m 2T 0  1 2T r Lm T i  T i s-0

33

where
Ef s

Fs As
 Ys;
F r s s

34

giving the gain value as

K

As
:
R1
Kp
0 ef t dt
K p K m 2T 0  1 2T r Lm T i  T i

35

The time constant can be determined by using the variable

euu t Kut yt

36

as it results (see (12))

Z t
euu v dv As L  T;
lim

37

t-1

that is,
T L

1
As

1
0

euu t dt:

38

If a load disturbance step response is evaluated, the dead time L is

determined as mentioned above and the amplitude Ad of the
disturbance can be estimated by calculating (as in (15)), the
integral of the process variable, that is,
Z t
T  K p K m 2T 0  1  2T r Lm T i
lim
yv dv Ad i
:
39
t-1 0
Kp

J.E. Normey-Rico et al. / Control Engineering Practice 27 (2014) 1122

the controller with a proper rule. Hereafter, these points are

discussed.

Hence, we have

R1
K p 0 yt dt
:
Ad
T i  K p K m 2T 0  1  2T r Lm T i

40

Regarding the process gain, this can be determined by means of

the integral of i(t) (see (18)):
Z t
A K m K p Lm 2T 0 T i  2T r  1  T i
lim
iv dv d
41
t-1 0
K
Kp
which yields
K

Ad K m K p Lm 2T 0 T i  2T r  1  T i
R1
:
K p 0 it dt

42

Finally, the time constant of the process can be obtained by

considering the variable
eui t  Kit  yt

43

so that we have
Z

tZ

lim

t- 1

eui d d

Ad
T  K m K p Lm 2T 0 T i  2T r  1 L  T
Kp i

44
and therefore
T L

R1 Rt
Kp
0
0 v d dt
:
Ad T i  K m K p Lm 2T 0 T i  2T r  1

15

45

4.1. Performance assessment

In order to evaluate the performance of a given control system,
it is usual to employ a performance index which provides a
comparison between the actual performance and a benchmark
one which represents the desired behavior of the control system.
The desired performance can be dened as that provided by
the control scheme of Fig. 1 by assuming that a perfect process
model is available (i.e., K m K, T m T, Lm L). In this context, a
signicant index is the integrated absolute error dened as
Z 1
jetj dt
46
IAE
0

as its minimization yields, in general, a low overshoot and a low

settling time at the same time (Shinskey, 1994). In order to
calculate the IAE benchmark value, the transfer functions (1) and
(2) can be exploited for the set-point following and load disturbance rejection tasks respectively.
Regarding the set-point following task, we have that, by
employing the design methodology described in the previous
section, the closed-loop transfer function becomes, for FOPDT,
IPDT and UFOPDT processes:
H r s

3.4. Discussion
At this point it is worth highlighting some relevant features of
the proposed estimation procedure. First, the result obtained does
not depend on the initial tuning of the dead time compensator if
the closed-loop system is stable, that is, of the evaluated transient
responses, as the process parameters are obtained by using
steady-state values of signals. This is a signicant difference from
the well-known standard least squares approach. Further, the use
of integrals of signals makes the overall methodology inherently
insensitive to measurement noise (with the exception of the dead
time estimation for self-regulating and unstable processes, for
which in any case the standard noise band technique has already
been proven to be effective). The only parameter whose estimation
depends on the initial tuning is the dead time for the FOPDT and
UFOPDT, as a very sluggish controller can yield an overestimation
of the dead time. It has to be in any case taken into account that
the overall procedure can be repeated (by evaluating new step
responses after the controller has been retuned) until a satisfactory performance is achieved.
Then, the overall methodology makes sense also in the presence of high-order dynamics. In fact, if additional time constants
are present in the process model, it can be easily shown that
formulae (13), (21), (25), (32) allow the estimation of the sum of
the time constants and of the dead time of the process, while
formula (37) allows the estimation of the difference between the
dead time and the sum of lags. This is in accordance with the halfrule proposed in Skogestad (2003) for the model reduction of the
process and therefore the estimated variable is useful for the
tuning of the control system. Example 2 of Section 5 illustrates
this case.

