Available online at www.sciencedirect.

com

ScienceDirect
IFAC PapersOnLine 55-22 (2022) 183–188
LQR-based
LQR-based MIMO
MIMO PID
PID control
control of
of a
a 2-DOF
2-DOF
LQR-based
LQR-based MIMO
helicopter
MIMO PID
system
PID control
with of a
uncertain
control of 2-DOF
LQR-based MIMO
helicopter PIDwith
system control of a 2-DOF
a
uncertain2-DOF
helicopter
helicopter system
system with
cross-coupled uncertain
gain
with uncertain
helicopter system with
cross-coupled
cross-coupled uncertain
gain
gain
cross-coupled
cross-coupled gain
gain Falguni Gopmandal, Arun Ghosh
Falguni Gopmandal, Arun Ghosh
Falguni
Falguni Gopmandal,
Gopmandal, Arun
Arun Ghosh
Ghosh
Falguni Gopmandal,
Falguni Department,
Gopmandal,IIT Arun
Arun Ghosh
Ghosh West Bengal,
Electrical Engineering Kharagpur,
Electrical Engineering Department, IIT Kharagpur, West Bengal,
Electrical Engineering
Electrical Engineering
Engineering Department, India-721302.
Department,
Department, IIT Kharagpur,
Kharagpur, West
IIT Kharagpur, West Bengal,
Bengal,
Electrical India-721302. IIT West Bengal,
(e-mails:
Electrical falgun.ece@gmail.com;
Engineering Department,
India-721302.
India-721302. arunghosh1@gmail.com)
IIT Kharagpur, West Bengal,
(e-mails: falgun.ece@gmail.com; India-721302. arunghosh1@gmail.com)
(e-mails:
(e-mails: falgun.ece@gmail.com; India-721302.
falgun.ece@gmail.com; arunghosh1@gmail.com)
arunghosh1@gmail.com)
(e-mails:
(e-mails: falgun.ece@gmail.com;
falgun.ece@gmail.com; arunghosh1@gmail.com)
arunghosh1@gmail.com)
Abstract: This work deals with designing multi-input multi-output (MIMO) proportional-
Abstract: This work deals with designing multi-input multi-output (MIMO) proportional-
integral-derivative
Abstract: This
This work (PID) deals controllers
with
with designing to achieve linear multi-output
multi-input quadratic (LQ) performance of a
Abstract:
integral-derivative
Abstract: This work
work(PID) deals
deals controllers
with designing
to achieve
designing linear multi-output
multi-input
multi-input quadratic (LQ)
multi-output
(MIMO)
(MIMO)
(MIMO)
proportional-
proportional-
performance
proportional- of a
laboratory-based
Abstract:
integral-derivative This
integral-derivativetwo two
work degree-of-freedom
(PID)
(PID) deals with
controllers designing
controllers to to(2-DOF)
achieve
to(2-DOF) helicopter
multi-input
linear
achievehelicopter system
multi-output
quadratic
linear quadratic
quadratic with
(LQ)
(LQ) norm-bounded
(MIMO) performance
performanceuncer- uncer-
proportional- of
of aaa
laboratory-based
integral-derivative degree-of-freedom
(PID) controllers achieve linear system with (LQ) norm-bounded
performance of
tainties in
integral-derivative
laboratory-based
laboratory-based the cross-coupled
two (PID) gain.
controllers
degree-of-freedom
two degree-of-freedom
degree-of-freedom It is shown
to(2-DOF)
(2-DOF) that,
achieve for
linear
helicopter
helicopter such an
quadratic
system
system uncertain
(LQ)
with system,
performance
norm-bounded
with norm-bounded
norm-bounded the MIMO
uncer-a
of
uncer-
tainties in the cross-coupled
laboratory-based two gain. It is shown (2-DOF) that, for suchsystem
helicopter an uncertain
with system, the MIMO uncer-
PID design
laboratory-based
tainties
tainties in
in problem
the
the two can
cross-coupled be recast
degree-of-freedom
cross-coupled gain.
gain. as is
It
It isa full-state
(2-DOF)
shown
shown that,
that,feedback
helicopter
for
for such
such control
system
an
an for an norm-bounded
with
uncertain
uncertain augmented
system,
system, uncertain
the
the uncer-
MIMO
MIMO
PID design
tainties in problem can be recast as isa full-state feedback control for an augmented uncertain
system.
tainties
PID
PID
system. design
designin the
the cross-coupled
Consequently,
cross-coupled
problem
problem
Consequently, can
cana simple
be
gain.
be recast
a simple
linear
gain.
recast
linear
It
It
as is
as
shown
matrix
aa full-state
full-state
matrix
that,
inequality
showninequality
that, for
feedback
feedback
such
problem
forproblem
such an
control
control isuncertain
an is solved
uncertain
for
for an
an to
solved
to system,
determine
system,
augmented
augmented
determine
the
the MIMO
the
MIMO
uncertain
uncertain
the
PID
PID
PID
PID design
parameters
design problem
in order
problem can
to
can be
achieve
be recast
robust
recast as
as a
aLQ full-state feedback
performance.
full-state feedback The control
efficacy
control for
forofan
an the augmented
proposed
augmented uncertain
controller
uncertain
system. Consequently,
system. Consequently,
parameters Consequently, a
in order to aaachieve simple linear
simple robust matrix
linear matrix inequality problem
inequality problem
LQ performance. The efficacy is solved
is solved
of the to determine
toproposed the
determinecontroller
the PIDPID
system.
is shownConsequently,
system. through extensive simple linear
linear matrix
simulations inequality
inequality problem
andperformance.
experiments. is
is solved to determine the
the PID
parameters
parameters
is shown through
parameters
in
in order
in order
order to aachieve
to
extensive
to
simple
achieve
achieve
robust
robust
simulations
robust
matrix
LQ
LQ
LQ andperformance.
experiments.
performance.
problem
The
The efficacy
The efficacy
efficacy
solved
of
of the
of the
the
toproposed
determinecontroller
proposed
proposed controller
controller
PID
parameters
is shown
Copyright
is shown in
through order
© 2023 The
through to
extensive
extensiveachieve robust
simulations
This
Authors.simulations is
simulations an LQ and
open
and performance.
experiments.
access article
experiments. The
under efficacy
the CC of the
BY-NC-ND proposed
license controller
is shown
shown through extensive and
and experiments.
is through
MIMOextensive
(https://creativecommons.org
Keywords: PID control, /licenses/by-nc-nd/4.0/)
simulations
2-DOF helicopter, experiments.
norm-bounded uncertainty.
Keywords: MIMO PID control, 2-DOF helicopter, norm-bounded uncertainty.
Keywords:
Keywords: MIMO
MIMO PID
PID control,
control, 2-DOF
2-DOF helicopter,
helicopter, norm-bounded
norm-bounded uncertainty.
uncertainty.
Keywords: MIMO
Keywords: MIMO PID PID control,
control, 2-DOF2-DOF helicopter,
helicopter, norm-bounded
norm-bounded uncertainty. uncertainty.
1. INTRODUCTION PID design problem is handled by transforming it into a
1. INTRODUCTION PID design problem is handled by transforming it into a
1.
1. INTRODUCTION
INTRODUCTION static
PID
PID output
design
design feedback
problem is (SOF) problem.
handled by In Pradhan
transforming it and
into a
1. INTRODUCTION
INTRODUCTION static
PID outputproblem
design feedback
problem is
is handled
(SOF) problem.
handled by
by transforming
In Pradhan
transforming it
it into
and
into a
a
The laboratory-based 1. two degree-of-freedom (2-DOF) he- static Ghosh
PID
static (2015),
design
output
output a
problem class
feedback
feedback of
is plants
handled
(SOF)
(SOF) forbywhich
problem.
problem. this
transforming
In
In SOF
Pradhan
Pradhan problem
it into
and
and a
The laboratory-based two degree-of-freedom (2-DOF) he- static Ghosh
static (2015),
output a class
feedback of plants
(SOF) for which
problem. this
In SOF
Pradhan problemand
The licopter system
laboratory-based is a benchmark
two multivariable
degree-of-freedom (2-DOF)unstable he- becomes
Ghosh
Ghosh outputcomplete
(2015),
(2015), a
a state
feedback
class
class of
of feedback
(SOF)
plants
plants for
for problem
problem.
which
which In
this
thisis Pradhan
obtained
SOF
SOF problem
