Proceedigs of the 15th IFAC Symposium on

Proceedigs
Proceedigs of
Information of the 15th
the
Control15th IFAC Symposium
IFAC
ProblemsSymposium on
on
in Manufacturing
Proceedigs
Information
Information of the 15th
Control
Control Problems
Problems in Available
IFAC Symposium on
in Manufacturing
Manufacturing
online at www.sciencedirect.com
Proceedigs
May of theOttawa,
11-13, 2015. 15th IFAC Symposium
Canada on
Information
May
May 11-13,
11-13, Control
2015.
2015. Problems
Ottawa,
Ottawa, in Manufacturing
Canada
Canada
Information Control Problems in Manufacturing
May 11-13, 2015. Ottawa, Canada
May 11-13, 2015. Ottawa, Canada
ScienceDirect
IFAC-PapersOnLine 48-3 (2015) 394–399
A
A Survey
Survey on
on Advanced
Advanced Control
Control Approaches
Approaches in
in Factory
Factory Automation
Automation
A
A Survey
Survey on
on Advanced
Advanced Control
Control Approaches
Approaches in
in Factory
Factory Automation
Automation
A Survey
Mariagrazia Dotoli*, Alexander Fay**, Marek Miśkowicz***, CarlaAutomation
on Advanced Control Approaches in Factory Seatzu****
Mariagrazia
Mariagrazia Dotoli*,
Dotoli*, Alexander
Alexander Fay**,
Fay**, Marek
Marek Miśkowicz***,
Miśkowicz***, Carla Carla Seatzu****
Seatzu****
Mariagrazia
Mariagrazia Dotoli*,
Dotoli*, Alexander
Alexander Fay**,
Fay**, Marek
Marek Miśkowicz***,
Miśkowicz***, Carla Carla Seatzu****
Seatzu****
*Dipartimento di Ingegneria Elettrica e dell’Informazione, Politecnico di Bari, Bari, Italy
*Dipartimento
*Dipartimento di Ingegneria
di Ingegneria Elettrica e dell’Informazione,
Elettricae-mail:
e dell’Informazione, Politecnico
Politecnico di Bari, Bari,
di Bari, Bari, Italy
Italy
*Dipartimento (Tel:
di +39 080 5963667;
Ingegneria Elettrica e mariagrazia.dotoli@poliba.it).
dell’Informazione, Politecnico
*Dipartimento (Tel: +39 080
di Ingegneria
(Tel: 5963667; e-mail:
Elettricae-mail:
+39 080 5963667; Politecnico di Bari, Bari,
mariagrazia.dotoli@poliba.it).
e dell’Informazione,
mariagrazia.dotoli@poliba.it).
di Bari, Bari, Italy
Italy
(Tel:
(Tel: +39
+39 080
080 5963667;
5963667; e-mail:
e-mail: mariagrazia.dotoli@poliba.it).
mariagrazia.dotoli@poliba.it).
**Institute of Automation Technology, Helmut Schmidt University Hamburg, Hamburg, Germany
**Institute
**Institute of
of Automation
Automation Technology,
Technology, Helmut
Helmut Schmidt
Schmidt University
University Hamburg,
Hamburg, Hamburg,
Hamburg, Germany
Germany
**Institute (Tel. +49 40 65412719, e-mail: alexander.fay@hsu-hh.de).
**Institute of
of Automation
(Tel.
Automation Technology,
+49 40 Helmut
65412719, Schmidt
e-mail: University Hamburg,
alexander.fay@hsu-hh.de).
(Tel. +49 40 65412719, e-mail: alexander.fay@hsu-hh.de).Hamburg, Germany
Technology, Helmut Schmidt University Hamburg, Hamburg, Germany
(Tel.
(Tel. +49
+49 4040 65412719,
65412719, e-mail: alexander.fay@hsu-hh.de).
e-mail: of
alexander.fay@hsu-hh.de).
*** Department of Electronics, AGH University Science and Technology, Kraków, Poland
*** Department of Electronics,
*** Department of Electronics, AGH
AGH University
University of
of Science
Science and
and Technology,
Technology, Kraków,
Kraków, Poland
Poland
*** (Tel: +48 12 6173034; e-mail: miskow@agh.edu.pl).
*** Department
Department of of Electronics,
(Tel: +48
(Tel: +48 12
Electronics, 12AGH University
6173034;
AGH of
of Science
e-mail:
University
6173034; e-mail:
and
and Technology,
Technology, Kraków,
miskow@agh.edu.pl).
Science
miskow@agh.edu.pl). Kraków, Poland
Poland
(Tel:
(Tel: +48
+48 12
12 6173034;
6173034; e-mail:
e-mail: miskow@agh.edu.pl).
miskow@agh.edu.pl).
****Dipartimento di Ingegneria Elettrica ed Elettronica, University of Cagliari, Cagliari, Italy
****Dipartimento di
****Dipartimento di Ingegneria
Ingegneria Elettrica
Elettrica eded Elettronica,
Elettronica, University
University ofof Cagliari,
Cagliari, Cagliari,
Cagliari, Italy
Italy
****Dipartimento (Tel.: +39 070 6755759; e-mail: seatzu@diee.unica.it).
****Dipartimento di Ingegneria
di(Tel.:
(Tel.: +39
+39 070
Ingegneria 070
Elettrica
6755759;
Elettrica
6755759;
ed Elettronica,
ed e-mail: University
University ofof Cagliari,
seatzu@diee.unica.it).
Elettronica,
e-mail: seatzu@diee.unica.it).Cagliari, Cagliari,
Cagliari, Italy
Italy
(Tel.:
(Tel.: +39
+39 070
070 6755759;
6755759; e-mail:
e-mail: seatzu@diee.unica.it).
seatzu@diee.unica.it).
