9th IFAC International Symposium on Advances in Automotive

9th IFAC International Symposium on Advances in Automotive

Control
9th
9th IFAC
IFAC International
International Symposium
Symposium onon Advances
Advances in
in Automotive
Automotive
Control
Orléans, France, 23-27, 2019 onAvailable
June Symposium online at www.sciencedirect.com
Control
9th IFAC
Control International Advances in Automotive
Orléans, France, June 23-27, 2019
Orléans, France,
Control
Orléans, France, June
June 23-27,
23-27, 2019
2019
Orléans, France, June 23-27, 2019
ScienceDirect
IFAC PapersOnLine 52-5 (2019) 673–678
Design
Design and Experimental Validation of an
Design and
and Experimental
Experimental Validation
Validation of of an
an
H
H ∞ Observer
Design and
Observer for
for Vehicle
Experimental
Vehicle Damper
Validation
Damper Force
of
Force an
H∞ for Vehicle Damper Force
∞ Observer Estimation
∞
H∞ Observer Estimation
for Vehicle Damper Force
Estimation
Estimation
∗,∗∗
 ∗ ∗
Thanh-Phong Pham ∗,∗∗ Olivier Sename ∗ Luc Dugard ∗
Thanh-Phong
Thanh-Phong Pham ∗,∗∗ Olivier Sename ∗ Luc Dugard ∗
Thanh-Phong Pham Pham ∗,∗∗ Olivier Olivier Sename
Sename ∗ Luc Luc Dugard
Dugard ∗
∗ Thanh-Phong Pham ∗,∗∗  ∗ Luc Dugard ∗
∗ Univ. Grenoble Alpes, CNRS, Olivier
GrenobleSename
INP  , GIPSA-lab, 38000
∗ Univ. Grenoble Alpes,  CNRS, Grenoble INP  ,, GIPSA-lab,
 38000
∗ Univ.
Grenoble, Grenoble
Univ. GrenobleFrance. Alpes,
Alpes, CNRS,
Institute
CNRS,
 Institute of Grenoble
Engineering
Grenoble INP Univ.
INPUniv. GIPSA-lab,
Grenoble
, GIPSA-lab, 38000
Alpes
38000
∗Grenoble,
Grenoble, France.
France. 
Institute of Engineering
of Grenoble
Engineering 
Univ. Grenoble
Grenoble Alpes Alpes
Alpes
Univ. (e-mail:
Grenoble, Grenoblethanh-phong.pham2,
France. Alpes,
 CNRS,
Institute of olivier.sename,
Engineering INPUniv. luc.dugard@
, GIPSA-lab,
Grenoble 38000
(e-mail:
(e-mail: thanh-phong.pham2,

