Preprints, 8th IFAC International Symposium on

Preprints,
Preprints, 8th
Advances IFAC
IFAC International
in Automotive
8th Control Symposium
International Symposium on
on
Preprints,
Advances 8thAutomotive
IFAC International Symposium on
June 19-23,in
Advances in2016. Control
Norrköping,
Automotive Sweden
Control Available online at www.sciencedirect.com
Advances
June in Automotive Control
June 19-23,
19-23, 2016.
2016. Norrköping,
Norrköping, Sweden
Sweden
June 19-23, 2016. Norrköping, Sweden
ScienceDirect
IFAC-PapersOnLine 49-11 (2016) 469–475
A Combustion Cycle Model for Stationary
A
A Combustion
Combustion Cycle
Cycle Model for
for Stationary
ModelOperation
Stationary
and Transient Engine
and Transient
and Transient Engine
Engine Operation
Operation
T. Makowicki ∗ M. Bitzer ∗ P. Kotman ∗ K. Graichen ∗∗
∗ ∗ ∗ ∗∗
T.
T. Makowicki ∗∗ M. Bitzer ∗∗ P. Kotman ∗∗ K. Graichen ∗∗
T. Makowicki
Makowicki M. M. Bitzer
Bitzer P. P. Kotman
Kotman K. K. Graichen
Graichen ∗∗
∗
Corporate Research (Control Engineering), Robert Bosch GmbH,
∗ Corporate Research (Control
GermanyEngineering), Robert Bosch GmbH,
∗
∗
Corporate Research
71272 Renningen,
Corporate Research (Control
(Control Engineering), Robert
(e-mail: {Thomas.Makowicki;
Engineering), Robert Bosch
Bosch GmbH,
GmbH,
71272
71272 Renningen,
Renningen, Germany
Germany (e-mail:
(e-mail:
Matthias.Bitzer2;Philipp.Kotman}@de.bosch.com). {Thomas.Makowicki;
{Thomas.Makowicki;
71272 Renningen, Germany (e-mail:
Matthias.Bitzer2;Philipp.Kotman}@de.bosch.com). {Thomas.Makowicki;
∗∗ Matthias.Bitzer2;Philipp.Kotman}@de.bosch.com).
Institute of Measurement, Control and Microtechnology, Ulm
Matthias.Bitzer2;Philipp.Kotman}@de.bosch.com).
∗∗ Institute
∗∗
of Measurement,
Measurement, Control and Microtechnology, Ulm Ulm
University,Institute
Institute 89081of Ulm, GermanyControl
of Measurement, (e-mail: and Microtechnology,
knut.graichen@uni-ulm.de)
GermanyControl and Microtechnology, Ulm
∗∗
University, 89081
University, 89081 Ulm, Germany
Ulm, (e-mail:
(e-mail: knut.graichen@uni-ulm.de)
knut.graichen@uni-ulm.de)
University, 89081 Ulm, Germany (e-mail: knut.graichen@uni-ulm.de)
Abstract: Since novel approaches of diesel engine control have to balance the emissions and the
Abstract:
Abstract: Since
Since novel approaches of diesel engine control have tooperation,
balance the emissions and the
generated torque
Abstract:
generated Since novel
torque novel
most
approaches
most accurately
approachesduring
accurately
of
of diesel
duringdiesel engine
highly
engine
highly
control
transient
control engine
transient
have
have to
engine balance
balance the
tooperation, the emissions
exact and
and the
cycle-by-cycle
emissions
exact cycle-by-cycle the
generated
estimates of
generated torque
thesemost
torque engine
most accurately
outputsduring
accurately during highly
are mandatory.
highly transient
Accordingly,
transient engine
engine operation,
this paper exact
operation, develops
exact cycle-by-cycle
a control-
cycle-by-cycle
estimates
estimates of these
of these engine outputs are mandatory. Accordingly, this paper develops aaa control-
control-
estimates
oriented of these engine
oriented combustion
combustion engine
cycle
outputs
cycle model
outputs
model
to are mandatory.
determine
to are mandatory.
determine
Accordingly,
the indicated
Accordingly,
the indicated
indicated
mean this
mean this paper
effective
paper
effective
develops
pressure
develops
pressure
as well
as i.e.
well
as the
control-
as gas
the
oriented
nitrogen
oriented combustion
oxide
combustion and cycle
the smoke
cycle model
model to
to determine
emissions. The
determine the
proposed
the model
indicated mean
meanhas effective
a hybrid
effective pressure
structure,
pressure as
as well
well as
the
as the
the
nitrogen
nitrogen
exchange oxide
oxide
and and
and the
the
compression smoke
smoke are emissions.
emissions.
modeled The
The proposed
proposed
physics-based model
model
with has
has
lumped aa hybrid
hybrid
mass structure,
structure,
and energy i.e.
i.e. the
the
balances,gas
gas
nitrogen
exchange oxide
and and the
compression smoke are emissions. The
modeled physics-basedproposed model
with has
lumped a hybrid
mass structure,
and energy i.e. the
balances,gas
exchange
whereas the
exchange and
and compression
combustion
compression are
are modeled
modeled physics-based
is approximated by a data-based
physics-based with
with lumped
approach
lumpeddue mass
due
massto theand energy
energy balances,
complex
andcomplex chemical
balances,
whereas
whereas
reaction the
the combustion
combustion
kinetics. Thus, is
is
the approximated
approximated
model outputs by
byarea
a data-based
data-based
described approach
approach
with respect due
to to
to
the the
the complex
time-varying chemical
chemical
signals
whereas
reaction the combustion
kinetics. Thus, is approximated
the model outputs
outputs byarea data-based
described approach
with respect due
to to
thethe complex chemical
time-varying signals
reaction
reaction kinetics.
of the transient
kinetics. airThus,
paththe
Thus, the model
conditions
model (intake are
outputs are described
pressure, oxygen
described with
with respect
concentration,
respect to
to the
the time-varying
and temperature)
time-varying signals
signalsas
of
of
wellthe
theastransient
transient
the inputs air
air path
path
adjusted conditions
conditions
by the (intake
(intake
fuel pressure,
pressure,
injection oxygen
oxygen
(start positionconcentration,
concentration,
and mass and
and
of eachtemperature)
temperature)
fuel injection as
as
of
welltheastransient
the inputs airadjusted
path conditions (intake pressure, oxygen concentration, and temperature) as
well
pulse).
well as
as the inputs
Measurement
the inputs data isby
adjusted
adjusted by
byused the
the
the
fuel
fuel
for
fuel
injection
injection
the (start
(start
parametrization
injection (start
position
position and
and
and validation
position and
mass
mass
mass
of
of
of
of
each
each
the
each
fuel
fuel
model.
fuel
injection
injection
injection
pulse). Measurement
pulse). Measurement data data is is used
used for the the parametrization
parametrization and and validation
validation of of the
the model.
model.
pulse).
© 2016, Measurement
IFAC (International is used for
dataFederation forof the parametrization
Automatic Control) Hosting and validation
by Elsevier of Ltd.the Allmodel.
rights reserved.
Keywords: Combustion modeling, Gaussian processes, Transient engine operation
Keywords: Combustion
Keywords:
Keywords: Combustion modeling,
Combustion modeling, Gaussian
modeling, Gaussian processes,
Gaussian processes, Transient
processes, Transient engine
Transient engine operation
engine operation
operation
1. INTRODUCTION predict the cylinder filling during transients and describe
1.
1. INTRODUCTION
INTRODUCTION predict
predict
the chemical the
the cylinder
cylinder
reaction filling
filling
kineticsduring
during duringtransients
transients and
and describe
the combustion describe of
1. INTRODUCTION predict
the chemical the cylinder
reaction filling
kineticsduring duringtransients
the and describe
combustion of