Abstract: The goal of this paper consists in providing a survey of the main advanced control
Abstract:
Abstract: The
The goal
goal of this
this paper
ofadopted paper consists
consists in
in providing
providing aa survey
survey of
of the
the main
main advanced
advancedto control
control
techniques
Abstract: currently
The goal of this in factory
paper consists automation.
in providingIn particular,
a survey attention
of the mainis devoted
advanced model
control
techniques
Abstract:
techniques currently
The goal
currently adopted
of this
adopted in
paper
in factory
consists
factory automation.
in providing
automation. In
In particular,
a survey
particular, attention
of the main
attention is
is devoted
advanced
devoted to
to model
control
model
based control,
techniques model
currently predictive
adopted in control,
factory intelligent
automation. andInadaptive control,
particular, discrete
attention is event
devoted and
to event-
model
based
based control,
techniques
control, model
currently
model predictive
adopted
predictive in control,
factory
control, intelligent
automation.
intelligent and
andInadaptive control,
particular,
adaptive discrete
attention
control, is
discrete event
devoted
event and
to
and event-
model
event-
triggered
based control.
control, Open
model issues and
predictive challenges
control, are pointed
intelligent and out, and
adaptive the needs
control, for event
discrete further
andresearch
event-
triggered
based
triggered control.
control,
control. Open
model
Open issues
predictive
issues and
and challenges
control,
challenges are
intelligent
are pointed
and out,
adaptive
pointed out, and the
control,
and the needs for
discrete
needs for further
event and
further research
event-
research
efforts arecontrol.
triggered discussed in detail.
Open issues
efforts
efforts are
triggered discussed
arecontrol.
discussed in issues and
in detail.
Open detail. and challenges
challenges are are pointed
pointed out,
out, and
and the
the needs
needs for
for further
further research
research
efforts
© 2015,
efforts are
Keywords: discussed
IFAC
are factory in detail.
automation,
(International
discussed in advanced
Federation
detail. of control,
Automatic model
Control)based control,
Hosting by model
Elsevier predictive
Ltd. All control,
rights reserved.
Keywords:
Keywords: factory automation,
factory adaptive
automation, advanced
advanced control,
control, model
model based
based control,
control, model
model predictive
predictive control,
control,
intelligent
Keywords: control,
factory automation,control, discrete
advanced event
control, control,
model event-triggered
based control, control.
model predictive control,
intelligent control,
Keywords:
intelligent control, adaptive
factory adaptive
automation,control, discrete
advanced
control, discrete event control,
control,
event control,
model basedevent-triggered control.
control, model
event-triggered predictive control,
control.
intelligent control,
intelligent control, adaptive
adaptive control,
control, discrete
discrete event
event control,
control, event-triggered
event-triggered control.
control.
1. INTRODUCTION system and evaluating its performance in a simulation
1.
1. INTRODUCTION
INTRODUCTION system
system and
and evaluating
evaluating its performance in
in aa simulation
1. INTRODUCTION environment
system and by way ofits
and evaluating
evaluating
performance model(s)
mathematical
itsmathematical
performance model(s)
simulation
of the
in aa simulation
simulation
Advanced control 1. approaches
INTRODUCTION are required to improve the system environment
environment by
by way
way of
its
of performance
mathematical in
model(s) of
of the
the
Advanced
Advanced control
control approaches
approaches are
areInrequired
required to improve
to improve the plant (Åstrom
the environment
environment and
by Kumar
way of 2014).
mathematical model(s) of the
the
efficiency
Advanced of production
control approaches sites.
are particular,
required to advanced
improve the plant
plant (Åstrom
(Åstrom and
by
and Kumar
way
Kumar of 2014).
mathematical
2014). model(s) of
efficiency
Advanced
efficiency of
of production
control approaches
production sites.
sites. In
In particular,
areproduction
required to
particular, advanced
improve the plant
advanced (Åstrom and Kumar 2014).
control
efficiencyapproaches may
of production
production lead to
sites. In particular, processes
particular, that
advanced 2.1 Model
plantModel predictive
(Åstrom and Kumar control approaches
2014).
control
efficiency
control
are more
approaches
of
approaches
competitive
may
may leadlead
and
to
sites. production
In
toautonomous
production processes
fromadvanced
processes that 2.1
that
human 2.1 Model predictive
predictive control
control approaches
approaches
control
are
control
are more
more
approaches
competitive
approaches
competitive
may
may lead
and
lead
and
to
to production
autonomous
production
autonomous
processes
from
processes
from
that
human
that
human 2.1
At an
2.1 Model
Model predictive
industrial
predictive control
level, approachesmost widespread
the recognized
control approaches
action,
are more butcompetitive
also have and positive effects in
autonomous from terms
humanof At
At an
an industrial
industrial level,
level, the
the recognized
recognized most
most widespread
widespread
action,
are more
action, but also
butcompetitive
also have
have positive
and effects
autonomous
positive effects in in
from terms
human
terms of technique
of technique
At in thislevel,
an industrial
industrial context isrecognized
the is Model Predictive Control
most widespread
widespread
optimization
action, but of
alsothe production
have positive distribution
effects inand support
terms of At an
technique in
in this
this context
level, the
context is Model
recognized
Model Predictive
most
Predictive Control
Control
optimization
action, but
optimization of
also
of the
the production
have positive
production distribution
effects
distribution inand
and support
terms of
support (MPC)
technique (Camacho
in this and
context Alba
is Bordons
Model 2007).
Predictive MPC
Control
decision-making.
optimization of theTheproduction
term advanced controland
distribution refers to a technique
support (MPC)
(MPC) (Camacho
in integrated
(Camacho and
this context
and Alba
is Model
Alba Bordons
Bordons 2007).
Predictive
2007). MPC
Control
MPC
decision-making.
optimization The term advanced control refers to a provides an solution for controlling systems
wide range ofoftechniques
decision-making.
decision-making.
theTheproduction
The
term
term applied distribution
advanced
advanced
controland
in industrial
control
support
refers
plants
refers
to
andaa (MPC)
to
(MPC) (Camacho
provides
provides
(Camacho
an
an integrated
integrated
andsolution
and Alba Bordons
Alba
solution
Bordons
for
for
2007).systems
controlling
2007).
controlling
MPC
MPC
systems
wide
wide range
decision-making.
range of
of techniques
The
techniques term applied
advanced
applied in
in industrial
control
industrial plants
refers
plants and
to
anda with
provides interacting
an variables,
integrated solution complex
for dynamics,
controlling and
systems
typically
wide range integrating
of techniques
techniquestoolsapplied
from various disciplines.
in industrial
industrial All provides
plants and
and with
with interacting
anAs
interacting variables,
integrated solution
variables, complex dynamics,
forthecontrolling
complex dynamics, and
systems
and
typically
wide range
typically integrating
of
integrating tools from
applied various
in disciplines.