problemand
The laboratory-based
system is a two degree-of-freedom (2-DOF) he- becomes
Ghosh complete
(2015), state feedback problem is SOF
obtained and
licopter
The The
system
licopter
laboratory-based
with strong
laboratory-based
system is
benchmark
two
cross-coupling.
aa two
multivariable
degree-of-freedom
It is an ideal
degree-of-freedom
benchmark multivariable
(2-DOF)unstable
system
(2-DOF)unstable
he-
he- nominal
to Ghosh
becomes
becomes linearaa quadratic
(2015),
complete
complete
class
class of
of plants
state
state (LQ)for
plants
feedback
feedback
which
which this
forperformance
problem
problem thisis SOF
is obtained
obtained
problem
is achieved.
problemand
and
licopter
systemlicopter system
with strong
system is benchmark
cross-coupling.
is aa benchmark
benchmark multivariable
It is an
multivariableideal unstable
system to
unstable becomes nominal
becomes linear
complete quadratic
state (LQ)
feedback performance
problem is is
obtainedachieved.
and
assess the
licopter
system performance
system
with strong is of a multivariable
cross-coupling. multivariable
It is an or multi-input
ideal unstable
system to Recently
nominal
nominal in Pradhan
complete
linear
linear state
quadratic
quadratic and Ghoshperformance
feedback
(LQ)
(LQ) (2022), the
problem
performance MIMO
is obtained
is
is PID
and
achieved.
achieved.
system
assesssystem with
the
with strong
performance
strong cross-coupling.
of a multivariable
cross-coupling. It
It is
is an
an ideal system
or multi-input
ideal system to
to Recently
nominal in
linear Pradhan
quadratic and Ghosh
(LQ) (2022),
performance the MIMO
is PID
achieved.
multi-output (MIMO) controller. Numerous decentralized
system to Recently controller
nominal
Recently is
linear
in
in designed
quadratic
Pradhan not
and only
(LQ)
Ghosh to achieve
performance
(2022), LQ
the performance
is
MIMO achieved.
PID
system
assess
assess withperformance
the
the strong
performance cross-coupling.
of
of It is an ideal
a multivariable
of aaa multivariable
multivariable or
or multi-input
multi-input controller
Recently is Pradhan
in designed not
Pradhan and
and Ghosh
only
Ghosh (2022),
to achieve
(2022), the
LQ
the MIMO
performance
MIMO PID
multi-output
assessassess
or multi-loop
multi-output the
(MIMO)
the performance
performance
control schemes
(MIMO)
controller.
of
controller.
Numerous
multivariable
haveNumerous
been proposedordecentralized
or multi-input
for this under
multi-input
decentralized Recently
controller
controller
under
nominal
nominal in
is condition,
is Pradhan
designed
designed
condition, andbut
not
not Ghosh
only
only
but
into
intopresence
(2022),of
achieve
achieve
presence ofLQ
LQ MIMO PID
norm-bounded
the performance
performance
norm-bounded PID
multi-output
or (MIMO)
multi-loop (MIMO)
multi-output control schemes controller.
controller. Numerous
haveNumerous
been proposed decentralized controller
for this controller
decentralized is designed not only to achieve LQ performance
system;
multi-output
or see
multi-loop Hernandez-Gonzalez
(MIMO)
control controller.
schemes have et al.
Numerous
been (2012); Wen
decentralized
proposed for and
this parametric
under
under nominal
nominal is uncertainties
designed
condition,
condition, not in the
only
but
but in
into plant
achieve
presence
presence asof
ofLQwell. However,
performance
norm-bounded
norm-bounded
or
system;
or multi-loop control schemes
see Hernandez-Gonzalez
multi-loop control schemes haveetbeen proposed
al. (2012); Wen for and parametric
this under nominal uncertainties
ofcondition,
derivativebut
in thein plant asof well.
presence However,
norm-bounded
or Lumulti-loop
(2008); Goyal
controlet al. (2020)have
schemes been
andetbeen
the proposed
references for this
this
therein. the
under inclusion
parametric
parametric nominal ofcondition,
derivativebut
uncertainties
uncertainties filter
in
in in dynamics
the
the presence
plant
plant as
asofin the controller
norm-bounded
inwell.
well. However,
However,
system;
system;
Lu (2008);
system;
see
see
see
Hernandez-Gonzalez
Hernandez-Gonzalez
Goyal et al.
Hernandez-Gonzalez (2020)have andetettheal.
al.
al.
proposed
(2012);
(2012);
references
(2012);
Wen
Wen
Wen
for and
and
therein.
and the
design
inclusion
parametric
parametric
the makes
inclusion
uncertainties
this approach
uncertainties
of derivative
filter
in
in the dynamics
plant
iterative,
the
filter plant
dynamics
as
even
as in
the controller
well.
for
well.
the the However,
class of
However,
controller
Lu However,
system;
Lu (2008);
(2008); the
see system
Goyal et being
Hernandez-Gonzalez
Goyal et al.
al. highly
(2020)
(2020) and
and coupled,
etthe
theal. such
(2012);
references
references controllers
Wen and
therein.
therein. the
design
the inclusion
makes
inclusion of
this
of derivative
approach
derivative filter dynamics
iterative,
filter dynamics even in
in the
for
the thecontroller
class of
controller
However,
Lu (2008); the system
Goyal et being
al. highly
(2020) and coupled,
the such controllers
references therein. plants
the
design as
inclusion
makes mentioned
of
this above.
derivative
approach Moreover,
filter dynamics
iterative, in
even this
in method,
the
for the classone
controller of
Lu yield
However,
However, poor
(2008); performance
Goyal
the
the system
system et al.
being
being in general.
(2020)
highly
highly and To
the
coupled,
coupled, get rid
references
such
such of the
controllers
controllerscou-
therein. design
plants
design makes
as
makes this
mentioned
this approach
above.
approach iterative,
Moreover,
iterative, even
in
even thisfor
for the
method,
the class
class of
one
of
yield However,poorthe performance
system being in highly
general. To get such
coupled, rid of the cou- plants
controllers positive
design makes
as scalar parameter,
this
mentioned approach
above. namely
iterative,
Moreover, ϵ, which
even
in thisforarises
the
method, dueone
class to
of
yield pling,
However,
yield the
poor
poor centralized
the system
performance
performance multivariable
being in
in highly
general.
general. controller
coupled,
To
To get
get such
rid
rid in
of
of Pradhan
controllers
the
the cou-
cou- plants
positive
plants as
as mentioned
scalar
mentioned above.
parameter,
above. Moreover,
namely
Moreover, ϵ, in
which
in this
this method,
arises
method, due one
to
one
pling,yield the
poor centralized
performance multivariable
in general. controller
To get rid in
of Pradhan
the cou- presence
plants
positive as of norm-bounded
mentioned
scalar above.
parameter, uncertainties,
Moreover,
namely ϵ, in
which has
this to be
method,
arises chosen
due one
to
yield and Ghosh
poor (2013) has multivariable
performance been designedcontroller
in general. via open-loop decou- positive scalar parameter, namely ϵ, which has arises due to
pling,
pling,
and pling,
the
the
Ghosh centralized
(2013) has multivariable
centralized
the centralized
centralized been designedTo
multivariable viaget
controller rid in
of Pradhan
in
open-loop
controller
the
Pradhan cou- presence
decou-
inapproach,
Pradhan positive
apresence
of norm-bounded
scalar
priori. Naturally
positive scalar parameter, uncertainties,
namely
it is interesting
parameter,
of norm-bounded
norm-bounded namely to ϵ, which
ϵ, see
uncertainties, whichifhas
has
to be chosen
arises
these
arises
to be
due
due to
problems
be chosen to
pling,
and
and plingGhosh
method.
the
Ghosh Besides
(2013)
(2013) has
has being
been
been a polynomial-based
multivariable
designed
designed controller
via
via in
open-loop