thanh-phong.pham2, olivier.sename,
olivier.sename, luc.dugard@
Grenoble,
(e-mail:France. gipsa-lab.grenoble-inp.fr).
Institute of Engineering
thanh-phong.pham2, olivier.sename, Univ.luc.dugard@
Grenoble Alpes
luc.dugard@
∗∗ gipsa-lab.grenoble-inp.fr).
gipsa-lab.grenoble-inp.fr).
∗∗ Faculty
(e-mail:ofthanh-phong.pham2,
Electrical and Electronic Engineering,
olivier.sename,
gipsa-lab.grenoble-inp.fr). University of
luc.dugard@
∗∗ Faculty
Faculty of
of Electrical
Electrical and
and Electronic
Electronic Engineering,
Engineering, University
University of
∗∗Technology
Faculty of and and Education,
Electrical The University
gipsa-lab.grenoble-inp.fr).
and Electronic of
Engineering, Danang, 550000of
University of
∗∗Technology
Technology and Education,
Education, The University
The University
University of Danang,
of Danang,
Danang, 550000
550000of
Faculty
TechnologyDanang,
of and Vietnam
Electrical
Education, and(e-mail:
Electronic
The ptphong@ute.udn.vn)
Engineering,
of University
550000
Danang,
Danang, Vietnam (e-mail: ptphong@ute.udn.vn)
TechnologyDanang, and Vietnam
Education,
Vietnam (e-mail: ptphong@ute.udn.vn)
The University
(e-mail: of Danang, 550000
ptphong@ute.udn.vn)
Danang, Vietnam (e-mail: ptphong@ute.udn.vn)
Abstract: The real-time estimation of damper force is crucial for control and diagnosis of
Abstract:
Abstract: The
The real-time estimation of damper force is crucial for control and diagnosis of
suspension
Abstract: systems
suspension The real-time
systems
in roadestimation
real-time
in road
vehicles. Inof
estimation
vehicles. Inofthisdamper
study,force
damper
this study, force
we
is
is crucial
we considercrucial
consider a
for
for control
a semi-active
semi-active
and
and diagnosis
diagnosis of
control electrorheological
electrorheological of
suspension
(ER)
Abstract:
suspension systems
suspensionThe
systems in road
system.
real-time
in road vehicles.
First,
estimation
vehicles.a In
Inof this
nonlinear study,
damper
this study, we
quarter-car
force
we consider
is model
crucial
consider a
a semi-active
is
for proposed
control
semi-active electrorheological
that
and takes the
diagnosis
electrorheological of
(ER)
(ER) suspension
suspension system.
system. First,
First, aa nonlinear
nonlinear quarter-car
quarter-car model
model is
is proposed
proposed that
that takes
takes the the
nonlinear
suspension
(ER) and
suspension dynamical
systems in road
system. characteristics
vehicles.
First, a of the
In this
nonlinear semi-active
study, damper
we consider
quarter-car model into account.
a semi-active
is proposed The estimation
electrorheological
that takes the
nonlinear
nonlinear and
and dynamical
dynamical characteristics
characteristics of
ofanthe
theHsemi-active
semi-active damper
damper into
into account.
account. The
The estimation
estimation
of
(ER)the suspension
nonlineardamperand force
dynamicalis developed
system. First, through
characteristicsa nonlinear
of the observer damper
quarter-car
semi-active
∞ whose
modelobjectives
is proposed
into account.are to minimize
that
The takes the
estimation
of the damper force is developed through an H ∞ observer whose objectives are to minimize the
of the
effects
nonlinear
of the damper
of bounded
damper force
and dynamical
force is
is developed
unknown road through
characteristics
developed through ofantheH
profile disturbances
an H ∞ observer
semi-active
∞ observer whose
and measurement
damper
whose objectives noises
into account.
objectives are
are onto
to minimize
the
The estimation
minimize the
the
effects
effects of bounded
bounded unknown
unknown road
road profile
profile disturbances
disturbances and
and measurement
measurement noises
noises on
on the
the estimation
estimation
errors
of the
effects of the
damper
boundedstate
force variables
is developed
unknown and
road nonlinearity
through
profile an H through
disturbances
∞ observer
anda Lipschitz
whose
measurement assumption.
objectives are
noises onThe The
to considered
minimize
the estimation the
errors
errors of of the
of bounded
the state variables
state unknown
variables and
and nonlinearity
nonlinearity through
throughand a Lipschitz
a Lipschitz
Lipschitz assumption.
assumption. The considered
considered
measured
effects
errors variables,
the state used asroad
variables inputs
and profilefor disturbances
nonlinearity the observer
through design,
a are the
measurement two
noises
assumption. accelerometers
onThe
the estimation
considereddata
measured
measured variables,
variables, used
used as inputs for the observer design, are the two accelerometers data
from the
errors
measured sprung
of the statemass
variables, andas
variables
used astheinputs
and unsprung
inputs for
for the
nonlinearity mass
the observer
of the design,
through
observer quarter-car
a Lipschitz
design, are
are the
system,two
assumption.
the accelerometers
two respectively.
accelerometers data
Finally,
The considered data
from
from the
the sprung
sprung mass
mass and
and the
the unsprung
unsprung mass
mass of
of the
the quarter-car
quarter-car system,
system, respectively.
respectively. Finally,
Finally,
the
from observer
measuredthe performances
variables,
sprung mass used
and are
as
the assessed
inputs
unsprung for experimentally
the
massobserver
of the using
design,
quarter-cartheareINOVE
the
system,twoplatform from
accelerometers
respectively. GIPSA- data
Finally,
the
the observer
observer performances
performances are
are assessed
assessed experimentally
experimentally using
using the
the INOVE
INOVE platform
platform from
from GIPSA-
GIPSA-
lab
from
the (1/5-scaled
the sprung real
observer performancesmassvehicle).
and areBoth
the simulation
unsprung
assessed massandof
experimentallyexperimental
the quarter-car results
using theresults system,
INOVE emphasize the
respectively.
platformthe from robustness
Finally,
GIPSA-
lab
labthe(1/5-scaled
(1/5-scaled real vehicle).
realmethod
vehicle). Both
Both simulation
simulation and
and experimental
experimental results emphasize
emphasize the robustness
robustness
of
the
lab estimation
observer
(1/5-scaled performances
real vehicle). against
areBothassessed measurement
experimentally
simulation and noises
using
experimental andtheroad INOVE
results disturbances,
platformthe
emphasize showing
from GIPSA-
robustness the
of
of the
the estimation
estimation method against measurement noises and road disturbances, showing the
effectiveness
lab
of (1/5-scaled
the estimation the method
in real ability
vehicle).
method ofagainst
estimating
Both
against measurement
simulation the damper
measurement noises
forceand
and experimental
noises and road
resultsdisturbances,
in real-time.
road emphasize the
disturbances, showing
robustness
showing the
the
effectiveness
effectiveness in
in the
the ability
ability of
of estimating
estimating the
the damper
damper force
force in
in real-time.
real-time.
of the estimation method against measurement
effectiveness in the ability of estimating the damper force in real-time. noises and road disturbances, showing the
© 2019, IFAC in
effectiveness (International
the ability Federation
of estimating of Automatic
the Control)
damper Hosting
force in by Elsevier Ltd. All rights reserved.
real-time.
Keywords: Semi-active suspension, H∞ observer, damping force estimation, Lipchitz condition,
Keywords:
Keywords: Semi-active
Semi-active suspension,
suspension, H H∞ observer, damping force estimation, Lipchitz condition,
Keywords: Semi-active suspension, H∞ ∞ observer,
observer, damping
damping force force estimation,
estimation, LipchitzLipchitz condition,
condition,
Keywords: Semi-active suspension, H∞ observer, mate
1. INTRODUCTION damping the force
damper estimation,
force without Lipchitz condition,the dynamic
considering
1. INTRODUCTION mate the damper force without considering the
1. INTRODUCTION
1. INTRODUCTION mate the
behavior
mate the of damper force without
the semi-active
damper force without considering
damper. Estrada-Vela
considering the dynamic
the dynamic
et al.
dynamic
behavior of the semi-active damper. Estrada-Vela et al.
Nowadays, semi-active suspensions are widely used in ve- behavior
1. INTRODUCTION (2018)the
mate
behavior
(2018)
of
introducedthe
damper
of
introducedthe semi-active
the
forceH∞
semi-active
the H
without damper.
damping
damping
Estrada-Vela
force
considering
damper. observer
Estrada-Vela
force
the dynamic
observer
et
et al.
using
using al.a
a
Nowadays, semi-active suspensions are widely used in ve- (2018)
dynamic introduced
nonlinear the
model H ∞
∞of damping
the ER force
damper, observer
while usinget al.a
requir-
Nowadays,
hicle
Nowadays, semi-active
applications
semi-active suspensions
due suspensions
to their advantagesare widely
are widely used in
compared
used to behavior
in ve-
ve- (2018)
dynamic
of the semi-active
introduced
nonlinear the
model H ∞ damper.
damping
of the ER
Estrada-Vela
force
damper, observer
while using
requir- a
hicle applications due to their advantages compared to dynamic
ing
to (2018) three nonlinear
sensors asthe
introduced model H∞of
inputs ofthe ER
observer. damper,
Inobserverwhile
order requir-
to using
reduce
hicle applications
active
hicle applications
Nowadays,and passive
semi-activedue
due to their
suspensions
to their(Savaresi
suspensions advantages
advantages et al.
are widely compared
(2010)
used
compared inand
ve-
to dynamic
ing three nonlinear
sensors asmodel
inputs ofdamping
ofthe ER force
observer. damper,
In while
order to requir-
reduce
a
active
active and
and passive
passive suspensions
suspensions (Savaresi
(Savaresi et
et al.
al. (2010)
(2010) and
and ing
the three
dynamicnumber sensors
of
nonlinear theas inputs
sensors
model of of
andobserver.
the maintain
ER In
damper, a order to reduce
consideration
while requir-
references
hicle
active therein).
applications
and therein). due Central
passive suspensionsto issues
their of
advantages
(Savaresi these applications
compared
et al.applications to ing three sensors as inputs
(2010) and the number of the sensors and maintain a consideration of observer. In order to reduce
references Central issues of
of these
references
include
active and
references therein).
dynamic
passivemodeling
therein). Centraland
suspensions
Central issues
control
(Savaresi
issues of these
designs
these applications
basedand
et al.applications
(2010) on the ing
the
of
number
of dynamic