plants All constraints. such, it has become standard approach
these techniques
typically integrating aretools
tools
from various
characterized
from various
disciplines.
by their ability All
disciplines. All
with interacting
to with interacting
constraints.
constraints. As
As such,
such,
variables,
it has
variables,
it has become
become
complex
complexthe
the
dynamics,
standard
dynamics,
standard
and
approach
and
approach
these
these techniques
typically integrating
techniques are characterized
aretools from various
characterized by their
by can ability
disciplines.
their to
ability All for
to for industryAs
constraints. process
such, control
it has today the
become (Jämsä-Jounela
standard 2007,
approach
design high performance controllers that be applied to industry
to constraints. process control today (Jämsä-Jounela 2007,
these techniques
design
these techniques
high are characterized
performance
are characterized
controllers by can
that
by their be ability to
be applied
for
Darby et al.As
industry such, In
process
2012). it has
control
MPC,become
today
a modelthe standard
(Jämsä-Jounela approach
of the process 2007, is
design
design
highand
high-order
high
performance
multivariable
performance
controllers
processes
controllers
that
that ortheir
can
can be
ability
applied
plants,
applied
to Darby
which
to
for industry
for
Darby
industry
et
et al.
al.
process
2012).
process
2012).
control
In MPC,
control
In MPC,
today
aa model
today (Jämsä-Jounela
model of
of the
(Jämsä-Jounela
the
2007,
process
2007,
process is
is
high-order
design high
high-order and multivariable
performance
and multivariable processes
controllers
processesthat or
can
or plants,
be which
applied
plants, to
which used
Darby to predict
etpredict its
al. 2012).
2012). evolution
In MPC,
MPC,over over a future time horizon.
are typicallyand
high-order
are typically
nonlinear and subject
multivariable
nonlinear and
to constraints.
processes
subject to or plants, The
plants,The five Darby
which used
used
Theseto
toet al.
predict its evolution
its evolution
In valuesover aa model
model of the
aa future
future
of the
time
process
timeprocess
horizon.
horizon.
is
is
high-order
are
maintypically
classes and
of multivariable
nonlinear
advanced andcontrol to constraints.
processes
subject techniquesor
constraints.may The five
which
five These
broadly used predicted
topredicted
predict output
its evolution
output values
are
over
are
then used
aa future
then used
to compute
time
to horizon.aa
compute
are typically
main typically
classes nonlinear
of andcontrol
subjecttechniques
to constraints.
constraints. The used
five These to predict
predicted its evolution
output values theover
are futurefuture
then used time horizon.
to compute
of theaa
are
main
be classes
singled ofasadvanced
outnonlinear
advanced
follows:and subject
control to
techniques may
mayThe broadly
five
broadly control signal that
These predicted
control predicted
signal that
optimizes
output
that values the
optimizes are future
then used behaviour
used to compute
behaviourcompute
of the
bemain classes
singled outofasadvanced
follows: control techniques may broadly These
control
plant signal
(Bemporad output
and values
optimizes
Morari are
the then
future
1999). To to
behaviour
determine thea
of the
main classes
be singled outofasadvanced
follows: control techniques may broadly plant control(Bemporad
signal that
that and
optimizes
Morari the futureTo
1999). behaviour
determine of the
the
be singled
• singled
Modelout out as follows:
based follows:
control techniques; control
plant
sequence signal
(Bemporad
of control optimizes
and Morari
moves, the future
1999).
a1999).
dynamic To behaviour
determine of
model of the the
the
be
•• Model as
based control techniques; plant
sequence(Bemporad
of controland Morari
moves, a dynamic To determine
model of the
• Model based techniques;or control techniques plant
controltechniques sequence
process,(Bemporad
of history
controlandof Morari
past aa1999).
moves, dynamic Tomoves
determine
model and of the
Intelligent
••• Model
Model based
Intelligent
based
control
controltechniques
control
control techniques;or control techniques sequence
techniques; sequence aaof
process, of history
control of
control moves,
past
moves, a
control
dynamicmoves
control
dynamic model and
model of the
of
an
the
an
Intelligent control techniques or control techniques process,
optimization a history
cost of
function past
are control
required moves
over the and
receding an
based on computational intelligence; process, aa historyhistory of past
past control over moves and an an
•• Intelligent
Intelligent
based on control techniques
computational
control techniques or control
intelligence;
or control techniques
techniques process,
optimization
optimization cost
cost function
of
function are required
are control
required moves
over the
the receding
and
receding
• based
Adaptive
based
on computational
on control techniques;
computational
intelligence;
intelligence;
prediction horizon.
optimization
prediction
optimization cost
horizon.
prediction horizon.
function are required over
cost function are required over the receding the receding
•• based
Adaptive control
control techniques;
on computational intelligence;
• Adaptive
Discrete event/hybrid techniques;
control techniques; prediction horizon.
After the horizon.
early diffusion of numerous MPC linear
••• Adaptive
Adaptive
Discrete control
event/hybridtechniques;
control
control techniques; techniques; prediction
After the
the early diffusion of of numerous MPC linear
• Discrete event/hybrid
Event-triggered control techniques;
and self-triggered control. After
techniques early
-e.g., diffusion
model predictive numerous MPCcontrol,
heuristic linear
••• Discrete
Discrete event/hybrid
event/hybrid
Event-triggered
Event-triggered and
control techniques;
and self-triggered
control
self-triggered
techniques;
control.
control.
After the
techniques
After
techniques
dynamic
the early
early
-e.g.,
-e.g.,
matrix
diffusion
model
diffusion
model
control,
of numerous
of numerous
predictive
predictive
extended
MPCcontrol,
heuristic
MPC
heuristic
prediction
linear
linear
control,
self-adaptive
••In this
Event-triggered
paper, for each
Event-triggered and
andof self-triggered
ofself-triggered
the above classes control. techniques
dynamic
of advanced techniques
control.of -e.g.,
matrix model
control,
-e.g.,control, predictive
extended
modelpredictive
predictive heuristic
prediction control,
self-adaptive
heuristic control,
In this dynamic
control, matrix
and generalized extended prediction
control- inself-adaptive
the ‘90s the
In this paper,
control paper, for
for each
techniques, eachwe of the
the above
above classes
describe classes
the main of advanced
advanced dynamicand
control,
control dynamic
control,
matrix
matrix
and
control, predictive
generalized
control,
generalized
extended prediction
extended
predictive
prediction
control-
control- in
in
self-adaptive
the ‘90s
self-adaptive
the ‘90s the
the
InIn this
control
this paper,
paper, for
techniques,
for each
each weof
of the
the above
describe
above classes
the
classes of
main
of advanced
control
advanced first applications
control,applications
and generalized
generalized of nonlinear
predictive MPC
control- in the appeared.
theappeared.
