Process

11th IFACSystems, including

Symposium Biosystems
on Dynamics and Control of
11th
JuneIFAC
Process Symposium
6-8,Systems,
2016. NTNU, on Dynamics
Trondheim,
including and Control of
Norway
Biosystems
Available online at www.sciencedirect.com
Process
June 6-8,Systems, including
2016. NTNU, Biosystems
Trondheim, Norway
June 6-8, 2016. NTNU, Trondheim, Norway
ScienceDirect
IFAC-PapersOnLine 49-7 (2016) 284–289
PID controller tuning for unstable
PID
PID controller
controller
processes usingtuning
tuning for
for unstable
unstable
a multi-objective
processes
processes
optimisation using
usingdesign a multi-objective
a multi-objective
procedure
optimisation
optimisation design
design procedure
procedure
G. Reynoso-Meza ∗
J.
Carrillo-Ahumada Y. Boada
∗∗ ∗∗∗

G. Reynoso-Meza ∗∗∗ J.
Carrillo-Ahumada
J. Picó ∗∗∗ ∗∗ Y. Boada ∗∗∗
∗∗ ∗∗∗
G. Reynoso-Meza J.
Carrillo-Ahumada
J. Picó ∗∗∗
∗∗∗
∗∗∗
∗∗
Y. Boada ∗∗∗
J. Picó
∗
Pontificia Universidade Católica do Paraná (PUCPR), Brazil.
∗ Pontificia Universidade
∗
Católica do Paraná (PUCPR), Brazil.
(e-mail: g.reynosomeza@pucpr.br).
∗
Pontificia
∗∗ Universidade
Universidad (e-mail: Católica do Paraná
g.reynosomeza@pucpr.br).
del Papaloapan, Instituto (PUCPR),
de Quı́mica Brazil.
Aplicada,
∗∗ Universidad (e-mail: g.reynosomeza@pucpr.br).
del Papaloapan, Instituto de Quı́mica Aplicada,
∗∗ Tuxtepec, Oax., México. (e-mail: jcarrillo@unpa.edu.mx)
∗∗
∗∗∗ Universidad del Papaloapan, Instituto
Tuxtepec,
Instituto Oax., México.
Universitario de (e-mail: ede Quı́mica Aplicada,
jcarrillo@unpa.edu.mx)
Automática Informática Industrial.
Tuxtepec, Oax., México. (e-mail: jcarrillo@unpa.edu.mx)
∗∗∗ Instituto
Universitat Universitario
Politènica de Automática
de València (e-mail: e{jpico,yaboa}@isa.upv.es)
Informática Industrial.
∗∗∗
∗∗∗
Instituto
Universitat Universitario
Politènica de Automática
de València (e-mail: e{jpico,yaboa}@isa.upv.es)
Informática Industrial.
Universitat Politènica de València (e-mail: {jpico,yaboa}@isa.upv.es)
Abstract: Multi-objective optimisation techniques have shown to be a useful tool for controller
Abstract: Multi-objective
tuning applications. optimisation
Such techniques aretechniques
useful when:have 1)shown to be a to
it is difficult useful
find tool for controller
a controller with
Abstract:
tuning
a desirable Multi-objective
applications.
trade-off Such
between optimisation
techniques
conflictive techniques
areobjectives;
useful when:have
or 2)1)shown to be a to
difficult
it is valuable useful
to find tool for additional
a controller
extract an controller
with
tuning
aknowledge applications.
desirable the Such
trade-off
from between
processtechniques
conflictive
by areobjectives;
analysing useful when:
trade-off 1) it
oramong
2) it is
ispossible
difficultcontrollers.
valuable to
to find a controller
extract an this
In with
additional
work,
a desirable
knowledge
we proposefrom trade-off between
the process by
a multi-objective conflictive objectives;
analysing trade-off
optimisation or 2) it is
among possible
design procedure valuable to extract
controllers.
for unstable process, an additional
In using
this work,
PID
knowledge
we proposefrom
controllers. The the process
a multi-objective
provided by
examples analysing
show thetrade-off
optimisation design among
of the possible
procedure
usability controllers.
for unstable
procedure process,
for this kindIn of
this
using work,
PID
process,
we proposedifficult
controllers.
sometimes a multi-objective
The provided
to control; optimisation
examples show the
comparison design procedure
usability
with existing oftuning for
ruleunstable
the procedure process,
for this
methods kind of
provide using PID
process,
promising
controllers.
sometimes
results for this The provided
difficult
tuning examples
to control;
procedure. show the usability of the procedure for
comparison with existing tuning rule methods provide promising this kind of process,
sometimes
results for this difficult
tuning to control;
procedure. comparison with existing tuning rule methods provide promising
© 2016, for
results IFACthis(International
tuning procedure.Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
Keywords: PID controller tuning, multi-objective optimisation, evolutionary algorithms.
Keywords: PID controller tuning, multi-objective optimisation, evolutionary algorithms.
Keywords: PID controller tuning, multi-objective optimisation, evolutionary algorithms.
1. INTRODUCTION As identified in Arrieta et al. (2011), efforts are particu-
1. INTRODUCTION As
larlyidentified
concentrated in Arrieta
in open et loop
al. (2011),
stable efforts
systems;areneverthe-
particu-
1. INTRODUCTION As
lessidentified
larly concentrated