three
number
dynamic
of
of the
and
sensors
and theas sensors
nonlinear
sensors
nonlinear
and
of
and maintain
inputs characteristics
observer.
maintain
characteristics
In ofaaorder
of
consideration
MRtodamper,
reduce
consideration
MR damper,
include dynamic modeling and control designs based on of dynamic
Tudon-Martinez and nonlinear
et al. (2018)characteristics
proposed a of MR
LPV-H damper,
ainclude
reduced
references
include dynamic
number
therein).
dynamic modeling
ofCentral
modelingsensors and
and control
to control
issuesimprove designs
of these vehicle
designs based
comfort
applications
based on of dynamic and nonlinear characteristics of consideration
on the number of the sensors and maintain a MR ∞ filter
damper,
aa reduced number of Tudon-Martinez et al. (2018) proposed aa LPV-H ∞ filter
and
a reduced
roaddynamic
include
reduced holding.
number of sensors
numbermodeling
Many
of sensors
control
sensors and
to
to
improve
to control
improvedesigns
designs
improve
vehicle
vehicle
were
vehicle
comfort
comfort
proposed
based
comfort in of
on Tudon-Martinez
to dynamic
estimate and
Tudon-Martinez et
et al.
al. (2018)
the nonlinear
damper force
(2018) proposed
using deflection
characteristics
proposed a ofLPV-H
LPV-H and
MR damper, filter
∞ filter
∞ de-
and road holding. Many control designs were proposed in to
to estimate
estimate the
the damper
damper force
force using
using deflection
deflection and
and de-
de-
aand
the
and road
literature
reduced holding.
(see
number
road holding. Many control
Poussot-Vassal
of sensors to
Many control designs designs
et al.
improve were
(2008), proposed
vehiclePriyandoko
were proposed comfort in flection
to estimatevelocity
Tudon-Martinez the signals,
et al.
damper which
(2018)
force are difficult
proposed
using
in flection velocity signals, which are difficult and expensive a
deflectionand
LPV-H expensive
and
∞ filter
de-
the literature (see Poussot-Vassal et al. (2008), Priyandoko
theal.
et
and
the literature
(2009)
road
literature (seeaPoussot-Vassal
and
holding.
(see Poussot-Vassal
reviewcontrol
Many et al.
al. (2008),
in Poussot-Vassal
designs
et (2008), Priyandoko
wereetproposed
al. (2012)).
Priyandoko in flection
to measure
estimate
flection
to measure
velocity
in
velocity
in
signals,
damperwhich
thepractice.
signals,
practice.
force are
Despite
which
Despite are difficult
these
using deflection
difficult and
achievements,
and
these achievements,
expensive
and the
expensive de-
the
et
et al. (2009) and a review in
in Poussot-Vassal et al.
al. (2012)). to measure
damping in
force practice.
observer Despite
based onthese
the achievements,
dynamic nonlinear the
Some
the
et al.
al. (2009)
controland
literature
(2009) and
(seeaaPoussot-Vassal
design review
methodologies
review in Poussot-Vassal
etconsidered
al. (2008),
Poussot-Vassal et the
et al. damper flection
(2012)).
Priyandoko
(2012)). to measure
damping
velocity
in
force
signals, which
practice.
observer Despite
based
are
on
difficult
these
the
and expensive
achievements,
dynamic nonlinear the
Some control design methodologies considered the damper damping
model force
of semi-active observer based
ER suspension onthese
the dynamic
system nonlinear
and using low-
Some
force
et
Some control
al. as
control design
the control
(2009) and
design methodologies
input in
a review
methodologies considered
of Poussot-Vassal
the suspension
considered the
etsystem, then to
damper
al. (2012)).
the damper measure
damping
model of
in practice.
force
semi-active observer ER
Despiteon
based
suspension the achievements,
dynamic
system and nonlinear
using
the
low-
force
force as
as the
the control
control input
input of
of the
the suspension
suspension system,
system, then
then model
cost
damping of
sensorssemi-active
forceis still in
observer ER suspension
demand.
based on system
the dynamic and using low-
nonlinear
using
Some
force an
as inverse
control
the model
design
control for
input implementation
methodologies considered
of the suspension (see for
the
system, instance
damper model of semi-active
then cost sensors is still in demand. ER suspension system and using low-
using an inverse model for implementation (see for instance
using
Do
force
using an
etas
an inverse
al.the(2010), model
control
inverse Nguyen
model for implementation
input
for implementation
et the
of al. suspension
(2015)). (see(see
Othersfor instance
system,
for instance
usethenthe costmodel
cost
To deal
sensors
sensors is still
of semi-active
withisthe still in
inER
above
demand.
suspension system and using low-
demand.
problem, the H∞ observer for Lip-
Do
Do et
et al.
al. (2010),
(2010), Nguyen
Nguyen et
et al.
al. (2015)).
(2015)). Others
Others use
use the
the cost
To sensors
deal with is
thestill in
above demand.
problem, the
forceetan
using
Do
force
tracking
al.inverse
tracking(2010),control
model
Nguyen
control
schemes
for
schemes
in(2015)).
in
order to(see
implementation
et al. order to
attain
Others
attain
control
for instance
use the
control To deal
To dealnonlinear
schitz with the
with the systems
above problem,
above problem, the H
is a potential
the
∞ observer
observer for
Hcandidate
H ∞ observersince
∞ for Lip-
for Lip-
the
Lip-
force
Do
forceet tracking
objectivesal.
tracking(2010),control
(Priyandoko Nguyen schemes
et
control schemes al.
et in order
(2009)).
al. to
Indeed,
(2015)).
in order attain
Others
to attainthe control
damper
use the schitz
control nonlinear
schitz nonlinear systems is a potential candidate since the
objectives (Priyandoko et al. (2009)). Indeed, the damper To dealnonlinear
schitz term
with
nonlinear theinsystems
the ER
above
systems is
is aa potential
model
problem, the Hcandidate
satisfies
potential ∞the observer
candidate since
Lipschitz
since the
for con-
Lip-
the
objectives
force signal
tracking
objectives (Priyandoko
plays an
control
(Priyandoko et al.
important
schemes (2009)).
role
in
et al. (2009)). orderIndeed,
in control
to
Indeed, the
attain
the damper
synthesis.
control
damper schitz nonlinear
nonlinear term
term in
in the
the ER
ER model
model satisfies
satisfies the
the Lipschitz
Lipschitz con-
con-
force signal plays an important role in control synthesis. dition.
nonlinear Over
nonlinear
term thein years,
systems
the ER many
is a
model LMI-based
potential
satisfies theobserver
candidate Lipschitz designs
since the
con-
forcethis
For signal
objectives
force signal playsseveral
reason,
(Priyandoko
plays an important
an important
damper
et role in
force
al. (2009)).
role in controlthe
estimation
Indeed,
control synthesis.
damper dition.
method-
synthesis. dition. Over
Over the
the years,
years, many
many LMI-based
LMI-based observer
observer designs
designs
For this reason, several damper force estimation method- for Lipschitz
nonlinear
dition. Overterm nonlinear
theinyears,
the ERmanysystems
model were widely
satisfies
LMI-based the investigated
Lipschitz
observer con-
designs
For this
force
For this
ologies reason,
were
signal plays
reason, several
presented damper
an important
several damper force
(see Estrada-Vela
role in
force estimation
et al.
control
estimation method-
(2018), for
synthesis.
method- for Lipschitz
Lipschitz nonlinear
nonlinear systems
systems were
were widely investigated
ologies were presented (see Estrada-Vela et al. (2018), in (Rajamani
dition.
for Over the
Lipschitz (1998),
years, many
nonlinear Zemouche
systems wereandwidely
LMI-based Boutayeb
widely investigated
observer (2013),
designs
investigated
ologies
For this
ologies were
Reichhartinger
reason,
were presented
et al.
several
presented (see
(2018),
damper
(see Estrada-Vela
Tudon-Martinez
force
Estrada-Vela et
estimation
et et al.
al.
al. (2018),
method-
(2018), in
in (Rajamani
(Rajamani (1998),
(1998), Zemouche
Zemouche and
and Boutayeb
Boutayeb (2013),
(2013),
Reichhartinger et al. (2018), Tudon-Martinez et al. (2018), Pertew
for
in et
Lipschitz
(Rajamani al. (2006),
nonlinear
(1998), Abbaszadeh
systems
Zemouche were and
and Marquez
widely
Boutayeb (2007),
investigated
(2013),
Reichhartinger
Koch
ologies et al.
were
Reichhartinger et
(2010)),al.
presented (2018),
since
et al. (2018),(see Tudon-Martinez
in practice
Estrada-Vela
Tudon-Martinez the et
damper
et al.
al. force
(2018),
et al. force in Pertew
Pertew et
et al.
al. (2006),
(2006), Abbaszadeh
Abbaszadeh and
and Marquez
Marquez (2007),
(2007),
Koch et al. (2010)), since in practice the damper Darouach
(Rajamani
Pertew et et
al. al. (2011),
(1998),
(2006), Koenig
Zemouche
Abbaszadeh (2006)).
and
and An
Boutayeb
Marquez interesting
(2013),
(2007),
Koch et et al.
measurement
Reichhartinger
Koch al. (2010)),
(2010)),
isetdifficultsinceand
al. (2018),
since in practice the
expensive.
Tudon-Martinez
in practice the damper
In damper
(Koch force
et
et al. (2018), Darouach
al. Pertew
force Darouach et
et al.
al. (2011),
(2011), Koenig
Koenig (2006)).
(2006)). An
An interesting
interesting
measurement is difficult and expensive. In (Koch et al. solution
Darouach of et
et designing
al. (2006),
al. (2011),observer
Abbaszadeh
Koenig for (2006)).
Lipschitz
and Marquez nonlinear
An interestingsys-
(2007),
measurement
(2010)),
Koch et parallel
measurement is difficult
difficult
al. (2010)),
is Kalman and expensive.
filters
sinceand In damper
were developed
in expensive.
practice theIn (Kochtoet
(Koch et al. solution
esti-
force
al. solution of
of designing
designing observer for Lipschitz nonlinear sys-
(2010)), parallel Kalman filters were developed to esti- tem in the
Darouach
solution of absence
et al. (2011),
designing ofobserver
unknown
Koenig
observer for
for Lipschitz
inputs
(2006)).
LipschitzwasAn nonlinear
presented
interesting
nonlinear sys-
by
sys-
(2010)),
measurement parallel
is Kalman
difficult filters
and were
expensive. developed
(2010)), parallel Kalman filters were developed to esti- solutionIn (Koch to etesti-
al. tem
tem in
in the
the absence
absence of
of unknown
unknown inputs
inputs was
was presented
presented by
by
 (Phanomchoeng
tem in of
the designing
absence and ofRajamani
observer
unknown for(2010)),
Lipschitz
inputs wasbased on
nonlinear
presented the S-
sys-
by
(2010)),
 This work parallel
has beenKalman
partiallyfilters were
supported bydeveloped to
the 911 scholarship esti- (Phanomchoeng
(Phanomchoeng and
and Rajamani
Rajamani (2010)),
(2010)), based
based on
on the S-
This
 This
from