‘90s the
the
control
techniques techniques,
with their we describe areas,
application the indicating
main control first
the control,
first and
applications of
of nonlinear
predictive
nonlinear MPC
control-
MPC in ‘90s
appeared.
control
techniques
control techniques,
with
techniques, their we describe
application the
areas, main
the indicating control
the Subsequently, MPC techniques for hybrid systems were
techniques
related emerging their we
with research describe
application
trends in areas,
industrialmain
indicatingcontrol
control and first
the first applications
Subsequently, MPC
applications of nonlinear
nonlinear
techniques
of for MPCsystems
hybrid
MPC appeared.
were
appeared.
techniques
related with research
emerging their application
application
trends areas, indicating
indicating the Subsequently,
developed MPC techniques
(Camacho and Alba for hybrid2007).
Bordons systems were
Typical
techniques
related
factory with
emerging
automation. their
research trends inin industrial
areas,
industrial control and
control and
the Subsequently,
developed
Subsequently,
developed
MPC techniques
(Camacho
MPC
(Camacho
techniques
and
and Alba
Alba
for hybrid
Bordons
for hybrid
Bordons
systems
2007).
systems
2007).
were
Typical
were
Typical
related
factory
related emerging
automation. research trends in industrial control and areas of industrial applications for the MPC strategy are
factory emerging
automation. research trends in industrial control and developed developed
areas
areas of
of
(Camacho
industrial
(Camacho
industrial
and
and Alba
applications
Alba
applications
Bordons
for the
Bordons
for the MPC
MPC
2007). Typical
strategy
2007).
strategy are
Typical
are
factory
factory automation.
2. MODEL
automation. BASED CONTROL TECHNIQUES chemical
areas of and
of industrial
industrialpetrochemical plants
applicationsplants for the operation,
theoperation,
MPC strategy together
strategy are
2. MODEL BASED CONTROL
2. MODEL BASED CONTROL TECHNIQUES TECHNIQUES chemical
areas
chemical and
and petrochemical
applications
petrochemical for
plants MPC
operation, together
are
together
2. MODEL BASED CONTROL TECHNIQUES with refinery
chemical and control (Jämsä-Jounela
petrochemical plants 2007). Emerging
operation, together
Model based control concerns with TECHNIQUES
designing a control chemicalwith refinery control (Jämsä-Jounela 2007). Emerging
Model 2. MODEL
Model based
based control
BASED
control concerns
CONTROL
concerns with with designing
designing aa control
with
areas for and
refinery
with refinery
control areas refinery
petrochemical
control
applications
control (Jämsä-Jounela
plants operation,
(Jämsä-Jounela
are: mining/ore
(Jämsä-Jounela
2007).
treatment,
2007).
together
Emerging
sheet
Emerging
with
areas for
for applications
control
applications are:
are: mining/ore
mining/ore treatment,
2007).
treatment, sheet
Emerging
sheet
Model based control concerns with designing
Model based control concerns with designing a control areas a control areas for applications
applications are: mining/ore treatment, sheet
2405-8963 © 2015, IFAC (International Federation of Automatic Control)
Copyright © 2015 IFAC 419
for
Hosting by Elsevier Ltd.are: mining/ore
All rights reserved.treatment, sheet
Peer review
Copyright
Copyright ©
© 2015
2015under
IFAC
IFAC responsibility of International Federation of Automatic
419
419 Control.
10.1016/j.ifacol.2015.06.113
Copyright © 2015 IFAC 419
Copyright © 2015 IFAC 419
INCOM 2015
Mariagrazia Dotoli et al. / IFAC-PapersOnLine 48-3 (2015) 394–399 395
May 11-13, 2015. Ottawa, Canada
processing, fuel cell, upstream oil production, power
generation, and pulp and paper industry. Moreover, MPC
is moving into pharmaceutical manufacturing, food
processing and has even penetrated discrete
manufacturing, most notably the automotive industry.
One of the limitations of MPC is that it relies on the
solution of an online optimization problem, which had
precluded its application to processes requiring high
sample rates. Recently, the impact of faster and multi-core
processors is being seen in MPC products. As a result,
several research directions for improved MPC application
in factories can be identified. First, the use of nonlinear
models is increasingly being applied in MPC, and it
would likely increase if the associated modelling costs
could be reduced. This motivates the idea of model
libraries, tailored to control, for important unit operations
(Darby et al. 2012). In this context, applications of
switched/hybrid MPC seem promising (Morari 2009), for
instant in the area of batch distillation columns, fed-batch
reactors and bioreactors profile control. Second, much
remains to be done in terms of robust MPC that currently
lacks a practically implementable method (Darby et al.
2012, Lee 2011). In this context, a further interesting
application which needs advances both in terms of
theoretical development and application is MPC for
uncertain systems (Lee 2011). Indeed, to facilitate tuning
in light of expected model errors, another area to consider
is incorporating into simulation the characteristics of
unmeasured (stochastic) disturbances that could be
obtained from historical data, test data, and/or from the
actual controller (Darby et al. 2012). Third, the
application of adaptive MPC, mainly in terms of plant
testing, is a natural area for further research (Morari
2009). Fourth, areas where improvements could be
realized are in the identication and modelling for MPC,
including enforcing model consistency and taking
advantage of prior knowledge. Benets would include
better handling of ill-conditioning and time savings in
model development (Darby et al. 2012). Fifth, an area
where a clear space for further contributions is available
regards detection and diagnosis of control performance
deterioration. The problem is evidently a challenging one
due to its multivariable aspect and the number of
constraint combinations (Darby et al. 2012). Sixth,
reducing complexity of online optimization and of
explicit solutions (i.e., number of regions) are other
interesting open research areas for MPC (Morari 2009).
Finally, decentralized and distributed MPC are an obvious
reply to address the control of large-scale systems in
factory automation (Cristofides et al. 2013, Morari 2009).