some in Arrieta
critical in open
processes et as
al.continuous
loop (2011),
stable efforts
systems;
stirredaretankparticu-
neverthe-
reac-
larly
less concentrated
tors some in
critical processes
and bioreactors, open
common loop stable
as continuous systems;
in chemicalstirred neverthe-
tank units
processing reac-
less
and some critical processes as unstable
continuous stirred
looptank reac-
Proportional - Integral - Derivative (PID) controllers re- tors tors
and
and
bioreactors,
biological
bioreactors,
common
processes,
common
are in chemical
in chemical
openprocessing
processing
units
systems.
units
Proportional
main as reliable - Integral - Derivative
and practical control(PID) controllers
solutions for severalre- and
Severalbiological
works processes,
have been are unstable
focused open loop
on PID-like systems.
controllers
Proportional
main as reliable
industrial - Integral
processes - Derivative
and practical
(Åström control (PID) controllers
solutions
and Hägglund, for several
2001). re- and
One Several
tuning biological
works
for suchprocesses,
have been are
processes; unstable
focused
nevertheless, open
on PID-like loop
efforts systems.
controllers
to merge
main as reliable and practical control solutions for several Several works have been focused on PID-like controllers
industrial
of the main processes
advantages (Åström
of PID andcontrollers
Hägglund,is2001). One tuning
their ease for suchoptimisation
multi-objective processes; nevertheless,
techniques have efforts
been tonotmerge
yet
industrial processes
of implementation
the main as(Åström
advantages wellofasPID andcontrollers
their Hägglund, is2001).
tuning, giving their One tuning
ease
a good applied for for such
multi-objective processes; nevertheless,
suchoptimisation
instances. techniques have efforts
been tonotmerge
yet
of
of the main
implementation
trade-off betweenadvantages
as wellofasand
simplicity PID their controllers
cost tuning, is their
giving
to implement ease multi-objective
a(Stew-
good applied for suchoptimisation
instances. techniques have been not yet
of implementation as well as their tuning, giving a good In this for
applied paper,such ainstances.
simple multi-objective problem state-
trade-off between2011).
art and Samad, simplicityOwingand to cost to implement
this, research for (Stew-
new In mentthisis paper,
defined afor simple multi-objective
unstable first order plus problemdeadstate-
time
trade-off
art between
andtechniques
Samad, 2011).simplicityOwingand cost to
toresearch implement
this, research (Stew-
for new In thisis paper, afor
simple multi-objective problem state-
tuning is an ongoing topic (Åström ment
(UFOPDT) defined
processes
art and Samad, 2011). Owing to this, research for new ment is defined for unstable first order plus dead time unstable
and first
compared order
with plus dead
existing time
tuning
tuning techniques
and Hägglund, 2005).is Current
an ongoing research
research pointstopic (Åström (UFOPDT)
to guarantee rules. The remainderprocesses and compared
of this paper withis as existing
follows: tuning
firstly
tuning techniques is margins
an ongoing researchas atopic (Åström
overall (UFOPDT) processes and compared with
and
and
Hägglund,
reasonable
Hägglund,
2005).
stability
2005).
Current
Current
research
as well
research
points
points
to guarantee
good
to guarantee rules. The 2remainder
in Section of this
it is presented apaper is as existing
brief backgroundfollows:on tuning
firstly
PID
reasonable
performancestability
for a wide margins
varietyasof well as a good overall rules.
processes. in The
Section
control tuning,2remainder
it UFOPDT of this
is presented apaper
process brief ismulti-objective
as follows:onfirstly
andbackground PID
op-
reasonable stability margins
performance for a wide variety of processes. as well as a good overall in Section 2Afterwards,
it UFOPDT
is presented a brief background on PID
New tuning techniques
performance for a wide are beingoffocused
variety processes. control
timisation.tuning, a process
MOO
on the fulfilment control tuning, UFOPDT process and multi-objective op- and
procedure multi-objective
for is presented op-
New tuning
of several techniques
objectives andarerequirements,
being focusedsometimes in con- timisation.
on the fulfilment in Section 3Afterwards,
timisation.
a MOO procedure
and it is compared and validatedfor is in
presented
Section
New
of tuning
several
flict among techniques
objectives
them (Ang andare being
etrequirements,
al., 2005;focused
Li et on
al.,the
sometimes fulfilment
2006). con- in
inSome Section some
4. Finally, 3Afterwards,
andconcluding a MOO
it is compared procedure
remarks andare for is in
validated
given. presented