open-loop Pradhan
decou-
decou- presence
a priori.
presence of
Naturally
of it
norm-bounded is uncertainties,
interesting to
uncertainties, see if
has to
these
to chosen
problems
be chosen
pling and method.
Ghosh Besides
(2013) has being
been the a polynomial-based
designed via open-loop
open-loop approach,
decou- can presence
a be avoided
priori. of with
norm-bounded
Naturally it is ideal PID controllers,
uncertainties,
interesting to see has
if atbe
to
these least
chosen
problems for
and
pling in this
Ghoshmethod,
method. (2013) however,
Besideshas been
being designed
a controller
via
polynomial-based order becomes
decou-
approach, a priori. Naturally it is interesting to see if these problems
pling
in this
pling method.
method,
method. Besides
however,
Besides being
beingthe a polynomial-based
controller orderapproach, becomes acan be avoided
priori. Naturally with
it is ideal PID controllers,
interesting to see
see if theseat problems
least for
pling
in
in
high.
high.
this
this Several
method.
method,
method,
Several
other
Besides
other
advanced
however,
however,being
advanced aa polynomial-based
the
the
polynomial-based
control
controller
controller
control
techniques
order
order
techniques
approach,
like H
approach,
becomes
becomes
like H∞
athe
can
can
the
can
class
priori.
be
be
class
be
of plants
Naturally
avoided
avoided
of
avoided plants itfor
with
with
withfor
which
isideal
interesting
ideal
which
ideal
PID
PID
PID
ideal
idealto PID
controllers,
controllers,
PID
controllers,
ifdesign
these
designat
at
at
becomes
problems
least
least
becomes
least
for
for
for
in
control,
in this
this method,
fuzzy other
method, however,
control,
however, the
predictive
the controller
control,
controller order
adaptive
order becomes
control,
becomes ∞ equivalent
can
the
the be
class
class ofto
avoided
of the
plants
plants complete
withfor
for ideal
which
which state
PID ideal
ideal feedback
controllers,
PID
PID problem,
design
designat least
becomes
becomes as
for
high.
high.
control, Several
Several other
fuzzy other
control, advanced
advanced
predictive control
control
control, techniques
techniques
adaptivelikelike
like H
H
control, ∞ equivalent
the class ofto the
plants complete
for which state
ideal feedback
PID problem,
design becomes as
high.
sliding
high. Several
mode
Several control,
other advanced
neural
advanced control
network
control techniques
control,
techniquesfeedback
like H
H
∞
∞
lin- identified
the class
equivalent
equivalent in
ofto
toPradhan
plants
the
the forand
complete
completewhichGhoshstate (2015).
ideal
state PID
feedback
feedback design becomes
problem,
problem, as
as
control,
control, fuzzycontrol,
fuzzy control,neural
control, predictive
predictive control,
control, adaptive
adaptive control,
control, identified
equivalent intoPradhan
the and Ghosh (2015).
the complete state feedback
feedback problem,problem, as
sliding mode network control, feedback ∞
lin-
control,
earizing
control, fuzzy
control,
fuzzy control,
etc.,
control, predictive
have also
predictive control,
been adaptive
employed;
control, adaptive see control,
López-
control, equivalent
identified
identified in
intoPradhan
Pradhan complete
and
and Ghosh
Ghoshstate(2015).
(2015). as
sliding
sliding mode
mode control,
control, neural
etc., neural network
network control,
control, feedback
feedback lin-
lin- In this paper,
identified in we attempt to solve (2015).
the above problem for a
earizing
sliding
Martı́nez
sliding
earizing
control,
mode
et
mode control,
al.
control, (2020);
control,
etc.,
have also
neural
Mendez-Monroy
neural
have
been employed;
network
network
also been
control,
and
control,
employed;
see López-
feedback
Benitez-Perez
feedback
see
lin-
lin-
López- In this paper,
identified in Pradhan
we attempt
Pradhan and
and to Ghosh
solve (2015).
Ghosh the above problem for a
earizing
Martı́nez
earizing control, etc.,
et al. (2020);
control, etc., have also
Mendez-Monroy
have also been
been employed; see
and Benitez-Perez
employed; see López-
López- laboratory-based
In this
In this paper,
paper, we 2-DOF helicopter
we attempt
attempt to solve
to solve the system.
the aboveFor
above simplicity,
problem
problem for a
for a
(2012);
earizing
Martı́nez
Martı́nez
(2012); Dutka
Dutka
et al. et
control,
et al. al. (2020);
etc.,
(2020);
(2020);
et have alsoNuthi
Mendez-Monroy
Mendez-Monroy
al. (2020); beenand
Nuthi and
Subbarao
employed;
and
and Benitez-Perez
Subbarao López- laboratory-based
see (2015);
Benitez-Perez
(2015);
In
we
In this paper,
consider
this paper,
laboratory-based
we
the
we
2-DOF helicopter
attempt
norm-bounded
attempt
2-DOF
to
to solve
solve
helicopter
system.
the above
uncertainties
the above
system.
For simplicity,
problem for
in simplicity,
problem
For the cross-
for
simplicity,
a
a
Martı́nez
Gao et et
al. al. (2020);
(2016); Mendez-Monroy
Boukadida et al. and
(2019); Benitez-Perez
Sadala and laboratory-based
we consider
laboratory-based the 2-DOF
norm-bounded
2-DOF helicopter
helicopter system.
uncertainties
system. For
Forin the cross-
simplicity,
Martı́nez
(2012);
(2012);
Gao et
Dutka
et Dutka al.
al. (2016);(2020);
et al.
et Mendez-Monroy
al.Boukadida
(2020); Nuthi
(2020); Nuthi
et al. and
and and
(2019); Benitez-Perez
Subbarao
SubbaraoSadala (2015);
(2015); coupled
and laboratory-based
we consider gainthe only.2-DOFThishelicopter
norm-bounded system being
system.
uncertainties inForthe
in class
simplicity,
the cross-as
(2012); Dutka et al. (2020); Nuthi and Subbarao (2015); we consider
coupled
we consider gain the norm-bounded
only. This system uncertainties
being in in
the the cross-
class as
Patre
(2012);
Gao
Gao (2018);
et
et Dutka
al.
al. Zhao
et al.et
(2016);
(2016); al. (2022);
(2020);
Boukadida
Boukadida Nuthi
et
etLopez-Martinez
and
al.
al. Subbarao
(2019);
(2019); Sadala
Sadalaand Ru-
(2015);
and
and mentioned
we consider inthe
the norm-bounded
above
Thisparagraph,
uncertainties
it is in shownin
in the thatcross-
the
Patre
Gao
bio
Gao
Patre
(2018);
et
(2003).
et al.
al.
(2018);
Zhao
(2016);
However,
(2016);
Zhao
et al. (2022);
Boukadida
the
Boukadida
et al. majority
(2022);
et
et
Lopez-Martinez
al.
of
al. (2019);
these
Lopez-Martinez
Sadala
controllers
(2019); Sadala
and
and
Ru-
and
have
and
Ru-
coupled
coupled
mentioned
coupled
gain
gain
gain inthe norm-bounded
only.
only.
the
only.
This
above
This
system
system
paragraph,
system
uncertainties
being
being
it is in
being in the
the
shown
the
the
thatcross-
class
class
class
as
as
the
as
Patre
bio
Patre (2018);
(2003).
(2018); Zhao
However,
Zhao et
et al.
the
al. (2022);
majority
(2022); Lopez-Martinez
of these controllers
Lopez-Martinez and
and Ru-
have
Ru- MIMO
coupled
mentioned
mentioned PID
gain in
in design
only.
the
the here
This
above
above cansystem
paragraph,
paragraph,be transformed
being
it
it is
is in the
shown
shown intothata full-
class
that as
the
the
complicated
Patre (2018); structures
Zhao etthe and
al. are
(2022); of higher order. Therefore, MIMO
mentioned PID in design
the here
above can
paragraph,be transformed
it is shown into a
that full-
the
bio
bio
complicated
bio
(2003).
(2003).
(2003).
However,
However,
structures
However,
the
the and are ofLopez-Martinez
majority
majority
majority
of these
of theseorder.
higher
of these
controllers
controllers andhave
Therefore,
controllers
Ru- mentioned
have
have state
MIMO
MIMO feedback
PID
PID in control
the
design
design above
here
herefor paragraph,
an augmented
can
can be
be it is
transformed