2.2 Multivariable control approaches
Multivariable control refers to control problems where
there are typically a number of process variables which
must be controlled and a number of variables which can
be manipulated (Skogestad et al. 2005). Multivariable
feedback control systems include numerous approaches,
e.g.: the so-called Relative Gain Array method,
decoupling control schemes, multi-loop controllers
(typically with PID regulators), optimal control methods
420
such as the well-known Linear Quadratic Gaussian (LQG)
control approach, and, in the case of uncertainty in the
plant, robust control, including H2 and H∞ control
(Skogestad et al. 2005).
In the industrial context, the following multivariable
control applications are notable: robust nonlinear
feedback control is being applied in pharmaceutical
companies; optimal LQG is applied in milling industries;
Kalman filtering is used for automatic gauge control;
extended Kalman filters are used to construct observers
for batch processes in chemical/petrochemical
applications; state feedback linearization has been
implemented in the glass and ceramic industries; LQG,
robust, and H∞ control trials have been performed in
experimental fusion main plasma control; applications of
multivariable decoupling control solutions have become
standard in pulp and paper mills.
Future challenges in the use of multivariable feedback
techniques in process industries are: improving energy
efficiency in industrial processes by monitoring,
optimization and control technologies; economic-
performance optimizing approaches, also known as
dynamic real time optimization approaches; and very
large scale integrated process control, i.e., integrated
hierarchical control approaches aiming at the combined
control of the classical layers of a process enterprise.
3. INTELLIGENT CONTROL APPROACHES
Intelligent control methods are computational intelligence
based approaches. They include a number of techniques
based on reasoning from data, including the following
three main classes and their integration: techniques based
on fuzzy systems, Artificial Neural Networks (ANNs),
evolutionary algorithms.
3.1 Fuzzy control techniques
Roughly speaking, fuzzy control algorithms consist of a
set of heuristic control rules, and fuzzy sets and fuzzy
logic are used, respectively, to represent linguistic terms
and to evaluate the rules (Feng 2006). Since the first
applications in the ‘70s, fuzzy logic control has attracted
great attention from the industrial community. As regards
factory automation, the main areas for application of
fuzzy control are chemical and petrochemical processes,
food processing industries, as well as nuclear reactors.
The industrial applications of fuzzy control are
characterized by the use of heuristic rules that can be a
viable alternative to classical crisp (non-fuzzy) control for
the control of ill-defined processes or systems with
functional nonlinearities subject to difcult mathematical
modelling (Precup and Hellendoorn 2011). Moreover,
compared to conventional control, fuzzy control can be
strongly based on the experience of a human operator, and
a Fuzzy Controller (FC) can model this experience in a
linguistic manner (Precup and Hellendoorn 2011).
In the majority of applications a FC is used either for
direct feedback control or on the low level in hierarchical
control system structures (Precup and Hellendoorn 2011).
In the former case, the related literature traditionally
INCOM 2015
396 Mariagrazia Dotoli et al. / IFAC-PapersOnLine 48-3 (2015) 394–399
May 11-13, 2015. Ottawa, Canada
classifies FCs into Mamdani FCs and Takagi-Sugeno
FCs. In the Mamdani case a model-free approach is
employed and numerous applications may be found in
factory automation (Precup and Hellendoorn 2011). In the
Takagi-Sugeno case, a model-based approach is
employed to determine a fuzzy non-linear model of the
controlled system. Applications to factory automation of
approaches in this second sub-class typically include
process control (Precup and Hellendoorn 2011). In the
second class of applications, typically adaptive and
predictive FC are designed (Precup and Hellendoorn
2011). Adaptive FCs include a supervisory module that
adjusts several components of the FC like the
size/position of the membership functions of the fuzzy
sets, and/or the rule weights/values. On the other hand, a
predictive FC does not change the parameters of the
underlying FC, but it chooses every time the best control
signal based on a performance measure expressed in
terms of an objective function. This subclass of FCs
includes the fuzzy MPC techniques, which are emerging
in the industrial context (Precup and Hellendoorn 2011).
Countless applications of adaptive and predictive FC have
been proposed in industrial control, especially in process
industries (textile industries, chemical/petrochemical
industries) (Mendes et al. 2013). Finally, applications of
type-2 fuzzy control to manufacturing operations and
industries are described in (Dereli et al. 2011).
3.2 Techniques based on artificial neural networks
ANNs are nonlinear input-output mappings that employ a
mathematical model to mimic biological neural networks.
Many different types of ANNs exist. A traditional
classification distinguishes between feedforward and
recurrent (i.e., implementing feedback) networks,
depending on how data is processed through the network.
Another way of classifying ANNs is by their learning
method (training), which leads to dividing them into
supervised and unsupervised (or self-organizing)
networks (Magali et al. 2003).
From the viewpoint of industrial applications, ANN
applications can be divided into four main categories
(Magali et al. 2003): modelling and identification;
optimization and classification; process control; pattern
recognition. The first class of applications refers to the
ability of ANNs to determine an approximation of the
input output plant model, which has been applied in
chemical processes, welding systems, and plate rolling
processes (Magali et al. 2003). The second class of
applications includes optimized planning of tasks and
distinguishing various classes of events/materials/objects
in functional form, e.g. for fault detection scopes (Magali
et al. 2003). More importantly, the third class refers to
applying the ANN nonlinearity, learning, parallel
processing, and generalization capabilities for application
to advanced intelligent control. Applications include
monitoring feed water flow rate and component thermal
performance of pressurized water reactors, temperature
regulation, control of manufacturing cells, and prediction
-from operating conditions and from features of a steel
sheet- of the thermal energy required to correct alloying
421
(Magali et al. 2003). Fourth and finally, some specific
ANN structures are studied and applied mainly for image
and voice recognition. Applications include transient
detection to enhance nuclear reactors’ operational safety
and papermaking industries (Magali et al. 2003). The
current trends in the use of ANNs at an industrial control
level show a clear increase of their applications in food
industrial processes. In the next decade, with further
advances of computer power, ANNs will be applied in the
following areas of factory automation: classification and
prediction of food quality and safety, field prescriptions
for variable rate chemical application (Huang 2009).