work
Vietnamese
work
has
has
been partially
government.
been partially The
supported by
authors also
supported by
the 911
thank
the 911
scholarship
the financial
scholarship
procedure
tem in
procedure
the
(Phanomchoeng lemma.
absence
lemma.
Here,
and of
Here,
we
unknown
Rajamani
we
aim
aim
to extend
inputs
(2010)),
to extend
was the
based
the on the S-
approach
presented
approach
the to
by
S-
to
This
from work
Vietnamese has been partially
government. Thesupported
authors by
also the 911
thank scholarship
the financial procedure
the Lipschitz
(Phanomchoeng lemma.system Here,
and in we
the
Rajamani aim to
presence extend
(2010)),of the
sensor
based approach
noises
on the to
and
S-

support
from
This
from of thehas
Vietnamese
work
Vietnamese ITEA 3, partially
been 15016 EMPHYSIS
government.
government. The
The authors project
supported
authors also
by
also thank
the 911
thank the financial
scholarship
the financial
procedure
the Lipschitz lemma.system Here, in we
the aim to
presence extend
of the
sensor approach
noises to
and
support of the ITEA 3, 15016 EMPHYSIS project
support of the ITEA 3, 15016 EMPHYSIS project the
the Lipschitz
procedure
Lipschitz lemma.system
system Here, in
in the
we
the presence
aim to
presence of
extend
of sensor
the
sensor noises
approach
noises and
to
and
from Vietnamese
support of the ITEA government. The authors project
3, 15016 EMPHYSIS also thank the financial
support of the ITEA 3, 15016 EMPHYSIS project the Lipschitz system in the presence of sensor noises and
Copyright
2405-8963 ©© 2019
2019, IFAC 673 Hosting by Elsevier Ltd. All rights reserved.
IFAC (International Federation of Automatic Control)
Copyright
Peer review© under
2019 IFAC 673 Control.
responsibility of International Federation of Automatic
Copyright
Copyright ©
© 2019
2019 IFAC
IFAC 673
673
10.1016/j.ifacol.2019.09.107
Copyright © 2019 IFAC 673
2019 IFAC AAC
674
Orléans, France, June 23-27, 2019 Thanh-Phong Pham et al. / IFAC PapersOnLine 52-5 (2019) 673–678

