Combustion Modeling ForDiesel Engine Control

Combustion Modeling for
Diesel Engine Control Design

Von der Fakultat f ur Maschinenwesen der
Rheinisch-Westfalischen Technischen Hochschule Aachen
zur Erlangung des akademischen Grades eines Doktors der
Ingenieurwissenschaften genehmigte Dissertation
vorgelegt von
Diplom-Ingenieur
Christian Felsch
aus Neuss
Berichter: Univ.-Prof. Dr.-Ing. Dr. h.c. Dr.-Ing. E.h. N. Peters
Univ.-Prof. Dr.-Ing. D. Abel
Tag der m undlichen Pr ufung: 27. August 2009
Diese Dissertation ist auf den Internetseiten
der Hochschulbibliothek online verf ugbar.
Shaker Verlag
Aachen 2009
Berichte aus der Energietechnik
Christian Felsch
WICHTIG: D 82 berprfen !!!
Bibliographic information published by the Deutsche Nationalbibliothek
The Deutsche Nationalbibliothek lists this publication in the Deutsche
Nationalbibliografie; detailed bibliographic data are available in the Internet at
http://dnb.d-nb.de.
Zugl.: D 82 (Diss. RWTH Aachen University, 2009)
Copyright Shaker Verlag 2009
All rights reserved. No part of this publication may be reproduced, stored in a
retrieval system, or transmitted, in any form or by any means, electronic,
mechanical, photocopying, recording or otherwise, without the prior permission
of the publishers.
Printed in Germany.
ISBN 978-3-8322-8545-6
ISSN 0945-0726
Shaker Verlag GmbH P.O. BOX 101818 D-52018 Aachen
Phone: 0049/2407/9596-0 Telefax: 0049/2407/9596-9
Internet: www.shaker.de e-mail: info@shaker.de
Vorwort/Preface
Diese Arbeit entstand wahrend meiner Tatigkeit als wissenschaftlicher Mitarbeiter
am Institut f ur Technische Verbrennung (ehemals Institut f ur Technische Mechanik)
der RWTH Aachen University. Sie wurde im Rahmen des Teilprojekts B1 Mo-
dellreduktion bei Niedertemperatur-Brennverfahren durch CFD-Simulationen und
Mehrzonen-Modelle im Sonderforschungsbereich 686 Modellbasierte Regelung der
homogenisierten Niedertemperatur-Verbrennung durchgef uhrt, der an der RWTH
Aachen University und der Universitat Bielefeld bearbeitet wird. Der Deutschen
Forschungsgemeinschaft danke ich in diesem Zusammenhang f ur die nanzielle
Forderung.
Mein besonderer Dank gilt Herrn Professor Dr.-Ing. Dr. h.c. Dr.-Ing. E.h. Norbert
Peters f ur seine Unterst utzung, seine kritischen Anmerkungen und die mir gewahrte
Freiheit. Herrn Professor Dr.-Ing. Dirk Abel danke ich f ur sein stetiges Interesse
an meiner Arbeit und f ur die Tatigkeit als weiterer Berichter. Herrn Professor
Dr.-Ing. (USA) Stefan Pischinger danke ich f ur die