3.3 Techniques based on evolutionary algorithms
Evolutionary Algorithms (EAs) are robust search and
optimization methods able to cope with ill-behaved
problem domains that exhibit multimodality,
discontinuity, time-variance, randomness, and noise
(Fleming and Purshouse 2002). The most popular among
EAs are certainly Genetic Algorithms (GAs). GAs draw
inspiration from the natural principles of the evolution of
the species and the survival of the fittest. They are
optimization techniques working iteratively on a
population of candidate solutions of the problem. The
fitness, i.e., the value of the objective function associated
with each solution in the population, rules the iterative
selection scheme, so that solutions with high fitness have
high likelihood to mate and form the offspring population.
Once the offspring population is selected, special
operators emulating genetic crossover and mutation are
randomly applied to create new solutions. These steps are
then iterated in order to obtain new solutions with
increased fitness until a stopping criterion is met. GAs,
and other EAs such as genetic programming, are
extensively applied to the off-line design of controllers,
particularly to parameters optimization and the automatic
synthesis of the controller topology (Fleming and
Purshouse 2002). Numerous applications in the related
literature regard the tuning and optimization by GAs of
industrial PID controllers as well as multivariable
controllers, such as H2 and H∞. On-line applications of
EAs to control are in principle possible but rare due to
practical constraints. Overall, applications of EAs to
factory automation and industrial control are motivated by
their ability to automatically tune the controller
parameters, without requiring expert operators.
3.4 Integrated techniques
Numerous combined applications of computational
intelligence techniques have been proposed in the related
literature and they are one of the emerging trends in
industrial control. Neuro-fuzzy control is in fact fuzzy
control with enhanced performance thanks to the learning
capabilities and parallel processing abilities of ANNs
(Precup and Hellendoorn 2011). The most widespread
neuro-fuzzy approach in this regard is ANFIS. Moreover,
EAs may be typically employed in the optimization of
fuzzy or neural controllers (Fleming and Purshouse 2002).
In fuzzy control, EAs can be used to generate the fuzzy
rule base and to tune the associated membership function
INCOM 2015
Mariagrazia Dotoli et al. / IFAC-PapersOnLine 48-3 (2015) 394–399 397
May 11-13, 2015. Ottawa, Canada
parameters. In neural control, an EA can constitute an
alternative choice to learn the weight values. EAs are also
capable of optimizing the topology of an ANN. Both
fuzzy control integrated with EAs and neural control
integrated with EAs received great attention in the
industrial area, for process control applications and
planning/scheduling applications in manufacturing. In
addition, in the context of the three integrated
computational intelligence techniques, countless
applications exist in industrial control.
4. ADAPTIVE CONTROL TECHNIQUES
Adaptive control techniques emerged in the process
control area and share as an objective the design of a
controller exhibiting time-varying parameters and a
mechanism for adjusting them to accommodate plant and
disturbance uncertainties (Åstrom and Kumar 2014,
Pathak and Adhyau 2012). The main classes of adaptive
control techniques with applications at an industrial level
may be singled out as follows: gain scheduling, Model
Reference Adaptive Control (MRAC), and self-tuning
regulators (Åstrom and Kumar 2014).
Gain scheduling is a control technique in which the
controller parameters are changed based on the plant
operating condition, i.e., according to the current value of
so-called scheduling signals, which may be internal or
external to the plant (Leith and Leithead 2000). The main
process control applications of gain scheduling techniques
are in the chemical processes area, with recent trends
toward the integration of the classical crisp gain
scheduling technique with fuzzy logic.
MRAC techniques are concerned with automatically
adjusting the parameters of a controller so that the
response to command signals is close to that of a
reference model (Åstrom and Kumar 2014). After the
development of the first linear MRAC methodologies,
MRAC was extended with nonlinear approaches, e.g.,
using backstepping or computational intelligence (fuzzy
control) techniques; decentralized MRAC approaches
were also proposed (Åstrom and Kumar 2014). MRAC
has found its main industrial application in process
control. In this context, nonlinear, fuzzy, as well as
decentralized MRAC approaches are still emerging trends
for research (Åstrom and Kumar 2014, Cerman 2013).
Self-tuning regulators are controllers in which the control
design is made online by determining the controller
parameters in real time based on some plant estimates,
which can be performed continuously or in batches. The
self-tuning regulator estimates the process parameters and
finds controller parameters that minimize a criterion, for
example the variance of the process output (Åstrom and
Kumar 2014). A number of applications have been
proposed, including rolling mills, continuous casting,
distillation columns, chemical reactors, distillation
columns and ore crushers, fermentation processes
(Åstrom and Kumar 2014). Stochastic variations and
integration with computational intelligence technique are
also a clear trend emerging in self-tuning regulators for
industrial control in recent years.
422
Emerging areas in the research of adaptive control
techniques for industrial applications are: adaptive control
rules for wireless control (with suitable modification of
data sampling rates for wireless measurement),
decentralized control, and integration of different adaptive
control approaches (e.g., self-tuning regulators and
MRAC), with renewed interest in the aerospace industry.
5. DISCRETE EVENT / HYBRID CONTROL
TECHNIQUES
Discrete event control techniques are approaches
employing the Discrete Event Systems (DESs) modelling
formalisms. DESs are dynamical systems whose
dynamics depends on the interaction of asynchronous
discrete qualitative changes called events (Cassandras and
Lafortune 2008). The DES formalism has found wide
application in the representation of discrete production
systems. In this context, a number of techniques have
been developed for DES real time control with particular
application to manufacturing systems, for sequencing the
operations in the system or preventing the system from
being in some undesired states. The exploitation of the
structural properties of DES models (typically automata
or Petri nets) in order to obtain computationally efficient
algorithms for computing controls is nowadays one of the
main goals of DES research.
The most widespread approach for DESs control is
supervisory control. In this approach a controller
(supervisor) is designed so that the closed loop system
satisfies some desired qualitative constraints
(specification). The plant is typically modelled as an
automaton or Petri net and its generated language is used
to describe its behaviour at the logical level. The control
task is formulated as designing a supervisor restricting the
plant dynamics so as to achieve the defined specification.
Many extensions of the basic supervisory control
problem, such as control with partial observations,
hierarchical and modular control, nondeterministic
supervisory control, and control of infinite behaviours
have been studied. Another emerging concept is
supervisory control with bounded time constraints in the
specification. Advantages and limitations of supervisory
control for industry application are discussed in (Morel et
al. 2007), where it is stressed that only small case study
tests have been carried out and the emerging challenge of
DES control under uncertainty. With particular reference
to industrial control and factory automation, a clearly
emerging concept is also decentralized and distributed
control of DES (Lafortune 2007), which allows
overcoming the recalled shortcoming of supervisory
control. The resulting techniques find their wide
application in the control of networked production and
manufacturing supply chains. In this context, research
challenges are uncertain dynamics and risk management.