4. Performance assessment and tuning

Having a methodology to estimate the model parameters, it is
important to dene a procedure to measure how far form the ideal
closed-loop performance the system is, and, if necessary, to retune

1
e  Lm s
T r s 1

47

Thus, by considering a step signal of amplitude Ar applied to the

set-point, and by considering that the step response of the system
(47) is monotonic, the integrated absolute error can be easily
calculated analytically as
IAEsp Ar T r Lm :

48

Regarding the load disturbance response, the transfer function to

be considered is (2) which can be rewritten, by considering the
design methodology outlined in the previous section, as
"
#
1 s 1  Lm s
e
H d s P n s 1 
:
49
T 0 s 12
Also in this case the value of the integrated absolute error when a
step signal of amplitude Ad is applied to the load disturbance input
can be therefore calculated analytically and in the three cases we
obtain the following expressions:
IAEload K m Ad 2T 0 Lm  T m T m 1 T 0 =T m 2 e  Lm =T m 

50

for FOPDT processes,

IAEload

K m Ad 2
Lm 4Lm T 0 2T 20
2

51

for IPDT processes, and

IAEload K m Ad 2T 0 Lm T m T m 1 T 0 =T m 2 eLm =T m :

52

for UFOPDT processes.

The performance index can be therefore determined after the
occurrence of a (set-point or load disturbance) step response (that
is, based on routine operating data) as the ratio between the target
integrated absolute error and the actual one. In this context, it has
to be stressed that the model parameters Km, Tm and Lm employed
in the performance assessment evaluation must be those determined after the parameters estimation procedure (see Section 3),
as they are the correct ones. The same reasoning applies to the
tuning parameters Tr and T0 which should be those eventually
selected for the retuning phase.

16

J.E. Normey-Rico et al. / Control Engineering Practice 27 (2014) 1122

For the set-point step response, we can therefore dene the

Set-Point Performance Index (SPPI) as
53

while for the load disturbance response we can dene the Load
Disturbance Performance Index (LDPI) as
IAE
LDPIR 1 load :
0 jetj dt

T0=Lm/2

magnitude

IAEsp
jetj dt

T =Lm/4
0

101

T0=3Lm/4
T0=Lm
P

100
101 2
10

54

101

100

101

102

frequency

In principle, the performance obtained by the controller is satisfactory if the determined performance index is equal to one, while
if it is less than one it means that the controller needs to be
retuned. From a practical point of view, however, as modelling
uncertainties and noise are unavoidable, the performance can be
considered to be acceptable if the performance index is greater
than a given threshold equal to 0.8. Obviously, this threshold value
can be modied by the user depending on how tight the control
requirements are.

performance
20
Lm=5

IAEload

SPPIR 1

robustness index
102

15

10

5
0.2

0.4

0.6

0.8

1.2

1.4

1.6

T0/Lm

4.2. Tuning for robustness performance trade-off

Fig. 3. Analysis of T0 tuning for the stable case.

As it has been pointed out, parameter T0 can be chosen to

achieve a compromise between robustness and disturbance rejection. Thus, this robustness performance trade-off can be studied
using the previous computed IAEs and the robust stability
condition.
For the robustness analysis consider that the process is represented by a family of plants P(s) such that Ps P n s1 Ps.
The multiplicative uncertainty, Pj, veries
jPjj r P

8 40

where P is the multiplicative norm-bound uncertainty (Morari &

Zariou, 1989). Considering that the nominal system is stable, the
robust stability condition for the FSP is
j1 CjGn jj
jCjGn jF r jj

P o Ri

8 4 0:

55

Using the proposed tuning rules a unied expression for Ri is

obtained:
Ri

j1 jT 0 2 j
j1 j 1 j

8 40:

56

Note that Ri increases if T0 increase, which yields a better

robustness. At the same time with bigger values of T0 the value of
IAEload increases, giving a slower disturbance rejection response. That
is, T0 can be used to handle the trade-off between robustness and
disturbance rejection capabilities of the system without affecting the
nominal set-point response (note that IAEsp does not depend on T0).
As an example of this robustness/performance trade-off one
case is analyzed hereafter. The effect of the tuning parameter T0 in
Ri and IAEload is studied for normalized plant models with
Km 1 and the delay normalized to the time constant Tm for the
stable case,2 that is Lm =T m . Thus, the analysis is done for different
normalized dead times, therefore the obtained tuning parameter is
also related to Tm. This analysis is interesting because Km does not
affect the relationship between Ri and IAEload but Lm =T m does.
To illustrate the results, model P n s e  5s =1 s is used. Fig. 3
shows Ri and IAEload for four different values of T 0 =Lm . As can be
seen, the impact on performance is almost linear but the robustness increases slowly when bigger values of T0 are used. In
the same gure the modelling error P is also shown for
Ps 1:2e  6s =1:3s 1.
2

Similar results are obtained for the integrative and unstable cases.