unknown input disturbances and apply it to estimate the

damping force of a semi-active suspension system.
In this paper, an H∞ observer using two accelerometers
is proposed in order to estimate the damper force in
the presence of unknown road input and measurement
noises. The design of the observer is based on a nonlinear
suspension model consisting of a quarter-car vehicle model,
augmented with a first order dynamical nonlinear damper
model. Such a model captures the main behaviour of the
ER dampers in an automotive applications. The major
contribution of this paper are as follows:
• An H∞ approach for Lipchitz nonlinear system is
developed to design an observer minimizing, in an L2 -
induced gain objective, the effect of unknown inputs
(road profile and measurement noises).
• The proposed observer has been implemented on
a real scaled-vehicle test bench, through the Mat-
lab/Simulink real-time workshop. The observer per-
formances are then assessed with experimental tests
The rest of this paper is as follows. Section 2 presents
about semi-active suspension modeling and 3 the design Fig. 1. 1/4 car model with semi-active suspension
of the observer. In section 4 is analysis of the observer
design in frequency and time domains. Section 5 presents
experimental validation. Section 6 gives some concluding
2.2 Quarter-car system description
remarks.

This section introduces the quarter-car model with the

2. SEMI-ACTIVE SUSPENSION MODELING AND semi-active ER suspension system depicted in Fig.1. The
QUARTER-CAR SYSTEM DESCRIPTION well-known model consists of the sprung mass (ms ), the
unsprung mass (mus ), the suspension components located
2.1 Semi-active suspension modeling between (ms ) and (mus ) and the tire which is modelled as
a spring with stiffness kt . From second law of Newton for
In the sequel, the dynamic nonlinear model for semi-active motion, the system dynamics around the equilibrium are
ER Damper is described. According to Guo et al. (2006), given as:
a phenomenological model of semi-active suspension can 
ms z̈s = −Fs − Fd
be represented by the following nonlinear equation: (4)
 mus z̈us = Fs + Fd − Ft
Fd = k0 xd + c0 ẋd + Fnl (x, u)
(1) where Fs = ks (zs − zus ) is the spring force, Ft = kt (zus −
Fnl (x, u) = fc .u.tanh(k1 xd + c1 ẋd )
zr ) is the tire force, and the damper force Fd is given as
where Fd is the damper force; c0 , c1 , k0 , k1 , fc are constant in (3) with deflection xd = zdef = zs − zus .
parameters. xd , ẋd are deflection and deflection velocity
of the damper, respectively. u is control input (in the Subtituting (3) into (4), one easily obtains
available test bench, this is the duty cycle of the PWM  1

 z̈s =− [(ks + k0 )(zs − zus ) + c0 (żs − żus ) + Fer ]
signal that controls the application). 
 m

 1
s

Now in order to take into account the dynamical behavior z̈us = [(ks + k0 )(zs − zus ) + c0 (żs − żus ) + Fer
of the ER fluid, it is important to complete the above mus

 −kt (zus − zr )]
model by including a first-order dynamical equation in the 


 1 fc
controlled part Fnl (x, u): Ḟ = − Fer + Φ(x, u)
er
τ τ
τ Ḟer + Fer = Fnl (x, u) (2) (5)
Therefore, the complete nonlinear damper dynamical where zs and zus are the displacements of the sprung and
model is given as unsprung masses, respectively; zr is the road displacement
 input.
Fd = k0 xd + c0 ẋd + Fer
1 f (3) By selecting the system states as x = [x1 , x2 , x3 , x4 , x5 ]T =
Ḟer = − Fer + c Φ(x, u) [zs − zus , żs , zus − zr , żus , Fer ]T ∈ R5 and the measured
τ τ variables y = [z̈s , z̈us ]T ∈ R2 , the system dynamics in the
where Φ(x, u) = u.tanh(k1 xd + c1 ẋd ) state-space representation can be written as follows

It is noted that linear and nonlinear indentification ẋ = Ax + BΦ(x, u) + D1 ω
methodologies are used to determine all the parameters of (6)
y = Cx + D2 ω
the above model (shown in table 1). They are not described
here since it is out of the scope of this paper. where

674
2019 IFAC AAC

Orléans, France, June 23-27, 2019 Thanh-Phong Pham et al. / IFAC PapersOnLine 52-5 (2019) 673–678 675

