10th IFAC Symposium on Fault Detection,

10th IFAC Symposium on Fault Detection,

Supervision and Safetyon
10th IFAC Symposium forFault
Technical Processes
Detection,
Supervision
10th
Warsaw, and Safety
IFACPoland,
Symposium
August for
on Technical
Fault
29-31, Processesonline at www.sciencedirect.com
Detection,
2018 Available
Supervision
10th IFAC and Safety
Symposium for
on Technical
Fault Processes
Detection,
Warsaw, Poland,
Supervision August
and Safety 29-31,
for 2018
Technical Processes
Warsaw, Poland,
Supervision August 29-31, 2018
Warsaw, Poland, August 29-31, 2018 Processes
and Safety for Technical
Warsaw, Poland, August 29-31, 2018
ScienceDirect
IFAC PapersOnLine 51-24 (2018) 389–394
Exhaust
Exhaust pressure
pressure LPV
LPV observer
observer for
for
Exhaust
Exhaust pressure
pressure LPV
LPV observer
observer for
for
turbocharged
turbocharged
Exhaust diesel
diesel
pressure engine
engine
LPV on-board
on-board
observer for
turbocharged
turbocharged diesel
diesel engine
engine on-board
on-board
diagnosis
turbocharged diagnosis
diesel engine on-board
diagnosis
diagnosis
diagnosis
∗,∗∗ ∗∗
Donatien Dubuc ∗,∗∗ Olivier Sename ∗∗
Donatien Dubuc
Dubuc ∗,∗∗
DonatienBresch-Pietri Olivier
∗∗∗ Sename ∗∗ ∗
Delphine ∗,∗∗ Olivier
∗∗∗ Adrien Sename Halle ∗∗ ∗
Donatien
Delphine
Delphine Dubuc
Bresch-Pietri
Bresch-Pietri Olivier
∗∗∗ Adrien Sename
∗∗∗∗ Halle ∗∗ ∗
∗∗∗ Adrien Halle
Christophe ∗,∗∗
Gauthier
Donatien
DelphineChristophe Dubuc
Bresch-Pietri Olivier Sename
∗∗∗∗ ∗
Christophe ∗∗∗ Adrien
Gauthier
Gauthier ∗∗∗∗ Halle ∗
DelphineChristophe Bresch-Pietri Adrien
Gauthier ∗∗∗∗ ∗∗∗∗ Halle
∗
∗ Volvo Group, Saint
Christophe Priest, France
Gauthier
∗ Volvo Group, Saint Priest, France
(e-mail:∗ Volvo Group, Saint
donatien.dubuc, Priest, France
adrien.halle@volvo.com)
∗∗ (e-mail:∗ Volvo Group, Saint
donatien.dubuc, Priest, France
adrien.halle@volvo.com)
∗∗ Univ. (e-mail:
GrenobleVolvodonatien.dubuc,
Alpes,Group,CNRS, adrien.halle@volvo.com)
Grenoble
Saint Priest,INP,France GIPSA-lab, 38000
∗∗ Univ. (e-mail:
Grenoble donatien.dubuc,
Alpes, CNRS, adrien.halle@volvo.com)
Grenoble INP, GIPSA-lab, 38000
∗∗ Univ. Grenoble
(e-mail: Alpes, CNRS,
donatien.dubuc, Grenoble
France INP, GIPSA-lab, 38000
Grenoble,adrien.halle@volvo.com)
∗∗ Univ. Grenoble Alpes, CNRS,
Grenoble,
Grenoble, Grenoble
France INP, GIPSA-lab,
France INP, 38000
Univ. Grenoble (e-mail:
Alpes,olivier.sename@gipsa-lab.fr)
CNRS, Grenoble GIPSA-lab, 38000
∗∗∗ (e-mail: Grenoble, France
olivier.sename@gipsa-lab.fr)
(e-mail: olivier.sename@gipsa-lab.fr)
∗∗∗ MINES Paristech, PSL Research
Grenoble, FranceUniversity, CAS -- Centre
∗∗∗ MINES Paristech, PSL
(e-mail: PSL Research
(e-mail: olivier.sename@gipsa-lab.fr)
∗∗∗ MINES Paristech, Automatique Research University,
University, CAS
et Systèmes
olivier.sename@gipsa-lab.fr) CAS - Centre Centre
∗∗∗ MINES Paristech, PSL
Automatique
Automatique Research
et University,
Systèmes
et Systèmes CAS - Centre
(e-mail:
MINES delphine.bresch-pietri@mines-paristech.fr)
Paristech, PSL Research University, CAS - Centre
∗∗∗∗ (e-mail:
(e-mail: Automatique et Systèmes
delphine.bresch-pietri@mines-paristech.fr)
delphine.bresch-pietri@mines-paristech.fr)
∗∗∗∗ JTEKT
(e-mail:
Europe, ZI du Broteau,
Automatique et BP1, 69540 Irigny, France
Systèmes
delphine.bresch-pietri@mines-paristech.fr)
∗∗∗∗ JTEKT Europe, ZI du
∗∗∗∗ JTEKT
(e-mail: Europe,
(e-mail: du Broteau,
Broteau, BP1,
BP1, 69540
ZIcgauthier2000@yahoo.fr)
delphine.bresch-pietri@mines-paristech.fr)69540 Irigny,
Irigny, FranceFrance
∗∗∗∗ JTEKT Europe,
(e-mail:
(e-mail: ZI cgauthier2000@yahoo.fr)
du Broteau, BP1,
cgauthier2000@yahoo.fr) 69540 Irigny, France
JTEKT Europe, ZI du Broteau,
(e-mail: cgauthier2000@yahoo.fr) BP1, 69540 Irigny, France
Abstract: This This paper
paper deals (e-mail:
deals with the cgauthier2000@yahoo.fr)
the estimation of of the exhaust
exhaust manifold
manifold pressure
pressure for for aa truck
truck
Abstract:
Abstract: This paper withdealsa with
with the estimation
estimation of the
the exhaust manifold pressure for ina order
truck
diesel
diesel engine
engine equipped
equipped turbocharger. The knowledge of this variable is essential
Abstract:
diesel engine This paper with
equipped dealsa turbocharger.
a with the estimation The knowledge
of the exhaust of
of this variable
thismanifold is
is essential
pressure for ina order
truck
to fulfill
fulfill
Abstract:
to functions
This
functions suchwith
paper
such as the
deals
as theturbocharger.
withexhaust
the estimation
exhaust
Thecontrol
brake
brake
knowledge
of theor
control orexhaust
on-board
on-board
variable
diagnosis
manifold
diagnosis
essential
(OBD)
pressure
(OBD) for ina order
of
of anti-
truck
anti-
diesel
to engine
fulfill equipped
functions with
suchwith a
as the turbocharger.
exhaust The
brake knowledge
control of
or on-boardthis variable
diagnosis is essential
(OBD) in order
ofsensor
anti-
pollution
diesel
pollution engine systems.
equipped
systems. However,
However, awhile in
turbocharger.
while in most cases
The the pressure
knowledge of is
thisdirectly
variable measured,
is essential the in order
to fulfill functions
pollution
mayfulfill
to systems.
encounter
functions
such as the
However,
failures inassome
such in while
some
the in most
exhaust
specific
exhaust most cases
brake
operating
brake
the
casescontrol
control
pressure
or on-board
theconditions.
pressure
or on-board
is
isItsdirectly
diagnosis