transformeduncertain
shown into
into system.
that
a
a the
full-
full-
it
bio is(2003).
complicated
complicated desirable to design
However,
structures
structures amajority
theand
and loware
are order
of
of ofMIMO
theseorder.
higher
higher controller
controllers
order. Therefore,
Therefore,that state
have MIMO
As
MIMO
feedback
PID
a feedback
result,
PID the
control
design
design PIDhere
for can
gains
an augmented
can be
be betransformeduncertain
computed into system.
a
a full-
by directly
it
can
is desirable
complicated
complicated
it is achieve
desirable
to design
structures
satisfactory
structures
to design
a low
and
and
a
are order
performance
loware of
of
order
MIMO
higher
higherand
MIMO
controller
order. Therefore,
robustness.
order. Therefore,
controller
that state
that
state
As
state a feedback
result,
feedback PIDhere
control
control
the
control
for
for
gains
for
can
an
an
an can betransformed
augmented
augmented
augmented
uncertain
uncertain
computed
uncertain
into
by directly full-
system.
system.
system.
it
can is desirable
achieve to design
satisfactory a low order
performance MIMO
and controller
robustness. that solving
state
As
As a a
feedback
result,
aa result, linear the
the matrix
control
PID
PID inequality
for
gains
gains an (LMI)
augmented
can
can be be in order
uncertain
computed
computed byto achieve
system.
directly
it is desirable to
to design
design a low
low order
order MIMO controller
controller that that solvingAs a linear matrix inequality (LMI) in orderby to directly
achieve
it
can
can
As
can
is achieve
desirable
isachieve satisfactory
satisfactory
well known,
achieve satisfactory
aperformance
performance MIMO
proportional-integral-derivative
performance
and robustness.
and
and
robustness.
robustness. (PID) robust As
solving
solving
robust a result,
LQ
result,
aaa linear
LQ linear
the PID
PID gains
performance.
thematrix
matrix
performance. gainsThecan
inequality
inequality
The
PIDbe
can
PID be
(LMI)
(LMI)
computed
controller
computed
in order
in
controller
thus
order
thus
by
byto
directly
todesigned
directly
achieve
achieve
designed
As isachieve
can well known,
satisfactoryproportional-integral-derivative
performance and robustness. (PID) solving solving
is also LQ a linear
implemented
linear matrix
matrix inequality
on thePID
inequality (LMI)
physical
(LMI) in order
setup.
in order to achieve
Extensive
to achieve
controllers
As
As is
is well
well are
known,
known,of low order and easy to
proportional-integral-derivative
proportional-integral-derivative implement in both
(PID)
(PID) robust
robust
is also LQ performance.
performance.
implemented on The
Thethe PID controller
controller
physical setup. thus
thus designed
designed
Extensive
controllers
As is well are
known,of low order and easy to
proportional-integral-derivative implement in both
(PID) robust
simulations
robust LQ
LQ performance.
and experiments
performance. The
The PID
are
PID controller
carried
controller outthus
to
thus designed
compare
designed
analog
As
controllers and are
is well digital
known, settings
order[Ang and et al. to
(2005)].
proportional-integral-derivative In literature,
in(PID) is
is also
also implemented
implemented on
on the
the physical
physical setup.
carriedsetup. Extensive
out toExtensive
controllers
analog
controllers and are
are
of low
of
digital
of
low
low order
settings
order and
[Ang
and
easy
easy
et
easyal. to
to
implement
implement
(2005)].
implement in
In literature,
in
both simulations
both
both is
its
is also
performance
also
simulations
and experiments
implemented
implemented
and with on
the
on PID
experiments
the are
the are physical
control
physical
carried
setup.
designed
setup.
out to to
compare
Extensive
achieve
Extensive
compare
acontrollers
analog
analog lot ofandefforts
and are
digital
digitalhave
of low been
order
settings
settings put
and
[Ang
[Ang inet
et the
easyal.
al. area
to
(2005)].
(2005)]. of In
(centralized)
implement
In in both
literature,
literature, simulations
its performance
simulations and
and experiments
with the
experiments PID are carried
control
are carried out
designed
out to
to to compare
achieve
compare
aanalog lot ofandefforts
digitalhave been put
settings [Ang inetthe
al. area of In
(2005)]. (centralized)
literature, only nominaland
simulations
its performance LQwith performance.
experiments
with the PID are carried
control out
designed to to compare
achieve
MIMO
analog
aaMIMO lot
lot of
of PID
and control
digital
efforts
efforts have
have design;
settings
been
been [Ang
put
put seein
in Mattei
et al.
the
the area
area (2001);
(2005)]. of
of In Lin
(centralized)
(centralized) et
literature, al. its
only
its performance
nominal
performance LQ the
performance.
with the PID control designed to achieve
aa(2004);
lot
lot
MIMO
of
of
PID
Bianchi
control
efforts
efforts
PID
have
et
have
control al.
design;
been
(2008);
been
design;
put
put
seeinMattei
the
Pradhan
seein the
Mattei
area
and
area
(2001);
of
ofGhosh
Lin et al. its
(centralized)
(2015);
(centralized)
(2001); Lin et al. only
only performance
nominal
nominal LQ
LQ with the PID control designed to achieve
performance.
performance.
PID control designed to achieve
MIMO
(2004);
MIMO PID
Bianchi
PID control
et al.design;
control (2008);see
design; Mattei
Pradhan
see Mattei and(2001);
Ghosh
(2001); Lin
Lin et
(2015);
et al.
al. The
only rest
nominal of theLQ paper is organized as follows. In Section 2,
performance.
Gopmandal
MIMO
(2004); PID and
Bianchi Ghosh
control
et (2021);
al.design;
(2008); see Pradhan
Matteiand
Pradhan and
and Ghosh
(2001);
Ghosh (2022)
Lin(2015);
et al. the
(2015); The
only rest
nominal of the LQ paper is organized
performance. as follows. In Section 2,
(2004);
Gopmandal
(2004); Bianchi
and
Bianchi et al.
Ghosh
et al. (2008);
(2021);
(2008); Pradhan
Pradhan
Pradhan and
and Ghosh
Ghosh
Ghosh (2022)
(2015); The
The state-space
rest of
rest of the model is
the paper
paper isof the system
organized
organized asis
as obtained.
follows.
follows. In Next, the
In Section
Section 2,
2,
and
(2004);references
Bianchi therein.
et In
al. (2008); most of the methods, the MIMO the
The state-space
rest of the model
paper of
is the
organizedsystem as is obtained.
follows. In Next,
Section the
2,
Gopmandal
Gopmandal
and references
Gopmandal
and
and
and
Ghosh
Ghosh
therein.
Ghosh In mostPradhan
(2021);
(2021);
(2021);
Pradhan
Pradhan and
of the methods,
Pradhan
andGhosh
and
and the (2015);
Ghosh
Ghosh
Ghosh
(2022) The
(2022)
MIMO
(2022) the
the
the
rest of the paper
state-space
state-space
state-space
model
model
model
isof
of
of
organized
the
the
the
system
system
system
asisfollows.
is
is
obtained.
obtained.
obtained.
In Section
Next,
Next,
Next,
2,
the
the
the
Gopmandal
and
and references
referencesand Ghosh
therein.
therein. (2021);
In
In most
most Pradhan
of
of the
the and
methods,
methods, Ghosh
the
the (2022)
MIMO
MIMO the state-space model of the system is obtained. Next, the
and
and references
references therein.
therein. In most of
of the methods, the MIMO
2405-8963 Copyright © 2023In Themost
Authors. the methods,
This is an openthe accessMIMO
article under the CC BY-NC-ND license.
Peer review under responsibility of International Federation of Automatic Control.
10.1016/j.ifacol.2023.03.031
 2 Modeling
2.1 Background
184 Falguni Gopmandal et al. / IFAC PapersOnLine 55-22 (2022) 183–188
2.1.1 Equations of Motion
The free-body diagram of the Quanser Aero Experiment is illustrated in Figure 2.1.