the chemical
multiple
the chemical fuel reaction
injections kinetics
(Warnatz
reaction kinetics during during
et al.,the combustion
2006).
the combustion of of
Modern diesel engines must accomplish complex and par- multiple multiple fuel
multiple fuel injections
fuel injections (Warnatz
injections (Warnatz et
(Warnatz et al.,
et al., 2006).
al., 2006).
2006).
Modern
Modern diesel engines
diesel engines
tially contrary engines must
must
goals in must accomplish
accomplish
stationary complex and
complexoperation,
and transient par-
and par-
par- The model presented in this paper predicts the indicated
Modern
tially diesel
contrary goals in accomplish
stationary and complex
transient and
operation, The
The
mean model
modeleffective presented
presented
pressure in
in this
this
(IMEP) paper
paper predicts
predicts
reflecting the indicated
the generated
the indicated
tially contrary goals
e.g. generating
tially contrary goals in stationary
the desired
in stationary
torque,andand transient
emitting
transient operation,
amounts The
lowoperation, model presented in (IMEP)
this paper predicts the the indicated
e.g. generating the desired torque, emitting low amounts mean
mean
torque, effective
effective
the pressure
pressure
nitrogen oxides (IMEP)
(NO ),reflecting
reflecting
and the the
smoke generated
generated
(SM) on
e.g. generating
of emissions
e.g. generating the
andthe desired
having
desired torque,
a high
torque, emitting
efficiency.
emitting low
Thus, amounts
amounts mean
auxiliary
low auxiliary effective pressure (IMEP)),reflecting
x the generated
of emissions and having aa high efficiency. Thus, torque,
torque, the
the nitrogen
a cycle-by-cycle nitrogenbasis.oxides
To do(NO
oxides so,xxxa),
(NO and the
and
cylinder the chamber
smoke
smoke (SM)
model
(SM) on
on
of emissions
components
of and
such
emissions such having
as
and having high
turbochargers, efficiency.
a high efficiency. exhaust Thus,
gas
Thus, auxiliary
recircula-
auxiliary ais torque, the nitrogen oxides (NO ), and the smoke (SM) on
components as turbochargers, exhaust gas recircula- a cycle-by-cycle
cycle-by-cycle
set up with a basis.
basis.
hybridTo
To do
do so,
so,
structure. aa cylinder
cylinder
This chamber
chamber
means model
model
that the
components
tion or
componentshigh such
pressureas turbochargers,
fuel injection
such as turbochargers, exhaust
are utilized gas recircula-
(Heywood,
exhaust gas(Heywood, recircula- is a cycle-by-cycle basis. To do so, a cylinder chamber model
is set
set up
up withand aa hybrid structure. This means that the
tion
1988;or
tion
tion or
high
high pressure
orMollenhauer
high pressureand
pressure
fuel
fuel
injection
fuelTschoke,
injection2010).
injection
are
are utilized
are utilized
utilized parts are gas
(Heywood,
These(Heywood, is
gas set up with
exchange
exchange withand a hybrid
hybrid
the
structure.
the compression
structure.phase
compression
This
This are
phase
means
means
are
that
that the
described
described
by
the
by
1988;
1988; Mollenhauer
Mollenhauer
summarized and
and
to subsystems Tschoke,
Tschoke, 2010).
2010).
such as 2010). These
These
the air, These parts
parts
the combustion, are
are gas
gas exchange
physics-based
exchange and
lumped
and the
the compression
mass
compression and phase
energy
phase are
are described
balances whereas
described by
by
1988; Mollenhauer
summarized to and
subsystems Tschoke,
such as the air, the parts
combustion, are physics-based
physics-based
the combustion lumped
lumped
phase mass
mass
is and
and
approximatedenergy
energy balances
balances
by a whereas
whereas
data-based
summarized
the injection,
summarized to subsystems
and the
to subsystems such
mechanical as the
system,air,
such as the air, the the
see combustion,
Fig. 1.
combustion, physics-based lumped mass and energy balances whereas
the
the combustion
combustion phase
phase is
is approximated by a data-based
the
the injection,
the injection, and
injection, and the
and the mechanical
the mechanical system,
mechanical system, see
system, see Fig.
see Fig. 1.
Fig. 1.
1.
modeling
the combustion
modeling
approach
approach phase is approximated
(Gaussian
(Gaussian
Process Regression)
approximated
Process
by
by aa data-based
Regression)
due to
data-based
due to
The control of the entire plant requires the coordination the modeling
complex
modeling approach
reaction
approach (Gaussian
kinetics.
(Gaussian Process
Thus,
Process Regression)
an efficient
Regression) due
due to
way of
to
The
The
of control
control
the various of
of the
the entire
entire
subsystems plant
plant
(Isermann,requires
requires the
the
2014). coordination
coordination
During tran- the
the complex
complex
describing reaction
reaction
thereaction
combustion kinetics.
kinetics. Thus,
Thus,
of multiple an
an efficient
efficient
fuelefficient
injectionsway
way of
of
uti-
The
of control
the various of the entire
subsystems plant
(Isermann,requires the
2014). coordination
During tran- the complex kinetics. Thus, an way of
of
of the
sients
the various
this
varioustask subsystems
becomes
subsystems (Isermann,
challenging
(Isermann, 2014).
since
2014). During
each
During tran-
exhibits
tran- describing
describing
lizing a globalthe
the combustion
combustion
non-parametric of
of multiple
multiple
regression fuel
fuel injections
injections
approach uti-
uti-
(Berger
sients describing the combustion of multiple fuel injections uti-
its ownthis
sients
sients this
this
task
task becomes
dynamics
task becomes
(Asprion
becomes
challenging
challenging
et al., 2014;
challenging
since
since Grahn
since
each
each exhibits
each et al., lizing
exhibits
exhibits al., aaa2011)
lizing
et
lizing
global
global
global
non-parametric
non-parametric
is combined
non-parametric with aregression
regression
regression
approach
approach (Berger
approach (Berger
precise physics-based filling
(Berger
its
its
2014;own
own dynamics
dynamics
Mrosek (Asprion
(Asprion
et al.,(Asprion
2010). Asetthe et
et al.,
al., 2014;
2014;
fuel Grahn
Grahn
injection et
et
(i.e. al.,
al.,
the et
et al.,
al.,
estimate, 2011)
2011) is
is
which combined
combined
extends with
with
the a
a precise
precise
well-known physics-based
physics-based
mean-value filling
filling
its
2014;own dynamics
Mrosek et al., 2010). As theal., 2014;
fuel Grahn
injection et
(i.e. al.,
the et al., 2011) is combined with a precise physics-based filling
2014;
positionMrosek
2014; Mrosekand et
mass al.,
et al.,of 2010).
the
2010). fuelAs the
injection
Asinjection fuel injection
pulses)
the fuel pulses)
injection (i.e.
affects the
(i.e. the estimate,
estimate,
estimationwhich
the estimate, which
which extends
extends
(Heywood, the
the
1988) well-known
well-known
and provides mean-value
mean-value filling
filling
an adequate
position and mass of the fuel affects extends the well-known mean-value filling
position
combustion andfast
mass ofdirectly,
andof the fuel
fuelitinjection
injection
represents pulses) affects the
a cycle-by-cycle the estimation
estimation (Heywood,
(Heywood, 1988)
1988) for and
and provides
provides an
an adequate
position
combustion and mass
fast and the