We conclude this section by recalling that numerous
control schemes are recently being developed for the
constrained control of systems integrating continuous and
discrete dynamics (Jämsä-Jounela 2007). In fact, the
interest in hybrid systems has considerably grown during
INCOM 2015
398 Mariagrazia Dotoli et al. / IFAC-PapersOnLine 48-3 (2015) 394–399
May 11-13, 2015. Ottawa, Canada
the last decade because of their potential impact on the
industrial applications of control, with natural
applications in areas such as automotive systems and
power systems (Camacho et al. 2010). In this context,
MPC methods for hybrid systems and processes are an
emerging challenge, with applications in process control,
manufacturing systems, automotive systems and power
systems (Camacho et al. 2010). Networked applications
of hybrid control techniques are also being developed.
6. EVENT-TRIGGERED AND SELF-TRIGGERED
CONTROL TECHNIQUES
The last decade has seen significant scientific interest in
the development of event-triggered control as an
alternative to classical digital control methods based on
equidistant sampling (Heemels 2012, Pawlowski et al.
2014). In event-based systems, the control action is
updated according to the plant state rather than due to
progression of time. Such a paradigm allows calling for
resources whenever they are indeed necessary. Therefore,
event-based control is especially attractive in resource-
constrained applications. The other motivation behind the
development of event-based control is that control over
networks with transmission delays and packet losses
cannot be modelled by conventional discrete-time control,
where data are processed and transmitted with a constant
sampling rate (Grune et al. 2014).
The objective of most research works on event-driven
control is positioned in the perspective of networked
systems (Heemels 2012, Pawlowski et al. 2014, Wang
and Lemmon 2011) especially wireless due to more
efficient bandwidth utilization and a reduction of
computational complexity (Miśkowicz 2014).
Event-based strategies enforce a generation of sporadic
executions of control tasks in the time domain. However,
compared to control systems with non-periodic sampling,
the use of event-triggered schemes provides some implicit
information on behaviour of the state of the plant between
sampling instants: as long as no event is triggered, the
actual measurement does not fulfil the event-sampling
criterion. In particular, in the send-on-delta scheme, the
sampled signal does not vary between sampling instants
more than by the threshold (delta) compared to the most
recent sample. This extra information on intersample
signal behaviour might be additionally exploited, for
example, to improve state estimation where a lack of
event occurrence is referred as “negative information”
(Sijs et al. 2013).
In Continuous Event-Triggered Control (CETC) systems,
the detection of passing the threshold that triggers
sampling operations is realized continuously in time
(Hemeels and Donkers 2013). On the other hand, in
Periodic Event-Triggered Control (PETC), the sensor is
sampled equidistantly and event occurrence is verified at
every sampling time in order to decide whether or not to
transmit a new measurement to a controller. The
periodically sampled sensor is communicated with
controller only when needed to guarantee stability or
desired performance properties. Thus, in PETC, the
423
periodicity of control system components is relaxed to
establish the event-driven communication between them.
The control laws in PETC can be computed in a
continuous-time framework or with event-triggered
controllers for discrete-time systems by disregarding the
intersample behaviour (Hemeels and Donkers 2013).
The implementation of the event-triggered
communication and processing activities on the top of
periodically sampled data has been applied in send-on-
delta scheme in LonWorks commercial control network
platform (ISO/IEC 14908), EnOcean energy harvesting
wireless technology (Ploennigs et al. 2010), and in event-
based PID controllers (Sànchez et al. 2012). Surveys on
event-triggered control strategies are provided in
(Heemels 2012, Grune et al. 2014).
To avoid event detection while keeping benefits of
resource-aware design, self-triggered control as cost-
effective approximation of event-triggered strategy
received a significant research attention in recent years
(Anta and Tabuada 2010). The self-triggered control uses
adaptive sampling as a class of time-triggered sampling
algorithms closely related to event-based schemes
(Ploennigs et al. 2010). In adaptive sampling, the next
sampling instant is planned to be triggered at the moment
when it is believed that the plant state will meet the
assumed event-based criterion (e.g., a change of the state
by delta). The belief is based on the prediction expanded
at the actual sampling instant. Therefore, the hardware for
detection of event occurrences is not required in self-
triggered control systems because the sampling instants
are controlled by the timer. Hence, unlike event-based
control, self-triggered strategies are classified as time-
triggered. The coordination method that combines ideas
from event- and self-triggered control has been proposed
in (Nowzari and Cortés 2013). It is shown that team-
triggered strategies act no worse than self-triggered ones
in terms of required communication while it is much less
costly than implementations of self-triggered strategy in
distributed systems (Nowzari and Cortés 2013).
7. CONCLUSIONS
The main contribution of this paper consists in a survey of
the main advanced control techniques currently adopted
in factory automation. Open issues and challenges are
pointed out, and we discuss in detail the needs for further
research efforts. Specifically, more efficient and advanced
control approaches should be derived for large-scale
industrial systems with special attention to decentralized
and distributed architectures. In addition, methods should
be further developed to allow factory systems to be more
autonomous and capable of self-organization and re-
configuration while still providing methods to ensure
transparency and reliability.
ACKNOWLEDGEMENTS
M. Miśkowicz was supported by Polish National Center
of Science under grant DEC-2012/05/E/ST7/01143.
INCOM 2015
Mariagrazia Dotoli et al. / IFAC-PapersOnLine 48-3 (2015) 394–399 399
May 11-13, 2015. Ottawa, Canada

REFERENCES S. Lafortune, “On Decentralized and Distributed Control

of Partially-Observed Discrete Event Systems”, in
A. Anta, P. Tabuada, “To sample or not to sample: Self-
Advances in Control Theory and Applications,
triggered control for nonlinear systems”, IEEE
Lecture Notes in Control and Information Sciences,
Transactions on Automatic Control, vol. 55, no.9, pp.
vol. 353, pp. 171-184, Springer-Verlag, Berlin
2030-2042, 2010.