Fig. 1. Quanser 2-DOF helicopter setup.

MIMO PID design methodology is described in Section 3.
The simulations and experimental results are presented in
Section 4. Finally, the conclusions are drawn in Section 5. Figure 2.1: Simple free-body diagram of Quanser Aero Experiment
Fig. 2. Free-body diagram.
The following conventions are used for the modeling:
2. SYSTEM DESCRIPTION Table 1. System parameters.
1. The helicopter is horizontal and parallel with the ground when the pitch angle is zero, i.e. θ = 0.
Parameters ˙ > 0, when the front rotor is Value/range
The laboratory-based helicopter model to be used is 2.shown
The pitch angle increases positively, θ(t)
Moment of about
inertia
moved upwards and the body rotate
0.0219 kgm2
counter-clockwise (CCW) the about
Y axis. pitch axis (Jp )
in Fig. 1. It is manufactured by Quanser Inc, Canada. Moment of inertia about 0.0220 kgm2
˙ > 0, yaw
3. The yaw angle increases positively, ψ(t)
axis (Jy )
when the body rotates counter-clockwise (CCW) about the
Unlike the normal helicopter model which has 3-DOF Damping about pitch axis (Dp ) 0.00711 Vs/rad
axis.
motions (namely, pitch, yaw, roll), it has 2-DOF motions Damping about yaw axis (Dy ) 0.0220 Vs/rad
4. Pitch increases, θ̇ > 0, when the front rotor voltage is positive Vp > 0.
(pitch and yaw) [see Inc. (2016)]. In this setup, the motion Stiffness about pitch axis (Ksp ) 0.0375 Nm/rad
about the pitch axis is controlled by the front rotor,5.which
Yaw increases,Torque
ψ̇ > 0, when the gain
thrust back (or
fromtail)pitch
rotor voltage
rotor is(K
positive,
pp ) V0.0011
y > 0. Nm/V
is horizontal to the ground, whereas the motion about the Torque thrust gain from yaw rotor (Kyy ) 0.0022 Nm/V
When voltage is applied to the pitch motor, Vp , the speed of rotation results in a force, Fp , that acts normal to th
yaw axis is controlled by the back or tail rotor. While
body atin a
a distance
Cross-torque thrust gain acting on pitch
rp from the pitch axis. The rotation of the propeller generates a torque about the pitch rotor mot
0.0021 Nm/V
the is in turnrotor
shaft which
normal helicopter, maneuvering is obtained by tilting from the
seen about yawyaw
rotor
axis.(K py )rotating the pitch propeller does not only cause motion abo
Thus
the pitch axis but also about the yaw thrust
Cross-torque axis. As gain
described
actingearlier,
on that’s
yaw why conventional helicopters include a tail,
rotor blade angle, in this setup, the same is carried out
anti-torque, rotor to compensate −0.0027 Nm/V
rotor fromfor the torque
pitch rotorgenerated
(Kyp ) about the yaw axis by the large, main rotor.
by changing the speed of the motors. The high-resolution
Voltage
Similarly, the yaw motor causesapplied
a forceto
Fy pitch
that actsrotor (Vpbody
on the ) at a distance ±24ry V from the yaw axis as well as
encoders are used to measure the pitch and yaw angles. ±24 V
torque about the pitch axis.
Voltage applied to yaw rotor (V ) y