directly, it represents pulses) affects
aa cycle-by-cycle
time-continuous
estimation
time-continuous (Heywood, interface
interface 1988) for
models
and
models provides an adequate
describing
describing
the air
adequate
the air
combustion
manipulated
combustion fast
input
fast andfordirectly,
the
andfordirectly, it represents
combustion
it represents system. cycle-by-cycle
In contrast,
a cycle-by-cycle time-continuous
system delay and interface
dead-time for models
effects describing
(Rowland, 1986; theMed-air
manipulated input the combustion system. In contrast, time-continuous
system delay and interface
dead-time foreffects
models describing
(Rowland, 1986; theMed-air
manipulated
the air
manipulatedsystem input
input for
conditions
for the
the combustion
(i.e.
combustion the system.
intake
system. In
and
In contrast,
exhaust
contrast, system
dahi et delay
al., and
2015; dead-time
Kotman et effects
al., (Rowland,
2009). The 1986;
hybrid Med-
model
the air system conditions (i.e. the intake and exhaust system
dahi et and delay
al., and
2015; dead-time
Kotman et effects
al., (Rowland,
2009). The 1986; Med-
the
the air system
pressures,
air system conditions
temperatures
conditions (i.e. the
or oxygen
(i.e. the intake and
concentrations)
intake and exhaust
exhaust
change scheme dahi
dahi etet and al.,
al., 2015; Kotman
the selection
2015; Kotmanof et al.,
etthe 2009).
al.,inputs
2009).of eachhybrid
The
The hybrid
submodel
hybrid
model
model
model is
pressures,
pressures,
slowly due temperatures
temperatures
to turbocharger or
or oxygen
oxygen
inertia concentrations)
concentrations)
and gas transport change
changede- scheme
scheme
motivated and bythe
thethe selection
selection
intrinsic of
of the
the inputs
inputs
structure ofof
of each
each
the submodel
submodel
cyclic cylinder is
is
pressures,
slowly due temperatures
to turbocharger or oxygen
inertia concentrations)
and gas transport changede- scheme and the selection of the inputs of each submodel is
slowly
lays
slowly due
(Nakayama
due to
to turbocharger
et al.,
turbocharger 2008; inertia
Hillion
inertia and
et
and gas
al.,
gas transport
2009).
transport For de-
the
de- motivated
motivated
operation by
by the
the
interacting intrinsic
intrinsic
with structure
structure
the air and of
of the
the cyclic
cyclic
injection cylinder
cylinder
system.
lays motivated by the intrinsic structure of the cyclic cylinder
lays (Nakayama
(Nakayama
combustion system et al.,
al., 2008;
2008; Hillion
et operating Hillion et
et al.,
al., 2009).
on a cycle-by-cycle For
2009).basis, the operation
For the
the operation interacting
interacting with with the
the air and injection system.
lays (Nakayama
combustion system et al., 2008;
operating Hillion
on et al.,
aatime-varying
cycle-by-cycle 2009). For
basis, the operation
The paper interacting
is structured with the air
as follows: air In and
and injection
injection
Section
system.
2 thesystem.
physics-
combustion
air system
combustion system
inputs
system operating
thus represent
operating on
on a cycle-by-cycle
cycle-by-cycle basis,
conditions.
basis, the
the The
The paper
paper is
is structured
structured as
as follows:
follows: In
In Section
Section 2
2 the
the physics-
air
air system
system the
Therefore,
air system
inputs
inputs thus
engine
inputs thus
represent
thuscontrol
represent
represent schemetime-varying
time-varying
has to determine
time-varying
conditions.
conditions.
conditions. the The based
based papercylinder
cylinder
chamber
is structured
chamber
model
asmodel
follows: is
is
introduced
In Section 2with
introduced the physics-
with respect
physics-
respect
Therefore,
Therefore,
fuel injectionthe
the engine
engine
parameters control
controlwith scheme
scheme
respect has
has to
to
totothedetermine
determine
air system the
the based
to
basedthe cylinder
engine
cylinder chamber
subsystem
chamber model
model is
interaction.introduced
is introduced In with
with respect
Section 3 the
respect
Therefore,
fuel the
injection engine
parameters controlwith scheme
respect has to determine
the air systemthe to
to
modelthe
the engine
engine
isengine
analyzed subsystem
subsystem interaction.
interaction.
from a system-theoretic In
In Section
Section
perspective 3 and
3 the
the
fuel
state
fuel injection
to achieve
injection parameters
an optimal
parameters with respect
trade-off
with respect to
between thetheair system
transient
to thetheairtransient
system model to the subsystem interaction. In Section 3 and
the
state to achieve an optimal trade-off between model
applied is
is analyzed
analyzed
to the phases from
fromof aathesystem-theoretic
system-theoretic
engine cycle, perspective
perspective
whereas the and
data-
state to
emissions,
state achieve
torque, an optimal
and
to achieve an optimal trade-off
efficiency. between
Thus,
trade-offThus, between the
control transient
oriented
the transient model is analyzed from a system-theoretic perspective and
applied
applied to
to the phases of
ofthetheengine cycle, whereas the data-
emissions,
models are torque,
emissions,
emissions, requiredand
torque,
torque, and efficiency.
efficiency.the
to determine
and efficiency. Thus,
engine
Thus,
control
control
outputs
control in de- based
oriented
oriented
oriented applied
based to the
the phases
approximation
approximationphases of of
of
the
the
the
engine
combustion
engine
combustion
cycle,
cycle, whereas
whereas
is
the
the data-
is discussed.
discussed.
Sec-
data-
Sec-
models
models
pendence are
are required
required
ofrequired to
to
the manipulated determine
determine inputsthe
the engine
engine
and outputs
outputs
theoutputs in
in
time-varying de-
de- based
tion
based 4 approximation
verifies
approximationthe modelof the
based
of based combustion
on
the combustion stationary is discussed.
and
is discussed. Sec-
transient
Sec-
models
pendence areof to determine the engine in de- tion
tion
test 4
4
benchverifies
verifies the
the
data. model
model
Finally, based
Section on
on stationary
stationary
5 concludes and
and
the transient
transient
paper.
pendence
signals. Such
pendence of the
of the manipulated
manipulated
models
the are challenging
manipulated
inputs
inputs as
inputs
and
and
andtheythe
themust
the
time-varying
time-varying
precisely tion
time-varying 4 verifies theFinally,
model based on 5stationary and transient
test bench data. Section concludes the paper.
signals.
signals. Such
signals. Such models
Such models are
models are challenging
are challenging as
challenging as they
as they must
they precisely test
must precisely
must precisely test bench
bench data. data. Finally,
Finally, Section
Section 55 concludes
concludes the the paper.
paper.
Copyright 2016 IFAC
2405-8963 © 2016, 479 Hosting by Elsevier Ltd. All rights reserved.
IFAC (International Federation of Automatic Control)
Copyright
Peer review©
Copyright 2016
2016 IFAC
©under IFAC 479
479Control.
Copyright © 2016 responsibility
IFAC of International Federation of Automatic
479
10.1016/j.ifacol.2016.08.069
IFAC AAC 2016
470
June 19-23, 2016. Norrköping, Sweden T. Makowicki et al. / IFAC-PapersOnLine 49-11 (2016) 469–475