 
0 1 0 −1 0
 (ks + k0 ) c0 c0 1 
− − 0 − 
 ms ms ms ms 
 0 0 0 1 0 
A=  (ks + k0 )

 c 0 k t c 0 1 
 − −
 mus mus mus mus mus  
1
0 0 0 0 −
τ
 
(ks + k0 ) c0 c0 1
− − 0 −
 ms ms ms ms 
C=
(ks + k0 ) c0 kt c0 1 
− −
mus mus mus mus mus
    Fig. 2. Block diagram of the H∞ damper force observer
0 0 0
0  0 0  
0   0 0.01 3.1 H∞ observer design
 
B =   , D1 = −1 0 , D2 =
0  0 0 0 0.01
fc The H∞ observer for the quarter-car system (6) is defined
0 0
τ as follows
 
żr x̂˙ = Ax̂ + L(y − C x̂) + BΦ(x̂, u) (9)
ω= , in which, żr is the road profile derivative and
n
where x̂ is the estimated states. The observer gain L will
n is the sensor noises. be determined in the next steps
The control input function Φ(x, u) of the system (6) can The estimation error is given as
be rewritten under the following form
e(t) = x(t) − x̂(t) (10)
Φ(x, u) = u.tanh(Γx) (7)
Differentiating e(t) with respect to time and using (6) and
where Γ = [k1 , c1 , 0, −c1 , 0] (9), one obtains
Since 0  u  1, Φ(x, u) satisfies the Lipschitz condition ė = ẋ − x̂˙
in x = (A − LC)e + B(Φ(x, u) − Φ(x̂, u)) + (D1 − LD2 )ω
Φ(x, u) − Φ(x̂, u)  Γ(x − x̂), ∀x, x̂ (8) (11)

Assuming the Lipschitz condition (8) for Φ(x, u), the H∞

Note that the measured outputs y = [z̈s , z̈us ]T can be observer design objective is stated below
obtained easily from on board sensors (accelerometers).
• The system (11) is stable for ω(t) = 0
• e(t)L2 < γω(t)L2 for ω(t) = 0
Table 1. Parameter values of the quarter-car
model equipped with an ER damper The following theorem solves the above problem into an
LMI framwork.
Parameter Description value Unit Theorem 1. Consider the system model (6) and the ob-
ms Sprung mass 2.27 kg
server (9) . Given positive scalars γ and l . The system
mus unsprung mass 0.25 kg e(t)
ks Spring stiffness 1396 N/m (11) is asymptotically stable for ω = 0 and ω(t)LL2 < γ
2
kt Tire stiffness 12270 N/m for ω(t) = 0 if there exist a symmetric positive definite
k0 Passive damper stiffness coefficient 170.4 N/m
matrix P and a matrix Y satisfying
c0 Viscous damping coefficient 68.83 N.s/m
k1 Hysteresis coefficient due to displacement 218.16 N.s/m  
c1 Hysteresis coefficient due to velocity 21 N.s/m Ω P B P D 1 + Y D2
fc Dynamic yield force of ER fluid 28.07 N  ∗ −l Id 0n,d <0 (12)
τ Time constant 43 ms
∗ ∗ −γ 2 I
where Ω = AT P + P A + Y C + C T Y T + l ΓT Γ + In
The observer matrix will be then L = −P −1 Y
3. OBSERVER DESIGN Proof. Consider the following Lyapunov function
V (t) = e(t)T P e(t) (13)
In this section, an H∞ observer is developed to estimate
the damping force accurately. The unknown input ω (road Differentiating V (t) along the solution of (11) yields
profile disturbance and measurement noise) is considered V̇ (t) = ė(t)T P e(t) + e(t)T P ė(t)
as an unknown disturbance. Therefore, an H∞ observer = [(A − LC)e + B(Φ(x, u) − Φ(x̂, u))
is proposed to minimize the effect of the accounting for
+ (D1 − LD2 )ω]T P e + eT P [(A − LC)e
unknown disturbance ω on the state estimation errors and
to bound the nonlinearity by Lipschitz constant. + B(Φ(x, u) − Φ(x̂, u)) + (D1 − LD2 )ω] (14)

675
2019 IFAC AAC
676
Orléans, France, June 23-27, 2019 Thanh-Phong Pham et al. / IFAC PapersOnLine 52-5 (2019) 673–678

 
e The resulting attenuation of the sensor noises (n) and road
For brevity, define η = Φ(x, u) − Φ(x̂, u) , then one profile disturbance (żr ) on the estimation error, subject to
ω the minimization problems, is shown in Figure 3. These
obtains results emphasize the attenuation level of measurement
noises and unknown road profile effect on the 5 estimation
errors, since the largest sensor noise and road profile
V̇ (t) = η T M η (15) disturbance amplification of the 5 errors, over the whole
where frequency range, are -30dB and -77dB, respectively.
 
Ω1 P B P (D1 − LD2 )
M = BT P 0 0 
Bode Diagram
T
(D1 − LD2 ) P 0 0 Road profile disturbance attenuation Measurement noise attenuation
-100
T
with Ω1 = (A − LC) P + P (A − LC)

e1
-120

From (8), the following condition is obtained -140

(Φ(x, u) − Φ(x̂, u))T (Φ(x, u) − Φ(x̂, u))  eT ΓT Γe -40

-60
⇔η Qη  0
T
(16) -80

e2
  -100

−ΓT Γ 0 0 -120
-140
where Q =  0 I 0 -110

Magnitude (dB)
0 0 0 -120

e3
-130
In order to satisfy the objective design w.r.t. the L2 gain
-140
disturbance attenuation, the H∞ performance index is -60
defined as: -80
-100
e4

J = eT e − γ 2 ω T ω -120
-140
= η T Rη (17) -160
  -20
-40
I 0 0 -60
where R = 0 0 0 
e5