directly
estimation
diagnosis
measured,
(OBD)
measured,
is(OBD)
then
the ofsensor
theof anti-
ofsensor
great
anti-
may
pollution
may encounter
systems.
encounter failures
However,
failures infault while
some specific
in
specificmost operating
cases
operating theconditions.
pressure
conditions. isIts estimation
directly
Itsdirectly
estimation is
measured, then
is models
thenthe of great
sensor
ofsensor
great
interest
pollution
interest for
for diagnosis
systems. and
However, tolerant
while in mostcontrol
cases objectives.
the pressure Basedis on mean value
measured, the of the
may
interest
turbocharger
may for diagnosis
encounter
encounter and
failures
diagnosisthe
failures
and
andinfault
exhaust
in
sometolerant
fault
some
specificcontrol
tolerant
manifold,
specific a
operating
control
Linear
operating
objectives.
conditions.
objectives.
Parameter
conditions.
Based
Based Itson
Varying
Its
mean
onestimation
mean
(LPV)
estimation
value
value is models
is
then of of
models
polytopic
then observer
of
the
ofgreat
the
great
turbocharger
interest
turbocharger and
for diagnosis
and the
the andexhaust
faultmanifold,
exhaust aa Linear
tolerant control Parameter
objectives. Varying
Based (LPV)
on mean value models
polytopic of the
observer
is designed
is designed
interest
turbocharger
to
for to provide
diagnosis
provide
and the and
an faultmanifold,
an estimation
estimation
exhaust tolerant
manifold,
of the
of the Linear
control
a Linear
Parameter
pressure. The merits
objectives.
pressure. The
Parameter
Varying
merits
Based
Varyingof
(LPV)
ofonthis
this
mean
(LPV)
polytopic
solution
value
solution are
models
are
polytopic
observer
illustrated
of the
illustrated
observer
is
withdesigned
turbocharger toandprovide
the high-fidelity
high-fidelity an estimation
professional
the professional
exhaust of the
simulator
manifold, pressure.
GT-POWER.
a Linear The merits
Parameter Varyingof this(LPV) solution are illustrated
polytopic observer
with
is
with the
designed to provide
the high-fidelity an estimation simulator
of the GT-POWER.
pressure. The merits of this solution are illustrated
is designed to provideprofessional
an estimation simulator GT-POWER.
of the pressure. The merits of this solution are illustrated
© 2018,
with theIFAC (International
high-fidelity Federation
professional of Automatic
simulator Control) Hosting by Elsevier Ltd. All rights reserved.
GT-POWER.
with
Keywords:the high-fidelity
LPV professional
observer, automotive simulator
systems, GT-POWER.
diagnosis
Keywords:
Keywords: LPV LPV observer,
observer, automotive
automotive systems,systems, diagnosis
diagnosis
Keywords: LPV observer, automotive systems, diagnosis
Keywords: LPV
1. INTRODUCTION observer, automotive systems, diagnosis
2010). TheThe latter estimators
estimators proposepropose to to directly estimate
estimate
1.
1. INTRODUCTION 2010).
INTRODUCTION 2010).
this The latter
variable latter
from estimators
the propose
information of to directly
the directly
exhaust estimate
air mass
mass
1. INTRODUCTION this
2010). variable
The latterfrom the information
estimators propose of the exhaust
to directly air
estimate
In most
most automotive 1. INTRODUCTION
automotive engines,engines, the the exhaust
exhaust manifold
manifold pres- pres- flow this
flow
2010). variable
through
The latterfrom
the the
orifice information
estimatorsflow equation
propose of the
or exhaust
from
to directly the air mass
turbine’s
estimate
In through
this variable
through from the
the orifice flow
flow equation
the information or
or from
of the exhaustthe
the turbine’s
air mass
In
sure most
is automotive
directly engines,
measured by the
a exhaust However,
sensor. manifold some pres- flow data-maps.
this variable
data-maps.
flow through from
the
orifice
the information
orifice flow
equation
equation of the
or
from
exhaust
from the
turbine’s
air mass
turbine’s
sure
In most
sure is
is directly
automotive
directly measured
engines,
measured by
by a
the
a sensor.
exhaust However,
sensor. manifold some
However, pres- data-maps.
some flow through the orifice flow equation or from the turbine’s
problems
In have been reported concerning the robustness
suremost
problems automotive
have
is directly been engines,
reported
measured bythe exhaust However,
concerning
a sensor. manifold
the pres- data-maps.
robustness
some In (Fredriksson
(Fredriksson and and Egardt, 2002), 2002), the the authors proposedproposed
problems
of the
sure
of
have In
theis sensor.
sensor.
directly been
In
reported
fact,
measured
fact, in by
in
concerning
addition
a sensor.
addition to its
to
the
itsHowever,
high
high
robustness
cost,
cost,some it In
it
data-maps.
In
a and Egardt,
(FredrikssonLuenberger
generalized Egardt, the authors
2002), based
observer authors
on meanproposed
value
problems have been reported concerning the robustness
cost, it aaIn generalized
(FredrikssonLuenberger observer
2002), based on
of
must
must
of
theface
problems
the
sensor.
face strong
have
strong
sensor.
In
been
In
fact,
pressure
pressure
fact,
in oscillations
reported
in
addition
concerning
oscillations
addition
toand
toand
itsthe
its
high
high
high
high
tempera-
robustness
tempera-
cost, it In generalized
models of
(Fredriksson the
and Egardt,
Luenberger
intake
and and
Egardt, observer
exhaust
2002),
the authors
based on mean
manifolds,
the authors mean the
value
proposed
value
tur-
proposed
must
turethe
of
ture
must
face strong