4.3. Tuning rules

Using the previous analysis and the results of Section 2, nal
tuning rules can be derived for T0 and Tr. In practice, if some
information of the modelling error is available, T0 can be tuned by
using (55). If not, a suitable default value of T0 can be dened by
using for instance, a robustness measurement such as the sensitivity peak.3 In practice, values of Ms between 1.5 and 3.0 can be
used for reasonable robustness. Thus, the main idea of the section
is to derive simple tuning rules for T0 that give nearly optimal
results, where the optimal values of T0 mean that the desired
value of Ms is obtained. The numerical method proposed to obtain
the optimal solutions consist of varying T0 until the desired Ms is
obtained for a set of model characteristics.
For the stable case, the set of models is dened using an
interval of Lm =T m A 0; 10, that is, considering lag-dominant processes and dead-time dominant ones. Thus, for each model, an
optimization algorithm nds the value of T0 needed to obtain
Ms 1.5. This value is considered the optimal solution. Fig. 4
shows the optimal values obtained by the tuning procedure
(dotted line) for the stable case. Note that they can be reasonably
approximated by the simple rule T 0 Lm =2 which can be used in
practice without any complex computation.
The same procedure can be applied in the other cases, however
considering that integrative and unstable models require a less
conservative tuning. For the integrative case, the simple tuning rule
T 0 Lm approximates very well the numerical solution for Ms 2.2 for
a wide range of Lm as can be seen in Fig. 5. Finally, the simple tuning
rule T 0 Lm also gives reasonable robustness for the unstable case for
small values of Lm =T m , that is, the common case in practice. In Fig. 6
the obtained values for Ms 3 are shown for the normalized unstable
case in dotted line, and the tuning rule T 0 Lm is shown in solid line.
Note that the approximation works well for Lm =T m o 0:6. For processes where Lm =T m 40:6 it is possible to derive more complex
approximations as the T0 increment is not linear.
Finally, Tr, which determines the set-point step response
performance, can be chosen in most cases smaller than T0, as
robust stability is not affected by the acceleration of this response.
Values between 1/4 and 1/2 of T0 can be used as initial tuning for
lag-dominant processes: T r A Lm =4; Lm =2. But also it is possible to

Ms is the peak of the sensitivity function in the frequency domain.

J.E. Normey-Rico et al. / Control Engineering Practice 27 (2014) 1122

Stable Case (Ms=1.5)

17

Unstable Case (Ms=3)

1.4

optimal solution
T0=Lm

optimal solution
T0=Lm/2

4.5

1.2

4
1

3.5

T0/Tm

T0/T

3
2.5

0.8

0.6
2
0.4

1.5
1

0.2
0.5
0

0
0

10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Lm/Tm

L/T
Fig. 4. T0 tuning for the stable case for Ms 1.5 (dotted) and the simple tuning rule
T 0 Lm =2 (solid).

Fig. 6. T0 tuning for the unstable case for Ms 3 (dotted) and the simple tuning rule
T 0 Lm (solid).

Integrative Case (Ms=2.2)

An example referred to a level control task is shown in Fig. 7.

The controller LIC implements the PI algorithm. Its output is sent
to the valve through the LV block. The model is implemented by
the series of the LAG and DLAY blocks while in the SUB block the
output of the model is subtracted from the process variable wired
to the LT input. The result of this difference enters into the series of
to lead/lag (LD/LAG) blocks where the lter is realized. Finally, the
output of the lter and that of the LAG block enter into the PID
where the difference between the two is internally computed and
used as the input to the PI controller.
The process parameters estimation is done in the CALCU block
where free computations can be implemented by using a C-like
programming language. The function of the block is executed at
each cycle time so that the integrals of the process variable and of
the controller output (wired to it) can be computed. Note that the
automatic tuning is done only when the operator runs the
AUTOTUN block, where, depending on which kind of process is
under control (this has to be selected in advance by the operator
by setting one of the global variables of the function), the integrals
computed in CALCU are used in different ways and then the PID
tuning formulae are implemented.
A different form of implementation can be also considered in
practice. In the experimental case study of Section 6 a Foundation
Fieldbus network widely used in industrial process control is
considered, which communicates with the SCADA (Supervisory,
Control And Data Acquisition System) through an OPC (OLE
(Object Linking and Embedding) for Process Control or also Open
Platform Communications) protocol (Mahnke, Leitner, & Damm,
2009). This protocol is nowadays used in several industrial
applications because of its communication facilities. Thus, in this
case the proposed controller (FSP) is implemented in the H1 level
of the Foundation Fieldbus network (FieldbusFoundation, 2010;
SMAR, 2010) while the process parameter estimation and automatic controller tuning are implemented in the SCADA, with the
interchange of data via OPC. The implementation of the FSP using
Foundation Fieldbus functional blocks is similar to the one
described in the DCS of the previous case and it is shown in Fig. 8.