Section
of several
flict
tuningamong objectives
them (Ang
procedures areand etrequirements,
basedal., on
2005; Li etsometimes
optimisational., 2006).
statementscon- in
inSome Section some
4. Finally, 3 andconcluding
it is compared remarks andare validated
given. in Section
flict 4. Finally, some concluding remarks are given.
(Ge among
tuning al., them
2002;(Ang
et procedures are et al.,2005;
based
Toscano, 2005;
on Li et al., et
optimisation
Åström 2006).
al., Some
statements
1998; 2. BACKGROUND
tuning
(Ge procedures
et al., 2002;
Panagopoulos et al.,are based
Toscano, on optimisation
2005; Åström
2002; Rajinikanth and et statements
al., 2012)
Latha, 1998;
(Ge et al.,cases
2002; Toscano, 2005; Åström 2. BACKGROUND
Panagopoulos
and some etthey
al., 2002;
are Rajinikanth
solved by means andof et al., 2012)
Latha, 1998;
evolutionary In order to describe
2. BACKGROUND
the tuning approach of this paper,
Panagopoulos
and some cases
algorithms etthey
al., are
(Reynoso-Meza 2002; Rajinikanth
solved
et al.,by2013b).
means andof Latha, 2012)
evolutionary
Recently Multi-
and some cases they are solved by means of evolutionary In
someorder to describein the
preliminaries tuning
control approach
tuning, of this
unstable openpaper,
loop
algorithms (Reynoso-Meza
objective Optimisation et al., techniques
(MOO) 2013b). Recently shown In
have Multi- someorder
process to EMOdescribe
preliminaries
and in the
are tuning
control
required. approach
tuning,
They of this
unstable
are provided open paper,
loop
below.
algorithms
objective (Reynoso-Meza
Optimisation
to be a valuable tool for et
(MOO) al.,
controller 2013b).
techniques Recently
tuning have Multi-
shown some
applications preliminaries in required.
control tuning, unstable open loop
objective
to Optimisation
be a valuable
(Reynoso-Meza et tool for(MOO)
controller
al., 2014b,a). techniques
They enable have
tuning shown process
applications
the designer
and EMO are They are
process and EMO are required. They are provided below.
provided below.
to be a valuable
(Reynoso-Meza
or decision makeret(DM)tool for controller
al., 2014b,a).
having a They tuning
enable the
close embedment applications
into the 2.1 Background on PID controller tuning and unstable
designer
(Reynoso-Meza
or decision
tuning maker
process et al.,
(DM)
since 2014b,a).
it ishaving
possible They
a close
to take enable
embedment the
into accountdesigner
the 2.1
intoeach Background on PID controller tuning and unstable
process
or decision
tuning
design maker
process
objective (DM)
since it ishaving
possible
individually; a close
they to also
takeembedment
into account
enable comparing the 2.1
intoeach Background on PID controller tuning and unstable
process
tuning process since it is possible to take into account each process
design objective
alternatives individually;
(i.e. different theycontrollers),
also enable in comparing
order to A basic control loop is depicted in Figure 1. It comprises
design
select aobjective
design tuning individually;
alternatives the they
(i.e. different
fulfilling also enable
controllers),
expected comparing
in
trade-off order
among to A basic control
transfer functions loopPis(s)depicted
and C(s) in Figure 1. It comprises
of a process and a
design
select aalternatives
conflicting tuning (i.e. different
fulfilling
objectives. the expectedcontrollers), in order
trade-off among to A basic control
transfer
controller functions loopPis
respectively. (s) depicted
The andmajor inaim
C(s) Figure 1. It
ofofathis
process comprises
control and a
loop
select a tuning
conflicting fulfilling the expected trade-off among transfer
objectives. controllerfunctions
respectively. P (s)The and C(s)aimofofathis
major process
control and a
loop
conflicting objectives. controller respectively. The major aim of this control loop
Copyright
2405-8963 ©© 2016,
2016 IFAC 284Hosting by Elsevier Ltd. All rights reserved.
IFAC (International Federation of Automatic Control)
Peer review©under
Copyright 2016 responsibility
IFAC 284Control.
of International Federation of Automatic
Copyright © 2016 IFAC 284
10.1016/j.ifacol.2016.07.287
 IFAC DYCOPS-CAB, 2016
June 6-8, 2016. NTNU, Trondheim, Norway
G. Reynoso-Meza et al. / IFAC-PapersOnLine 49-7 (2016) 284–289 285
is to keep the desired output Y (s) of the process P (s) in
the desired reference R(s).
Fig. 1. Basic control loop.
For this work, process P (s) in figure 1 represents a
UFOPDT process:
K Jj (θ 2 ) < Jj (θ 1 ) for at least one j, j ∈ [1, 2, . . . , m].
P (s) = e−Ls
Ts − 1
where K is the process gain; T the time constant and L
the time lag of the system. Equation (2) shows the transfer
function of the selected structure of the PID controller:
 