conditions.
sensor.
conditions.
face strong In pressure
Indeed,
Indeed,
pressure in oscillations
these
fact, these conditions
addition
conditions
oscillations toand
and
high
have
its
have high
high
tempera-
accused
cost, for
accused
tempera- it models
for a generalized
models
bocharger
a
of the
of and
generalized
intake
theLuenberger and
intake dynamics
engine
Luenberger
exhaust
andobserver
exhaust
observer which
manifolds,
based on mean
manifolds,
based leads
on to
mean
the
the
a
tur-
value
tur-
fourth
value
ture
example
must conditions.
tube
facetube Indeed,
clogging
strong pressure these
problems conditions
to theand
oscillations have
sensor.
high accused
Therefore
tempera- for bocharger
models of and the engine
intake dynamics
and exhaust which leads
leads to
manifolds, a
a fourth
to the tur-
example
ture conditions.
example tube clogging
Indeed,
clogging problems
these
problems to
to the
conditions sensor.
have
the conditions
sensor. Therefore
accused
Therefore for bocharger
order
bocharger of and
order nonlinear
models nonlinear the engine
and intake dynamics
observer.
observer.
engine and exhaust
dynamics
which
which manifolds,
leads to a
fourth
the tur-
fourth
it
ture
it is
is not
not reliable
conditions.
reliable in
Indeed,
in all
all the
these
the operating
conditions
operating have
conditionsaccused of
of the
for
the order nonlinear
bocharger and observer.
engine dynamics which leads to a fourth
example
it tube
is notAlthough clogging
reliable problems
in allpressure to
the operating the sensor. Therefore orderthisnonlinear
In this paper, we observer.
we propose for for thethe first
first time,
time, up up toto the
the
engine.
example
engine. tube
it is notAlthough
reliable
this
clogging
this problems
in allpressure to the conditions
information
information
the operating sensor.is difficult
is
conditions
of the
difficult
Therefore
of the
to In
to order paper,
nonlinear propose
observer.
engine.
get,is it
it itnotAlthough
is reliable this
essentialinfor pressure
forallengine
engine information
control. Among
the operating Among is
conditions difficult
others,
of theto In
it authors’this
authors’ paper,
knowledge,we propose
a Linearfor the first
Parameter time,
Varyingup to
(LPV)the
get, In this paper,knowledge,we proposeaa Linear forParameter
the first time, Varyingup (LPV)
to the
engine.
get,
is usedit is
used
engine. is
to
essential
Although
essential this
control this
Although the engine control.
pressure
for exhaust
pressure
information
control.
pressure
information Among
withisthethe
others,
is difficult
others,
exhaust
difficult
it
to
it authors’
to observer
In knowledge,
based
this paper,
observer
authors’ based
knowledge,
on mean
we mean
on propose
a
Linear
value Parameter
models
formodels
value
Linear the first
Parameter
of the
of theVarying
time, up (LPV)
turbocharger
to the
turbocharger
Varying (LPV)
is
get,
is it is
used to control
to essential
control the
the exhaust
for exhaust pressure
engine control.
pressure with
Among
with exhaust
theothers,
exhaust it observer
and the based
exhaust onmanifold
mean value
to models
estimate of
the the turbocharger
pressure. Since
flap
get, in
it order
is to
essential get an
for engine
engine brake
control. and to
Among estimate
others, the
it authors’
and the knowledge, a Linear Parameter Varying (LPV)
flap
is used
flap
burned
is
in
in order
used order
fraction
to
to
to get
to control
control get
to
the an
ensure
the
engine
anexhaust
engine
exhaustan
brake
pressure
brake
on-board
pressure
and
andwithto
to estimate
the exhaust
estimate
diagnosis
with the (OBD)
exhaust
the
the and the the exhaust
observer based onmanifold
exhaust
equipment
observer based of
on
mean value
manifold
the
mean engine
to
value
estimate
toundermodels of
estimate the
the the
consideration,
models of the
pressure.
pressure.and
Since
turbocharger
Since
turbochargerthus
burned
burned fraction
flap in order
fraction to
to get
to ensure
an engine
ensure an
an on-board
brake anddiagnosis
on-board to estimate
diagnosis (OBD)
(OBD) the the and equipment
the the exhaust
equipment
measurement
of
of the
theat
engine
manifold
engine
our
tounder
under
disposal,
consideration,
estimate the pressure.
consideration,
are not the
and
and
same
thus
Since
thus
as
for
flap the anti-pollution
in order to get system.
an system.
engine Since
brake it
and is mandatory
tomandatory
estimate to
the and
the the exhaust
measurement
equipment manifold
of the at our
engine to estimate
disposal, are thenot
under consideration, pressure.
the same
and Since
as
thus
for
burned
for the
the anti-pollution
fraction to ensure
anti-pollution system. Since
an on-board
Since it
it is
diagnosis
is mandatory (OBD) to
to the
in measurement
(Fredriksson and at our
Egardt, disposal,
2002), are
it not
leads the
to asame
secondas
propose
burned an OBD solution (Mohammadpour et al., 2012), the equipment of the engine under consideration, and thus
for the fraction
propose an
an OBD
tosolution
anti-pollution ensure an on-board
Since it diagnosis
(Mohammadpour
system. et
et al.,
is mandatory (OBD)
al., 2012), to in the(Fredriksson
measurementand at Egardt, 2002),
our disposal, it
it leads
are not the to
to a second
asame as
propose
its estimation
for
its estimation OBD
the anti-pollution
propose an OBD is
solution
is then
then
solution
of (Mohammadpour
great
system.
of great interestit for
Since
interest
(Mohammadpour