Ubernahme des Vorsitzes der
Pr ufungskommission.
Ausdr ucklich mochte ich mich bei Jeremy Weckering, Vivak Luckhchoura, Michael
Gauding, Bernhard Jochim, Bruno Kerschgens, Christoph Glawe, Johannes Kepp-
ner, Hanno Friederichs, Benedikt Sonntag und Abhinav Sharma bedanken, die
mich durch ihre Studien- oder Diplomarbeiten bzw. durch ihre Arbeit als wissen-
schaftliche Hilfskrafte tatkraftig unterst utzt haben. Auerdem gilt mein Dank
allen jetzigen und ehemaligen Mitarbeitern des Instituts f ur Technische Verbren-
nung, die ebenfalls zum Gelingen dieser Arbeit beigetragen haben. Ich bedanke
mich bei Christian Hasse, Stefan Vogel, Jost Weber, Frank Freikamp, Jens Ewald,
Peter Spiekermann, Sylvie Honnet, Elmar Riesmeier, Christoph Kortschik, Sven
Jerzembeck, Rainer Dahms, Anyelo Vanegas, Olaf Rohl, Peng Zeng, Joachim
Beeckmann, Hyun Woo Won, Klaus-Dieter Stoehr, Jens Henrik Gobbert, Tomoya
Wada, Lipo Wang, Stefanie Bordihn, Sonja Engels, Yvonne Lichtenfeld, Dieter
Ostho, Bernd Binninger und G unter Paczko.
Dar uber hinaus bedanke ich mich sehr bei Peter Drews und vor allem bei Kai
Homann vom Institut f ur Regelungstechnik der RWTH Aachen University f ur
die hervorragende Zusammenarbeit im Sonderforschungsbereich 686.
iii
I would like to thank Andreas Lippert for making my research stays at General
Motors R&D in Warren, MI, USA, possible. Furthermore, I would like thank
Tom Sloane, Jaehoon Han, Mark Huebler, Brian Peterson, Vinod Natarajan, and
Carl-Anders Hergart for our good collaboration there. Thanks are also due to
Nicole Wermuth of General Motors R&D for providing the experimental data on
the HCCI engine. My most sincere thanks go to Hardo Barths of General Motors
Powertrain, who provided me with valuable advice and ideas in numerous helpful
discussions.
Bei meinen Freunden Tobias Krockel, Torsten Moll, Simon Frischemeier, Adam
Gacka und Martin Ross bedanke ich mich f ur unseren gemeinsamen Spa am
Maschinenbau.
Ein groer Dank gilt meinen Eltern, die mich in jeder Phase meiner Ausbildung
unterst utzt und motiviert haben.
Meiner Freundin Kathrin Borggrebe danke ich ebenfalls sehr f ur ihre aueror-
dentliche Unterst utzung, auf die ich mich immer verlassen konnte.
Aachen, im August 2009
Christian Felsch
iv
Zusammenfassung
Gegenstand dieser Arbeit ist zunachst die Entwicklung eines konsistenten Mi-
schungsmodells f ur die interaktive Kopplung eines CFD-Codes und eines auf
mehreren nulldimensionalen Reaktoren basierenden Mehrzonenmodells. Das in-
teraktiv gekoppelte Modell ermoglicht eine rechenzeit-eziente Modellierung von
HCCI- und PCCI-Verbrennung. Der physikalische Bereich im CFD-Code wird mit-
tels dreier Phasenvariablen (Mischungsbruch, Verd unnung und totale Enthalpie)
in mehrere Zonen unterteilt. Die Phasenvariablen reichen aus, um den ther-
modynamischen Zustand jeder Zone zu denieren, da diese denselben Druck
aufweisen. Jede Zone im CFD-Code wird durch eine korrespondierende Zone im
nulldimensionalen Code abgebildet. Das Mehrzonenmodell lost die Chemie f ur
jede Zone, und die Warmefreisetzung wird zum CFD-Code zur uckgef uhrt. Die
Schwierigkeit bei dieser Methodik liegt darin, den thermodynamischen Zustand
jeder Zone zwischen CFD-Code und nulldimensionalem Code konsistent zu hal-
ten, nachdem die Initialisierung der Zonen im Mehrzonenmodell stattgefunden
hat. Der thermodynamische Zustand jeder Zone (und daher auch die Phasen-
variablen) verandert sich mit der Zeit aufgrund von Mischung und Quelltermen
(z.B. Verdampfung des Brennstos, Wandwarmetransfer). Der Fokus dieser Ar-
beit liegt auf einer einheitlichen Beschreibung der Mischung zwischen den Zonen
im Phasenraum des nulldimensionalen Codes, basierend auf der Losung des CFD-
Codes. Zwei Mischungsmodelle mit unterschiedlichen Stufen von Genauigkeit,
Komplexitat und numerischem Aufwand werden beschrieben. Das am besten
ausgearbeitete Mischungsmodell (sowie eine angemessene Behandlung der Quell-
terme) halt den thermodynamischen Zustand der Zonen im CFD-Code und im
nulldimensionalen Code identisch. Die Modelle werden auf einen Testfall der
HCCI-Verbrennung in einem Benzinmotor angewandt. Von dort ausgehend wird
ein Simulationsmodell f ur PCCI-Verbrennung erstellt, welches f ur die Entwicklung
geschlossener Regelkreise verwendet werden kann. F ur den Hochdruckteil des Mo-
torzyklus wird das interaktiv gekoppelte CFD-Mehrzonenmodell systematisch zu
einem eigenstandigen Mehrzonenmodell reduziert. Das eigenstandige Mehrzonen-
modell wird um ein Mittelwertmodell erganzt, welches die Ladungswechselverluste
ber ucksichtigt. Das resultierende Modell ist in der Lage, PCCI-Verbrennung mit
stationarer Genauigkeit zu beschreiben und ist gleichzeitig sehr ezient bez uglich
der benotigten Rechenzeit. Das Modell wird weiterhin um eine identizierte Sys-
temdynamik erweitert, welche die stationaren Stellgroen beeinusst. Zu diesem
v
Zweck wird ein Wiener-Modell erstellt, welches das stationare Modell als eine
nichtlineare Systemabbildung verwendet. Auf diese Weise wird ein dynamisches
nichtlineares Modell zur Abbildung der Regelstrecke Dieselmotor entworfen.
vi
Abstract
This thesis deals at rst with the development of a consistent mixing model for
the interactive coupling (two-way-coupling) of a CFD code and a multi-zone code
based on multiple zero-dimensional reactors. The interactive coupling allows for
a computationally ecient modeling of HCCI or PCCI combustion, respectively.
The physical domain in the CFD code is subdivided into multiple zones based on
three phase variables (fuel mixture fraction, dilution, and total enthalpy). These
phase variables are sucient for the description of the thermodynamic state of
each zone, assuming that each zone is at the same pressure. Each zone in the CFD
code is represented by a corresponding one in the zero-dimensional code. The
zero-dimensional code solves the chemistry for each zone, and the heat release is
fed back into the CFD code. The diculty in facing this kind of methodology is
to keep the thermodynamic state of each zone consistent between the CFD code
and the zero-dimensional code after the initialization of the zones in the multi-
zone code has taken place. The thermodynamic state of each zone (and thereby
the phase variables) changes in time due to mixing and source terms (e.g., va-
porization of fuel, wall heat transfer). The focus of this work lies on a consistent
description of the mixing between the zones in phase space in the zero-dimensional
code, based on the solution of the CFD code. Two mixing models with dierent
degrees of accuracy, complexity, and numerical eort are described. The most
elaborate mixing model (and an appropriate treatment of the source terms) keeps
the thermodynamic state of the zones in the CFD code and the zero-dimensional
code identical. The models are applied to a test case of HCCI combustion in a
gasoline single-cylinder research engine. Following from there, a simulation model
for PCCI combustion is derived that can be used in closed-loop control develop-
ment. For the high-pressure part of the engine cycle, the interactively coupled
CFD-multi-zone approach is systematically reduced to a stand-alone multi-zone
chemistry model. This multi-zone chemistry model is extended by a mean value
model accounting for the gas exchange losses. The resulting model is capable of
describing PCCI combustion with stationary exactness, and is at the same time
very economic with respect to computational costs. The model is further extended
by identied system dynamics inuencing the stationary inputs. For this purpose,
a Wiener model is set up that uses the stationary model as a nonlinear system
representation. In this way, a dynamic nonlinear model for the representation of
the controlled plant Diesel engine is created.
vii
Publications
This thesis is mainly based on the following publications; some material is updated,
together with some new introduced results:
H. Barths, C. Felsch, and N. Peters. Mixing models for the two-way-coupling
of CFD codes and zero-dimensional multi-zone codes to model HCCI com-
bustion. Combust. Flame, 156:130-139, 2009.
C. Felsch, R. Dahms, B. Glodde, S. Vogel, S. Jerzembeck, N. Peters, H.
Barths, T. Sloane, N. Wermuth, and A. M. Lippert. An Interactively Cou-
pled CFD-Multi-Zone Approach to Model HCCI Combustion. Flow, Turbu-
lence and Combustion, 82:621-641, 2009.
C. Felsch, K. Homann, A. Vanegas, P. Drews, H. Barths, D. Abel, and
N. Peters. Combustion model reduction for Diesel engine control design.
International Journal of Engine Research, 2009, accepted for publication.
viii
Contents
Title i
Vorwort/Preface iii
Zusammenfassung v
Abstract vii
Publications viii
Contents ix
1 Introduction 1
2 Multi-Zone Chemistry Model 5
2.1 Multi-Zone Modeling Overview . . . . . . . . . . . . . . . . . . . . 5
2.2 Multi-Zone Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Chemistry Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3 Modeling of Chemically Reacting Turbulent Two-Phase Flows 9
3.1 Governing Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2 Scales of Turbulent Motion . . . . . . . . . . . . . . . . . . . . . . . 12
3.3 Averaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.4 Turbulent Flow and Mixing Field . . . . . . . . . . . . . . . . . . . 14
3.5 CFD Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.6 Liquid Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4 Interactive Coupling of the CFD Code and the Multi-Zone Model 21
4.1 Initialization of the Interactive Coupling . . . . . . . . . . . . . . . 21
4.2 Mixing Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.2.1 Neighboring Zone Mixing . . . . . . . . . . . . . . . . . . . 22
4.2.2 The Global Mass Exchange Rate . . . . . . . . . . . . . . . 24
4.2.3 The Individual Mass Exchange Rate . . . . . . . . . . . . . 24
4.3 Consistency of Heat Release . . . . . . . . . . . . . . . . . . . . . . 25
4.4 Treatment of Source Terms . . . . . . . . . . . . . . . . . . . . . . . 26
ix
Contents
4.5 Consistent Modeling of Source Terms . . . . . . . . . . . . . . . . . 26
4.6 Concept Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5 Validation of the Interactive Coupling 29
5.1 Experimental and Numerical Setup . . . . . . . . . . . . . . . . . . 29
5.1.1 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . 29
5.1.2 Numerical Setup . . . . . . . . . . . . . . . . . . . . . . . . 30
5.2 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.2.1 Sensitivity Study . . . . . . . . . . . . . . . . . . . . . . . . 32
5.2.2 Importance of Incorporating Mixing . . . . . . . . . . . . . . 34
5.3 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 42
6 Systematic Reduction towards a Stand-alone Multi-Zone Model 43
6.1 Extended Interactive Coupling for PCCI Combustion . . . . . . . . 44
6.2 Reduction Methodology . . . . . . . . . . . . . . . . . . . . . . . . 46
6.2.1 Heat Transfer to the Walls . . . . . . . . . . . . . . . . . . . 46
6.2.2 Fuel Injection and Evaporation . . . . . . . . . . . . . . . . 46
6.2.3 Simplied Mixing Model . . . . . . . . . . . . . . . . . . . . 46
6.2.4 Further Reduction Potential . . . . . . . . . . . . . . . . . . 47
6.3 Computational Accuracy and Eciency . . . . . . . . . . . . . . . . 47
6.3.1 Interactively Coupled CFD-Multi-Zone Approach . . . . . . 47
6.3.2 Stand-alone Multi-Zone Model . . . . . . . . . . . . . . . . . 48
6.3.3 Run Time Comparison . . . . . . . . . . . . . . . . . . . . . 48
7 Validation of the Systematic Reduction 51
7.1 Experimental and Numerical Setup . . . . . . . . . . . . . . . . . . 51
7.1.1 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . 51
7.1.2 Numerical Setup . . . . . . . . . . . . . . . . . . . . . . . . 53
7.2 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . 55
7.2.1 Interactively Coupled CFD-Multi-Zone Approach . . . . . . 56
7.2.2 Stand-alone Multi-Zone Model . . . . . . . . . . . . . . . . . 60
7.3 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 66
8 Gas Exchange 67
9 System Identication 71
9.1 Wiener Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
9.2 Step Response Experiments . . . . . . . . . . . . . . . . . . . . . . 72
9.3 Dynamic Transfer Functions . . . . . . . . . . . . . . . . . . . . . . 73
10 Integrated Model 79
10.1 Suitable Test Environment for Control Applications . . . . . . . . . 79
x
Contents
10.2 Transient Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . 80
10.3 Potential within Closed-Loop Control Development . . . . . . . . . 81
11 Summary and Conclusion 85
12 Future Work 87
Bibliography 89
Denitions, Acronyms, Abbreviations 99
Lebenslauf 101
xi
1 Introduction
There is a strong demand to explore new combustion concepts capable of meet-
ing the stringent future emission standards, such as Tier 2 Bin5 on the North
American market or EURO VI on the European market. Due to its potential
to simultaneously achieve both high eciency and low pollutant engine-out emis-
sions, many engine researchers have therefore investigated Homogeneous Charge
Compression Ignition (HCCI) in the recent past. From a modeling perspective, it
ideally represents a zero-dimensional auto-ignition problem that could be solved in
only one zone, because of its homogeneity. In reality, the temperature is stratied
in the engine due to wall heat transfer. Depending on the operating strategy of
HCCI engines, the dilution level (Exhaust Gas Recirculation, EGR, internal or
external), as well as the fuel/air equivalence ratio, is stratied (e.g., due to direct
injection), too. In addition, since the ow in engines is turbulent, the turbulent
uctuations of the quantities mentioned above (temperature, dilution, and fuel/air
equivalence ratio) have to be accounted for in order to achieve a precise descrip-
tion of the HCCI combustion process. Thus, a closure problem for the chemical
source terms in the coupled partial dierential equations describing this process
is encountered, similar to the closure problem for conventional spark-ignited and
Diesel combustion systems [66].
Several models for the description of HCCI combustion with varying degrees of
accuracy and complexity have been proposed. References [1, 14, 15, 21, 31, 34, 36,
39, 45, 57, 60, 72, 76, 97, 103, 105] give a good overview of the modeling work in
this eld up-to-date. The overall progress and recent trends in HCCI engines are
discussed in the work by Yao et al. [104].
In the rst part of this thesis, an interactively coupled CFD-multi-zone approach
for HCCI combustion is developed that is precise enough for the description of the
chemistry, but is at the same time economical enough to allow application in
an industrial environment. The basic idea is similar to that of a Representative
Interactive Flamelet model or a PDF method, where the chemistry is solved in
phase space instead of the physical space in the CFD code. In the PDF method,
for example, each particle in phase space can be considered as a single zone. The
concept analogies and dierences with respect to the Representative Interactive
Flamelet model are discussed at length.
It is assumed that the transport of all scalars (e.g., species mass fractions, to-
tal enthalpy) can be represented by a smaller set of phase variables (fuel mixture
1
1 Introduction
fraction, EGR mass fraction, and total enthalpy). The distribution function of the
phase variables is subdivided into discrete zones in phase space, and the chem-
istry is solved in these zones. Because the focus lies on the interactive coupling
(two-way-coupling) between a CFD code and a zero-dimensional code, the mass
weighted distribution function is obtained, for reasons of simplicity, by assuming
that the variance in each computational cell is negligible and the local PDF is a
delta function. The error introduced by this simplifying assumption depends on
the homogeneity of the charge and is small for near homogeneous cases.
The coupling of a CFD code to a zero-dimensional multi-zone code poses two
major challenges to keep the thermodynamic states in both codes consistent:
1. the description of the mass and energy exchange between the zones, and
2. the treatment of the source terms (e.g., vaporization of fuel, wall heat trans-
fer).
The emphasis is on the modeling of the interactive coupling of the codes and, in
particular, the proper modeling of the exchange between zones. The exchange be-
tween the zones is based on the three-dimensional fuel mixture fraction distribution
in the CFD domain. Two exchange models are discussed:
1. an exchange model using a least square t to approximate the exchange
between zones, and
2. an exchange model that requires the solution of a system of algebraic equa-
tions equal to the number of zones, but is exact (i.e., it keeps the value of
the phase variables and the thermodynamic state identical for each corre-
sponding zone in both codes).
The rst part of this thesis closes with the validation of the interactive cou-
pling against a test case of HCCI combustion in a gasoline single-cylinder research
engine.
Premixed Charge Compression Ignition (PCCI), or Premixed Compression
Ignition (PCI), as it is sometimes referred to, has emerged as another inter-
esting alternative to conventional Diesel combustion in the part-load operating
range [37, 47, 90]. PCCI combustion is conceptually similar to HCCI combus-
tion. It involves relatively early injection timings, high external EGR rates, and
cooled intake air, leading to a low-temperature combustion (LTC) process with
low NO
x
and particulate emissions. A general review of LTC processes, including
both HCCI and PCCI combustion, can be found in [17]. However, in contrast to
conventional Diesel combustion, with early direct-injection LTC it can be dicult
to prevent combustion from occuring before top dead center (TDC), which often
2
increases noise and reduces engine eciency. Sophisticated closed-loop control
of the combustion process is one means to overcome this diculty. Against this
background, the thesis further deals with a rst step towards the development of
a controller for the PCCI combustion process.
The most important characteristics for the operation of an internal combustion
engine are the engines load and its combustion eciency. The former is directly
dependent on the indicated mean eective pressure (IMEP), the latter can be
characterized by the centre of combustion, where 50 % of the total injected fuel
mass is burned (CA50). The engines load is set by the driver in an automobile
application. For CA50, a set point can be determined which is dependent on the
operating point and gives the best combustion eciency. These two variables are
the main focus of the controller to be developed. For inuencing the process, the
start of injection (SOI), the external EGR rate, and the total fuel mass injected
are suitable actors.
In the recent past, several eorts have been reported in the literature that aim at
controlling engine combustion. Among these, [9, 42, 43, 56, 58, 87, 94] can be men-
tioned. The standard procedure for creating a controller includes the modeling part
as the rst step [4, 10, 57, 86, 88]. Often these models dier in several aspects from
models widely used for gathering a deeper understanding of combustion details,
like three-dimensional computational uid dynamics (CFD) models [8, 13, 71, 77].
From the viewpoint of automatic control, the dynamics describing the dependency
of the systems outputs (IMEP, CA50) on the actors (SOI, external EGR rate, and
total fuel mass injected) is of highest priority. Nevertheless, stationary exactness
of the model is important, too. Another requirement is an acceptable calculation
speed, as it is usually applied in dynamic closed-loop simulations.
The second part of this thesis presents a new approach to the development of a
simulation model for the use in closed-loop control development. The interactively
coupled CFD-multi-zone approach formulated in the rst part is reduced to a
computationally ecient stand-alone multi-zone model. The stand-alone multi-
zone model covers the nonlinear dependencies within the high-pressure part of the
engine cycle with stationary exactness. This model is extended by a physically
inspired description of the gas exchange part of the engine cycle. For the use
in closed-loop simulations, the systems dynamics have to be covered. For this
reason, the stationary model is further extended by identied system dynamics
inuencing the stationary inputs. In this manner, a stationary exact model is
extended to a Wiener-type model with a static part describing the nonlinearities
and an upstream part describing the systems dynamics. This novel procedure
integrates the detailed knowledge from combustion simulation tools into closed-
loop control and establishes a broad eld of possibilities for testing completely new
controlled process variables.
3
1 Introduction
This thesis is arranged as follows: Chapters 2, 3, and 4 deal with the combus-
tion modeling approach employed in the present investigation. The theory and
assumptions underlying the interactive coupling of a CFD code and a multi-zone
model are reviewed. Chapter 5 contains its application to a test case of HCCI
combustion in a gasoline single-cylinder research engine. In the following Chap-
ter 6, the reduction of the three-dimensional CFD model to a computationally
ecient stand-alone multi-zone chemistry model is described. After this, Chap-
ter 7 presents the validation of the reduction against PCCI combustion in a Diesel
engine. Subsequently, the physically inspired description of the gas exchange part
of the engine cycle is put forward in Chapter 8. Afterwards, Chapter 9 identies
the systems dynamics. Following this, the integrated model composed of all three
model parts (stand-alone multi-zone model, gas exchange model, and dynamic
time response) is validated against transient experimental data in Chapter 10.
Finally, the conclusions and major ndings from the thesis are summarized and
an outlook to future work is given. Frequently used denitions, acronyms, and
abbreviations are included at the end of this thesis, following the references.
4
2 Multi-Zone Chemistry Model
2.1 Multi-Zone Modeling Overview
In HCCI or PCCI engines, where substantial premixing occurs, the cylinder charge
may be divided into several distinct regions, each characterized by a certain ther-
modynamic state. A major requirement is for the variance within each individual
zone to be small, since the multi-zone chemistry code solves only for the average
values provided.
Several multi-zone approaches have been proposed in the literature. Aceves et
al. [2, 3] used zones based on the temperature to account for thermal gradients
inside the cylinder of heavy-duty truck engines running in HCCI mode on natural
gas and propane.
Babajimopoulos et al. [6] used equivalence ratio, temperature, and EGR to
dene the zones in investigating various variable valve actuation (VVA) strategies
in a heavy-duty truck engine fueled by natural gas. In all these studies, the zones
were initialized based on three-dimensional CFD calculations of the ow up to a
certain crank angle at which chemical reactions start to occur. The mass of each
zone was primarily governed by the given temperature distribution in the cylinder,
and no mixing occurred between the zones in the multi-zone chemistry code. In
the study by Babajimopoulos et al. [6], each temperature zone was divided into
three equivalence ratio zones of equal mass.
In a more recent publication, Babajimopoulos et al. [5] extended the sequential
approach presented in [6] to a fully coupled computational uid dynamics and
multi-zone model with detailed chemical kinetics. The multi-zone model commu-
nicated with the CFD code at each computational time step, and the composition
of the CFD cells was mapped back and forth between the CFD code and the
multi-zone model. This approach was validated against experimental data in [30]
and [40].
5
2 Multi-Zone Chemistry Model
2.2 Multi-Zone Model
The multi-zone model employed in this work is X0D, a zero-dimensional chemistry
solver based on multiple zero-dimensional reactors. X0D was developed internally
at General Motors R&D by Hardo Barths, Tom Sloane, and Christian Hasse, and
was rst described in Hergart et al. [39] and Felsch et al. [26].
The equations governing species mass fractions, temperature, and pressure
change in the multi-zone chemistry code are given below:
dY
ij
dt