1
C(s) = kp 1 + + TD · s
TI · s
where kp is the proportional gain, TI the integral time
(s), TD the derivative time (s); this controller will send
a control signal to the process, according with the error
E(s) = R(s) − Y (s).
The control problem consists in selecting proportional,
k
integral and derivative gains (kp , kI = TpI and kD = kp ·
TD respectively) for the PID controller C(s), in order to
achieve a desirable performance of the process P (s) in the
control loop as well as robust stability margins. Conflictive
objectives may appear, when seeking for a desirable trade-
off between performance and robustness; for this reason,
EMO techniques could be appealing for PID controller
tuning.
2.2 Multi-objective optimisation statement
As referred in Miettinen (1998), a multi-objective problem
(MOP) with m objectives 1 , can be stated as follows:
min J(θ) = [J1 (θ), . . . , Jm (θ)]
θ
subject to:
K(θ) ≤ 0
L(θ) = 0
θi ≤ θi ≤ θi , i = [1, . . . , n]
where θ = [θ1 , θ2 , . . . , θn ] is defined as the decision vector
with dim(θ) = n; J(θ) as the objective vector and K(θ),
L(θ) as the inequality and equality constraint vectors
respectively; θi , θi are the lower and the upper bounds in
the decision space.
It has been noticed that there is not a single solution in
MOPs, because there is not generally a better solution in
all the objectives. Therefore, a set of solutions, the Pareto
1
A maximisation problem can be converted to a minimisation
problem. For each of the objectives that have to be maximised, the
transformation: max Ji (θ) = −min(−Ji (θ)) could be applied.
set, is defined. Each solution in the Pareto set defines an
objective vector in the Pareto front. All the solutions in
the Pareto front are a set of Pareto optimal and non-
dominated solutions:
• Pareto optimality (Miettinen, 1998): An objective
vector J(θ 1 ) is Pareto optimal if there is not another
objective vector J(θ 2 ) such that Ji (θ 2 ) ≤ Ji (θ 1 ) for
all i ∈ [1, 2, . . . , m] and Jj (θ 2 ) < Jj (θ 1 ) for at least
one j, j ∈ [1, 2, . . . , m].
• Dominance (Coello and Lamont, 2004): An objective
vector J(θ 1 ) is dominated by another objective vector
J(θ 2 ) iff Ji (θ 2 ) ≤ Ji (θ 1 ) for all i ∈ [1, 2, . . . , m] and
(1) This is denoted as J(θ2 ) � J(θ1 ).
To successfully implement the multi-objective optimisa-
tion approach, three fundamental steps are required: the
MOP definition, the multi-objective optimisation (MOO)
process and the multi-criteria decision making (MCDM)
stage. This integral and holistic process will be de-
(2) noted hereafter as a multi-objective optimisation design
(MOOD) procedure.
Fig. 2. Multi-objective optimisation design (MOOD) pro-
cedure.
Next, this MOOD procedure will be used in order to find
suitable PID parameters for UFOPDT processes.
3. MOOD PROCEDURE FOR UNSTABLE SYSTEMS
As commented before, for a successful implementation
of the MOOD procedure, the following steps should be
carried out: the MOP definition, the optimisation process
and the MCDM stage. All of them will be clarified within
(3) the PID controller tuning for a UFOPDT framework.
3.1 Multi-objective problem definition
(4) Within this context, the decision variables for the opti-
(5) misation statement are θ = [kp , TI , TD ]. A total of three
design objectives will be stated: one related to perfor-
(6) mance and two related with robustness. In the first case,
the settling time St[s] for a step response will be used;
in the latter case, the inverse 2 of gain and phase mar-
gins, Gm P m respectively. It is possible to incorporate
more design objectives, nevertheless this would lead to
which is known as a many-objectives optimisation instance
(Ishibuchi et al., 2008). Such instances represent a partic-
ular challenge for MOO algorithms, since convergence and
spreading capabilities are usually in conflict. This instance
will be addressed in a future work, and we will focus
on this paper in a multi-objective problem with 3 design
objectives.
2 in order to use an overall minimisation problem statement.
285
IFAC DYCOPS-CAB, 2016
286 G. Reynoso-Meza et al. / IFAC-PapersOnLine 49-7 (2016) 284–289
June 6-8, 2016. NTNU, Trondheim, Norway