for diagnosis
is mandatory
diagnosis
et al.,
2012),to in
and
and
2012),
the (Fredriksson
order
order
in
LPV observer.
measurement
LPV
(Fredriksson
and
observer.
and at Egardt,
Besides,
Besides,
Egardt,
2002),
our disposal, our method
our
2002),
method
are
it
leads
leads
encompasses
notencompasses
the
to a
second
same
second
a
as
a
its
faultestimation
propose tolerant
an OBD is then
control of great
objectives.
solution interest for diagnosis
(Mohammadpour et al., and order
2012), in LPV calibration
systematic
(Fredriksson
systematic
observer.
and
calibration
order LPV calibration
Besides,
observer. Besides,
procedure,
Egardt,
procedure,
ourcontrary
2002), method
it
contrary
ourcontrary leads
method to
encompasses
to
to the
to
the aprevious
second
previous
encompasses
a
a
fault
fault tolerant
tolerant control
its estimation is thenobjectives.
control of great interest for diagnosis and systematic
objectives. one where
where tuning procedure,
theBesides,
observer parameters is the
not previous
an easy
easy
its estimation is then of great interest forestimation
diagnosis rep- and one order LPV tuning
systematic observer.
calibration the observer
procedure, our method to
parameters
contrary encompasses
is not
the an
previous a
fault
To tolerant such
overcome control objectives.
problems, model-based one
task. where
Moreover,tuning thethe observer
observer is parameters
proposed is not
considering an easy
two
To
faultovercome
tolerant such
control problems,
objectives.model-based estimation rep- systematic calibration procedure, contrary to the previous
To overcome
resents such
an efficient
efficient problems,
alternative. model-based
Therefore, estimation
several authors
authorsrep- task. Moreover,
one where tuningthe theobserver
observeris proposed
is parameters considering
is not an easy two
resents
To overcome
resents an
an alternative.
such problems,
efficient alternative. Therefore,
model-based
Therefore, several
estimation
several authorsrep- task.
one Moreover,
different
where
different
task. tuningthe
models
models
Moreover,
forthe
for
the
observer
the
the turbine
observer
turbine
observer is
proposed
mass
mass flow
proposed
considering
flow rate:
parameters rate:
is nota
considering
two
a standard
standard
an easy
two
have
To
have proposed
overcome such different
problems, methods to
model-based estimate
estimation the ex-
rep- different
orifice
task. models
equation,
Moreover, for
and
the the a turbine
observernew mass
identified
is flow
proposed rate:
black-box a standard
considering model.
two
resentsproposed
have an efficient
proposed different
different methods
alternative.
methods to
Therefore,
to estimate
several the
estimate authors
the ex-
ex- orifice
different equation,
models and
for the a new
turbine identified
mass flow black-box
rate: a model.
standard
haust
resents
haust manifold
an efficient
manifold pressure.
pressure. One
alternative.
One can
can categorize
Therefore,
categorize them
several
them in
authors
in two
two orifice
The meritsequation,
merits
different of the
models thefor and theaa turbine
developed newsolution
identified flowblack-box
are
mass are then model.
validated
rate:validated
a standard on
have
haust proposed
manifold different methods to estimate the ex- The
orifice of
equation, developed
and new solution
identified then
black-box model.on
types: nonlinearpressure.
types: proposed
have nonlinear observer-based
different
observer-based
One canapproaches
methods categorize
to estimate
approaches
themthe in two
(Fredriksson
(Fredriksson ex- The a high
orificemerits of the and
industrial
equation, developed
complex a new solution
simulator
identified are
with then validated
realistic
black-box engine
model.on
haust manifold pressure. One can categorize them in two
types:
and Egardt,
haust
and
types:
nonlinear
Egardt,
manifold2002)
2002)
nonlinear
observer-based
and inverse
pressure.
and inverse
observer-basedOne can approaches
model
model approaches
categorize
approaches
approaches them in two aaThe
(Fredriksson
(Castillo
(Castillo
(Fredriksson
high
high
cycles.
The
industrial
merits
industrial
merits
complex
complex simulator
of the developed
of the developed
solution are
simulator
solution are
with
with then
then
realistic
realistic engine
validated
engine
validated
on
on
and
et Egardt,
al.,
types: 2013; 2002)
nonlinear Olin, and inverse
2008;
observer-based model
Yue-Yun approaches
Wang
approaches and (Castillo
Haskara,
(Fredriksson cycles.
a high
cycles. industrial complex simulator with realistic engine
et
and al., 2013;
Egardt, Olin,
2002) 2008;
and Yue-Yun
inverse model Wang
approaches and Haskara,
(Castillo a high industrial complex simulator with realistic engine
cycles.
et
and al., 2013;2002)
Egardt, Olin, 2008; Yue-Yun
model Wang and Haskara,
et al., 2013; Olin,and inverse
2008; Yue-Yun approaches
Wang (Castillo cycles.
and Haskara,
et al.,
Copyright 2013; Olin,
2018 IFAC
2405-8963 © 2018, 2008; Yue-Yun Wang and
IFAC (International Federation of Automatic Control)Haskara, 389 Hosting by Elsevier Ltd. All rights reserved.
Copyright © 2018 IFAC 389
Copyright
Peer review©under
2018 responsibility
IFAC 389Control.
of International Federation of Automatic
Copyright © 2018 IFAC
10.1016/j.ifacol.2018.09.606 389
Copyright © 2018 IFAC 389
IFAC SAFEPROCESS 2018
390
Warsaw, Poland, August 29-31, 2018 Donatien Dubuc et al. / IFAC PapersOnLine 51-24 (2018) 389–394