-80

0 0 −γ 2 I -100
-120
10-1 100 101 102 10-1 100 101 102
By applying the S-procedure Boyd et al. (1994) to the two Frequency (Hz)

contraints (16) and J  0, V̇ (t) < 0 if there exists a scalar

l > 0 such that

Fig. 3. Transfer e/ω- Bode diagrams
V̇ (t) −
l (η T Qη) + J < 0
⇔η T (M −
l Q + R)η < 0 (18)
The condition (18) is equivalent to 4.2 Simulation
M −
l Q + R < 0
  To demonstrate the effectiveness of the proposed design,
Ω1 +
l ΓT Γ + I P B P (D1 − LD2 ) the simulations are made with the nonlinear quarter-
⇔ BT P −
l I 0 <0 (19) car model (6). The block-scheme given in Figure 2 il-
T 2
(D1 − LD2 ) P 0 −γ I lustrates how the simulations of the H∞ observer are
done. The following initial conditions of the proposed
Let define Y = −P L and substitute into (19), the LMI T
observer are considered: x0 = [0 0 0 0 0] , x̂0 =
(12) is obtained.  T
[0.01, −0.4, 0.001, −0.15, 2]
4. ANALYSIS OF THE OBSERVER DESIGN: Two simulation scenarios are used to evaluate the perfor-
FREQUENCY AND TIME DOMAINS mance of the observer as follows:

In this section, the synthesis result of the H∞ observer is Scenario 1:

presented and some simulation results are given. • The road profile is a sequence of sinusoidal bumps
zr = 15sin(4πt)(mm).
4.1 Synthesis results and frequency domain analysis • The duty cycle u = 0.1 is chosen
Scenario 2:
Solving theorem 1 with
l = 2, we obtain the L2 gain
γ = 1.1938 and the observer gain • An ISO 8608 road profile signal (Type C) is used.
  • The duty cycle u = 0.1
−0.1947 −0.0051
 −1.1537 −0.0321  The simulation results of two tests are shown in the Fig. 4a
 
L = −12.7490 −167.4689 and Fig. 4b. It can be clearly observed in Fig. 4a and Fig.
 −0.2154 0.9968  4b that the damping force is estimated with a satisfactory
−87.9502 −1.4818 accuracy.

676
2019 IFAC AAC

Orléans, France, June 23-27, 2019 Thanh-Phong Pham et al. / IFAC PapersOnLine 52-5 (2019) 673–678 677

Force Estimation Force Estimation

Simulated force Simulated force
Estimated force Estimated force
10 10

5
5
Force (N)

Force (N)
0
0

-5 0
-2
-5
-4
-10 -6
0 0.05
-10
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
Time(s) Time(s)

Fig. 4. Estimated force v.s real force: (a) Simulation

scenario 1, (b) Simulation scenario 2 Fig. 5. (a) Uncertain system for robust stability analysis,
(b) Uncertain system for robust performance analysis

Both robust stability and performance analyses are based

4.3 Robustness analysis on the LFT representations of the uncertain systems
(shown in Figure 5a and Figure 5b, respectively). These
results are presented in the Fig.6, where the upper and
In this section, the µ-tool is used to analyse the robustness lower bounds of µ are always less than 1. Therefore, the
of the H∞ observer (9) associated to the system model (6) robust stability and robust performance of the observer
leading to the estimation error dynamics (11). system are guaranteed. Note that the observer is stable
The two main types of uncertainties are the Lipschitz even for larger uncertainties.
nonlinear matrix Γ and the time-varying parametric un- 0.2
plot of robust stability margins
1
Robust Performance Plot

certainty, concerning the unsprung mass, the sprung mass, 0.18

0.9908

upper/lower bounds
upper/lower bounds
0.8 0.9906

the spring stiffness and the damper model coefficients. In 0.16

0.14
0.6
0.9904
0.9902
0.29 0.292

particular, the parametric uncertainties are shown in the 0.12

0.4

Table 2 and a Lipschitz nonlinear uncertainty is assumed

0.2
0.1

0.08 0

as follows 10-1 100

Frequency (rad/sec)
101 102 10-1 100
Frequency (rad/sec)
101 102

Φ∆ (x, u) = Φ(x, u) + ∆Φ(x, u) (20) Fig. 6. (a) Robust stability analysis, (b) Robust perfor-
where mance analysis

∆Φ(x, u) − ∆Φ(x̂, u)  ∆Γ(x − x̂) (21) 5. EXPERIMENTAL VALIDATION

with ∆Γ = Γ∆ − Γ
To validate the effectiveness of the proposed algorithm,
The estimation error dynamics (11) in the uncertain do- experiments have been performed on the 1/5 car scaled
main is rewritten as follows car INOVE available at GIPSA-lab, shown in Fig. 7.
ė∆ = (A∆ − LC∆ )e + B∆ (Φ∆ (x, u) − Φ∆ (x̂, u))
+ (D1∆ − LD2∆ )ω (22)
where A∆ , B∆ , C∆ , D1∆ , D2∆ are the system matrices
considering the uncertainties.
The system (22) is written into the LFT representations
(see Figure 5a and Figure 5b). In which, Σ(s) is the system Fig. 7. The experimental testbed INOVE at GIPSA-lab
(22) with nominal values of parameters, shown in the Table (see www.gipsa-lab.fr/projet/inove)
2. ∆r (s) represents the parametric uncertainties shown in
the Table 2. ∆L (s) is the uncertain Lipschitz condition
(21). The performance objectives are represented by the
fictitious uncertainties ∆f (s).

Table 2. Parameter values of the quarter-car

model equipped with an ER damper
Fig. 8. Block diagram for implementation of the H∞
Uncertain parameters Variation damper force observer
ms 2.27 ∓ 50%(kg)
mus 0.25∓ 20%(kg) Table 3. Normalized Root-Mean-Square Errors
ks 1396∓ 20%(N/m)
kt 12270 ∓ 20%(N/m) Road Profile Sequence Experiment
k0 170.4∓ 20%(N/m) Sinusoidal bumps 0.0896
c0 68.83∓ 20%(N.s/m) ISO 8608 road 0.1015
k1 218.16∓ 20%(N.s/m)
c1 21∓ 20%(N.s/m) This test-bench which involves 4 semi-active ER sus-
fc 28.07∓ 20% pensions is controlled in real-time using using Mat-
τ 43∓ 20%(ms) lab/simulink workshop and a host PC. The target PC is