Injection System Combustion System Air System

uFI Diesel Intake Throttle Charge
Intake Injector Exhaust Manifold Valve Air Cooler
Common Diesel Valve Valve
Rail Injector uTV

Ambience
ṁIV
E ṁEV uRV
Compressor
High-Pressure Q̇W

EGR
Pressure Q̇B  
Engine
Pump Sensor m Turbine
...
pr mO2  Lambda
uHP p Sensor
... Exhaust
ṁI Mechanical Manifold uWG
...

nE System Wastegate Exhaust Aftertreatment

Fig. 1. Scheme of the internal combustion engine with the combustion, air, injection, and mechanical subsystems.

2. PHYSICS-BASED CYLINDER MODEL 2
ṁv = αv Av pu ψ(pu , pd , κ) with pu > pd (3)
R Tu
The control-oriented cylinder chamber model requires on    2   κ+1 

the one hand an appropriate description of the inputs  κ
 pd κ pd κ
and internals to accomplish the desired accuracy and and ψ(pu , pd , κ) = −
κ−1 pu pu
on the other hand needs to be set up as a low-order
model with respect to implementation conditions and depending on the up- and downstream pressure pu and pd
controller design. A corresponding cylinder chamber model as well as on the upstream temperature Tu . Further pa-
is described in Subsection 2.1, whereas Subsection 2.2 rameters are the adiabatic index κ, the cross-sectional
discusses the calculation of the IMEP pmi , the nitrogen area Av and the flow coefficient αv (Guzzella and Onder,
oxide emissions ENOx , and the smoke emissions ESM . 2004). It is worth noting that equation (3) changes sign
for an inverse pressure gradient. With respect to (1)-(3)
2.1 First-Principle Cylinder Chamber Model the intake and exhaust pressures pIM , pEM , the tempera-
O2 O2
tures TIM , TEM and the oxygen mass fractions XIM , XEM
The model considered in this paper describes the cylin- are assumed to be known e.g. by measurements as depicted
der chamber by means of a single, ideally mixed ther- in the test bench scheme in Fig. 6 or by respective models.
modynamic zone (Pischinger et al., 2009). This lumped The fuel mass flow ṁI consisting of two pulses, see Fig. 2,
parameter approach comprises an overall and a component is approximated by a sequence of heavy-side functions Θ(·)
mass balance as well as an energy balance to describe the 2
cylinder state variables mass m(t), oxygen mass mO2 (t), mIi  
ṁI ≈ Θ(t − tSIi ) − Θ(t − (tSIi + ∆tSIi )) . (4)
and pressure p(t), see Fig. 1. Here, two chemical fractions i=1
∆t SIi