Dc Dt Table 1. Nomenclature

Turb VGT
Comp.
Air filter

P1 ; T 1 P4 Exhaust flap Notation Description Unit

ωtc
Neng Engine speed rpm
EGR cooler
ωtc Rotor speed of the turbocharger rad.s−1
P3 ; T 3
Jtc Shaft moment of inertia of the kg.m2
Exhaust manifold
turbocharger
P Power W
Charge air cooler

T Temperature of a subsystem K
Degr P Pressure of a subsystem Pa
cp Specific heat J.kg−1 .K−1
Intake manifold R Ideal gas constant for the air J.kg−1 .K−1
γ Specific heat ratio -
Preheater P2 ; T 2
EGR valve D Mass flow rate kg.s−1
ITV
Subscript
1 Upstream the compressor
Fig. 1. Scheme of the air path in the considered engine 2 Inside the intake manifold
3 Inside the exhaust manifold
The paper is organized as follows. In Section 2, a mean 4 Downstream the turbine
value model is presented. In Section 3, based on this model, c Related to the compressor
a LPV polytopic observer is designed to estimate the t Related to the turbine
exhaust pressure. The observer is synthesized in order to egr Related to the EGR loop
f Related to the fuel
minimize a H∞ criterion associated with a pole placement
by Linear Matrix Inequalities (LMIs) regions. Then, in where m3 is the total air mass inside the volume V3 .
Section 4, the performances of this observer are illustrated
in a realistic simulator designed with GT-POWER. Fi- By derivating this equation, one obtains:
nally, conclusions are stated in Section 5. ṁ3 RT3 m3 RṪ3
Ṗ3 = + (2)
V3 V3
2. MEAN VALUE MODEL OF THE ENGINE
where ṁ3 represents the mass rate of gas flowing through
The architecture of the diesel engine under consideration the exhaust manifold and can be expressed, from a balance
is depicted in Fig. 1. This engine is a medium-duty 4- equation, ṁ3 = Dasp + Df − Degr . Besides, assuming that
cylinders 5L diesel one equipped with a Variable Geometry the temperature varies slowly in comparison to P3 , we
Turbocharger (VGT) and a Exhaust Gas Recirculation consider Ṫ3  0 and (2) becomes:
(EGR) loop. In this section and in all the paper, we will
use data provided by a high-fidelity simulator designed RT3
with GT-POWER 1 . This software, developed by Gamma Ṗ3 = (Dasp + Df − Degr − Dt ) (3)
V3
Technology, consists in a set of simulation libraries for
analyzing the engine behavior and is largely used in the In (3), the mass flow rates Df and Degr are known input
automotive industry. Therefore all the reference data are variables, and the air mass flow aspired by the cylinders
provided by this simulator. can be expressed as:
The measurements considered for this study are: Dc , Degr , ηv Vcyl Neng
Dasp = P2 (4)
P1 , T1 , P2 , T2 , T3 , P4 and ωtc . These variables are typically RT2 120
measured or estimated in the automotive industry. Their
where ηv is the volumetric efficiency defined as the ratio
nomenclature is given in Table 1.
between the actual volume flow rate of air and the the-
In the following, the exhaust manifold and turbocharger oretical volume flow rate of air displaced by the pistons.
dynamics are modeled using a mean value approach such It is in general given by a map in function of the engine
as in (Isermann, 2014; Moulin, 2010). speed and the intake manifold pressure: ηv (Neng , P2 ). See
(Isermann, 2014) for a more detailed model and about the
2.1 Exhaust manifold dynamics efficiency protocol measurement. The mass flow rate Dt
will be expressed in Section 2.3.
Located just after the engine block, this manifold permits
to collect all the gases from the cylinders into one pipe 2.2 Turbocharger dynamics
which is directly connected to the turbine.
Located just after the EGR loop and the exhaust manifold,
The exhaust manifold can be represented as an open the turbocharger is a combination of a turbine and a
thermodynamical system, where the quantity of gas can compressor. The main function of the turbocharger is
increase or decrease. It is called a ”filling and emptying” to increase the air density in the intake manifold by
system. Inside this volume, the ideal gas law can be applied recovering energy from the exhaust gases. Its secondary
and the pressure P3 can be expressed as: functions are: increase the exhaust pressure P3 in order to
drive the EGR flow or allow engine brake.
m3 RT3
P3 = (1) There are several studies proposing a model of the tur-
V3 bocharger. Some are control-oriented such as (Salehi et al.,
1 www.gtisoft.com 2013) or modeling-oriented (Jung et al., 2002; Isermann,