677
2019 IFAC AAC
678
Orléans, France, June 23-27, 2019
connected to the host computer via Ethernet communi-
cation standard. The proposed observer system is imple-
mented with the sampling time Ts = 0.005s. Note that the
experimental platform is fully equipped sensors to measure
its vertical motion. Each corner of the system has a DC
motor to generate the road profile.
In this study, the damping force estimation algorithm is
applied for the rear-left corner. As previously mentioned,
the unsprung mass z̈us and the sprung mass z̈s accelerom-
eters are used as inputs of the proposed observer. The
following block-scheme illustrates the experiment scenario
of the observer (shown in Fig. 8)
In this experiment senario, the duty cycle of PWM signal
is constant u = 0.1 and the real road profiles are sequence
of sinusoidal bumps and ISO 8608 road. The experiment
results of the observer are presented in Fig. 9(a) and Fig.
9(b). The result illustrates the accuracy and efficiency of
the proposed observer. To further describe this accuracy,
Table 3 presents the normalized root-mean-square errors,
considering the difference between the estimated and mea-
sured forces, for the experimental results presented in the
Fig. 9 (a) and Fig. 9 (b).
Damping force
15 Real experiment force
Estimated force
10
Force (N)
5 5
0
-5
-10
0 2 4 6 8
Time (s)
10 12
(a)
Fig. 9. Real force vs. Estimated force in experiments: (a)
with sinsoidal road profile, (b) with ISO road profile
6. CONCLUSION
This paper presented an LMI-based H∞ observer to esti-
mate the damper force, based on the dynamic nonlinear
model of the ER damper. For this purpose, the quarter-
car system is represented by considering a phenomenolog-
ical model of damper. Using two accelerometers, an H∞
observer is designed, providing a good estimation result
of the damping force. The estimation error is minimized
accounting for the effect of unknown inputs (road profile
disturbance and measurement noises) and the nonlinearity
term bounded by a Lipchitz condition. The robust statbiliy
and robust performance properties are ensured by using
the µ tool. Both simulation and experiment results assess
the ability and the accuracy of the proposed models to
estimate the damping force of the ER semi-active damper.
REFERENCES
Abbaszadeh, M. and Marquez, H.J. (2007). Robust H∞
observer design for a class of nonlinear uncertain sys-
tems via convex optimization. In American Control
Conference, 2007. ACC’07, 1699–1704. IEEE.
Boyd, S., El Ghaoui, L., Feron, E., and Balakrishnan, V.
(1994). Linear matrix inequalities in system and control
theory, volume 15. Siam.
Darouach, M., Boutat-Baddas, L., and Zerrougui, M.
(2011). H∞ observers design for a class of nonlinear
singular systems. Automatica, 47(11), 2517–2525.
Thanh-Phong Pham et al. / IFAC PapersOnLine 52-5 (2019) 673–678
Do, A.L., Sename, O., and Dugard, L. (2010). An LPV
control approach for semi-active suspension control with
actuator constraints. In American Control Conference
(ACC), 2010, 4653–4658. IEEE.
Estrada-Vela, A., Alcántara, D.H., Menendez, R.M.,
Sename, O., and Dugard, L. (2018). H∞ observer
for damper force in a semi-active suspension. IFAC-
PapersOnLine, 51(11), 764–769.
Guo, S., Yang, S., and Pan, C. (2006). Dynamic modeling
of magnetorheological damper behaviors. Journal of
Intelligent material systems and structures, 17(1), 3–14.
Koch, G., Kloiber, T., and Lohmann, B. (2010). Nonlinear
and filter based estimation for vehicle suspension con-
trol. In Decision and Control (CDC), 2010 49th IEEE
Conference on, 5592–5597. IEEE.
Koenig, D. (2006). Observers design for unknown input
nonlinear descriptor systems via convex optimization.
IEEE Transactions on Automatic control, (06), 1047–
1052.
Nguyen, M.Q., da Silva, J.G., Sename, O., and Dugard, L.
(2015). Semi-active suspension control problem: Some
new results using an LPV/H∞ state feedback input
constrained control. In Decision and Control (CDC),
Damping force
2015 IEEE 54th Annual Conference on, 863–868. IEEE.
10
Real force
Estimated force Pertew, A.M., Marquez, H.J., and Zhao, Q. (2006). H∞
observer design for Lipschitz nonlinear systems. IEEE
Force (N)
0
Transactions on Automatic Control, 51(7), 1211–1216.
-5 Phanomchoeng, G. and Rajamani, R. (2010). Observer
14 -10
0 2 4 6
Time (s)
8 10 12 14
design for Lipschitz nonlinear systems using riccati
equations. In American Control Conference (ACC),
(b)
2010, 6060–6065. IEEE.
Poussot-Vassal, C., Sename, O., Dugard, L., Gaspar, P.,
Szabo, Z., and Bokor, J. (2008). A new semi-active
suspension control strategy through LPV technique.
Control Engineering Practice, 16(12), 1519–1534.
Poussot-Vassal, C., Spelta, C., Sename, O., Savaresi, S.M.,
and Dugard, L. (2012). Survey and performance evalua-
tion on some automotive semi-active suspension control
methods: A comparative study on a single-corner model.
Annual Reviews in Control, 36(1), 148–160.
Priyandoko, G., Mailah, M., and Jamaluddin, H. (2009).
Vehicle active suspension system using skyhook adap-
tive neuro active force control. Mechanical systems and
signal processing, 23(3), 855–868.
Rajamani, R. (1998). Observers for Lipschitz nonlinear
systems. IEEE transactions on Automatic Control,
43(3), 397–401.
Reichhartinger, M., Falkensteiner, R., and Horn, M.
(2018). Robust estimation of forces for suspension sys-
tem control. IFAC-PapersOnLine, 51(25), 328–333.
Savaresi, S.M., Poussot-Vassal, C., Spelta, C., Sename, O.,
and Dugard, L. (2010). Semi-active suspension control
design for vehicles. Elsevier.
Tudon-Martinez, J.C., Hernandez-Alcantara, D., Sename,
O., Morales-Menéndez, R., and Lozoya-Santos, J.d.J.
(2018). Parameter-dependent H∞ filter for LPV semi-
active suspension systems. IFAC-PapersOnLine, 51(26),
19–24.
Zemouche, A. and Boutayeb, M. (2013). On LMI condi-
tions to design observers for Lipschitz nonlinear systems.
Automatica, 49(2), 585–591.
678