1
m
i
nz
k=1
m
ik
Y
ex
kj
Y
ij

ij

s
ij
i
= 0 , (2.1)
dT
i
dt

1
m
i
c
pi
nz
k=1
m
ik
h
ex
k
h
i
+
1
m
i
c
pi
nsp
j=1
h
ij
nz
k=1
m
ik
Y
ex
kj
Y
ij
+
1
i
c
pi
nsp
j=1
h
ij

ij
i
c
pi
dp
dt

1
i
c
pi
nsp
j=1

s
ij
h
vj
+
1
m
i
c
pi
Q
wall,i
= 0 , (2.2)
and
dp
dt
=
p
V
dV
dt
+
p
V
nz
i=1
V
i
1
m
i
dm
i
dt
+
1
T
i
dT
i
dt
+ W
i
nsp
j=1
1
W
j
dY
ij
dt
. (2.3)
In Eq. (2.1), Y
ij
denotes the mass fraction of species j in zone i, and
ij
is the
corresponding chemical source term.
s
ij
accounts for the source term due to fuel
vaporization. It is zero for all species except the fuel itself (
s
ij
= 0, j = fuel).
The second term on the left-hand side of Eq. (2.1) describes the mass exchange
between the zones, where m
ik
is the rate at which mass is transported between
zones i and k. Thereby, nz stands for the total number of zones. Similarly, in
Eq. (2.2) the second and third term on the left-hand side represent the enthalpy
exchange between zones due to enthalpy stratication between zones and due to
species stratication between zones, respectively. h
vj
denotes the latent heat
of vaporization of species j, and

Q
wall,i
is the wall heat transfer of zone i. In
Eqs. (2.2) and (2.3), nsp accounts for the number of species employed in the
underlying chemical mechanism. The equation of state is used to derive Eq. (2.3),
which solves for the pressure across the zones. Through this equation, all zones
are thermodynamically coupled with each other.
6
2.3 Chemistry Model
Wall heat transfer is described as
Q
wall,i
=
n
wall
l=1
A
wall,l,i
h
wall,l,i
(T
i
T
wall,l
) , (2.4)
where n
wall
denotes the total number of walls in the engine, A
wall,l,i
the area of wall
l belonging to zone i, h
wall,l,i
the heat transfer coecient of zone i to wall l, T
i
the
temperature of zone i, and T
wall,l
the temperature of wall l.
The mixing in the multi-zone model is accounted for by allowing the dierent
zones to exchange mass and energy with each other, in addition to the interaction
through the pressure. They exchange their scalar quantities (species composition
and enthalpy) based on the rate at which they exchange their mass. The mass of
each zone is kept constant and the model can accommodate any number of zones.
The multi-zone model is written in FORTRAN 77. Equations (2.1), (2.2),
and (2.3) are numerically solved using the dierential/algebraic system solver
DASSL [68].
2.3 Chemistry Model
A major aspect in modeling engine combustion is the treatment of the chemistry.
The advantage of using a zero-dimensional multi-zone model is that complex chem-
ical mechanisms can be used. In addition to its own very fast chemical solver, X0D
has an interface to CHEMKIN [49] and can therefore use any mechanism available
in this format. This is important in state-of-the-art applications (to investigate
pollutant formation, for example).
In the gasoline HCCI combustion simulations (see Chapter 5), combustion chem-
istry in X0D is described by a detailed chemical kinetic mechanism that incorpo-
rates low, intermediate, and high temperature oxidation chemistry of n-heptane
- iso-octane mixtures. This mechanism, which consists of 115 species and 482
reactions, was constructed by Advanced Combustion GmbH [70]. All these calcu-
lations were performed using a PRF82 mixture, which consists of 82 % iso-octane
and 18 % n-heptane by liquid volume.
In the PCCI combustion simulations (see Chapter 7), combustion chemistry
in X0D is described by a detailed chemical kinetic mechanism that comprises
59 elementary reactions among 38 chemical species. This mechanism describes
low-temperature auto-ignition and combustion of n-heptane, which serves as a
surrogate fuel for Diesel in these simulations. Furthermore, it accounts for thermal
NO formation. The chemical mechanism for n-heptane was constructed by Peters
et al. [67]. The NO-submechanism, which is part of the full mechanism, is the
extended Zeldovich mechanism [41].
7
3 Description and Modeling of
Chemically Reacting Turbulent
Two-Phase Flows
This chapter rst provides a short description of the governing equations for a
chemically reacting two-phase system with a gas phase and an evaporating liquid
phase. Following the introduction of the scales of turbulent motion and the averag-
ing procedure for the governing equations, the favre-averaged governing equations
for the turbulent ow and mixing eld are presented. After this, the implemen-
tation of these equations into the CFD code AC-FluX is depicted. Finally, this
chapter closes with a brief description of the modeling of the liquid phase.
3.1 Governing Equations
Fluid ows are described by a system of coupled, nonlinear, partial dierential
equations. In cases where chemical reactions occur, additional equations for the
species mass fractions, as well as an underlying chemical mechanism, are required.
The governing equations for the gas phase are presented in the following for a
two-phase system with a gas phase and an evaporating liquid phase.
The equation for the gas phase density reads
t
+

x
(v
) =
s
, (3.1)
where
s
denotes the source term due to the presence of the evaporating liquid
phase.
The rate of change of the gas phase momentum in each direction is given by
t
(v
) +

x
(v
) =
p
x
+ f
s
, = 1, 2, 3 , (3.2)
where f
s
is the rate of momentum gain per unit volume due to interaction with the
liquid phase. Gravitational inuences are neglected.
is the symmetric stress

tensor. Assuming a Newtonian uid, it is usually expressed as
+
v
2
3
, (3.3)
9
3 Modeling of Chemically Reacting Turbulent Two-Phase Flows
with
denoting the Kronecker delta. is the laminar viscosity.

The equation for the mixture enthalpy h, which includes the species heat of
formation h
0
f
according to
h =
nsp
j=1
Y
j
h
0
fj
+
T
T
0
c
pj
dT
, (3.4)
is given by
t
(h) +

x
(v
h) =
Dp
Dt
+
j
q
+ q
s
q
r
q
wall
. (3.5)
In Eq. (3.5),
/x
denotes the viscous heating term. This term is small

for low-speed ows and is therefore not considered further on. q
s
and q
r
describe
changes due to interaction with the liquid phase and due to radiative heat losses,
respectively. The latter are neglected in this work. q
wall
denotes the heat transfer
to the walls. Equation (3.5) does not contain a chemical source term as the heat
of formation of all species is included in the enthalpy. The heat ux j
q
accounts
for thermal diusion and enthalpy transport by species diusion, yielding
j
q
=
T
x
+
nsp
j=1
j
j
h
j
. (3.6)
The second term on the right-hand side is identical zero, because all Lewis numbers
are assumed equal to unity in this work.
The composition of the charge is represented by four so-called active species
streams, which are the mass fraction of fuel Y
fuel
, the mass fraction of air Y
air
, the
mass fraction of combustion products (carbon dioxide and water) Y
products
, and the
mass fraction of residuals Y
EGR
. The corresponding transport equation reads
t
(Y
j
) +

x
i
(v
i
Y
j
) =

x
i
D
j
Y
j
x
i
+
j
+
s
j
. (3.7)
In Eq. (3.7),

Y
j
denotes active species stream j, D
j
the diusion coecient of
active stream j, and
j
the chemical source term of active stream j. The source
term for the residuals is identical zero (
EGR
= 0), because they are assumed to
be chemically inert. The residuals are therefore an indicator for dilution. The
determination of the remaining source terms (
fuel
,
air
, and
products
) is described
in Sec. 4.3. Here again, the term
s
j
represents the source term due to evaporation
of liquid fuel. It is zero for all species except the fuel itself (
s
j
= 0, j = fuel).
The link between pressure, temperature, active species streams, and density is
established by means of the ideal gas law, given by
10
3.1 Governing Equations
p
=
nsp
j
Y
j
W
j
RT , (3.8)
where R is the ideal gas constant and W
j
the molecular weight of species j. Using
the mean molecular weight W dened as
W =
nsp
j
Y
j
W
j
1
, (3.9)
Eq. (3.8) is reduced to
p =