390
 IFAC SAFEPROCESS 2018
Warsaw, Poland, August 29-31, 2018 Donatien Dubuc et al. / IFAC PapersOnLine 51-24 (2018) 389–394 391

2014). From the mechanical power balance, one can obtain P3

T3 Input Linear Output Dt
the rotor speed dynamic of the turbocharger 2 : P4
uvgt
Nonlinearity Block Nonlinearity

1 Fig. 2. Hammerstein-Wiener model block diagram

Jtc (ω˙tc
2)=P −P
t c (5)
2
where Pc and Pt are respectively the compressor and
turbine powers. FIT = 0.7698

Now, from the formula of the isentropic compression, the

consideration of heat loses through the efficiency maps
and the first law of thermodynamics, these powers can
be expressed as:
  validation data training data
  γ1γ−1
1 P 2 1
FIT = 0.9275 FIT = 0.9627
Pc = T1 cp1 Dc  − 1 (6)
ηc P1
 
  γ3γ−1
P4 3
Pt = ηt T3 cp3 Dt 1 −  (7)
P3
Fig. 3. Turbine mass flow rate obtained with orifice flow
As mentioned in (Isermann, 2014), one can consider the
equations (8)-(10) and a Hammerstein-Wiener model
parameters γ1 , γ3 , cp1 , cp3 and R as constant. In addition,
(FIT computed with (11))
due to their low dispersion, we assume that γ1 = γ3 = γ.
HW model is a combination of three blocks as depicted in
The efficiencies ηc and ηt are given by interpolated maps
Fig. 2: a static input nonlinearity, a linear dynamic system,
from data provided by the manufacturer of the tur-
and a static output nonlinearity. For more information
bocharger.
about this model see for example (Zhu, 2002). In our
case, sigmoid networks have been chosen for the input and
2.3 Turbine flow modeling output nonlinear functions, which can be defined as a sum
of weighted sigmoid functions (ez +1)−1 . The linear system
Since mass flow rate passing through the turbine Dt is not is a polynomial model with 1 zero, 2 poles with a delay set
measured, one can use the classical orifice flow equation to to 1. The identification was performed with the System
model the mass flow rate of the turbine (Moulin, 2010): Identification Toolbox of Matlab using the same inputs
  as the previous model (i.e: P3 , T3 , P4 and uvgt as in Fig.
P3 P4 2).
Dt = A(uvgt ) √ Ψ (8)
RT3 P3 This black-box model improves significantly the fit of the
   data (Fig. 3).
P4 2γ  γ2 γ+1

Ψ = Π −Π γ (9)
P3 γ−1 2.4 Considered system
where Π represents the pressure ratio in normal and Finally, combining equations (3), (4), (5), (6) and (7), the
critical conditions, which is defined by: following nonlinear differential equations are obtained:
  γ 
 γ−1  
P4 2γ   γ3γ−1
Π = max , (10) 2 P 3
ω˙tc
4
P3 γ + 1 2 = ηt T3 cp3 Dt 1 − 
Jtc P3
In our study, the effective area A(uvgt ) is identified as a  
  γ1γ−1
third order polynomial function of the command uvgt . 2 1 P 2 1 (12)
− T1 cp1 Dc  − 1
Jtc ηc P1
In the sequel, the following fit performance index will be
used: RT3
Dt − Dt (M odel)2 Ṗ3 = (Dasp + Df − Degr − Dt )
F IT = 1 − (11) V3
Dt − mean(Dt )2
where Dt , for comparison purpose, will be calculated
The results of this model for Dt are depicted in Fig. 3 using the two models defined in the previous Section
and compared with the reference data provided by GT- 2.3: the orifice flow equations (8)-(10) and the identified
POWER. Hammerstein-Wiener model.
Even if the model seems to be good, the results could be 3. OBSERVER DESIGN
improved especially in the last hundreds seconds. To do so,
a Hammerstein-Wiener (HW) model has been identified. A To take into account the nonlinearities of the system de-
2 In this study the mechanical friction is neglected but can be easily scribed in Section 2.4, a quasi-LPV approach is considered
added via a constant efficiency to design an observer for the exhaust manifold pressure P3 .

391
IFAC SAFEPROCESS 2018
392
Warsaw, Poland, August 29-31, 2018 Donatien Dubuc et al. / IFAC PapersOnLine 51-24 (2018) 389–394

3.1 LPV modeling The objective is to estimate the state vector x of (15) while
minimizing a H∞ criterion with respect to disturbance
Let denote [x1 x2 ]T = [ωtc2
P3 ]T the state vector. We terms. The proposed polytopic LPV observer has the
choose to transform the nonlinear system described by (12) following structure:
into the following quasi-LPV form: 4

        x̂˙ = (µi (ρ̂)(Ai x̂ + Li (y − ŷ))) + Bu

x˙1 0 ρ1 x1 10 u1 i=1
(19)
= + ŷ = C x̂; ẑ = Cz x̂
x˙2 0 ρ2 x2 01 u2 (13)
= A(ρ)x + Bu where Li are unknown matrices to be determined.
y = [1 0] x = Cx Let Twez denote the closed-loop transfer function from w
to the state error estimation ez = z − ẑ. The objective is
with, in the case where Dt is given by (8)-(10),
to design an observer (19) such that:
   γ−1    - the poles of Twez are in a desired region to ensure
P4 γ
1 P4 2 both convergence performance and stability
ρ1 = ηt T3 cp3 1− A(uvgt ) √ Ψ
P3 RT3 P3 Jtc - the H∞ norm of Twez is minimized
 