10
optimal solution
T0=Lm

9
8
7

T0

6
5
4
3
2
1
0

10

L
Fig. 5. T0 tuning for the integrative case for Ms 2.2 (dotted) and the simple tuning
rule T 0 Lm (solid).

relate Tr with Tm (dening the relationship between the open and

closed-loop time constants). This second tuning is in general more
appropriate for dead-time dominant processes, being Tr Tm a
conservative rule (Hagglung, 1996). Oscillations in the set-point
response but not in the disturbance rejection response are
observed when small values of Tr are used together with big
modelling errors. This indicates that bigger Tr values should be
used for a less oscillatory set-point response.
As it will be shown in the next section these simple tuning
rules for T0 and Tr give good results in the three case studies.
4.4. Implementation
The dead time compensator architecture and the automatic tuning
procedure can be implemented by means of standard software in a
Distributed Control System (DCS). Hereafter the most common case of
stable processes, for which an implementation by considering a
continuous-time domain can be done, is addressed. Simple modications can be done for integral and unstable processes for which the
discrete-time domain should be explicitly considered (see Section 2.1).

5. Simulation results
Illustrative examples are given in this section in order to show
the application of the proposed automatic tuning methodology.

18

J.E. Normey-Rico et al. / Control Engineering Practice 27 (2014) 1122

process variable

Fig. 7. Implementation of the automatic tuning procedure in a DCS (courtesy of Yokogawa Italy).

SCADA

OPC

0.5

10

20

In_1

sp
out

in

control variable

LT.AI out

Bkcal_in
V1.PID
out

V1.PRED

out
LT.TF
in

50

In_2
Cas_in

30

40

50

40
20
0
20
0

10

20

time

Bkcal_out

Fig. 9. Set-point step response for process P 1 s before (dashed line) and after
(solid line) performing the automatic tuning.

Fig. 8. Implementation of the FSP in H1 level of Foundation Fieldbus.

The considered process models are taken from Normey-Rico and

Camacho (2009) so that the results can be better evaluated by
comparing them with others already proposed in the literature.
Note that the presence of noise is not considered in the following
simulations because it will be obviously considered in Section 6
where experimental results will be shown.
5.1. Self-regulating process
As a rst illustrative example, consider a heat exchanger which
can be modelled by means of the following transfer function:
0:12  3s
e
:
6s 1

40

60

V1.AO

P 1 s

30

time

H1-FF

57

Thus, for this case, K 0.12, T 6 and L 3. Initially, it is assumed

that the FSP has been tuned with a wrong model of the plant, thus,
an estimation uncertainty of 10% in the process gain and of
20% in the time constant and in the dead time is introduced,
yielding the following controller parameters: Km 0.132, Tm 7.2
and Lm 3.6. As a consequence (see Section 4.3), T r Lm =4 0:9
and T 0 Lm =2 1:8 and therefore K p T m =T r K m 60:61,
T i T m 7:2 and 1 4:74. The unit set-point step response
obtained with this controller is shown in Fig. 9 as a dashed line.
The resulting integrated absolute error is IAE 6:66. By applying
the parameter estimation procedure, the following parameters are
obtained: Km 0.12, Lm 3.03, Tm 5.96. Based on these values, the
performance index can be calculated as SPPI 0:57. A new tuning