W
RT . (3.10)
The mixing eld is described by two additional scalars. These are the fuel
mixture fraction Z
fuel
and an articial scalar S.
The fuel mixture fraction Z
fuel
is the dominant quantity for the description of
non-premixed combustion and can be related to the commonly used equivalence
ratio or the combustion-air ratio , respectively, according to
=
1
=
Z
fuel
1 Z
fuel
(1 Z
fuel,st
)
Z
fuel,st
, (3.11)
where Z
fuel,st
is the fuel mixture fraction at stoichiometric mixture [66]. The cor-
responding transport equation reads
t
(Z
fuel
) +

x
(v
Z
fuel
) =

x
D
Z
Z
fuel
x
+
s
fuel
. (3.12)
Similar to Eq. (3.1), the term
s
fuel
represents the source term due to fuel evapora-
tion.
In addition to the transport equation for the fuel mixture fraction Z
fuel
, a generic
convection diusion equation without source terms is considered for the articial
scalar S. This transport equation reads
t
(S) +

x
(v
S) =

x
D
S
S
x
. (3.13)
It is assumed that the diusion coecients D
j
, D
Z
, and D
S
are identical.
Accounting for the fuel mixture fraction and an articial scalar is mandatory
in order to establish the interactive coupling of the CFD code AC-FluX and the
multi-zone model X0D. For HCCI combustion, the interactive coupling is described
in Chapter 4. For PCCI combustion, it is described in Sec. 6.1.
11
3.2 Scales of Turbulent Motion
Equations (3.1), (3.2), and (3.5), along with appropriate initial and boundary
conditions, are sucient to describe even turbulent ow elds. In general, these
equations have to be solved numerically, since analytical solutions can only be
obtained for simple ows. However, turbulent ow elds show a large range of
length, time, and velocity scales and the large scales are usually of the order of the
geometric dimensions
1
. These large structures are usually referred to as integral
scales. The smallest occuring ow structures are associated with the Kolmogorov
length, time, and velocity scales. For high Reynolds numbers, several orders of
magnitude lie between the integral scales and the Kolmogorov scales. The latter
are also called dissipative scales, as molecular diusion destroys these smallest
eddies, and their energy is dissipated. Kolmogorovs rst similarity hypothesis
states that for turbulence with suciently high Reynolds numbers, the statistics
of the small scale motions have a universal form that is uniquely determined by
the laminar viscosity and the viscous dissipation rate . The Kolmogorov length
scale is dened by
=
1/4
. (3.14)
The viscous dissipation rate itself scales as
=
v
3
l
0
, (3.15)
where v
and l
0
are the mean velocity uctuation and the integral length scale,
respectively [91]. Equations (3.14) and (3.15) can be used to further illustrate the
scale dierence between the large and the small scales. Introducing the turbulent
Reynolds number Re according to
Re =
v
l
0
, (3.16)
the ratio of the Kolmogorov length scale and the integral length scale l
0
is
l
0
Re
3/4
. (3.17)
Similar scaling laws are found for the velocities,
v
v
0
Re
1/4
, (3.18)
1
In internal combustion engines, for example, the largest vortices (eddies) can have diameters
of up to one tenth of the combustion chamber bore.
12
3.3 Averaging
and the time scales,
0
Re
1/2
, (3.19)
where
0
can be interpreted as the turnover time of an eddy with the size l
0
, which
has a characteristic velocity v
0
[73]. The characteristic velocity v
0
is of the same
order of magnitude as v
. It can be seen from Eqs. (3.17), (3.18), and (3.19) that

for a suciently high Reynolds numbers a wide range of length, time, and velocity
scales exists in a turbulent ow. Solving Eqs. (3.1), (3.2), and (3.5) numerically
therefore requires suciently ne meshes and small time steps, as even the small
scales have to be resolved completely. However, with current computer capabilities,
this direct numerical simulation (DNS) is only possible for simple geometries and
moderate Reynolds numbers. Since engineering applications generally have high
Reynolds numbers, signicant parts of the small scale motions have to be modeled.
3.3 Averaging
Modeling parts of the small scale motion is usually achieved by averaging the
original equations.
According to Reynolds, each variable f is split into a mean component

f and a
uctuating component f
, leading to
f =

f + f
. (3.20)
Ensemble averaging is most commonly used for obtaining the mean component

f.
This yields
f
N
=
1
N
N
i=1
f
i
, (3.21)
where N is the number of realizations, over which the instantaneous values f
i
are
averaged.
For ows with large density changes as occur in combustion, it is often convenient
to introduce a density-weighted average

f, called the Favre average, by splitting f
into

f and f
as
f =

f + f
. (3.22)
This averaging procedure is dened by requiring that the average of the product
of f
with the density (rather than f
itself) vanishes:
f
= 0 . (3.23)
13
Reynolds averaging and Favre averaging are related to each other according to
f =
f

and f

. (3.24)
3.4 Turbulent Flow and Mixing Field
The splitting operations described above can also be applied to the governing
equations presented in Sec. 3.1 instead of a single quantity only. The ensemble
averaging procedure then yields the Reynolds Averaged Navier-Stokes (RANS)
equations. These equations only describe the motion on the integral length, time,
and velocity scales, while every motion on smaller scales needs to be modeled, even
if the grid size is much smaller than the integral length scale. Filtering with lter
widths smaller than the integral scales allows to resolve more of the small scale
motion. This is typically done in Large Eddy Simulations (LES) or Very Large
Eddy Simulations (VLES), for more details see [28] or [73]. In this work, a RANS
approach is used.
The continuity equation reads, after averaging,

t
+

x
( v
) =
s
. (3.25)
Equation (3.25) is very similar to Eq. (3.1). All instantaneous quantities are re-
placed by the average values and no additional terms occur.
The averaged momentum equations are
t
( v
) +

x
( v
) =
p
x
+ f
s
. (3.26)
with the averaged symmetric stress tensor
being

+
v
2
3
. (3.27)
In Eq. (3.26), in addition to the original contributions the so-called Reynolds
stresses v
appear. They represent the classical closure problem of turbu-

lent ows. Reynolds stresses are second-order correlations, and they describe the
convective momentum transport by turbulent uctuations. Transport equations
can be derived for the Reynolds stress tensor; however, these equations contain
triple correlations of the sort v
. These triple correlations are also unclosed.

They therefore require a modeling approach. Furthermore, unclosed correlations
of pressure uctuations and velocity gradients, the so-called pressure-rate-of-strain
tensor, appear. First modeling suggestions were made by Rotta [81]. Instead of
14
3.4 Turbulent Flow and Mixing Field
modeling the triple correlations, it is also possible to derive equations for them,
which then contain correlations of fourth order. This indicates an innite modeling
hierarchy and thus no closed equations can be obtained directly.
The averaged enthalpy equation is obtained as
+

x
=
Dp
Dt

j
q
+ q
s
q
wall
. (3.28)
In Eq. (3.28), the term v
is similar to the Reynolds stresses in Eq. (3.26).

It represents the convective enthalpy transport by turbulent uctuations and it
needs to be modeled, as it is unclosed, too.
If no equations for the turbulent transport terms are solved, a model is required
which relates these terms to known quantities. Boussinesq [11] proposed the con-
cept of a turbulent viscosity
t
, which is often referred to as an eddy viscosity. It
is used to relate the turbulent stresses to the mean eld, yielding
v
=
t,
=
t
+
v
2
3
v
2
3
k
=
t
S
ij
2
3
v
2
3
k
, (3.29)
where

S
ij
denotes the rate-of-strain tensor and
k the mean turbulent kinetic energy

dened by
k =
1
2
v
. (3.30)
The turbulent viscosity
t
is not a uid property such as the laminar viscosity . It
depends on the turbulence in the vicinity. It can be considered to be the product
of a velocity scale and a length scale. Several models for the turbulent viscosity
are available, and they are usually classied according to the number of additional
equations that need to be solved. Prandtl [74] related the turbulent viscosity to
a mixing length l
m
and the absolute gradient of the mean velocity eld. As no
additional equations are solved, this model belongs to the class of zero-equation
models. One of the rst one-equation models was proposed by Prandtl [75]. The
turbulent viscosity is determined using the turbulent kinetic energy, for which an
equation is solved, and a length scale, which is determined empirically. Rotta [82]
derived an equation for the length scale by integrating the two-point correlation
over the correlation coordinate. This leads to a k l model, which then belongs to
the class of two-equation models. Rodi and Spalding [80] rst used an l-equation
15
to compute a turbulent jet ow. The most popular two-equation model is the k
model, where is the turbulent dissipation. The standard k model was rst
developed by Jones and Launder [48] and improved model constants were provided
by Launder and Sharma [51]. Turbulent length, time, and velocity scales can be
easily formed using these two quantities. The length scale, for example, can be
expressed as
l k
3/2
/ . (3.31)
The turbulent viscosity is modeled according to
t
= C
k
2
, C
= 0.09 . (3.32)
The modeling constant C
was already proposed by Launder and Sharma [51]. It

is usually not changed. The equation for the turbulent kinetic energy that is used
in this work reads
t
( k) +

x
( v
k) =

x
+

t
Pr
t,k
k
x
+
t,
v
+

W
s
k
, (3.33)
where

W
s
k
describes the eect of turbulent dispersion of droplets on the turbulent
kinetic energy. For constant-density ows, the equation for the turbulent kinetic
energy can be derived systematically. From this derivation follows the denition
of the viscous dissipation rate, reading
=
+
v
. (3.34)
An -equation is dicult to derive and to close in a systematic manner. Instead,
a model equation, which is partly empirically, is solved according to
t
( ) +

x
( v
) =

x
+

t
Pr
t,
+ C
1
k

t,
v
C
2

2
k
+ C
3

v
+

k
C
s

W
s
k
. (3.35)
The model constants are given as Pr
t,k
= 1.0, Pr
t,
= 1.22, C
1
= 1.44, C
2
= 1.92,
C
3
= 0.33, and C
s
= 1.5.
16
3.5 CFD Code
After closing the turbulent transport term in Eq. (3.28) with a gradient ux
approximation,
v
=

t
Pr
h
x
, (3.36)
the nal equation for turbulent mean enthalpy reads
+

x
=
Dp
Dt
+

x
c
p
+

t
Pr
t
h
x
+ q
s
q
wall
. (3.37)
Averaging Eq. (3.7) representing the composition of the charge leads to
Y
j
+

x
i
v
i
Y
j
=

x
i

Sc
e
Y
j
+

t
Sc
t,
e
Y
j
Y
j
x
i
+
j
+
s
j
. (3.38)
The density-weighted averaged mixture eld is accounted for by
Z
fuel
+