2.1 Modelling We can develop a simple linear model that takes this coupling into account, and represents the motions of th
Quanser Aero about the horizontal, i.e. when the body is parallel with the ground. The equations of motion are:
Jp θ̈ + Dp θ̇ + Ksp θ = τp , (2.
The free-body diagram of the system is shown in Fig. 2.
Jy ψ̈ + Dy ψ̇ = τy (2.
In this figure, Fp is the aerodynamic force acting on the
pitch rotor and the corresponding torque τp about the
QUANSER AERO Laboratory Guide 2
pivot causes angular motion θ in pitch plane. Similarly
the aerodynamic force Fy causes torque τy which in turn
creates angular motion ψ in yaw plane. Again, because Fig. 3. Representation of the system with cross-coupled
of the self-rotational motion of pitch rotor a cross-torque uncertainties.
is produced, which prevents the yaw motion. Similarly,
the cross-torque produced by the yaw rotor rotational Now, consider the state vector x = [θ ψ θ̇ ψ̇]T , control
motion prevents the pitch motion. These cross-torques input up = [Vp Vy ]T and output y = [θ ψ]T . Then
cause significant cross-coupling between pitch and yaw from (1), (2), the state-space representation of the system
channels. In Inc. (2016), a simple spring-mass-damper type becomes
of linear model that takes this coupling into account, is ẋ = Ax + Bup , y = Cx (3)
considered. The equations of motion are described by where,
 
Jp θ̈ + Dp θ̇ + Ksp θ = τp , Jy ψ̈ + Dy ψ̇ = τy (1) 0 0 1 0
where τp , τy are related to their respective motor voltages  0 0 0 1 
A= ,
by −Ksp /Jp 0 −Dp /Jp 0
0 0 0 −Dy /Jy
τp = Kpp Vp + Kpy Vy , τy = Kyp Vp + Kyy Vy . (2)  
0 0  
Table 1 lists the model parameters and their respective
 0 0  1 0 0 0
values. In above, the term Ksp is because of the gravita- B= ,C = . (4)
Kpp /Jp Kpy /Jp  0 1 0 0
tional force acting on the vertical (i.e., pitch) motion. The Kyp /Jy Kyy /Jy
yaw motion being in the horizontal plane, no such term is
considered there. This makes the system to have a pole at
s = 0. It is noteworthy that in the above modelling, the Next, to explicitly consider the uncertainties in cross-
coupling between pitch and yaw motions is represented coupled gain we add a full-block, time-varying uncertainty
by the cross-terms Kpy and Kyp . As |Kpy | ≈ 2|Kpp |, but ∆(t) as shown in Fig. 3, where
 
|Kyp | ≈ |Kyy | (see Table 1), the coupling from Vy to θ is δ (t) δ12 (t)
∆(t) = 11 , ∥∆(t)∥2 ≤ δ. (5)
significantly more than the one from Vp to ψ. δ21 (t) δ22 (t)
 Falguni Gopmandal et al. / IFAC PapersOnLine 55-22 (2022) 183–188 185

Then, the uncertain system shown in Fig. 3 becomes  

θ
ψ 
 
ẋ = Ax + B[I + ∆(t)]u, y = Cx. (6) u = [Kp Kd Ki ]  θ̇  =⇒ u = KP ID x̄ (10)
 
The above system can be represented in standard form [see ψ̇
Petersen and Tempo (2014)] as ξ

where,
ẋ = Ax + [B + D∆B (t)E]u, y = Cx (7)
KP ID = [Kp Kd Ki ], x̄ = [xT ξ T ]T . (11)
where, D = B, E = δI2 and ∆B (t) is a norm-bounded Next, augmenting ξ with x, the augmented system be-
time-varying uncertain matrix satisfying ∥∆B (t)∥2 ≤ 1. comes
x̄˙ = Āx̄ + B̄(t)u + Gr, y = C̄ x̄ (12)
3. MIMO PID CONTROL DESIGN
where,
   
A 04×2 04×2
Ā = , B̄(t) = B̄0 + D̄∆B (t)Ē, G = ,
−C 02×2 I2
   
B D
C̄ = [C 02×2 ] , B̄0 = , D̄ = , Ē = E. (13)
02×2 02×2
Now, the control law (10) can be thought of as a state
es feedback control law for the augmented uncertain system
(12). Thus, designing the gains Kp , Ki , Kd for the original
eZ
uncertain system (7) becomes the design of state feedback
Fig. 4. MIMO PID controller. gain KP ID for the augmented uncertain system (12).

The MIMO PID controller to be used for the uncertain
he plant (7) is shown in Fig. 4. The control law is
tor  t
out dy
or u = Kp y + K d + Ki (r − y)dt (8)
dt 0 plant (12), such that some LQ performance is achieved.
sa
where Kp , Ki , Kd ∈ R2×2 . It may be noted that the
above controller structure is I-PD type, where the PD part
is kept in the feedback path and I-part in the forward
he
path. As is well known, the PD part in the feedback
.1)
path eliminates the derivative and proportional kicks on
output and control input, which occur in case of a standard
.2)
parallel PID controller, when a step reference input is
applied. The I-PD structure also makes the PID controller
2-DOF in nature, which helps one achieve improved r-to-y
and r-to-u responses as compared to the standard (i.e., 1-
DOF) PID structure; see Ang et al. (2005). Moreover, the
I-PD structure helps one transform the controller into a
state feedback control, as presented next. The presence
of pure derivative part, however, makes the feedback
system prone to measurement noise. Also, the same being
improper, it is difficult to implement. To this end, a
derivative filter is employed in cascade with the derivative
gain Kd , which is given by
1
F (s) = I2 (9)
τd s + 1
where the filter time constant is chosen as τd <<
∥Kd ∥2 /∥Kp ∥2 , so that it does not affect the behaviour of
the ideal PID control much Skogestad and Postlethwaite
(2007).
3.1 State feedback transformation of the control law
In this section, it is shown that the MIMO PID design for
the uncertain system (7) can be transformed into a state
feedback design for an augmented uncertain system.
t
Let, ξ = 0 (r −y)dt. As y = [θ ψ]T , one has ẏ = [θ̇ ψ̇]T .
Then, the control law in (8) can be rewritten as
3.2 Robust LQ performance
In this section, an LMI-based method is given to design
the state feedback gain KP ID for the uncertain augmented
To this end, we assume the pair (Ā, B̄0 ) stabilizable and
consider the LQR cost as
 ∞
J= (x̄T Qx̄ + uT Ru)dt, Q ∈ R6×6 , R ∈ R2×2 (14)
0
with Q = QT > 0, R = RT > 0 and initial state
x̄(0) = x̄0 . Then, the gain KP ID can be determined
following the result given below.
Proposition 1. The feedback system consisting of the
uncertain plant (7) and the PID controller (8) is quadrat-
ically stable 1 with cost J < γ iff there exist matrices
P̄ = P̄ T > 0, Y of appropriate dimensions and a real
scalar ϵ > 0 such that the following LMIs hold:
 