is then applied by setting T r Lm =4 0:757, T 0 Lm =2 1:515,

Kp 65.53, Ti 5.93 and 1 3:96. The step response obtained with
the retuned controller is shown in Fig. 9 as a solid line where the
signicant improvement of the performance appears. This is
conrmed by the value of IAE 3:82 and by the new performance
index SPPI 0:99.
If a unit load disturbance step response occurs with the initial
controller, the results are that plotted in Fig. 10 (dashed line),
where IAE 0:35. In this case, the estimated parameters are Ad 1,
Km 0.12, Lm 3.06, Tm 6.25 which yield LDPI 0:71. The new
tuning is then determined by selecting T r Lm =4 0:765,
T 0 Lm =2 1:53, Kp 67.31, Ti 6.25 and 1 4:06. The unit load
disturbance step response obtained with the automatically tuned
controller is shown again in Fig. 10 (solid line) where it is
IAE 0:25 and LDPI 1. The effectiveness of the proposed automatic tuning technique appears also in this case.
5.2. Integral process
As a second illustrative example, consider a level control in a
chain of evaporators. The process is modelled by means of the
following transfer function:
P 2 s 

0:1
s2s 15

58

An IPDT transfer function is initially considered with K m  0:08

and Lm 16. Hence (see Section 4.3), T r Lm =4 4 and T 0 Lm
16 and therefore K p 1=K m T r 3:125 and 1 48. The unit

0.06

process variable

process variable

J.E. Normey-Rico et al. / Control Engineering Practice 27 (2014) 1122

0.04
0.02
0
0.02

1
0.8
0.6
0.4
0.2
0

10

20

30

40

19

50

50

100

150

control variable

control variable

time

0.5
1
1.5

10

20

30

40

50

process variable

1
0.8
0.6
0.4
0.2
50

100

150

200

150

200

time
control variable

2
1
0
1

50

100

time
Fig. 11. Set-point step response for process P 2 s before (dashed line) and after
(solid line) performing the automatic tuning.

400

250

300

350

400

4
2
0
0

50

100

150

200

Fig. 13. Set-point step response for process P 3 s before (dashed line) and after
(solid line) performing the automatic tuning.

technique applied to this step response yields K m  0:10 and

Lm 13.0 so that the performance index results SPPI 0:66. The
new tuning is therefore T r Lm =4  3:25 and T 0 Lm 13:0 and
K p  3:05. The obtained improvement can be evaluated by seeing
the step response obtained with the retuned controller plotted in
Fig. 11 as a solid line. It results IAE 18:14 and SPPI 0:90.
The case of a step load disturbance of amplitude Ad  0:1 is
then considered. The result obtained with the initial controller is
shown in Fig. 10 (IAE7.18). The estimated parameters are
Ad 0:1, K m  0:1 and Lm 10.0, which yield LDPI 0:49. Thus,
a new tuning is obtained by setting Tr 2.5, T0 10.0, K p  4:0 and
1 30:0, and the corresponding load disturbance step response is
shown as a solid line in Fig. 12. It results IAE 3:50 (and LDPI 1)
which conrms the improvement of the performance.
5.3. Unstable process
In order to consider the case of an unstable process, the
following linearized model of a chemical reactor is employed:
P 3 s

process variable

350

2
3

0.1

0.05

0
0

50

100

150

200

150

200

time
0.15

control variable

300

time

Fig. 10. Load disturbance step response for process P 1 s before (dashed line) and
after (solid line) performing the automatic tuning.

250

time

200

time

0.1
0.05
0

50

100

time
Fig. 12. Load disturbance step response for process P 2 s before (dashed line) and
after (solid line) performing the automatic tuning.

set-point step response obtained with this controller is somewhat

oscillatory, as it appears in Fig. 11 (dashed line). The resulting
integrated absolute error is IAE 24:73. The parameter estimation

3:433
e  20s
103:1s  1

59

(note that K 3.433, T 103.1 and L 20). Initially, an estimation

uncertainty of  10% in the process gain, of 50% in the time
constant and of 10% in the dead time is introduced, yielding the
following model parameters: Km 3.087, Tm 51.55 and Lm 22. By
tuning the controller according to the rules mentioned in Section
4.3, the following values are obtained: T r Lm =4 5:5 and
T 0 Lm 22
and
therefore
K p T r 2T m =K m T r 6:40,
T i T r 2 T r =T m 11:59 and 1 109:2. The unit set-point step
response obtained with this controller is shown in Fig. 13, where
IAE 67:49. The application of the parameter estimation procedure yields Km 3.43, Lm 20.2, Tm 103.1. Based on these values,
the performance index results SPPI 0:37. The new tuning is then
applied by setting Tr 5.05, T0 20.2, Kp 12.2, Ti 10.35 and
1 76:27. The corresponding set-point step response is plotted
in Fig. 13 where it appears that both the overshoot and the rise
time are signicantly reduced. The IAE value is also reduced to
25.22 and the new performance index SPPI is equal to one.
If a unit load disturbance step response occurs with the initial
controller, the results are that plotted in Fig. 14, where IAE 89:20.
In this case, the estimated parameters are Ad 1, Km 3.44,
Lm 20.7, Tm 99.1 which yields LDPI 0:66. Based on the estimated process parameters, the controller is designed by setting
Tr 5.175, T0 20.7, Kp 11.44, Ti 10.62 and 1 79:36. The unit