x
i
v
i
Z
fuel
=

x
i

Sc
e
Z
fuel
+

t
Sc
t,
e
Z
fuel
Z
fuel
x
i
+
s
fuel
(3.39)
and
+

x
i
v
i
=

x
i

Sc
e
S
+

t
Sc
t,
e
S
S
x
i
. (3.40)
3.5 CFD Code
The averaged equations for the gas phase described in the previous section are
solved by the CFD code AC-FluX (formerly known as GMTEC). AC-FluX is
a ow solver based on Finite Volume methods [29] that employs unstructured,
mostly hexahedral meshes. AC-FluX is mainly used for internal combustion en-
gine simulations, for gasoline as well as for Diesel enginges. The code is able to
treat moving meshes and non-conforming internal mesh motion boundaries that
facilitate generating a realistic geometric model of intake ports, exhaust ports, and
the in-cylinder in combination with valve motion. In order to provide high spatial
accuracy, adaptive run time controlled mesh renement can be used optionally.
17
This section briey covers the numerical algorithms and the general code struc-
ture of AC-FluX. More details (e.g., regarding spatial and temporal discretization
schemes) are given in Khalighi et al. [50], Ewald et al. [24], and Freikamp [32].
AC-FluX uses an iterative implicit pressure-based sequential solution procedure
to solve the coupled system of governing partial dierential equations. The equa-
tions are solved sequentially rather than simultaneously; coupling is achieved via an
iterative updating procedure. The procedure accommodates incompressible and/or
compressible ows, as well as steady and/or transient ows. It is applicable for es-
sentially arbitrary Mach numbers, although for Mach numbers much greater than
unity the eciency of the approach decreases signicantly. AC-FluXs pressure
algorithm is patterned after SIMPLE (SemiImplicit Method for PressureLinked
Equations, [62]) and PISO (PressureImplicit Split Operator, [46]). PISO origi-
nally was conceived as a predictor-corrector method to be used with a xed number
of passes through the equations on each time step; however, a pure PISO method
generally is neither suciently ecient nor suciently robust for the highly dis-
torted computational meshes and complex three-dimensional time-dependent ows
that characterize practical engineering applications. The algorithm used in AC-
FluX can be thought of as a modied PISO scheme, where both the numbers of
outer and inner iterations are variable. In the case of a single outer loop (momen-
tum predictor) and a single inner loop (pressure/velocity corrector) per time step,
this algorithm reduces to a SIMPLE-like method.
The essential steps in the pressure/momentum/continuity coupling to advance
the solution over one computational time step are:
1. momentum predictor compute a new velocity eld using the current pres-
sure eld; this velocity eld does not satisfy continuity.
2. pressure/velocity correctors compute corrections to the pressure and ve-
locity elds to enforce continuity.
The momentum predictor and pressure corrector each require the solution of
a sparse implicit linear system that corresponds to a linearised discretised form
of the governing partial dierential equations. The velocity corrector is explicit.
Equations for additional quantities (e.g., total enthalpy, species mass fractions)
are included in each pressure/velocity corrector step to maintain tight coupling
among the equations. At the end of the pressure/velocity corrections, equations
requiring a lesser degree of coupling are solved (e.g., turbulence model equations).
The process then is repeated as necessary, starting from the momentum predictor,
to obtain a converged solution for the current time step or global iteration. Three
levels of iteration are thus employed on each time step (each global iteration for
a steady solution algorithm): an outer loop or outer iteration, an inner loop or
18
3.6 Liquid Phase
inner iteration, and iterations within the linear equation solvers. The outer it-
eration corresponds to the momentum predictor step, the inner iteration to the
pressure/velocity corrector step. The basic sequence is displayed in Fig. 3.1.
The calculation of the source terms in the gas-phase equations is a preparatory
step to the sequential solution procedure described above. The mathematical
formulation of the liquid phase is subject of the following section. The formulation
of the chemical source terms is elaborated on in Sec. 4.3.
3.6 Liquid Phase
The previous sections discuss the governing equations of the gas phase and their
implementation into the CFD code AC-FluX. During the injection period in an
internal combustion engine an additional liquid phase is present, which must be
adequately described to obtain the spray-related source terms in the gas phase
equations (see Eqs. (3.25), (3.26), (3.33), (3.35), (3.37), (3.38), and (3.39)). Solv-
ing for the dynamics of a spray with a wide distribution of drop sizes, velocities,
and temperatures is a complicated problem. A mathematical formulation capable
of describing this distribution is the spray equation proposed by Williams [101]. It
is an evolution equation for the probability density function f with the independent
variables droplet position, droplet velocity, temperature, radius, distortion from
sphericity y, and its time derivative dy/dt. Depending on the specic problem,
additional variables can be introduced. A direct solution of this equation is ex-
tremely dicult due to its high dimensionality; the aforementioned variables alone
constitute a 10-dimensional space and associated storage and computing time re-
quirements. Instead, a suciently large number of particles are introduced, which,
according to Crowe et al. [16], are called parcels. This model is usually referred
to as the discrete droplet model (DDM). Each parcel represents an ensemble of
droplets. Within one parcel, all droplets have the same properties, which corre-
spond to the independent variables described above. The ensemble of all parcels
provides the statistical information on the spray. All the subprocesses that are not
resolved on the parcel level are modeled using a Monte-Carlo method [19]. Impor-
tant subprocesses, which need to be described, are breakup, collision, coalescence,
evaporation, and dispersion. A detailed description of the models for these sub-
processes, the formulation of the source terms for the gas phase equations, and its
implementation into the CFD code AC-FluX can be found in [89].
19
total enthalpy

h
velocity corrector
w
viscous dissipation
turb. kinetic energy

k
add. scalars

Z
fuel
and

S
momentum predictor
w v u
active streams

Y
j
pressure corrector p
m
c
c
p
m
t
u
r
b
m
o
u
t
e
r
Figure 3.1: Schematic owchart illustrating three levels of iteration. This
owchart corresponds to a compressible case (enthalpy equation included in
the inner iterations) using an enthalpy predictor (versus explicit corrector) for
each inner iteration (adopted from [32])
20
4 Interactive Coupling of the CFD Code
and the Multi-Zone Model
The interactive coupling (two-way-coupling) of the three-dimensional CFD code
AC-FluX and the multi-zone model X0D is outlined in this chapter. After ini-
tialization, the interactive coupling comprises one information stream which is fed
from the CFD code to the multi-zone model. Mixing eects are introduced into the
multi-zone model based on the CFD solution. Vice versa, one information stream
is fed from the multi-zone model to the CFD code in order to keep the heat release
consistent between both codes. The initialization of the interactive coupling and
the two information streams are addressed in Secs. 4.1, 4.2, and 4.3. Section 4.4
deals with the treatment of source terms within the interactive coupling. A method
how to account for source terms in a completely consistent manner is described in
Sec. 4.5. At the end of this chapter, Sec. 4.6 discusses this modeling concept in
the context of the Representative Interactive Flamelet model.
4.1 Initialization of the Interactive Coupling
In order to construct the mass distribution function and subdivide it into discrete
zones in phase space, the computational cells in the physical domain in the CFD
code are sorted in ascending order in terms of the phase variables (fuel mixture
fraction, EGR mass fraction, and total enthalpy). The computational cells are then
assigned to the zones so that these have equal mass. It is not required that the
zones have equal mass, but it simplies the derivation and implementation. This
procedure is repeated at each time step of the CFD code. Only once, however, at a
certain crank angle prior to any chemical reactions starting to occur, a correspond-
ing set of zones dened by specic fuel masses, temperatures, and residual mass
fractions is initialized in the multi-zone model X0D. From this transition point on,
both codes interact back and forth throughout the engine cycle. The two codes
run independently but simultaneously, without any reinitialization of the zones
(and the numerical solver for the dierential equations) in the multi-zone model.
Figure 4.1 schematically illustrates how the AC-FluX CFD code is used to ini-
tialize the zones in X0D, for which the chemistry is solved. For the gasoline HCCI
engine test case investigated (see Chapter 5), the left hand side of Fig. 4.1 shows
two injector cut planes colored by zone aliation at time of X0D initialization.
21
4 Interactive Coupling of the CFD Code and the Multi-Zone Model
The zone aliation is the result of the sorting procedure outlined above. Upon ini-
tialization, the specic fuel masses, the mean temperatures, and the mean residual
mass fractions of the zones in the physical domain are used to dene a correspond-
ing set of zones in the multi-zone model. This set of zones is shown in the right
hand side of Fig. 4.1. The zones are colored according to their initial temperature.

700 720 740 760 780
3 3
21
12
7
24
8
4
6
9
2
1
27
27
26
25
9
9
6
5
7
8
18
18
16
17
15
Temperature [K]
Zone
numbers
Zone affiliation [-]
Definition of a corresponding
set of zones in the multi-zone model
Upon X0D initialization, sorting of the physical
domain in ascending order in terms of the
phase variables (fuel mix. frac.,
EGR mass frac.,
and total
enthalpy)
Crank Angle = 660.0 degs.
Figure 4.1: Schematic illustration of the zone initialization in the multi-zone
model for HCCI combustion
It is assumed in the following that the injected fuel mass is completely evaporated
at the transition point. Nevertheless, this is no limitation of the model. Section 4.5
below outlines how to deal with more advanced operating conditions where fuel
vaporization happens after initialization of the interactive coupling. Furthermore,
such advanced operating conditions are subject of Chapters 6 and 7.
4.2 Mixing Models
4.2.1 Neighboring Zone Mixing
Assuming that, rstly, the combustion chamber is a closed system, meaning that
the valves are closed; secondly, all fuel is vaporized; and thirdly, the total mass is
constant, then the distribution of the phase variables in the CFD code only changes
due to convection and diusion. This is true for the fuel mixture fraction and the
EGR mass fraction, because both are conserved scalars. The total enthalpy is an
exception, as it also changes due to volume change, turbulent dissipation, and wall
heat transfer. These source terms are addressed later. The conserved scalars are
22
4.2 Mixing Models
considered rst. The change for the mean fuel mixture fraction from CFD time
step n 1 to CFD time step n in the individual zones can be formulated as
m
i
Z
n
fuel,i

Z
n1
fuel,i
t
n
=
6
k=1
m
ik
Z
n1
fuel,k

Z
n1
fuel,i
, (4.1)
where t
n
is the time increment, m
i
the mass of zone i, and m
ik
the mass exchange
rate of zone i with neighboring zone k (see Eqs. (2.1) and (2.2)). Since the phase
space is three-dimensional, each zone in phase space can have up to six neighbors
to exchange their scalar quantities with.
Allowing mixing only between neighboring zones in phase space is a logical
choice, because transitions in phase space have to be continuous. If a state B is
intermediate between two states A and C, then a zone of state A cannot transit
to state C without passing through state B during a diusive process. The zones
in phase space are therefore arranged in ascending (or descending) order, as il-
lustrated in Fig. 4.1. Consequently, neighboring zone mixing is used throughout
the remainder of the thesis. This is analogous to a Representative Interactive
Flamelet. In the amelet equations, the scalar dissipation rate is an exter-
nal parameter that is imposed on the amelet structure. It can be thought of
as a diusivity in mixture fraction space that accounts for an exchange between
adjacent reactive scalars (temperature and species mass fractions) in mixture frac-
tion space. The laminar amelet equations were derived by Peters [64, 65, 66]
and integrated into the concept of Representative Interactive Flamelets by his
co-workers [8, 25, 38, 69].
The challenge in developing a mixing model is to extract the rate of exchange
between each of the zones from the CFD code based on the mass distribution
function of the current and the previous time step. Why this is a challenge becomes
apparent when the number of unknowns and equations is considered. If n
zones
is the
number of zones in each of the three directions of the independent phase variables,
then
n
m
ik
= 3n
2
zones
(n
zones
1) (4.2)
unknown mass exchange rates exist between the n
3
zones
zones. It is easy to show
that for n
zones
2, the number of unknown mass exchange rates n
m
ik
exceeds
the total number of zones n
3
zones
, and it follows that the system of equations is
underdetermined and not directly solvable. In the following, two approaches to
overcome this diculty are presented. These two approaches allow for a calculation
of the mass exchange rates m
ik
so that they can be updated in the multi-zone code
at every CFD time step.
23
4.2.2 The Global Mass Exchange Rate
Since the system of equations is underdetermined, either the number of equations
needs to be increased or the number of unknowns must be reduced. The latter
is readily done, assuming that all the mass exchange rates m
ik
are equal and
can be represented by a global mass exchange rate m
global
. By doing this, the
composition between the CFD code and the zero-dimensional multi-zone code is
not identical anymore, but still be coupled. Now there is one one unknown and
n
3
zones
equations. Hence, the system is overdetermined. A common approach to
overcome overdetermined systems of equations is to employ a least square t to
nd the solution that has the least error, leading to
m
global
= m
av. zone
n
3
zones
i=1