f ∈ {O2 , O2 } are distinguished, namely oxygen and non- The variables tSIi and ∆tSIi define the start and duration
oxygen. Since the data-based approximation of the com- of the fuel pulses with ∆tSIi = f∆t (mIi , pr (t)) depending
bustion phase does not describe the fuel oxidation over on the fuel mass mIi and the fuel pressure pr (t), see Fig. 1.
time and it is assumed that no fuel remains after the The start time tSIi refers to the piston top dead center
stoichiometric combustion, diesel fuel (fl) has no balance (TDC). It is worth noting that the parametrization of the
equation. Further, the impact of the emissions ENOx and fuel mass flow ṁI (4) may equivalently be expressed by
ESM on the overall cylinder mass balance is neglected. means of the crank angle ϕSIi instead of tSIi .
Nevertheless, their generation is considered within the
equation of the outputs discussed in Subsection 2.2. The ODE (ordinary differential equation) of the cylinder
pressure p(t) is derived from the energy balance as well as
The overall mass balance is given by the ideal gas law and holds as
dm(t)
= ṁIV (t) − ṁEV (t) + ṁI (t) (1) dp(t) R  cp 
dt = Q̇B + Ψ + Ω − p V̇ , (5)
dt V cv     R 
with the mass flows ṁv , v ∈ {IV, EV}, through the intake (A) (B)
(C)
valve (IV) and exhaust valve (EV) as well as the injected  
fuel mass flow ṁI . The oxygen mass balance is fl fl cp Rfl T
ψ =ṁI h (T̃ ) − h (T ) + ,
dmO2 (t) R
= ṁO O2 O2
(2)  
IV (t) − ṁEV (t) − ṙ (t)
2

dt   f f f cp R f T
Ω = v f ṁv h (T̃ ) − h (T ) + .
with the oxygen mass flow ṁO R
v = Xv ṁv and the oxygen
2 O2

mass fraction Xv which depend on the flow direction

O2
tSI2
of ṁv . The oxygen demand of the combustion is described ṁI (t)
tSI1 Injection 1
by ṙO2 = µO2 ṁI,b (t) with the stoichiometric factor µO2 ∆tSI1 ∆tSI2 (Pilot)
and the burned fuel mass ṁI,b .
Injection 2
An accurate prediction of the overall and the oxygen mass (Main)
requires an exact calculation of the mass flows through the TDC t
inlet and the outlet valves. Therefore, both mass flows are
described by means of detailed throttle equations Fig. 2. Fuel mass flow profile with two injection impulses.

480
 IFAC AAC 2016
June 19-23, 2016. Norrköping, Sweden T. Makowicki et al. / IFAC-PapersOnLine 49-11 (2016) 469–475 471

 T
Thereby, term (A) is the heat Q̇B released by the combus- m(t) mO2 (t) p(t) p(t)
Air
tion. Term (B) summarizes the enthalpy flows through the System  T
v IM (t) v EM (t) O
fuel injector and the valves with the specific enthalpy h... , Combustion X 2 (t)
System
the in-cylinder temperature T and the gas temperature T̃ Injection

ENOx ,k
chosen in dependence of the mass flow direction (Kotman System T  dx
tSI1 tSI2 mI1 mI2 = f (. . . ) ESM,k
et al., 2009). Term (C) accounts for the volumetric work pr (t)
k
dt
Mech. pmi,k
with the first derivative V̇ of the cylinder volume. The vari- System nE (t)
ables cv and cp are the specific isochoric and isobaric heat
capacities whereas R stands for the specific gas constant.
The heat loss Q̇W through the cylinder walls is neglected Fig. 3. Overview of the input-output-structure of the
for reasons of simplicity. combustion system and neighboring subsystems.
Since Q̇B results from complex chemicals reaction kinetics,
dx(t)
it is hard to describe in a control-oriented manner. To = f (x(t), u(t), v(t), nE (t)), t > 0 (9)
this end, Section 3 derives a data-based combustion phase dt
 T
approximation including the effects of the heat release Q̇B . with x(t) = m(t) mO2 (t) p(t) and x(0) = x0 .
The variable nE is the engine speed. The time-dependent
2.2 Cycle-by-Cycle Outputs of the Combustion System
signal u(t) considers the fuel mass flow ṁI (t) of (4), i.e.
 
Unlike the continuous-time states discussed in Section 2.1, u(t) = ṁI (t) = fI t, uk , pr (t) (10a)
the IMEP and the emissions are single-valued outputs with uk = [tSI1 tSI2 mI1 mI2 ]k . (10b)
computed once per cycle. Thereby, the IMEP is given by
 tk+1 Thereby, the parameters uk of the basis function (4)
1
pmi,k = p(t) V̇ (t) dt (6) are the actual cycle-by-cycle manipulated variables. The
Vcyl tk rather slowly time-varying conditions in the air system are
based on the cylinder pressure p(t), the first derivative V̇ (t) summarized by the vector
of the chamber volume and the cylinder displacement Vcyl . T
v(t) = [v IM (t) v EM (t)] (11)
The start and end time of the k th engine cycle are O2
defined by tk and tk+1 , respectively. Furthermore, the with v IM (t) = [pIM (t) TIM (t) XIM (t)]T
exhaust emissions ENOx and ESM , which are generated in and v EM (t) = [pEM (t) O2
TEM (t) XEM (t)]T
dependence of the in-cylinder conditions, are calculated by
 tk+1 with variables v IM (t) and v EM (t) representing the condi-
 
ENOx ,k = ṙNOx m(t), mO2 (t), p(t) dt, (7) tions in the intake and exhaust manifold, respectively.
tk
 tk+1 The continuous-time measurement y(t) comprises the
 