Heidelberg, 2007.
K.J. Åstrom, P.R. Kumar, “Control: A perspective”,
J.H. Lee, “Model predictive control: Review of the three
Automatica, vol. 50, no. 1, pp. 3-43, Jan. 2014.
decades of development”, International Journal of
A. Bemporad, M. Morari, “Robust model predictive
Control Automation and Systems, vol. 9, no. 3, pp.
control: A survey,” in Robustness in Identification
415-424, 2011.
and Control, A. Garulli, A. Tesi, and A. Vicino, Eds.,
D.J. Leith, W.E. Leithead, “Survey of gain-scheduling
Lecture Notes in Control and Information Sciences,
analysis and design”, Int. Journal of Control, vol. 73,
vol. 245, pp. 207-226, Springer-Verlag, 1999.
no. 11, 2000, pp. 1001-1025.
E.F. Camacho, C. Alba Bordons, Model Predictive
R.G. Magali, P.E.M. Meireles, M.G.S. Simões, “A
Control, Springer-Verlag London, UK, 2007.
Comprehensive Review for Industrial Applicability
E.F. Camacho, D.R. Ramirez, D. Limon, D. Muñoz de la
of Artificial Neural Networks”. IEEE Trans. on
Peña, T. Alamo, “Model predictive control
Industrial Electronics, Vol. 50, No. 3, June 2003
techniques for hybrid systems”, Annual Reviews in
J. Mendes, R. Araújo, F. Souza, “Adaptive fuzzy
Control, vol. 34, no. 1, Apr. 2010, pp. 21-31.
identification and predictive control for industrial
C. Cassandras, S. Lafortune, Introduction to Discrete
processes”, Expert Systems with Applications, vol.
Event Systems (2nd Ed.), Kluwer Academic
40, no. 17, pp. 6964-6975, 2013.
Publishers, Boston, MA, 2008.
M. Miśkowicz, “Event-based sampling strategies in
O. Cerman, “Fuzzy model reference control with
networked control systems”, Proc. IEEE Workshop
adaptation mechanism”, Expert Systems with
on Factory Communication Systems, pp. 1-10, 2014.
Applications, vol. 40, no. 13, 2013, pp. 5181-5187.
M. Morari, “Predicting the future of model predictive
P.D. Christofides, R. Scattolini, D. Muñoz de la Peña, J.
control”, Workshop in celebration of David Clarke’s
Liu, “Distributed model predictive control: A tutorial
contribution to MPC, Univ. of Oxford, Jan. 2009.
review and future research directions”, Computers &
G. Morel, P. Valckenaers, J.-M. Faure, C.E. Pereira, C.
Chemical Engineering, vol. 51, pp. 21-41, Apr. 2013.
Diedrich, “Manufacturing plant control challenges
M.L. Darby, M. Harmse, M. Nikolaou , “MPC: Current
and issues”, Control Engineering Practice, vol. 15,
practice and challenges”, Control Engineering
no. 11, Nov. 2007, pp. 1321-1331.
Practice, vol. 20, no. 4, Apr. 2012.
C. Nowzari, J. Cortés, “Team-triggered coordination of
T. Dereli, A. Baykasoglu, K. Altun, A. Durmusoglu, I. B.
networked systems”, Proc. American Control
Türksen, “Industrial applications of type-2 fuzzy sets
Conference, pp. 3821-3826, 2013.
and systems: A concise review”, Computers in
K.B. Pathak, D. M. Adhyaru, “Survey of Model
Industry, vol. 62, no. 2, pp. 125-137, Feb. 2011.
Reference Adaptive Control”, 2012 NIRMA Univ.
G. Feng, “A survey on analysis and design of model-
Int. Conf. on Eng., Ahmedabad, India, Dec. 2012.
based fuzzy control systems”, Fuzzy Systems, IEEE
A. Pawlowski, A. Cervin, J. L. Guzmán, M. Berenguel,
Trans. on, vol.14, no.5, pp.676,697, Oct. 2006
“Generalized predictive control with actuator
P.J Fleming, R.C Purshouse, “Evolutionary algorithms in
deadband for event-based approaches”, IEEE Trans.
control systems engineering: a survey”, Control
on Industrial Informatics, Vol. 10, no.1, 523-537,
Engineering Practice, vol. 10, no. 11, pp. 1223-1241,
2014.
Nov. 2002.
J. Ploennigs, V. Vasyutynskyy, K. Kabitzsch,
L. Grune, S. Hirche, O. Junge, P. Koltai, D. Lehmann, J.
“Comparative study of energy-efficient sampling
Lunze, et al., “Event-based control”, In: Control
approaches for wireless control networks”, IEEE
Theory of Digitally Networked Dynamic Systems,
Transactions on Industrial Informatics, vol. 6, no. 3,
Ed. J. Lunze, pp. 169-261, Springer, 2014.
pp. 416-424, 2010.
W.P.M.H. Heemels, M.C.F. Donkers, “Model-based
R.-E. Precup, H. Hellendoorn, “A survey on industrial
periodic event-triggered control for linear systems”,
applications of fuzzy control”, Computers in
Automatica, vol. 49, pp. 698-711, 2013.
Industry, vol. 62, no. 3, pp. 213-226, Apr. 2011.
W.P.M.H. Heemels, K.H. Johansson, and P. Tabuada,
J. Sánchez, A. Visioli, S. Dormido, “Event-based PID
“An introduction to event-triggered and self-triggered
control”, in: PID Control in the Third Millennium.
control”, Proc. IEEE Conference on Decision and
Advances in Industrial Control, Ed. R. Vilanova, A.
Control, pp. 3270-3285, 2012.
Visioli, pp. 495-526, Springer, 2012.
Y. Huang, “Advances in artificial neural networks –
S. Skogestad, I Postlethwaite, Multivariable Feedback
Methodological development and application.”,
Control: Analysis and Design (2nd Ed.), Wiley 2005.
Algorithms, vol. 2, no. 3, pp.973-1007, 2009.
J. Sijs, B. Noack, U.D. Hanebeck, “Event-based state
S.-L. Jämsä-Jounela, “Future trends in process
estimation with negative information”, Proc. IEEE
automation”, Annual Reviews in Control, vol. 31, no.
Int. Conf. on Information Fusion, pp. 2192-2199,
2, 2007, pp. 211-220.
2013.

424