J.E. Normey-Rico et al. / Control Engineering Practice 27 (2014) 1122

load disturbance step response obtained with the retuned controller is shown in Fig. 14 where IAE 59:76 and LDPI 0:99. The
effectiveness of the proposed automatic tuning technique is therefore demonstrated also for unstable processes.

process variable

20

0.5

6. Experimental results

50

100

150

200

250

300

350

400

250

300

350

400

time
0

control variable

To illustrate the application of the presented re-tuning procedure, the level control of a pilot plant of the Automation and
Systems Department (DAS) of Federal University of Santa Catarina
(UFSC) in Brazil is shown in this section. Fig. 15 shows a photo of
the pilot plant. All the instrumentation and control devices are
connected by a Foundation Fieldbus network to the SCADA that
runs in a PC. In the H1 level of the Foundation Fieldbus network
runs the FSP controller while the estimation parameters and
automatic tuning runs in the SCADA. Both systems communicate
using OPC protocol.
As can be seen in the scheme of Fig. 16 the inlet water is driven
by pump B1 from the storage tank to tanks 1, 2 and 3 that are
connected in series. A controlled pneumatic valve (V1) is used to
control the inlet ow and manual valves regulate the output ow.
The water passes from tank 1 to tank 2 before entering in tank
3 where the level sensor is installed. Tank 3 is a conic tank,
therefore volume varies nonlinearly with level, making level
control more difcult. Manual valves can be used to change the
operating conditions of tanks 2 and 3 and to include unmeasurable
disturbances. In the case of level control analyzed in this work the
manipulated variable is the aperture of valve V1 and the process
variable is the level at tank 3. The open-loop behavior of the level
can be approximated by a FOPDT model near an operating point.
Because of the conic shape of the tank this model varies with the
operating level.
Two experiments were performed, one using a change in the
level set-point and a second one introducing a step disturbance in
the process. These tests were performed in different days and also
with different operation conditions of the plant, which is used for
several control applications in the laboratory. For the rst case, the
FSP controller was initially tuned using the proposed tuning rules
and a set of approximated values of the model parameters:
Lm 80 s, Km 2.7 and Tm 165 s, obtaining T 0 Lm =2 40 s,
T r Lm =2 40 s which gives Kc 1.53, Ti 165 s and 1 106:88.
The obtained response for a change in the set point of the level of
10% (from 50% to 60%) is shown in dashed lines of Fig. 17 obtaining
IAE 2419. As can be seen, as the used model is not a good
approximation of the real process, the obtained responses can be
improved. Although, the robust tuning of the controller allows for
a stable behavior. Using the data of this rst experiment and
applying the parameter estimation procedure, new model parameters are obtained: Km 1.61, Lm 100, Tm 261. Based on these
values, the performance index can be calculated as SPPI 0:62.
A new tuning is then applied by setting T r Lm =2 50,
T 0 Lm =2 50, Kp 3.23, Ti 261 and 1 114:7. The step
response obtained with the retuned controller is shown in
Fig. 17 as a solid line where the signicant improvement of the
performance appears, in spite of the fact that output and control
signals exhibit noise. This is conrmed by the value of IAE 1840
(76% of the initial value) and by the new performance index
SPPI 0:81.
In the second experiment, the initial tuning of the FSP controller was dened using the following approximated values of the
model parameters: Tm 110 s, Km 1.96 and Lm 144 s, obtaining
T 0 Lm =2 72 s, T r Lm =4 36 s which gives Kc 1.5646, Ti 110s
and 1 106:77. The obtained response for a step disturbance
applied in the input ow (a 10% step disturbance was applied
manually in the control action) is shown in dashed lines of Fig. 18

0.5
1
1.5
2

50

100

150

200

time
Fig. 14. Load disturbance step response for process P 3 s before (dashed line) and
after (solid line) performing the automatic tuning.

Fig. 15. Pilot plant where experiments have been performed.