ESM,k = ṙSM m(t), mO2 (t), p(t) dt. (8) pressure p and the oxygen fraction X O2 , i.e.
tk  T
It is worth mentioning, that analytical and accurate ex- y(t) = p(t) X O2 (t) =: h(x(t)) (12)
pressions of the reaction rates ṙ(NOx ) and ṙ(SM) are hard whereas the cycle-discrete outputs (6)-(8) are summarized
to derive. Accordingly, a hybrid model of the engine cycle in the vector
is discussed in Subsection 3 which comprises data-based T
y k+1 = [pmi ENOx ESM ]k =: g(x(tk )). (13)
approximations of this model part.
The block diagram of Fig. 3 visualizes the discussed system
3. HYBRID MODEL OF THE ENGINE CYCLE theoretic structure of the cylinder model (9)-(13).
In this section, a hybrid model structure is proposed which Exhaust Intake
contains physics- and data-based model parts in order to Stroke EC/IO Stroke
obtain the cycle-discrete outputs of the control-oriented cycle
cylinder chamber model. Therefore, the inherent cyclic EO start tEC IO
engine operation is analyzed jointly with the physical phe- k /tk IC
cycle
nomena occurring during the combustion. In a first step, end tEO tIC
k Fuel k
a general system-theoretic representation of the overall pmi Injection
engine model (1)-(8) is derived in Subsection 3.1. Since ENOx
the engine cycle is a periodic sequence of phases, i.e. gas ESM
exchange, compression and combustion, the representation tSC
k
tSI1
k
can also be set up as a series connection of submodels as
discussed in Subsection 3.2. Therefore, the gas exchange SC SI1
Combustion & Compression
phase and the compression phase are described by physics Stroke
Expansion
while the combustion phase is approximated by a data-
based model as discussed in Subsection 3.3.
Fig. 4. Combustion cycle with EO = exhaust valve open-
3.1 State Space Representation of the Cylinder Model ing, EC = exhaust valve closing, IO = intake valve
opening, IC = intake valve closing, SI1 = start of the
From a system-theoretic perspective, the cylinder chamber first fuel injection, SC = start of combustion, EO =
model (1)-(5) has the form exhaust valve opening.

481
IFAC AAC 2016
472
June 19-23, 2016. Norrköping, Sweden T. Makowicki et al. / IFAC-PapersOnLine 49-11 (2016) 469–475

Combustion System
 EO      Table 1. Active (•) and inactive (◦) inputs of
u(t) x tk x tIC
k x tSI1
k y(t) the combustion system model (9)-(13).
Phase 1 Phase 2 Phase 3
Gas Combustion Variable Phase 1 Phase 2 Phase 3
Compression
Exchange & v IM (t), v EM (t) • ◦ ◦

SI1
EO

v(t) Stroke yk

IC
Phase Expansion ṁI (t), ṙ O2 (t), ṙ NOx (t), ṙ SM (t) ◦ ◦ •

discrete outputs (17b), phases 1 and 2 are described by

Fig. 5. Structure of the cyclic combustion system model. the continuous-time, physics-based model (9)-(13) which
3.2 Inherent Cyclic Engine Operation and Model Structure easily enables the coupling to a physics-based air system
model (Kotman et al., 2009) or measured air system sig-
The cyclic engine operation, which is assumed to start at nals. During phase 3, the state x(t) and the output y SI1,EO
k+1
exhaust valve opening (EO), see Fig. 4, depends on the are described according to the discrete-time, start-to-end
air system impact during the gas exchange phase (blue mapping (16c), i.e. an algebraic calculation. The required
section) and remains unaffected by external influences data-based models are discussed in the following Subsec-
during the compression phase (gray section) after closing tion 3.3.
the valves. During the combustion & expansion phase
(yellow section), the initial cylinder charge state x(tSI1k ) 3.3 Approximation of the Combustion & Expansion Phase
and the fuel injection parameters influence the combustion
characteristics. This periodic sequence of the phases can
be set up as a series connection of respective submodels. Gaussian Process Regression (GPR) (Rasmussen and
For the proposed model three phases are considered as Williams, 2006; ETAS GmbH, 2015) in combination with
depicted in Fig. 5. Thereby, the currently active submodel simple physics relations are utilized to approximate (16c)
depends on the time events tlk , l ∈ {EO, IC, SI1}, which and the output y SI1,EO
k+1 (17b). Thus, the complex phenom-
correspond to exhaust valve opening EO, intake valve ena during the combustion are represented by functions
closing IC and the start of the first fuel injection SI1. which describe the state x(tEO
k+1 ) at the end of the combus-
tion & expansion in dependence of the state x(tSI1 k ) after
The relation between the cyclic operation, the continuous- the compression, the fuel injection parameters uk (10b)
time evolution of the model (9)-(12) and the cycle-by-cycle as well as the engine speed nE and the fuel pressure pr .
outputs (13) can be described by the flow (Palis and Melo, Therefore, one obtains
1982) of the ODEs (9). Thus, during the k th engine cycle,    
the evolution of the state x(t) is generally described by m(tEO
k+1 ) m(tSI1
k ) + mk
IΣ

 t x(tEO  O2 (tEO
k+1 ) = m
  O2 (tSI1
k+1 ) ≈ m k )−µ
O2 IΣ 
(18)
EO   mk
x(t) = x(tk ) + f (·) dt =: Φt,u,v (x(tEO
k )) (14) EO
p(tk+1 ) Mp SI1,EO
γ
tEO  
k
SI1 T
with γ = x(tSI1
k )
T
uT SI1
k pr (tk ) nE (tk ) . (19)
with t ∈ (tEO
k , tk+1 ]. Accordingly, the state trajectory x(t)
EO

is determined based on the initial state xk = k ) of the

ˆ x(tEO Thus, the cylinder mass m(tSI1 k ) (18) is increased by
k cycle and the time-varying signals u(t) (10a) and v(t)
th the injected fuel quantity mIΣ k whereas the oxygen
(11). Further, the mapping Φ(·) relates the initial state xk mass mO2 (tEO k+1 ) (18) depends on the oxygen demand of
to the final state the burned fuel mass mIΣ k assuming a complete combus-
xk+1 =ˆ x(tEO
k+1 ) = ΦtEO ,u,v (xk ) (15) tion. Since the cylinder pressure p(tEO k+1 ) (18) depends
on the complex heat release of the combustion, a GPR
k+1

and is thereby defining a cycle-by-cycle relation in accor- model MpSI1,EO (γ) with a squared exponential covariance
dance to (13). With respect to the previously mentioned
hybrid model structure of physics- and data-based parts it function is derived which considers the inputs γ (19) as
is worth noting that the time-varying signals u(t) and v(t) discussed above. In the same way, the outputs y SI1,EO k+1
as wells as the reaction rates in (2), (7)-(8) are only of (17b) are set up by GPR models according to
relevant during certain phases of the engine cyclic, see  T
Table 1. Thus, based on (14)-(15), the engine cycle model y SI1,EO
k+1 = pSI1,EO
mi E SI1,EO
NOx E SI1,EO
SM (20a)
proposed according to Fig. 5 is expressed as  T
SI1,EO SI1,EO
x(tIC EO
(16a) ≈ MpSI1,EO (γ) MNO (γ) MSM (γ) . (20b)
k ) = ΦtIC ,v (xk = x(tk )),
k
mi x

x(tSI1 IC
k ) = ΦtSI1 (x(tk )), (16b) O2
k XIM−2 O2
XIM−1
xk+1 = x(tEO
k+1 ) = ΦtEO ,u (x(tSI1
k )). (16c) pIM
k+1

The cycle-by-cycle outputs y k+1 (13) are handled equally, TIM−1 TIM−2
i.e. the integrals (6)-(8) are calculated for each submodel as p
ESM
y k+1 = g(Φt,u,v (xk )), (17a) O2
TEM XEM ENOx
= y EO,IC
k+1 + y IC,SI1 SI1,EO
k+1 + y k+1 . (17b)
pEM
The integral limits in (6)-(8) are set according to the
time events tlk depicted in Fig. 5. The proposed par-
titioning allows the application of modeling approaches
tailored for each phase. In order to determine the cycle- Fig. 6. Sensor locations in the intake and exhaust manifold.