Z
n
fuel,i
Z
n1
fuel,i
t
n
k=1
Z
n1
fuel,k

Z
n1
fuel,i
n
3
zones
i=1
k=1
Z
n1
fuel,k

Z
n1
fuel,i
2
. (4.3)
In Eq. (4.3), m
av. zone
represents the average zone mass. The global exchange rate
m
global
is then used in place of the individual mass exchange rates m
ik
in Eq. (4.1).
4.2.3 The Individual Mass Exchange Rate
Obviously, it is desirable to keep all the mixing time scales and distributions be-
tween X0D and AC-FluX consistent at all times. The problem stated above is
that there are more unknown mass exchange rates (or cell faces) between the zones
(n
m
ik
= 3n
2
zones
(n
zones
1), see Eq. (4.2)) than there are zones (n
3
zones
). The so-
lution to this problem is not to solve for the mass exchange rates m
ik
directly, but
to substitute the mass exchange rates m
ik
by the average of the neighboring zones
i and k,
m
ik
= 0.5 ( m
i
+ m
k
) , (4.4)
where m
i
and m
k
denote individual mass exchange rates of zones i and k, re-
spectively. Inserting Eq. (4.4) into Eq. (4.1) yields a system of n
3
zones
unknowns
and n
3
zones
+ 1 equations. The system is overdetermined, because there is no ex-
change over the boundaries of the phase space and, therefore, the summation of
the exchange (see Eq. (4.1)) over all zones is zero. This constitutes an additional
equation. The remedy is to simply x the value for one of the individual mass
exchange rates m
i
(e.g., m
1
= 1). The values for the mass exchange rates m
ik
are dened by the two solutions for the fuel mixture fractions

Z
fuel,i
at CFD time
steps n and n1. No matter which value is assigned to one of the individual mass
exchange rates m
i
, it does not change the resulting values for the mass exchange
24
4.3 Consistency of Heat Release
rates m
ik
.
4.3 Consistency of Heat Release
The chemical source terms
j
(see Eq. (3.38)) couple the CFD code AC-FluX
to the multi-zone model X0D and keep the heat release consistent between both
codes. The fuel mass fraction

Y
fuel
acts as a progress variable for the heat release.
In the CFD code, the average rate of change for the fuel mass m
fuel,i
in zone i is
coupled to the gross heat release in the multi-zone model through
m
fuel,i
=
Q
ch,X0D,i
Q
LHVp
. (4.5)
In Eq. (4.5),

Q
ch,X0D,i
denotes the gross heat-release rate of zone i that is taken
from the multi-zone model. Q
LHVp
is the lower heating value of the fuel.
The actual value of the mean fuel mixture fraction in each CFD cell varies from
the average value of the zone it belongs to. For this reason, in order to properly
distribute the conversion of fuel over the CFD cells, the average rate of change for
the fuel mass m
fuel,i
in zone i is weighted with the ratio of the actual mean fuel
mixture fraction in CFD cell ijk and the average mean fuel mixture fraction of
the corresponding zone i. This yields
m
CFD
fuel,ijk
=
Z
CFD
fuel,ijk
Z
fuel,i
m
fuel,i
, (4.6)
where m
CFD
fuel,ijk
denotes the rate of change for the fuel mass in CFD cell ijk and
Z
CFD
fuel,ijk
the actual mean fuel mixture fraction in CFD cell ijk.
The other species (air and products of combustion) which are accounted for
in AC-FluX are computed via a one-step global reaction corresponding to the
stoichiometry.
25
4.4 Treatment of Source Terms
The conservation equation for the mean total enthalpy

h, reading
+

x
=
Dp
Dt
+

x
c
p
+

t
Pr
t
h
x
+ q
s
q
wall
(cf. Eq. (3.37) , (4.7)
has two source terms ( q
s
and q
wall
) that need to be accounted for in the multi-zone
model X0D.
As mentioned above, it is assumed that the fuel is fully vaporized and the source
term for the spray is identical zero ( q
s
= 0).
In X0D, wall heat transfer (cf. Eqs. (2.2) and (2.4)) is described as
q
wall,i
V
i
=

Q
wall,i
=
n
wall
l=1
A
wall,l,i
h
wall,l,i
(T
i
T
wall,l
) , (4.8)
The area A
wall,l,i
of wall l belonging to zone i is taken from the CFD solution
and updated in the multi-zone model at every CFD time step. The heat transfer
coecient h
wall,l,i
of zone i to wall l is continuously recalculated based on the CFD
solution and updated in the multi-zone model at every CFD time step such that
the wall heat transfer in the multi-zone model exactly matches the one in the CFD
code. The recalculation of h
wall,l,i
uses the source term q
wall
in Eq. (4.7), averaged
for each zone in the CFD code.
The pressure change in X0D is described by Eq. (2.3), where the spatial pressure
uctuations are neglected and only the average cylinder pressure change, which
is due to the volume change and heat release, is taken into account. The volume
change in both codes is computed by using the same slider-crank-equation [41].
Thereby, the full interactive coupling of AC-FluX and X0D is accomplished.
4.5 Consistent Modeling of Source Terms
A further improvement over the current modeling may be achieved by a more so-
phisticated treatment of phase variables whose partial dierential equations con-
tain source terms. This is the case for the total enthalpy and the fuel mixture
fraction (if we allow for the vaporization of fuel after initializing the interactive
coupling). As discussed previously, the mixing coecients for those phase variables
cannot be computed in the current approach, because of the presence of source
terms, which would cause disturbances.
26
4.6 Concept Analysis
Consider a generic convection diusion equation with source terms representing
the phase variables according to
+

x
i
v
i
=

x
i

Sc
e
S
+

t
Sc
t,
e
S
S
x
i
+

S . (4.9)
Then, with each CFD time step, the solution for the variable

S is numerically
advanced from its state

S
n
at time t
n
to its new state

S
n+1
at time t
n+1
. By
introducing an additional conservation equation for the variable

S omitting the
source term

S,
S
c
+

x
i
v
i
S
c
=

x
i

Sc
e
Sc
+

t
Sc
t,
e
Sc
S
c
x
i
, (4.10)
and initializing

S
n
c
=

S
n
at time t
n
, then advancing it numerically to

S
n+1
c
at time
t
n+1
, a solution for variable

S is obtained that is again purely based on mixing.
The two solutions

S
n
c
=

S
n
and

S
n+1
c
can now be used to compute the mixing
coecients for variable

S with the help of Eqs. (4.1) and (4.4). Furthermore, the
dierence

S
n+1
c

S
n+1
divided by the time increment t
n
= t
n+1
t
n
,
S

=
S
n+1
c

S
n+1
t
n
, (4.11)
is the numerical representation of the source term for variable