V3

LT

Tank 2

B1

V1

Tank 1

Tank 3

V2

Fig. 16. Scheme of the pilot plant.

obtaining IAE 3596:8. It can be observed that a sluggish response

is obtained, thus, to improve the performance a new model will be
estimated. Using the data of this rst experiment and applying the
parameter estimation procedure, new model parameters are
obtained: Km 1.92, Lm 105, Tm 156.55. Note that, as expected,
because of the different operation plant conditions, the estimated
process model is different from the one computed in the set-point
case. In this procedure the estimated value of the step disturbance
was 10.0675. Based on these values, the performance index can be

process variable (%)

J.E. Normey-Rico et al. / Control Engineering Practice 27 (2014) 1122

Acknowledgments

65
60

J.E. Normey-Rico and R. Sartori were funded by CNPq-Brasil and

Agncia Nacional do Petrleo, Gs Natural e Biocombustiveis
(ANP), Brazil, under Project aciPG-PRH No. 34 ANP/MCT.

55
initial tuning
final tuning

50
45

200

400

600

800

1000

1200

1400

References

control variable (%)

time (s)
100
initial tuning
final tuning

80
60
40

200

400

600

800

1000

1200

1400

time (s)
Fig. 17. Initial set-point response (dashed) and nal one (solid) for the FSP
controller.

process variable

15
initial tuning
final tuning

10
5
0
5

200

400

600

800

1000

time(s)

control variable

5
initial tuning
final tuning

0
5
10
15

21

200

400

600

800

1000

time(s)
Fig. 18. Initial disturbance response (dashed) and nal one (solid) for the FSP
controller.

calculated as SPPI 0:4743. A new tuning is then applied by

setting T r Lm =4 26:25, T 0 Lm =2 52:5, Kp 3.10, Ti 156.55
and 1 121:18. The step response obtained with the retuned
controller is shown in Fig. 18 as a solid line. A signicant
improvement of the performance is obtained, conrmed by the
value of IAE 1754:9 (approximately 50% of the initial value) and
by the new performance index SPPI 0:9722.

7. Conclusions
This work has presented an automatic tuning methodology for the
ltered Smith predictor control scheme. The tuning procedure uses
data extracted from the closed-loop plant operation and can be used
with stable, integral and unstable processes with dead-time. Simple
calculations based on the evaluation of the integral of signals are used
to estimate the model parameters which are necessary for the
predictor of the FSP. Three simulation case studies have been used
to illustrate the proposed method and experimental results demonstrate the effectiveness of the methodology.