482
 IFAC AAC 2016
June 19-23, 2016. Norrköping, Sweden T. Makowicki et al. / IFAC-PapersOnLine 49-11 (2016) 469–475 473

Table 2. Range of the parameter variation.

Range Range
nE 1500 RPM − 3000 RPM ϕSI1 setpoint ± 6◦ CA
mIΣ 5 mg − 35 mg ϕSI2 setpoint ± 6◦ CA
pIM setpoint ± 0.2 bar pr setpoint ± 300 bar
O2
XIM setpoint − 5% to fresh air mI1 1 mg − 3.5 mg

Reference data which is required for the above mentioned

data-based modeling approaches is derived from test bench
measurements as discussed in Section 4.1.
4. MODEL VERIFICATION
In this section, the combustion cycle model discussed in
Section 3 is parametrized and evaluated using test bench
measurements. Subsection 4.1 describes the test setup and
measurement strategies for stationary and transient engine
operation. Subsection 4.2 evaluates the physics-based sub-
models, i.e. gas exchange and compression phase, whereas
Subsection 4.3 focuses on the data-based approximation of
the combustion. In Subsection 4.4, the predicted cycle-by-
cycle outputs y k+1 are evaluated for input data represent-
ing transient engine operation.
4.1 Setup for Stationary and Transient Engine Operation
Measurements are done on a two-cylinder diesel engine
(1.2l) with a conventional combustion mode and the topol-
Fig. 8. Correlation of the simulated / measured cylinder
pressure p(tSI1
k ) at end of phase 2 (left plot) and the
IMEP of phase 1 and 2 (right plot); red dashed line:
relative error of 10%
4.2 Gas Exchange and Compression Phase
The upper plot of Fig. 7 shows the simulated and measured
cylinder pressure p(t) during two consecutive engine cycles,
whereas the phases are numbered according to Fig. 5. Due
to the approximation of the combustion no simulation re-
sults are plotted during phase . The lower two plots show
zooms of the gas exchange phase  and the compression
phase . During both phases, the simulated and measured
pressure fits well, whereas the deviation at IO/EC in
the lower left plot (at t = 0.105s) can be explained by
slightly inaccurate valve lift curves. Further, in the left
plot of Fig. 8, the predicted end-pressure p(tSI1
k ) of the

ogy depicted in Fig. 1. Sensor positions in the intake

and exhaust manifold are shown in Fig. 6. Redundant
measurements are averaged. Based on this test setup the
parameters listed in Table 2 are varied independently
following a space-filling test plan (Siebertz et al., 2010),
whereat the setpoints correspond to the standard appli-
cation of the engine. The 8-dimensional input space is
covered by 955 variations at stationary engine operation.
This data set represents the calibration and test data for
the hybrid combustion cycle model (16)-(20b). Transient
engine operation is tested via fuel mass steps with various
heights at constant engine speed.

Fig. 7. Comparison of the simulated and measured cylinder Fig. 9. Correlation of the predicted cylinder pres-
pressure p(t) during two consecutive engine cycles; sure p(tEO SI1,EO
k+1 ) and the outputs y k+1 with measured
phase 3 is described by discrete-time mappings, see references; red dashed line: relative error of 10%;
Subsection 3.3; nE = 2500 RPM, mIΣ k = 15 mg. FSN = ˆ Filter Smoke Number.

483
IFAC AAC 2016
474
June 19-23, 2016. Norrköping, Sweden T. Makowicki et al. / IFAC-PapersOnLine 49-11 (2016) 469–475

Fig. 10. Model test via a sequence of fuel mass steps at constant engine speed showing the transient response of the
measured and modeled outputs. The region marked by the dashed lines is depicted in detail in the plots of Fig. 11.

Fig. 11. Zoom-in of the region marked by the red dashed lines in the plots of Fig. 10

compression phase  is correlated with the measured one
for all data points. The red dashed line indicates a relative
error of ±10%. At high pressure levels the error increases,
whereas the measured pressure is persistently smaller than
the predicted one. This effect results of the increasing
heat loss through the cylinder walls which is neglected in
the pressure ODE (5). Nevertheless, the model quality is
proofed by the low RMSE (root mean squared error) and
the high coefficient of determination R2 . Further, in the
right plot of Fig. 8, the predicted IMEP pmi of phase 1
and 2 is compared with the measured reference. The model
accuracy is high as indicated by the relative error bound.
4.3 Combustion and Expansion Phase
The plots depicted in Fig. 9 compare the predicted cycle-
by-cycle outputs y SI1,EO
k+1 and the cylinder pressure p(tEO
k+1 )
at exhaust valve opening with respective reference mea-
surements. The left plots visualize the training data error
whereas the right ones show test data results. Thereby,
the cylinder pressure p(tEO
k+1 ) at exhaust valve opening is
estimated very well for the training as well as the test
data. Concerning the IMEP pSI1,EO mi , the relative error
stays within the range of ±10% for training and test data.
Although the test data RMSE is slightly higher, the model
accuracy is very high. Focusing on the nitrogen oxide
emissions ENOx , only the training data remains within
the limit of the relative error. Although the test data
has an increased RMSE, the model quality is good as
indicated by the R2 measure. In case of the smoke ESM , the
relative error of the training and test data exceeds ±10%.
The measurements are obviously affected by an exogenous
impact which is not considered by the model inputs, e.g.
the in-cylinder mixing dynamics. However, the trend of the
smoke emissions is predicted well as indicated by the R2
measure.
4.4 Model Test at Transient Engine Operation
The plots in Fig. 10 and Fig. 11 show the performance of
the model for input data which represents transient engine
operation via a sequence of fuel mass steps. The overall
sequence is visualized in the plots of Fig. 10 whereas the
region marked by the red dashed lines is shown in detail
in the plots of Fig. 11. As depicted in the upper left plot
of Fig. 10, the fuel mass is varied step-wise at constant