1.4 1.4 1.4

JSt (θ) = SettlingT ime98%[s] (7)
1.3 1.3 1.3
1
JGm−1 (θ) = [dB −1 ] (8) 1.2 1.2 1.2
Gm

1 1.1 1.1 1.1

JP m−1 (θ) = [degrees−1 ] (9)
Pm

J(θ)2
1 1 1

ˆ
Therefore, the following optimisation (minimisation) state-
ment is defined: 0.9 0.9 0.9

min J(θ) = [JSt (θ), JGm−1 (θ), JP m−1 (θ)] (10) 0.8 0.8 0.8
θ

0.7 0.7 0.7

subject to:

2 2.5 3 3.5 40 60 80 100 2 4 6 8

kp ∈ [kp , kp ] X1: kp X2: Ti X3: Td

TI ∈ [TI , TI ] (a) Pareto Set Approximation

TD ∈ [TD , TD ] 1.4 1.4 1.4

JSt (θ) ≤ St
1.3 1.3 1.3
JGm−1 (θ) ≤ Gm−1
1.2 1.2 1.2
JP m−1 (θ) ≤ P m−1 (11)
1.1 1.1 1.1
Last constraints are known as pertinency bounds in the
objective space, in order to guarantee practical and reason-
J(θ)2

1 1 1
able controllers in the approximated Pareto front. In order
ˆ

to state such pertinency bounds, an idea on the tolerable

0.9 0.9 0.9
value on design objectives is required. In every instance,
internal stability of the closed loop is also included as
0.8 0.8 0.8
constraint.
0.7 0.7 0.7
3.2 Multi-objective optimisation process
50 100 150 0 0.2 0.4 0 0.02 0.04 0.06
For the optimisation stage, the sp-MODE algorithm J1: Settling time [s] −1
J2: 1/(Gain Margin) [dB ]
−1
J3: 1/(Phase Margin) [deg ]
(Reynoso-Meza et al., 2010) will be used 3 . It is an
algorithm based on the differential evolution algorithm (b) Pareto Front Approximation
(Storn and Price, 1997; Das and Suganthan, 2010) which
uses a spherical grid in order to prune the approximated Fig. 3. Pareto set and front approximations for Cholette
set of solutions, and thus promoting diversity along the example. Selected soluction after the MCDM process
Pareto front approximation. It includes also a mechanism appears within a circle.
to improve pertinency (described in Reynoso-Meza et al. 4. COMPARATIVE STUDY AND VALIDATION
(2012)). Therefore, the algorithm will seek actively for a
Pareto front approximation inside the pertinency bounds. In order to validate the MOOD procedure exposed here, an
Default evolutionary parameters are used in this work. unstable process will be used. With the aim of compensat-
ing the stochasticity added by the optimisation algorithm
3.3 Multi-criteria decision making stage (an evolutionary algorithm), a total of 51 runs were carried
out; afterwards, the Pareto front approximation with the
In order to visualise and analyse the Pareto front ap- median value of the Hypervolume (HypV ˜ ) index 4 is se-
proximation, Level diagrams are used (Blasco et al., 2008; lected for further analysis. The Hypervolume measure is an
Reynoso-Meza et al., 2013a); they are used due to its capa- usual choice to evaluate the performance of a given Pareto
bilities to represent trade-off in the objective and decision front approximation (Zitzler et al., 2003). This will be done
space simultaneously. In order to compare specific con- in order to work with the expected performance of this
trollers (up to 5), parallel coordinates are used (Inselberg, approach. Simulations 5 and Pareto front approximations
1985) due to its simplicity. Interested readers may refer to were performed in a standard personal computer with Intel
Tušar and Filipič (2015) for a review on visualisation tools 4 that is, the volume enclosed by the Pareto front approximation
and techniques. and the reference point Jref . In this case, pertinency bounds are
stated with such reference point.
3 Scripts and tutorial available in Matlab Central at .../ 5 Simulink,c with ode3 (Bogacki-Shampine) with fixed-step size of
matlabcentral/fileexchange/39215 10 ms