S. This source term
can then be averaged for each zone and passed on to X0D. Therefore, using the
mixing coecients and source terms generated in this manner makes the coupling
between AC-FluX and X0D completely consistent in the limit of the numerical
error.
4.6 Concept Analysis
The similarities between the multi-zone equations for species mass fractions and
temperature (see Eqs. (2.1) and (2.2)) and the corresponding laminar amelet
equations are noteworthy. Both sets of equations feature a transient term, terms
describing the mixing, and a chemical source term. This may seem somewhat
surprising, considering the fact that no assumptions of fast chemistry went into
deriving the multi-zone equations.
The laminar amelet equations could alternatively be interpreted as describing a
series of coupled ow reactors (or zones), where the scalar dissipation rate, through
its functional dependence on the mixture fraction, governs the mixing between
them. As discussed in Subsec. 4.2.1, only ow reactors adjacent to each other
27
in phase space, i.e. mixture fraction space, exchange mass and energy with each
other. Conceptually, this interpretation does not rely on the amelet assumptions.
In situations where both methods could be applied, the interactively coupled
CFD-multi-zone approach clearly oers an advantage in terms of computational
eciency. Several amelets are required to properly account for various regions of
mixing and heat transfer. Each amelet calculation typically involves solving ap-
proximately 100 coupled ow reactors. As shown in Chapters 5 and 7, signicantly
less zones are needed in the CFD-multi-zone approach. It should be emphasized
that the Representative Interactive Flamelet model, when interpreted as solving a
series of ow reactors, should no longer be referred to as a amelet concept, but
rather a multi-zone approach.
Hergart et al. [39] investigated a single-cylinder heavy-duty Diesel engine that
was operated in PCCI mode. In the simulations, the authors rst applied a se-
quential treatment of the in-cylinder processes, where the injection processes and
the uid dynamics were modeled using the CFD code AC-FluX and the chemi-
cal reactions were treated in the multi-zone model X0D. Secondly, they applied
the Representative Interactive Flamelet model. Both modeling approaches proved
successful in predicting auto-ignition, subsequent heat release, and pollutant for-
mation. This work also contains a detailed discussion about the relative merits of
both modeling approaches.
28
5 Validation of the Interactive Coupling
In this chapter, the interactive coupling of the three-dimensional CFD code AC-
FluX and the multi-zone model X0D is validated against a test case of HCCI
combustion in a gasoline single-cylinder research engine.
5.1 Experimental and Numerical Setup
5.1.1 Experimental Setup
The experimental data were generated at General Motors Corporation in a direct
injection gasoline single-cylinder research engine with four valves and negative
valve overlap (NVO). The main characteristics of the engine are summarized in
Table 5.1.
Engine type: Single-cylinder research engine
Bore: 86.0 mm
Stroke: 94.6 mm
Connecting rod length: 152.2 mm
Piston pin oset: -0.8 mm
Compression ratio: 12.0:1
Injection system: 8-hole injector
Included spray angle: 60.0 degs.
Hole diameter: 0.4 mm
Fuel: Standard gasoline
Table 5.1: Engine specications
The engine was operated in HCCI mode. The valve timings given in Table 5.2
were chosen such that the exhaust valves closed before top dead center of the
exhaust stroke, and the intake valves opened after top dead center of the same
stroke. This provided an NVO interval of 182.0 degs. CA, in which both the
exhaust and intake valves were closed. NVO is a variable valve actuation (VVA)
strategy that helps in assisting with combustion phasing and to extend operation
to low speed and loads. Closing the exhaust valve before top dead center of the
exhaust stroke, thus trapping and recompressing some of the hot residual, raises
29
the enthalpy level enough to enable the charge to auto-ignite at the subsequent
combustion cycle. The use of VVA was suggested by Willand et al. [100], and
several examples of its exploitation appeared [20, 52, 61, 84, 96].
EVO: 148.0 degs. CA aTDC
EVC: 268.0 degs. CA aTDC
IVO: 450.0 degs. CA aTDC
IVC: 570.0 degs. CA aTDC
NVO: 182.0 degs. CA
Table 5.2: Valve timings
The engine was operated at a speed of 3000 rpm. A quantity of 6.0 mg fuel was
injected at 270.0 degs. CA before top dead center (bTDC) ring. With injection
early in the intake stroke, the charge is fairly homogeneous at the end of the main
compression when auto-ignition occurs. The operating conditions are summarized
in Table 5.3.
Engine speed: 3000 rpm
Global air/fuel ratio: 25.0
External EGR: 0 %
Injected fuel mass: 6.0 mg
Start of Injection (SOI): 285.2 degs. CA bTDC
End of Injection (EOI): 270.0 degs. CA bTDC
Spark Adv.: No spark
PIVC: 1.26 bar
TIVC: 529.0 K
Table 5.3: Engine operating point
5.1.2 Numerical Setup
The simulation was run from 197.0 degs. CA after top dead center (aTDC) until
exhaust valve opening (EVO). This captured almost the whole gas exchange and
the complete heat release, including pollutant formation. In the rst part of the
CFD simulation, AC-FluX was run alone. At 660.0 degs. CA aTDC, the coupling
to X0D was initiated.
The computational domain included the intake runner and the exhaust system
up to the rst closely coupled muer in addition to the cylinder. The transient
30
boundary condition for the orices at intake and exhaust, as well as the initial
conditions, were taken from a one-dimensional simulation of the complete system
in the test cell with the WAVE code by Ricardo [78].
The in-cylinder region of the computational mesh consisted of 91454 cells at top
dead center (TDC) ring. An outline of the in-cylinder region of the computational
mesh is given in Figs. 5.1 and 5.2. In Fig. 5.1, the piston bowl and the cylinder
wall are removed for illustration purposes. Vice versa, in Fig. 5.2, the cylinder
head and the cylinder wall are removed for the sake of a clear insight into the
bowl.
Figure 5.1: Computational grid at TDC ring showing the cylinder head surface.
Piston bowl and cylinder wall are removed for illustration purposes
Figure 5.2: Computational grid at TDC ring showing the piston bowl surface.
Cylinder head and cylinder wall are removed for illustration purposes
31
5.2 Results and Discussion
5.2.1 Sensitivity Study
A sensitivity study of the local in-cylinder quantities, the ignition timing, and
the overall heat-release rate on the number of zones was performed rst. For this
purpose, simulations with number of zones of n
3
zones
= 5
3
= 125, n
3
zones
= 4
3
= 64,
n
3
zones
= 3
3
= 27, and n
3
zones
= 2
3
= 8 were performed. It was determined that
beyond n
3
zones
= 3
3
= 27, the change of the local in-cylinder quantities (e.g.,
temperature, fuel mass fraction), the ignition timing, and the subsequent heat-
release rate was insignicant.
The distribution of the fuel mixture fraction, the EGR mass fraction, and the
total enthalpy in two injector cut planes at time of X0D initialization (660.0 degs.
CA aTDC) are shown in Fig. 5.3. The EGR mass fraction stratication and the
enthalpy stratication are qualitatively similar in comparison to the fuel mixture
fraction stratication, indicating a coupling of the phase variables for the specic
test case investigated. Due to this, simulations with reduced numbers of phase
variables were also carried out. This included simulations in which the computa-
tional cells were subdivided only based on fuel mixture fraction and EGR mass
fraction (two-dimensional binning, n
2
zones
= 5
2
= 25) and in which the computa-
tional cells were subdivided purely based on fuel mixture fraction (one-dimensional
binning, n
1
zones
= 27
1
= 27).
Figure 5.4 displays the heat-release rates from X0D for the cases with n
3
zones
=
3
3
= 27, n
2
zones
= 5
2
= 25, and n
1
zones
= 27
1
= 27 zones, as well as the heat-release
rate obtained from the experiment. All simulations yielded similar heat-release
rates and combustion eciencies. These results reveal that considering the fuel
mixture fraction stratication due to the fuel injection and evaporation process is,
in this case, sucient to accurately describe HCCI auto-ignition and subsequent
heat release. For the purpose of investigating the importance of incorporating
mixing into multi-zone models, the one-dimensional binning with a resolution of
n
1
zones
= 27
1
= 27 zones is sucient and, for the sake of simplicity, was thus chosen
for all subsequent comparisons of the various mixing models.
32
Fuel mixture fraction [-]
EGR mass fraction [-]
Total Enthalpy [J/kg]
Figure 5.3: Fuel mixture fraction, EGR mass fraction, and total enthalpy distri-
bution at 660.0 degs. CA aTDC in two injector cut planes
33
690 700 710 720 730 740 750
Crank Angle [degs.]
0
5
10
15
20
25
30
35
40
H
e
a
t
-
R
e
l
e
a
s
e

R
a
t
e

[
J
/
d
e
g
.
]
Experiment
X0D, n
2
zones
= 5
2
= 25
X0D, n
1
zones
= 27
1
= 27
X0D, n
3
zones
= 3
3
= 27
Figure 5.4: Average heat-release rate comparison for X0D with dierent binning
dimensions to the experiment
5.2.2 Importance of Incorporating Mixing
In the following, three cases are presented in order to investigate the importance
of incorporating mixing into the multi-zone model:
1. a simulation applying the individual mass exchange rates (IME),
2. a simulation using the least square t for the global mass exchange rate
(GME),
3. a simulation without mixing between the zones.
The run time increase for the more complex models was small compared to the
overall run time of the CFD code.
34
Figure 5.5 compares the pressure traces for AC-FluX and X0D with the individ-
ual mass exchange rates to the experiment. The simulation slightly underestimates
the early heat release, followed by a steeper pressure gradient. The peak pressures
match very well. All in all, the agreement between experiment and simulation
is very good. However, the experimental data primarily serves as a reference.
The main point of this plot is the agreement between the pressures of AC-FluX
and X0D over the whole crank angle range, but especially during the heat-release
phase. Both pressures are identical except for negligible dierences.
705 710 715 720 725 730 735 740 745
Crank Angle [degs.]
20.0
22.5
25.0
27.5
30.0
32.5
35.0
37.5
P
r
e
s
s
u
r
e

[
b
a
r
]
Experiment
AC-FluX
X0D
Figure 5.5: Average cylinder pressure comparison for AC-FluX and X0D with
the individual mass exchange rates to the experiment
Figure 5.6 shows a comparison of the experimental heat-release rate, the heat-
release rate of X0D with the individual mass exchange rates, and X0D with the
global mass exchange rate from the least square t, respectively. The crank angle
of 50 % burnt mass (CA50) obtained from the experiment is 722.9 degs. CA
aTDC. This is captured very well by the IME model revealing a CA50 of 723.5
degs. CA aTDC. The GME model leads to a CA50 of about 1.9 degs. CA earlier
compared to the experiment. In Fig. 5.6, the heat-release rate for a simulation
without mixing is also shown. The simulation without mixing exhibits a CA50 of
about 2.7 degs. CA earlier in comparison to the experiment. Thus, the maximum
heat-release rate is overpredicted for the GME model and the case without mixing.
The corresponding pressure curves shown in Fig. 5.7 conrm these ndings.
35
690 700 710 720 730 740 750
Crank Angle [degs.]
0
5
10
15
20
25
30
35
40
45
H
e
a
t
-
R
e
l
e
a
s
e

R
a
t
e

[
J
/
d
e
g
.
]
Experiment
X0D, IME
X0D, GME
X0D, no mixing
Figure 5.6: Average heat-release rate comparison for X0D with dierent mixing
models to the experiment
705 710 715 720 725 730 735 740 745
Crank Angle [degs.]
20.0
22.5
25.0
27.5
30.0
32.5
35.0
37.5
40.0
P
r
e
s
s
u
r
e

[
b
a
r
]
Experiment
X0D, IME
X0D, GME
X0D, no mixing
Figure 5.7: Average cylinder pressure comparison for X0D with dierent mixing
36
Table 5.4 summarizes the combustion eciency, as well as the emission data for
HC and CO in the exhaust gas. The results obtained from the simulations with the
dierent mixing models are compared to the experiment. The experiment revealed
a combustion eciency of 95.3 %. This is only reproduced well by the simulation
with the IME model. The case with the GME model reveals a combustion eciency
of only 92.1 %. An even lower combustion eciency of 91.6 % is obtained with
no mixing. As shown in Table 5.4, the amount of HC and CO in the exhaust
gas is reproduced best by the IME model. The CO emissions, which are rather
challenging to predict, are slightly overestimated in all the dierent simulations.
This is possibly related to a post-oxidation of CO in the exhaust manifold in the
experiment, considering the good match between simulation and experiment in
regard to HC emissions and combustion eciency. For the test case investigated,
the experiment yielded no soot and just 0.21 g/kg fuel NO
x
in the exhaust gas,
which is typical for HCCI combustion [12, 55, 93]. Therefore, soot and NO
x
formation were not considered in the simulations.
IME model GME model No mixing Exp.
Comb. e. 95.6 % 92.5 % 91.6 % 95.3 %
HC 24.6 g/kg fuel 26.9 g/kg fuel 61.2 g/kg fuel 30.1 g/kg fuel
CO 185.0 g/kg fuel 300.0 g/kg fuel 204.0 g/kg fuel 68.6 g/kg fuel
Table 5.4: Combustion eciency in percent and emission data for HC and CO in
g/kg fuel in the exhaust gas. Comparison for simulations with dierent mixing
The IME model matches the experiment best in terms of pressure and heat-
release rate curves, as well as in terms of combustion eciency and pollutant
formation. The causes for these deviations and their implications is discussed in
the following.
The evolution of the fuel mixture fraction in AC-FluX and X0D for the IME
and GME model for six representative zones is shown in Fig. 5.8. The fuel mixture
fractions for the case without mixing were omitted, because they are constant in
X0D. The zones were initialized at 660.0 degs. CA aTDC. Therefore, all distribu-
tions are identical at this point. Subsequently, the distribution narrows down due
to mixing. The fuel mixture fraction in each zone in X0D with the GME model
follows the trend of the evolution of the corresponding zone in AC-FluX. However,
the solution in X0D with the GME model increasingly deviates from the solution
of AC-FluX. The reason for this behavior is that the application of only one global
mass exchange rate combined with the equal mass of all zones actually drives the
solution to an equally spaced distribution in phase space. The distribution in X0D
37
evolves towards equal dierences between the zones on the globally averaged mix-
ing time scale of the CFD solution. In summary, the mixing model using the least
square t for the global mass exchange rate yields the correct average mixing time
scale, but drives the solution to an articial distribution. In contrast, the evolu-
tion of the fuel mixture fraction in X0D with the IME model is identical to that
of AC-FluX. The mixing model keeps the mixing and distribution of the phase
variable consistent between AC-FluX and X0D at all times. Therefore, the mixing
model exactly replicates the mixing of the phase variable between the zones in
phase space.
660 680 700 720 740 760 780 800 820 840 860
Crank Angle [degs.]
0.010
0.015
0.020
0.025
F
u
e
l