484
 IFAC AAC 2016
June 19-23, 2016. Norrköping, Sweden T. Makowicki et al. / IFAC-PapersOnLine 49-11 (2016) 469–475 475
engine speed. The upper middle plot shows the oxygen
mass fraction and the pressure in the intake manifold,
whereas the applied main and pilot injection position as
well as the rail pressure are depicted in the upper right
plot. The lower plots show the cycle-by-cycle outputs (13)
compared with measured references. The IMEP pmi is
predicted well during stationary and transient sections as
indicated by the absolute error. The predicted trajectory
of ENOx deviates slightly from the measured profile during
transients. This is explained by low-pass characteristic of
the measurement apparatus (MEXA-7100D), see t = 85s
in Fig. 11. Nevertheless, the oscillation at t = 77s are
predicted well. The deviation of ENOx at the beginning
and the end of the sequence reflects the validity range of
the model, since the overall fuel mIΣ k is near the lower
limit of permissible input space, see Table 2. Further, the
smoke emissions ESM fit well with the measured reference 1
whereas the deviations e.g. at t = 110s are as expected due
to the training and test data error range.
5. CONCLUSION & OUTLOOK
The engine model proposed in this paper predicts the
overall IMEP pmi , the nitrogen oxides ENOx , and the
smoke ESM on a cycle-by-cycle basis. The model scheme
is oriented on the combustion cycle and has a hybrid
structure with the gas-exchange and compression phases
described by physics-based and the combustion phase
modeled by a data-based approach (Gaussian Process
Regression). All in all, the IMEP and the nitrogen oxides
are predicted well at stationary as well as transient engine
operation. The smoke model shows a higher error and
requires further investigation e.g. an analysis of exogenous
impacts which have to be considered as an additional
model input to improve the quality. Further topics of
interest are the coupling of the cylinder model with an
adequate air system description, its real time capability
and the utilization of the model scheme in the context
of engine control, e.g. by means of optimization-based
approaches (Asprion et al., 2014).
REFERENCES
Asprion, J., Chinellato, O., and Guzzella, L. (2014).
Optimal Control of Diesel Engines: Numerical Meth-
ods, Applications, and Experimental Validation. Math-
ematical Problems in Engineering, 21 pages, doi:
10.1155/2014/286538.
Berger, B., Rauscher, F., and Lohmann, B. (2011).
Analysing Gaussian Processes for Stationary Black-Box
Combustion Engine Modelling. In Proceedings of the
18th IFAC World Congress, 10633–10640. Milano, Italy.
ETAS GmbH (2015). ASCMO V4.5. URL http://www.
etas.com/de/products/ascmo.php.
Grahn, M., Johansson, K., and McKelvey, T. (2014).
Model-based diesel Engine Management System opti-
mization for transient engine operation. Control Engi-
neering Practice, vol. 29, 103–114.
1 As the smoke concentration for model calibration is determined
by an AVL Smokemeter (FSN - Filter Smoke Number) and transient
smoke emissions are measured by an AVL Opacimeter (opacity), the
predicted smoke concentration is converted into opacity.
485
Guzzella, L. and Onder, C. (2004). Introduction to Mod-
eling and Control of Internal Combustion Engine Sys-
tems. Springer, Berlin and Heidelberg.
Heywood, J.B. (1988). Internal Combustion Engine Fun-
damentals. McGraw-Hill series in mechanical engineer-
ing. McGraw-Hill, New York.
Hillion, M., Buhlbuck, H., Chauvin, J., and Petit, N.
(2009). Combustion Control of Diesel Engines Using
Injection Timing. SAE Technical Paper, 2009–01–0367.
Isermann, R. (2014). Engine Modeling and Control: Model-
ing and Electronic Management of Internal Combustion
Engines. Springer, Berlin and Heidelberg.
Kotman, P., Bitzer, M., Graichen, K., and Kugi, A. (2009).
Hybrid Modeling of a Two-Stage Turbocharged Diesel
Engine Air System. In Proceedings of the 6th Vi-
enna Symposium on Mathematical Modeling, 2015–2024.
Wien, Austria.
Meddahi, F., Fleck, C., Grodde, S., Charlet, A., and
Chamaillard, Y. (2015). Modeling Waves in ICE Ducts:
Comparison of 1D and Low Order Models. SAE Tech-
nical Paper, 2015–24–2386.
Mollenhauer, K. and Tschoke, H. (2010). Handbook of
Diesel Engines. Springer, Berlin and London.
Mrosek, M., Sequenz, H., and Isermann, R. (2010). Control
Oriented NOx and Soot Models for Diesel Engines. In
Proceedings of the 6th IFAC Symposium Advances in
Automotive Control, 234–239. Schwabing, Germany.
Nakayama, S., Ibuki, T., Hosaki, H., and Tominaga, H.
(2008). An Application of Model Based Combustion
Control to Transient Cycle-by-Cycle Diesel Combustion.
SAE Technical Paper, 2008–01–1311.
Palis, J. and Melo, W. (1982). Geometric Theory of
Dynamical Systems. Springer, New York and Heidelberg
and Berlin.
Pischinger, R., Klell, M., and Sams, T. (2009). Thermo-
dynamik der Verbrennungskraftmaschine. Der Fahrzeu-
gantrieb. Springer, Wien and New York, 3rd edition.
Rasmussen, C. and Williams, C. (2006). Gaussian Pro-
cesses for Machine Learning. Adaptive computation and
machine learning. MIT Press, Cambridge and Mass.
Rowland, B. (1986). The Thermodynamics and Gas
Dynamics of Internal Combustion Engines. Clarendon,
Oxford.
Siebertz, K., van Bebber, D., and Hochkirchen, T. (2010).
Statistische Versuchsplanung. Springer, Berlin and Hei-
delberg.
Warnatz, J., Maas, U., and Dibble, R. (2006). Com-
bustion: Physical and Chemical Fundamentals, Model-
ing and Simulation, Experiments, Pollutant Formation.
Springer, Berlin and New York, 4th edition.