Sym(ĀP̄ + B̄0 Y ) + ϵD̄D̄T P̄ Y T (ĒY )T
 P̄ −Q −1
0 0 
 ≤0
Y 0 −R−1 0 
ĒY 0 0 −ϵI
  (15)
γ x̄T0
≥ 0. (16)
x̄0 P̄
where Sym(X) means X + X T . Then, KP ID becomes
KP ID = Y P̄ −1 . (17)
Proof of Proposition 1. Here S0 of Pradhan and Ghosh
(2022) becomes identity matrix. Therefore, on substituting
KP ID P̄ = Y , the bilinear matrix inequality (BMI) in
Theorem 2 of Pradhan and Ghosh (2022) becomes LMI.
Also, as no uncertainty is considered in the state matrix
A, the matrix Ē1 of Theorem 2 of Pradhan and Ghosh
1 A time-varying system ẋ(t) = f (x(t), t) is quadratically stable
if there exists a constant matrix P = P T > 0 such that with
Lyapunov function V (x(t)) = x(t)T P x(t), x(t) ̸= 0, the system
satisfies V̇ (x(t)) < V (x(t)), ∀ t; see Petersen and Tempo (2014).
186 Falguni Gopmandal et al. / IFAC PapersOnLine 55-22 (2022) 183–188
(2022) is set to zero in this case. Finally, (17) follows from
the relation KP ID P̄ = Y .
Remark 1. If there is no uncertainty, i.e., δ = 0, then
one may consider D̄ = 0, Ē = 0. Therefore the LMI (15)
is satisfied for any ϵ > 0. This implies, for nominal LQR
performance design, it suffices to consider the first 3 × 3
block of the LMI (15) with D̄ = 0.
3.3 Some design details
In this section, we first present some numerical details
of the proposed PID design that achieves robust LQ
performance. Numerical details are also provided for the
PID design that achieves only nominal LQ performance.
The latter results will be used for performance comparison
in Section 4.
To ensure robust LQ performance, we set the uncertainty
limit as δ = 0.7. Next, the LQR weights for the system
(12) are chosen as Q = diag{30, 30, 1, 1, 200, 200}, R =
diag{0.001, 0.001}. Note, in order to reduce tracking error,
more weight is given on ξ. Further, to reduce the overshoot,
some weights are put on θ and ψ. The derivatives of
outputs, i.e., θ̇, ψ̇ are given comparatively less weights. The
weight R is chosen to keep the control voltage well within
±24 V. Further, to give more emphasis on the tracking
error, the initial state of the augmented system is taken
as x̄0 = [0, 0, 0, 0, 1, 1]T . With these settings, the optimal
PID gains are obtained following Proposition 1 as
   
−367.17 481.28 −137.13 134.38
Kp = , Kd =
−326.86 −185.02 −124.65 −53.26
 
484.12 −605.31
Ki = . (18)
415.51 212.58
Finally, as ∥Kd ∥2 /∥Kp ∥2 = 0.3330, we choose the deriva-
tive filter time-constant τd = 0.0333 s.
Next, the MIMO PID controller is designed to achieve LQ
performance for the nominal plant (i.e., with δ = 0) only.
To this end, the LQR weights are kept same as before, so
that the nominal responses become similar. In this case,
the optimal PID gains are obtained following Remark 2 as
   
−328.61 400.42 −77.51 77.20
Kp = , Kd =
−279.25 −143.36 −69.01 −30.52
529.50 −649.27
Ki = . (19)
420.10 205.06
As ∥Kd ∥2 /∥Kp ∥2 = 0.2190, here, we choose τd = 0.0219 s.
4. SIMULATION AND EXPERIMENTAL RESULTS
In this section, we study the behaviour of both the nom-
inal and robust PID designs through simulations and ex-
periments. The real-time implementation diagram of the
controller is shown in Fig. 5. The controller is implemented
on MATLAB/Simulink (2021b) of the computer and the
built-in data acquisition (DAQ) card with sampling rate
of 500 Hz is used to interface with the physical system via
QFLEX 2 USB.
4.1 Frequency response study
To study the behaviour of the compensated system us-
ing frequency response, we consider sensitivity function
Table 2. Robustness margins for the uncertain plant.
PID control method ∥S∥∞ ∥T ∥∞ ∥SK∥∞
Proposed robust LQ control 1.67 1.38 3.34
Nominal LQ control 2.18 2.07 11.22
S(s), complementary sensitivity function T (s), and control
sensitivity function S(s)K(s), where S(s) = [I + L(s)]−1 ,
T (s) = L(s)[I +L(s)]−1 , L(s) = K(s)P (s) with P (s) being
the plant and K(s) the PID controller (along with filter). It
is well known that ∥S∥∞ is a measure of robustness against
parameter variations and ∥T ∥∞ is a measure of robustness
against worst-case multiplicative type unmodelled dynam-
ics of the plant. On the other hand, ∥SK∥∞ is used as a
measure of maximum control input magnitude as well as a
measure for robustness against unmodelled additive type
uncertainties; see Skogestad and Postlethwaite (2007).
To see the performance in presence of uncertainty, 50
randomly chosen plants P (s) are taken from the considered
uncertainty set. The proposed as well as nominal PID
controller have been employed for all these sample plants.
The singular value plots are shown in Figs. 6, 7, 8. The
peaks of those plots are mentioned in Table 2. It clearly
shows that with robust PID control, these magnitudes,
in particular ∥SK∥∞ , are much less as compared to the
nominal PID design, as expected. The large value of
∥SK∥∞ of the nominal PID signifies that it is very much
sensitive to uncertainties in cross-coupled gain. Also, it
is seen from Figs. 6, 7 that with both the designs, the
low frequency magnitude of S plot and high frequency
magnitude of T plot are more or less same. This implies
that the disturbance and noise attenuation capability of
the robust PID is not deteriorated w.r.t the nominal PID.
4.2 Time response
The real-time responses for both the robust and nominal
LQ performance-based PID designs are shown in Figs.
9, 10. Here, to create a worse situation, we consider
δ11 = δ12 = δ21 = δ22 = −0.3 (which satisfies δ = 0.6).
Fig. 10-(a) shows that the response with the nominal
LQ performance-based PID control deviates significantly
from the unperturbed one, when the uncertainty in the
cross-coupled gain occurs. On the other hand, as shown
in Fig. 9-(a), the perturbed responses with the proposed
robust LQ performance-based PID design remains close
to the unperturbed one, as expected. From Fig. 10-(b),
it is clear for the case of nominal LQ performance-based
PID design that, when the yaw command is changed,
because of the cross-coupling effect, Vp saturates for some
time duration (see in between 10 and 15 s and at 25 s),
which in turn causes significant deviations in the perturbed
outputs from the unperturbed ones. Note, as can be seen
from Fig. 9-(b), for the robust LQ performance-based
PID design, the above saturation occurs for comparatively
Fig. 5. Real-time implementation diagram.
 Falguni Gopmandal et al. / IFAC PapersOnLine 55-22 (2022) 183–188 187

15
20
10
0 5

θ (deg)
−20
0
Singular Values (dB)

−5
−40
−10
Proposed robust control
−60 Nominal control −15
0 5 10 15 20 25 30
−80
10 Reference input
−100 Nominal response
Perturbed response

ψ (deg)
−120 0

−140 −2 −1 0 1 2 3
10 10 10 10 10 10 −10
Frequency (rad/s)
0 5 10 15 20 25 30
Time (s)
Fig. 6. Singular value plot of S.
(a) Output.
10

0 20
−10

Vp (V)
Singular Values (dB)

Proposed robust control 0

−20
Nominal control
−30
−20
−40
0 5 10 15 20 25 30
−50

−60
20
−70
Vy (V)

−80 −2 −1 0 1 2 3
0
10 10 10 10 10 10
Frequency (rad/s)
−20

Fig. 7. Singular value plot of T . 0 5 10 15 20 25 30

Time (s)

20 (b) Control input.