286
 IFAC DYCOPS-CAB, 2016
June 6-8, 2016. NTNU, Trondheim, Norway
G. Reynoso-Meza et al. / IFAC-PapersOnLine 49-7 (2016) 284–289 287

Setpoint response Load response

2.5 1
DM21
2 DM22
0.5 Huang et al.
Output [−] 1.5
Jhun. et al.
Rodr. et al.
1
0
0.5

0 −0.5
0 100 200 300 0 100 200 300

6 0

4 −0.5
Control signal [−]

2 −1

0 −1.5

−2 −2

−4 −2.5
0 100 200 300 0 100 200 300
Time [s] Time [s]
(a) Original Control loop

Setpoint response Load response

2.5 1
DM21
2 DM22
0.5 Huang et al.
Output [−]

1.5
Jhun. et al.
Rodr. et al.
1
0
0.5

0 −0.5
0 100 200 300 0 100 200 300

5 0

−0.5
Control signal [−]

−1

0 −1.5

−2

−2.5

−5 −3
0 100 200 300 0 100 200 300
Time [s] Time [s]
(b) Gain increase of 1.2

Fig. 4. Time response comparison for Cholette process.

Core i5-4210U, 1.7GHz, 4GB RAM. PID implemented weighting for the derivative action equals to zero, in order
for time response simulations uses the standard setpoint to avoid the derivative kick in step response.

287
IFAC DYCOPS-CAB, 2016
288 G. Reynoso-Meza et al. / IFAC-PapersOnLine 49-7 (2016) 284–289
June 6-8, 2016. NTNU, Trondheim, Norway

An approximation for the Cholette reactor (Liou and 1

DM21
Yu-Shu, 1991) is selected for evaluation; such process 0.9 DM22

has been widely used in order to validate several con- 0.8

Huang et al.
Jhun. et al.
trol techniques as PI controllers (Ibarra-Junquera and Rodr. et al.

0.7
Rosu, 2007) and Pareto-optimality techniques (Carrillo-

Coordinate Value
Ahumada et al., 2011). For this work, an identified pro- 0.6

cess around a stationary state with the feed concentra- 0.5

tion cf s = 3.2888[kmol · m3 ] and substrate concentration 0.4
cs = 1.0439[kmol · m3 ] is used:
0.3

0.2
1.121
P2 (s) = e−10s (12) 0.1
33.635s − 1
0
IAE sp IAE load ITAE sp ITAE load Max sp Max load TV sp TV load
Pertinency constraints are defined as JSt (θ) ≤ 150,
JGm−1 (θ) ≤ 0.5 and JP m−1 (θ) ≤ 0.05 while bound con- (a) Original Control loop
straints as kp ∈ [0, 5], TI ∈ [1, 100], TD ∈ [0, 10]. The 1
DM21
Pareto front approximation with the HypV ˜ measure is 0.9 DM22
Huang et al.
depicted in Figure 3 using the Level Diagrams visualisa- 0.8 Jhun. et al.
Rodr. et al.
tion. For the decision making process, for this case, we
0.7
are looking for controllers with JSt (θ) ≤ 80[s] (depicted

Coordinate Value
as dark solutions in Figure 3a, 3b ). Using the color 0.6

coding provided by the toolbox (black and gray in the 0.5

same Figure), it is possible to identify the exchange of 0.4

such solutions regarding gain and phase margin. Controller
0.3
selected is identified within a circle (DM21 ).
0.2
In order to validate the performance of this controller, it
0.1
will be compared with controllers from Huang and Chen
(1999) (H.etAl), Jhunjhunwala and Chidambaram (2001) 0
IAE sp IAE load ITAE sp ITAE load Max sp Max load TV sp TV load