strm, K. J., & Hgglund, T. (2006). Advanced PID control. Research Triangle Park,
USA: ISA Press.
strm, K. J., Hgglund, T., Hang, C. C., & Ho, W. K. (1993). Automatic tuning and
adaptation for PID controllers a survey. Control Engineering Practice, 1,
699714.
strm, K. J., Hang, C. C., & Lim, B. C. (1994). A new Smith predictor for controlling a
process with a integrator and long dead time. IEEE Transactions on Automatic
Control, 39(2), 343345.
Chien, I.-L., Peng, S. C., & Liu, J. H. (2002). Simple control method for integrating
processes with long deadtime. Journal of Process Control, 12(3), 391404.
Fieldbus Foundation (2010). Foundation technical specications index.
Hagglung, T. (1996). An industrial dead-time compensating PI controller. Control
Engineering Practice, 4(6), 749756.
Hang, C. C., Wang, Q.-G., & Yang, X.-P. (2003). A modied Smith predictor for a
process with an integrator and long dead time. Industrial and Engineering
Chemistry Research, 42, 484489.
Kaya, I. (2003). Obtaining controller parameters for a new PI-PD Smith predictor
using auto tuning. Journal of Process Control, 13, 465472.
Kaya, I. (2004). Imc based automatic tuning method for pid controllers in a Smith
predictor conguration. Computers and Chemical Engineering, 28, 281290.
Kwak, H., Whan, S., & Lee, I.-B. (2001). Modied Smith predictor for integrating
processes: Comparisons and proposition. Industrial and Engineering Chemistry
Research, 40, 15001506.
Liu, T., Cai, Y. Z., Gu, D. Y., & Zhang, W. D. (2005). New modied Smith predictor
scheme for integrating and unstable processes with time delay. IEE Proceedings
Control Theory and Applications, 152(2), 238246.
Liu, T., Zhang, W. D., & Gu, D. Y. (2005). Analytical design of two-degree-of-freedom
control scheme for open-loop unstable processes with time delay. Journal of
Process Control, 15(5), 559.
Lu, X., Yang, Y.-S., Wang, Q.-G., & Zheng, W.-X. (2005). A double two-degree-offreedom control scheme for improved control of unstable delay processes.
Journal of Process Control, 15(5), 605614.
Mahnke, Wolfgang, Leitner, Stefan Helmut, & Damm, Matthias (2009). OPC unied
architecture. London: Springer.
Majhi, S., & Atherton, D. P. (2000). Obtaining controller parameters for a new Smith
predictor using autotuning. Automatica, 36, 16511658.
Matausek, M. R., & Kvascev, G. S. (2003). A unied step response procedure for
autotuning of PI controller and Smith predictor for stable processes. Journal of
Process Control, 13, 787800.
Matauek, M. R., & Mici, A. D. (1996). A modied Smith predictor for controlling a
process with a integrator and long dead time. IEEE Transactions on Automatic
Control, 41(8), 11991203.
Matauek, M. R., & Mici, A. D. (1999). On the modied Smith predictor for
controlling a process with a integrator and long dead-time. IEEE Transactions
on Automatic Control, 44(8), 16031606.
Morari, M., & Zariou, E. (1989). Robust process control. Englewood Cliffs, NJ:
Prentice Hall.
Normey-Rico, J. E., Bordons, C., & Camacho, E. F. (1997). Improving the robustness of
dead-time compensating PI controllers. Control Engineering Practice, 5(6),
801810.
Normey-Rico, J. E., & Camacho, E. F. (2002). A unied approach to design dead-time
compensators for stable and integrative processes with dead-time. IEEE
Transactions on Automatic Control, 47(2), 299305.
Normey-Rico, J. E., & Camacho, E. F. (2007). Control of dead-time processes. Berlin:
Springer.
Normey-Rico, J. E., & Camacho, E. F. (2009). Unied approach for robust dead-time
compensator design. Journal of Process Control, 19, 3847.
Palmor, Z. J. (1996). The control handbook. Time delay compensation: Smith predictor
and its modications. Boca Raton, USA: CRC Press and IEEE Press.
Rao, A. S., & Chidambaram, M. (2005). Enhanced Smith predictor for unstable
processes with time delay. Industrial and Engineering Chemistry Research, 44,
82918299.
Shinskey, F. G. (1994). Feedback controllers for the process industries. New York, NY:
McGraw-Hill.
Skogestad, S. (2003). Simple analytic rules for model reduction and PID controller
tuning. Journal of Process Control, 13, 291309.
SMAR (2010). Fieldbus tutorial. A foundation eldbus technology overview.
Smith, O. J. M. (1957). Close control of loops with deadtime. Chemical Engineering
Progress, 53(5), 217.
Tan, K. K., Chua, K. Y., Zhao, S., Yang, S., & Tham, M. T. (2009). Repetitive control
approach towards automatic tuning of Smith predictor controllers. ISA Transactions, 48, 1623.
Tan, W., Marquez, H. J., & Chen, T. (2005). IMC design for unstable processes with
time delay. International Journal of Control, 45, 8291.

22

J.E. Normey-Rico et al. / Control Engineering Practice 27 (2014) 1122

Veronesi, M., & Visioli, A. (2009). Performance assessment and retuning of PID
controllers. Industrial and Engineering Chemistry Research, 48, 26162623.
Veronesi, M., & Visioli, A. (2010a). An industrial application of a performance
assessment and retuning technique for PI controllers. ISA Transactions, 49,
244248.
Veronesi, M., & Visioli, A. (2010b). Performance assessment and retuning of PID
controllers for integral processes. Journal of Process Control, 20, 261269.
Veronesi, M., & Visioli, A. (2011). A simultaneous closed-loop automatic tuning
method for cascade controllers. IET Control Theory and Applications, 5, 263270.

Visioli, A. (2006). Practical PID control. London: Springer.

Watanabe, K., & Ito, M. (1981). A process-model control for linear systems with
delay. IEEE Transactions on Automatic Control, 26(6), 12611269.
Zhong, Q. C., & Normey-Rico, J. E. (2002). Control of integral processes with dead
time. Part 1: Disturbance observer-based 2DOF control scheme. Control Theory
and Applications, IEE Proceedings, 149(4), 285290.