10 Fig. 9. Responses with the robust LQ optimal MIMO PID
Singular Values (dB)

controller (real-time).
0
PID controller has also been designed following the same
−10
Proposed robust control approach. Both the controllers have been implemented in
−20
Nominal control real-time. The simulation and experimental results show
that, in presence of uncertainties, the behaviour of the
−30 PID controller designed to achieve robust LQ performance
−2 −1 0 1 2 3
is much superior to the one designed to achieve only the
10 10 10 10 10 10
Frequency (rad/s) nominal LQ performance, as expected.

Fig. 8. Singular value plot of SK. ACKNOWLEDGEMENTS

small duration of time in between 10 and 15 s, which
results in minute deviation of the perturbed outputs. Note, The authors acknowledge the financial support from Sci-
as already mentioned in the paragraph above (3), the ence and Engineering Research Board (SERB), New Delhi,
coupling from the other way is not that much significant. India, through the sponsored project with Project No.
CRG/2020/00843, to carry out this research.
5. CONCLUSIONS
REFERENCES
In this paper, an LQR-based MIMO PID controller has Ang, K.H., Chong, G., and Li, Y. (2005). PID control
been designed for a laboratory-based 2-DOF helicopter system analysis, design, and technology. IEEE Trans-
system with uncertain cross-coupled gain. The design actions on Control Systems Technology, 13(4), 559–576.
has been carried out by first transforming the problem Bianchi, F.D., Mantz, R.J., and Christiansen, C.F. (2008).
into a state feedback control for an augmented uncertain Multivariable PID control with set-point weighting via
system and then computing the state feedback gains by BMI optimisation. Automatica, 44(2), 472–478.
solving an LMI problem, so that robust LQ performance Boukadida, W., Benamor, A., Messaoud, H., and Siarry,
is achieved. For comparison, a nominal LQR-based MIMO P. (2019). Multi-objective design of optimal higher
188 Falguni Gopmandal et al. / IFAC PapersOnLine 55-22 (2022) 183–188

Hernandez-Gonzalez, M., Alanis, A.Y., and Hernández-

Vargas, E.A. (2012). Decentralized discrete-time neural
10
control for a quanser 2-DOF helicopter. Applied Soft
Computing, 12(8), 2462–2469.
θ (deg)

0
Inc., Q. (2016). Quanser aero laboratory guide. Technical
Report, ON, Canada.
−10
Lin, C., Wang, Q.G., and Lee, T.H. (2004). An improve-
0 5 10 15 20 25 30 ment on multivariable PID controller design via iterative
LMI approach. Automatica, 40(3), 519–525.
10 Reference input Lopez-Martinez, M. and Rubio, F.R. (2003). Control
Nominal response of a laboratory helicopter using feedback linearization.
ψ (deg)

Perturbed response
0 Proceedings of the European Control Conference (ECC),
3430–3435.
−10 López-Martı́nez, M., Vivas, C., and Ortega, M.G. (2020).
0 5 10 15 20 25 30 A multivariable nonlinear H∞ controller for a laboratory
Time (s) helicopter. Proceedings of the 44th IEEE Conference on
Decision and Control, 4065–4070.
(a) Output. Mattei, M. (2001). Robust multivariable PID control for
linear parameter varying systems. Automatica, 37(12),
30 1997–2003.
20 Mendez-Monroy, P. and Benitez-Perez, H. (2012). Fuzzy
10 control with estimated variable sampling period for non-
V (V)

0 linear networked control systems: 2-DOF helicopter as

p

−10 case study. Transactions of the Institute of Measurement

−20 and Control, 34(7), 802–814.
−30 Nuthi, P. and Subbarao, K. (2015). Experimental verifi-
0 5 10 15 20 25 30
cation of linear and adaptive control techniques for a
two degrees-of-freedom helicopter. Journal of Dynamic
20 Systems, Measurement, and Control, 137(6), 064501.
Petersen, I.R. and Tempo, R. (2014). Robust control of
Vy (V)

0 uncertain systems: Classical results and recent develop-

−20
0 5 10 15
Time (s)
(b) Control input.
Fig. 10. Responses with LQ optimal MIMO PID controller
(real-time).
order sliding mode control for robust tracking of 2-DOF
helicopter system based on metaheuristics. Aerospace
Science and Technology, 91, 442–455.
Dutka, A.S., Ordys, A.W., and Grimble, M.J. (2020). Non-
linear predictive control of 2 DOF helicopter model.
Proceedings of the 42nd IEEE International Conference
on Decision and Control, 3954–3959.
Gao, W., Huang, M., Jiang, Z.P., and Chai, T. (2016).
Sampled-data-based adaptive optimal output-feedback
control of a 2-degree-of-freedom helicopter. IET Control
Theory & Applications, 10(12), 1440–1447.
Gopmandal, F. and Ghosh, A. (2021). A hybrid search
based H∞ synthesis of static output feedback controllers
for uncertain systems with application to multivariable
PID control. International Journal of Robust and Non-
linear Control, 31(12), 6069–6090.
Goyal, J.K., Aggarwal, S., Ghosh, S., Kamal, S., Dworak,
P., Singh, B., and Pal, A.K. (2020). Experimental
design of robust decentralized PI controller for TRMS
through polytopic modeling. Proceedings of the IEEE
International Conference on Industrial Technology, 47–
52.
ments. Automatica, 50(5), 1315–1335.
Pradhan, J.K. and Ghosh, A. (2015). Multi-input and
20 25 30 multi-output proportional-integral-derivative controller
design via linear quadratic regulator-linear matrix in-
equality approach. IET Control Theory & Applications,
9(14), 2140–2145.
Pradhan, J.K. and Ghosh, A. (2013). Design and imple-
mentation of decoupled compensation for a twin rotor
multiple-input and multiple-output system. IET Con-
trol Theory & Applications, 7(2), 282–289.
Pradhan, J.K. and Ghosh, A. (2022). Multivariable ro-
bust proportional-integral-derivative control for linear
quadratic compensation of a class of norm-bounded
uncertain systems. Journal of Dynamic Systems, Mea-
surement, and Control, 144(10), 101003.
Sadala, S. and Patre, B. (2018). A new continuous sliding
mode control approach with actuator saturation for
control of 2-DOF helicopter system. ISA Transactions,
74, 165–174.
Skogestad, S. and Postlethwaite, I. (2007). Multivariable
feedback control: Analysis and design, volume 2. Wiley
New York.
Wen, P. and Lu, T.W. (2008). Decoupling control of a
twin rotor MIMO system using robust deadbeat control
technique. IET Control Theory & Applications, 2(11),
999–1007.
Zhao, Z., Zhang, J., Liu, Z., Mu, C., and Hong, K.S.
(2022). Adaptive neural network control of an uncertain
2-DOF helicopter with unknown backlash-like hysteresis
and output constraints. IEEE Transactions on Neural
Networks and Learning Systems. (Early Acess).