(J.etAl) and Rodrı́guez-Mariano et al. (2015) (R.etAl), (b) Gain increase of 1.2
depicted in Table 1. Time performance is depicted in
Figure 4 for two cases: (1) the original control loop and (2) Fig. 5. Normalized controllers performance: integral of
a control loop with gain increased by a factor of 1.2; some the absolute value of the error (IAE), integral of the
additional time response indicators are shown in Figure 5. time weighted absolute value of the error (ITAE),
As it is possible to notice, the selected controller DM11 maximum deviation (Max) and total variation (TV)
has a good performance on the nominal model, but a poor of control action are depicted for setpoint response
performance when compared with other techniques when (sp) and load response.
gain is increased. Therefore, the decision maker might
ponder his/her decision on selecting this controller from of process, sometimes difficult to control. The main advan-
the approximated set; for example, it is more preferable to tage rely on the capabilities of analysing trade-off between
have a controller with the same level of performance in the conflictive design objectives, and therefore, this enables
nominal model, but with better robustness. This selection the designer to select a controller, fulfilling her/his wishes
is depicted in Figure 3 with a square (DM22 ), where a and preferences, regarding the typical trade-off between
controller with almost the same performance in JSt (θ) performance and robustness. Future work will focus on
than DM21 but a better JGm−1 (θ) and worse JP m−1 (θ). dealing with many-objectives optimisation instance and
As it is possible to notice, this solution has a slightly defining properly feasibility regions for the search process.
different performance in the nominal model (Figure 5a),
but a more acceptable performance when gain is increased ACKNOWLEDGEMENTS
(Figure 5b).
This work was partially supoted by projects FEDER-
Table 1. PID Parameters for Cholette example. CICYT DPI2014-55276-C5-1 (Spain) and the fellow-
ships FPI/2013-3242 (UPV, Spain), BJT-304804/2014-2
Tuning kp Ti Td (CNpQ, Brazil). Second author wishes to thank to Uni-
DM21 3.24 41.66 4.11 versidad del Papaloapan by the approved project entitled
DM22 2.68 55.72 2.34 Sintonizacin de controladores lineales óptimo-robustos me-
H.etAl 2.19 35.35 2.89 diante algoritmos evolutivos y técnicas de decisión multi-
J.etAl 3.15 53.85 5.86
criterio. Third author thanks the Santander Scholarship
R.etAl 2.19 35.35 2.00
program (Investigación-JPI).

5. CONCLUSIONS AND FUTURE WORK REFERENCES

Ang, K.H., Chong, G., and Li, Y. (2005). PID control
In this work, a multi-objective optimisation design (MOOD) system analysis, design, and technology. IEEE Transac-
procedure for unstable process, using PID controllers, was tions on Control Systems Technology, 13(4), 559 – 576.
provided. This procedure is a promising tool for this kind doi:10.1109/TCST.2005.847331.