1 P 4 T3 R 3.3 Synthesis
ρ2 = −A(uvgt ) √ Ψ
RT3 P3 V3
2 T3 R First let us notice that, even if the parameter vector is
u1 = − Pc ; u2 = (Dasp + Df − Degr )
Jtc V3 using estimated state variables (due to the quasi-LPV
(14) model), we assume here that ρ̂ = ρ for the synthesis of
the observer (19). As presented in (Heemels et al., 2010),
It is worth noting that (13) is a quasi-LPV model since a robustness study with respect to uncertainty on the
ρ1 and ρ2 depend on x2 = P3 . Therefore, in the LPV scheduling variables could be done in the future.
observer form, ρ1 and ρ2 will be computed on-line using
Proposition 1. Given (15)-(17) and the observer (19). If
the estimated pressure P̂3 .
there exists symmetric positive-definite matrix P and
3.2 Problem formulation matrices Yi such that:

For the synthesis problem, consider the following LPV

system: P Ai − Yi C + ATi P − C T YiT + 2αP < 0 (20)
 
ẋ = A(ρ)x + Bu + Ew −rP P Ai − Yi C
(15) <0 (21)
y = Cx; z = Cz x
∗ −rP
where x ∈ Rnx is the state vector, u ∈ Rnu the known
 sin θ(P A 
input vector, y ∈ Rny the measure vector, z ∈ Rnz the T T T
i −Yi C+Ai P −C Yi ) cos θ(P Ai −Yi C−AT T T
i P +C Yi )

variable to estimate and w ∈ Rnw represents additive

∗ sin θ(P Ai −Yi C+AT T T
i P −C Yi )
uncertainties that we want to attenuate. The parameter
T <0
vector ρ consists of N varying parameters [ρ1 ... ρN ]
where each component ρi ∈ [ρi , ρi ]. (22)

As introduced in (Apkarian et al., 1995), if the parameter min γ∞

dependence of A(ρ) is affine (as in (13)) and if the  PA T T T
P CzT

i −Yi C+Ai P −C Yi E
parameter vector ρ varies in a polytope Υ of 2N vertices (23)
such that, s.t  ∗ −γ∞ I 0 <0
ρ ∈ Υ := Co{ω1 , ω2 , ..., ω2N } ∗ ∗ −γ∞ I
  (16)
ωi ∈ (1 , 2 , ..., N ) | i ∈ {ρi , ρi } for all i = 1, 2, . . . , 2N .

then, the matrix A(ρ) can be transformed into a convex Then the poles of the transfer function Twez are in the
interpolation such that: considered area depicted in Fig. 4 and its H∞ norm
N N
is minimized by γ∞ . Besides, the observer gains Li are
2
 2
 deduced as Li = P −1 Yi .
A(ρ) = µi (ρ)Ai , µi (ρ) ≥ 0, µi (ρ) = 1 (17)
i=1 i=1 Proof: The proof follows the steps as described in (Chilali
and Gahinet, 1996). If a single Lyapunov function can
where the matrices Ai = A(ωi ) are time-invariant and
be found for all the vertices of the polytopic system, the
correspond to the image of a vertex of Υ.
derived LPV observer stabilizes the LPV system for all
In our case, with 2 parameters bounded in [ρ1 , ρ1 ] and possible variations of the scheduling parameters in the
[ρ2 , ρ2 ], (see (Bara et al., 2001) for the general case) the bounded set. 
corresponding polytope is: Applying Proposition 1 to system (13)-(14) for: α = 4;
Υ = Co{ω1 , ω2 , ω3 , ω4 } θ = π/6; r = 80 associated with E = I2 and Cz = [0 1],
(18)
= Co{(ρ1 , ρ2 ), (ρ1 , ρ2 ), (ρ1 , ρ2 ), (ρ1 , ρ2 )} gives the roots locus depicted in Fig. 4 and γ∞ = 0.8902.

392
 IFAC SAFEPROCESS 2018
Warsaw, Poland, August 29-31, 2018 Donatien Dubuc et al. / IFAC PapersOnLine 51-24 (2018) 389–394 393

Table 2. N RM S obtained for the different

cases
r Model for Dt
Orifice eq. HW
Cycle
WHSC 0.0610 0.0255
θ WHTC 0.0499 0.0189

(8)-(10), are respectively presented in Fig. 6 and 7. The

Considered area α
observer manages to follow the variations of the pressure
with a reasonable error for both scenarios.

Fig. 4. Roots locus of Twez

GT-POWER Simulink

ωtc
Cycle
Plant Observer

Inputs
Orifice Interpolation
eq. functions

HW P̂3

Fig. 5. Scheme of the simulation tests

The observer gains at the 4 vertices of the polytope are:
L1 = [70.60 1.10]T , L2 = [96.79 2.01]T Fig. 6. WHSC cycle with Dt defined by (8)-(10)
T T
L3 = [70.60 1.93] , L4 = [101.91 2.36]

4. SIMULATION RESULTS

The validation will be based on two standard types of test:

- The World Harmonized Stationary Cycle (WHSC),
which is a succession of stationary points in the engine
speed and torque.
- The World Harmonized Transient Cycle (WHTC),
which is a transient test based on the pattern of heavy
duty commercial vehicles.
These two cycles are used as inputs for the GT-POWER
simulator, then the needed data are collected to feed the
observer (19) of the system (13). The scheme presented
in Fig. 5, summarizes the considered simulation tests. In
addition to both cycles, we will compare the two different
models for Dt defined in Section 2.3. Fig. 7. WHTC cycle with Dt defined by (8)-(10)
For comparison purpose, let us define a performance
estimation criterion as the normalized root mean square 4.2 Case 2: Dt defined by HW model
N RM S: 
1 Ne In this case Dt is defined by the Hammerstein-Wiener
|P̂3 (n) − P3 (n)|2
Ne n=1 model identified in Section 2.3. The results for the WHSC
N RM S = (24) cycle are depicted in Fig. 8 and for the WHTC in Fig. 9.
maxn=1,Ne P3 (n) − minn=1,Ne P3 (n)
One can observe that the estimation is improved and, for
where N e is the number of data points.
some stationary points, the error is close to zero. This is
The performance results for the different cases and scenar- confirmed by the NRMS results shown in Table 2.
ios are summarized in Table 2.
5. CONCLUSION
4.1 Case 1: Dt defined by the orifice equations (8)-(10)
In this paper, we have designed a LPV observer in order to
The results obtained for the two scenarios (WHSC/WHTC) estimate the exhaust manifold pressure of a turbocharged
in the case where Dt is defined by the orifice flow equations diesel engine. In Section 2, a mean value model of the

393
IFAC SAFEPROCESS 2018
394
Warsaw, Poland, August 29-31, 2018 Donatien Dubuc et al. / IFAC PapersOnLine 51-24 (2018) 389–394
Fig. 8. WHSC cycle with Dt defined by the Hammerstein-
Wiener model
Fig. 9. WHTC cycle with Dt defined by the Hammerstein-
Wiener model
turbocharger and pressure dynamics has been developed.
In this section, we also proposed two models of the turbine
mass flow rate. A standard model from the orifice flow
equations and a Hammerstein-Wiener model.
Then, in Section 3, thanks to the LPV framework, a
polytopic LPV observer has been designed to estimate the
state space vector of the system. It is worth mentioning
that this quasi-LPV observer internally depends on the
estimated state variable x̂2 , and further works must be
performed to study the effects of the uncertainties in the
parameters due to the estimation of P3 as investigated in
(Heemels et al., 2010).
In Section 4, the observer has been tested in a high-fidelity
simulator in GT-POWER in different scenarios. It has
been shown that the observer has a low estimation error in
every tested scenarios. As shown in Table 2, the HW model
has the best estimation results. This observation could be
expected because it has been established in Section 2.3
that this model has the best fit. However, this model seems
to be too complex for a real-time implementation. Then, in
practical implementations, the orifice flow equations may
be preferred.
394
Even if the estimation results are good, it appears that the
estimation error is more important at certain points. It has
been established in Section 4.1 and 4.2 that the modeling
improvement on the variable Dt corrects most of these
errors. However, some still remain. To attenuate them, one
can investigate in further works the constant assumption
made on the thermal coefficients, improve the precision
of the efficiency maps of the turbocharger or use the
pressure measurement just after the compressor instead of
the intake manifold one to compute the compressor power
(6).
REFERENCES
Apkarian, P., Gahinet, P., and Becker, G. (1995). Self-
scheduled H∞ control of linear parameter-varying sys-
tems: a design example. Automatica, 31(9), 1251–1261.
Bara, G.I., Daafouz, J., Kratz, F., and Ragot, J. (2001).
Parameter-dependent state observer design for affine
LPV systems. International Journal of Control, 74(16),
1601–1611.
Castillo, F., Witrant, E., Dugard, L., and Talon, V. (2013).
Exhaust Manifold Pressure Estimation Diesel Equipped
with a VGT Turbocharger. SAE Technical Paper.
Chilali, M. and Gahinet, P. (1996). H∞ design with
pole placement constraints: an LMI approach. IEEE
Transactions on Automatic Control, 41(3), 358–367.
Fredriksson, J. and Egardt, B. (2002). Estimating exhaust
manifold pressure in a turbocharged diesel engine. In
Conference on Control Applications, volume 2, 701–706.
Heemels, W.P.M.H., Daafouz, J., and Millerioux, G.
(2010). Observer-Based Control of Discrete-Time LPV
Systems With Uncertain Parameters. IEEE Transac-
tions on Automatic Control, 55(9), 2130–2135.
Isermann, R. (2014). Engine modeling and control: model-
ing and electronic management of internal combustion
engines. Springer.
Jung, M., Ford, R.G., Glover, K., Collings, N., Christen,
U., and Watts, M.J. (2002). Parameterization and Tran-
sient Validation of a Variable Geometry Turbocharger
for Mean-Value Modeling at Low and Medium Speed-
Load Points. SAE Technical Paper.
Mohammadpour, J., Franchek, M., and Grigoriadis, K.
(2012). A survey on diagnostic methods for automotive
engines. International Journal of Engine Research,
13(1), 41–64.
Moulin, P. (2010). Modélisation et Commande des
Systèmes d’Air des Moteurs Suralimentés. Ph.D. thesis,
MINES ParisTech.
Olin, P.M. (2008). A Mean-Value Model for Estimating
Exhaust Manifold Pressure in Production Engine Ap-
plications. SAE Technical Paper.
Salehi, R., Shahbakhti, M., Alasty, A., and Vossoughi,
G.R. (2013). Nonlinear observer design for turbocharger
in a SI engine. In Proceedings of the American Control
Conference (ACC), 5231–5236.
Yue-Yun Wang and Haskara, I. (2010). Exhaust pressure
estimation and its application to variable geometry
turbine and wastegate diagnostics. In Proceedings of
the American Control Conference (ACC), 658–663.
Zhu, Y. (2002). Estimation of an N–L–N Hammer-
stein–Wiener model. Automatica, 38(9), 1607–1614.