288
 IFAC DYCOPS-CAB, 2016
June 6-8, 2016. NTNU, Trondheim, Norway
G. Reynoso-Meza et al. / IFAC-PapersOnLine 49-7 (2016) 284–289 289
Arrieta, O., Vilanova, R., and Visioli, A. (2011).
Proportional-integral-derivative tuning for
servo/regulation control operation for unstable and
integrating processes. Industrial & Engineering
Chemistry Research, 50(6), 3327–3334.
Åström, K.J. and Hägglund, T. (2005). Advanced PID
Control. ISA - The Instrumentation, Systems, and
Automation Society, Research Triangle Park, NC 27709.
Åström, K. and Hägglund, T. (2001). The future of PID
control. Control Engineering Practice, 9(11), 1163 –
1175. doi:10.1016/S0967-0661(01)00062-4.
Åström, K., Panagopoulos, H., and Hägglund, T. (1998).
Design of PI controllers based on non-convex optimiza-
tion. Automatica, 34(5), 585 – 601.
Blasco, X., Herrero, J., Sanchis, J., and Martı́nez, M.
(2008). A new graphical visualization of n-dimensional
Pareto front for decision-making in multiobjective opti-
mization. Information Sciences, 178(20), 3908 – 3924.
Carrillo-Ahumada, J., Rodrı́guez-Jimenes, G., and Garcı́a-
Alvarado, M. (2011). Tuning optimal-robust linear
mimo controllers of chemical reactors by using pareto
optimality. Chemical Engineering Journal, 174(1), 357–
367.
Coello, C.A.C. and Lamont, G.B. (2004). Applications of
Multi-Objective evolutionary algorithms. World scien-
tific publishing, advances in natural computation vol. 1
edition.
Das, S. and Suganthan, P.N. (2010). Differential evolution:
A survey of the state-of-the-art. IEEE Transactions on
Evolutionary Computation, PP(99), 1 –28. doi:10.1109/
TEVC.2010.2059031.
Ge, M., Chiu, M.S., and Wang, Q.G. (2002). Robust PID
controller design via LMI approach. Journal of process
control, (12), 3 – 13.
Huang, H.P. and Chen, C.C. (1999). Auto-tuning of pid
controllers for second order unstable process having
dead time. Journal of chemical engineering of Japan,
32(4), 486–497.
Ibarra-Junquera, V. and Rosu, H. (2007). Pi-controlled
bioreactor as a generalized liénard system. Computers
& chemical engineering, 31(3), 136–141.
Inselberg, A. (1985). The plane with parallel coordinates.
The Visual Computer, 1, 69–91.
Ishibuchi, H., Tsukamoto, N., and Nojima, Y. (2008). Evo-
lutionary many-objective optimization: A short review.
In Evolutionary Computation, 2008. CEC 2008. (IEEE
World Congress on Computational Intelligence). IEEE
Congress on, 2419 –2426.
Jhunjhunwala, M.K. and Chidambaram, M. (2001). Pid
controller tuning for unstable systems by optimiza-
tion method. Chemical Engineering Communications,
185(1), 91–113.
Li, Y., Ang, K.H., and Chong, G. (2006). PID control
system analysis and design. IEEE Control Systems.
Liou, C.T. and Yu-Shu, C. (1991). The effect of nonideal
mixing on input multiplicity in a cstr. Chemical engi-
neering science, 46(8), 2113–2116.
Miettinen, K.M. (1998). Nonlinear multiobjective opti-
mization. Kluwer Academic Publishers.
Panagopoulos, H., Åström, K., and Hägglund, T. (2002).
Design of PID controllers based on constrained optimi-
sation. Control Theory and Applications, IEE Proceed-
ings -, 149(1), 32 – 40.
289
Rajinikanth, V. and Latha, K. (2012). Tuning and retun-
ing of pid controller for unstable systems using evolu-
tionary algorithm. ISRN Chemical Engineering, 2012.
Reynoso-Meza, G., Blasco, X., Sanchis, J., and Herrero,
J.M. (2013a). Comparison of design concepts in multi-
criteria decision-making using level diagrams. Informa-
tion Sciences, 221, 124 – 141. doi:10.1016/j.ins.2012.09.
049.
Reynoso-Meza, G., Blasco, X., Sanchis, J., and Martı́nez,
M. (2013b). Evolutionary algorithms for PID controller
tuning: Current trends and perspectives (in spanish).
Revista Iberoamericana de Automática e Informática
Industrial, 10(3), 251 – 268.
Reynoso-Meza, G., Sánches, H.S., Vilanova, R., and
Blasco, X. (2014a). Reliability based multiobjective
optimization design procedure for pi controller tuning.
In Memories of the 19th World Congress IFAC 2014.
Reynoso-Meza, G., Sanchis, J., Blasco, X., and Herrero,
J.M. (2012). Multiobjective evolutionary algortihms for
multivariable PI controller tuning. Expert Systems with
Applications, 39, 7895 – 7907. doi:10.1016/j.eswa.2012.
01.111.
Reynoso-Meza, G., Sanchis, J., Blasco, X., and Martı́nez,
M. (2010). Multiobjective design of continuous con-
trollers using differential evolution and spherical prun-
ing. In C.D. Chio, S. Cagnoni, C. Cotta, M. Eber,
A. Ekrt, A. I.Esparcia-Alcarz, C.K. Goh, J. J.Merelo,
F. Neri, M. Preuss, J. Togelius, and G. N.Yannakakis
(eds.), Applications of Evolutionary Computation, Part
I, volume LNCS 6024, 532–541. Springer-Verlag.
Reynoso-Meza, G., Sanchis, J., Blasco, X., and Martı́nez,
M. (2014b). Controller tuning using evolutionary multi-
objective optimisation: current trends and applications.
Control Engineering Practice, 1, 58 – 73.
Rodrı́guez-Mariano, A., Reynoso-Meza, G., Páramo-
Calderón, D., Chávez-Conde, E., Garcı́a-Alvarado, M.,
and Carrillo-Ahumada, J. (2015). Comparative perfor-
mance analysis of different linear controllers tuned for
several choletette’s bioreactor steady states using multi-
criteria decision making techniques. Revista Mexicana
de Ingenierı́a Quı́mica, 14(1), 167–204.
Stewart, G. and Samad, T. (2011). Cross-application
perspectives: Application and market requirements. In
T. Samad and A. Annaswamy (eds.), The Impact of
Control Technology, 95 – 100. IEEE Control Systems
Society.
Storn, R. and Price, K. (1997). Differential evolution: A
simple and efficient heuristic for global optimization over
continuous spaces. Journal of Global Optimization, 11,
341 – 359.
Toscano, R. (2005). A simple robust PI/PID controller
design via numerical optimization approach. Journal of
process control, (15), 81 – 88.
Tušar, T. and Filipič, B. (2015). Visualization of
pareto front approximations in evolutionary multiobjec-
tive optimization: A critical review and the prosection
method. Evolutionary Computation, IEEE Transactions
on, 19(2), 225–245.
Zitzler, E., Thiele, L., Laumanns, M., Fonseca, C., and
da Fonseca, V. (2003). Performance assessment of
multiobjective optimizers: an analysis and review. IEEE
Transactions on Evolutionary Computation, 7(2), 117 –
132. doi:10.1109/TEVC.2003.810758.