M
i
x
t
u
r
e

F
r
a
c
t
i
o
n

[
-
]
AC-FluX
X0D, IME
X0D, GME
Figure 5.8: Comparison of fuel mixture fraction for AC-FluX to X0D with the
individual mass exchange rates and X0D with the mass exchange rates from
least square t for six selected zones
In Fig. 5.9, the temperature evolution in the same six representative zones for
AC-FluX and X0D for the simulation with the IME model, the GME model, and
no mixing is displayed. As one would expect, starting from the IME model, the
temperature dierences between minimum and maximum temperatures are more
extreme with the GME model and most extreme for the case without mixing. The
case without mixing ignites rst, followed by the case with the GME model and the
one with the IME model. The temperature increase for the coldest zones, however,
is lesser the less elaborate the treatment of mixing is. This results in overall
decreasing combustion eciencies, as already shown in Table 5.4. In summary, the
IME model captures the correct ignition timing and the subsequent heat release
best (see also Fig. 5.6), and keeps the temperature between AC-FluX and X0D
identical. This shows the signicance of the mixing even in this example with a
fairly homogeneous charge. In order to get the heat release captured correctly, the
temperature and species mass fractions have to be correct, which again can only
38
be achieved when the mixing is described properly.
The interactive coupling of the three-dimensional CFD code AC-FluX and the
multi-zone model X0D further allows to locate the physical position of all the zones
in the CFD domain. In the context of Fig. 5.9, Fig. 5.10 shows a temperature
isosurface of 1200.0 K colored by zone aliation at four dierent crank angles for
the simulation with the IME model. Shortly after auto-ignition at 718.00 degs.
CA aTDC, the temperature isosurface consists only of CFD cells belonging to
the rst three hottest zones. Subsequently, CFD cells of more and more zones
compose the temperature isosurface until every zone contributes at 728.00 degs.
CA aTDC. In case this model is applied in the framework of an HCCI engine
development programme, this feature may in particular lead to some interesting
observations regarding the sources of incomplete combustion in HCCI engines.
For the test case investigated, three dierent hot spots can be located where auto-
ignition takes place between 718.00 degs. CA aTDC and 720.00 degs. CA aTDC.
Incomplete combustion occurs in the coldest zones.
39
680 690 700 710 720 730 740 750 760
Crank Angle [degs.]
800
1000
1200
1400
1600
1800
2000
T
e
m
p
e
r
a
t
u
r
e

[
K
]
AC-FluX
X0D, IME
(a) Individual mass exchange rates
680 690 700 710 720 730 740 750 760
Crank Angle [degs.]
800
1000
1200
1400
1600
1800
2000
T
e
m
p
e
r
a
t
u
r
e

[
K
]
AC-FluX
X0D, GME
(b) Mass exchange rates from least square t
680 690 700 710 720 730 740 750 760
Crank Angle [degs.]
800
1000
1200
1400
1600
1800
2000
T
e
m
p
e
r
a
t
u
r
e

[
K
]
AC-FluX
X0D, no mixing
(c) No mixing
Figure 5.9: Comparison of temperature for AC-FluX to X0D with dierent mixing
models for six selected zones
40
Crank angle = 718.0 degs.
Figure 5.10: Temperature isosurface of 1200.0 K colored by zone aliation at
four dierent crank angles
41
5.3 Concluding Remarks
A zero-dimensional multi-zone code (X0D) was interactively coupled to a CFD
code (AC-FluX) to solve the chemistry for HCCI combustion. Two models for
the mass exchange between zones in phase space based on the CFD solution were
presented and applied to an HCCI engine test case. The IME model, where in-
dividual mass exchange rates for each zone are computed, replicates the CFD
solution identically in terms of mixing. The GME model is less accurate with de-
viations from the ideal solution. The strong deviations of the results without any
mixing emphasize the importance of incorporating mixing. The dierences in the
results for the dierent models are expected to be more pronounced for cases with
greater stratication. An approach for the consistent modeling of the exchange
of source terms between the CFD code and the zero-dimensional multi-zone code
was proposed.
42
6 Systematic Reduction of the
Interactively Coupled
CFD-Multi-Zone Approach towards a
Stand-alone Multi-Zone Model
Chapter 4 depicts the exchange of information between the CFD code and the
multi-zone model. Two information streams and an appropriate treatment of the
source terms establish the interactive coupling of both codes. Chapter 5 contains
its validation against HCCI combustion in a gasoline single-cylinder research en-
gine. In this chapter, the interactive coupling is rst extended to account for PCCI
combustion in Sec. 6.1. Following this, a methodology is presented in Sec. 6.2 of
how the information that is fed from the CFD code to the multi-zone model can be
accounted for within the multi-zone model. This allows for decoupling the multi-
zone model from the CFD code and leads to a systematic reduction of the interac-
tively coupled CFD-multi-zone approach towards a stand-alone multi-zone model
for PCCI combustion. In detail, Subsecs. 6.2.1 and 6.2.2 describe an appropriate
modeling of the sink and source terms (heat transfer to the walls, fuel injection
and vaporization) within the multi-zone model. In addition, Subsec. 6.2.3 presents
a simplied mixing model within the multi-zone model. Finally, Subsec. 6.2.4 out-
lines further reduction potential. Section 6.3 then discusses the computational ac-
curacy and eciency of both the interactively coupled CFD-multi-zone approach
and the stand-alone multi-zone model.
43
6 Systematic Reduction towards a Stand-alone Multi-Zone Model
6.1 Extended Interactive Coupling for PCCI
Combustion
Correctly capturing the physics and chemistry underlying partially premixed
compression ignition in Diesel engines poses a tremendous modeling challenge.
Whereas all combustion modeling relies on adequate descriptions of uid dynam-
ics, liquid-phase, and chemistry, capturing the phenomena pertaining to partially
premixed Diesel combustion stretches the limits in each of these areas. Ignition
timing is governed both by turbulent mixing and chemistry. Since Diesel fuel dis-
plays signicant low-temperature chemistry in the thermodynamic ranges relevant
for engine operation, it is necessary that the reaction mechanism used to model
heat release is capable of capturing this. Moreover, the very advanced injection
timings required to yield a suciently homogeneous mixture at the time of ignition
are likely to lead to substantial wall impingement. The latter is a rather complex
process to model.
For PCCI combustion, the computational cells in the physical CFD domain are
sorted in ascending order in terms of the mean fuel mixture fraction

Z
fuel
, meaning
that the phase space is one-dimensional in this case. At the initialization point, the
distribution of the mean fuel mixture fraction

Z
fuel
is mapped on the mean articial
scalar

S. From this transition point on, the sorting procedure is continued based on
the mean articial scalar

S. This is done in order to account for fuel injection and
vaporization being moderately coupled with combustion chemistry during PCCI
combustion (in contrast to being completely uncoupled during HCCI combustion).
Fuel vaporization also takes place after initialization of the interactive coupling.
In this case, the mean fuel mixture fraction

Z
fuel
is not conserved. Tracing a
conserved scalar is, however, a prerequisite for the mixing model (see Subsec. 4.2.1).
As mentioned above, the conservation equation for the mean artical scalar does
not include any source terms (see Eq. (3.40)). Nevertheless, most of the fuel
mixture formation process is already captured with the mean articial scalar

S,
when mapping the distribution of the mean fuel mixture fraction

Z
fuel
on it at the
initialization point. This is due to the fact that the initialization of the interactive
coupling usually takes place close to end of injection during PCCI combustion (see
also Subsec. 7.1.2). Figure 6.1 schematically illustrates how the AC-FluX CFD
code is used to initialize the zones in X0D.
44
6.1 Extended Interactive Coupling for PCCI Combustion
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

740
760
780
800
820
840
860
880
900
Upon X0D initialization, sorting of
the physical domain in ascending
order in terms of the phase
variable (fuel mix. fraction).
Definition of a corresponding
set of zones in the multi-zone model
Crank Angle = -19.0 degs. aTDC
Zone
numbers
T
e
m
p
e
r
a
t
u
r
e

[
K
]
Figure 6.1: Schematic illustration of the zone initialization in the multi-zone
model for PCCI combustion
Mixing is incorporated into the multi-zone model based on the mean articial
scalar

S. The change for the mean articial scalar

S
i
from CFD time step n 1
to CFD time step n in the individual zones can be described as
m
i
S
n
i

S
n1
i
t
n
=
2
k=1
m
ik
S
n1
k

S
n1
i
. (6.1)
Equation (6.1) is then used instead of Eq. (4.1).
Within the extended interactive coupling, vaporization of fuel in the multi-zone
model is accounted for by averaging the source term
s
fuel
in Eq. (3.39) for each
zone in the CFD code. This averaging procedure yields the source term
s
ij,j=fuel
in Eqs. (2.1) and (2.2) and is linked to the source term for the spray in Eq. (4.7)
according to
q
s
i
=
nsp
j=1

s
ij
h
vj
with
s
ij
= 0, j = fuel . (6.2)
45
6.2 Reduction Methodology
6.2.1 Heat Transfer to the Walls
In the multi-zone model, heat transfer to the walls is accounted for by Eq. (2.4).
Instead of using CFD information to calculate the heat transfer coecient h
wall,l,i
,
it can also be modeled with the Woschni correlation [102], reading
h
wall,l,i
= h
wall
= 3.26 b
0.2
p
0.8
T
0.53
av.
(C
1
v
piston
)
0.8
, (6.3)
where b is the bore of the engine, p the mean cylinder pressure, T
av.
the average
in-cylinder temperature, and v
piston
the mean piston speed. The constant C
1
is
adjustable and is set to 0.5. Since the average in-cylinder temperature T
av.
is used
within the Woschni correlation and h
wall,l,i
does not depend on any specic wall, the
heat transfer coecient is equal for all zones and walls, leading to h
wall,l,i
= h
wall
.
For the specic engine investigated, the area A
wall,l,i
of wall l belonging to zone
i has to be taken from a preceding simulation with the interactively coupled CFD-
multi-zone approach so that this information can be incorporated into a successive
simulation with the stand-alone multi-zone model.
6.2.2 Fuel Injection and Evaporation
A preceding simulation with the interactively coupled CFD-multi-zone approach
also allows to determine the amount of fuel that is injected and vaporizes in each
CFD zone. This information can then be incorporated into a successive simulation
with the stand-alone multi-zone model (see source term
s
ij,j=fuel
in Eqs. (2.1)
and (2.2)).
6.2.3 Simplied Mixing Model
The simplied mixing model within the stand-alone multi-zone model assumes
that the previously introduced mass exchange rate m
ik
of zone i with neighboring
zone k (see Eqs. (2.1) and (2.2)) is proportional to the wall heat transfer coecient
h
wall
and the area A both zones share [39]. This is an analogy to the well known
empirical Lewis law which assumes a similarity between heat and mass transfer [7].
The heat transfer to the walls and the mass exchange between the zones are both
induced by the turbulence intensity so that they can be related to each other. The
corresponding equation reads
m
ik
= r
MB
A
m
i
+ m
k
(m
i
c
pi
+ m
k
c
pk
)
h
wall
. (6.4)
46
6.3 Computational Accuracy and Eciency
In Eq. (6.4), c
pi
is the average heat capacity at constant pressure of zone i. r
MB
is a parameter that is adjusted to yield the desired level of mixing. The exact
choice of r
MB
is discussed in Subsec. 7.2.2 below. The area A is computed from
the volumes V
i
and V
k
of the two zones i and k according to
A = 4
3
4
(V
i
+ V
k
)
2
3
. (6.5)
6.2.4 Further Reduction Potential
A possible further reduction beyond the decoupling of the multi-zone model from
the CFD code could be achieved by reducing the chemical mechanism applied in
the PCCI simulations (see Sec. 2.3). The work by M uller et al. [54] describes a
reduced chemical mechanism for n-heptane with only four global reactions and ve
global chemical species. Two of these four global reactions account for the high-
temperature branch and the other two for the low-temperature branch of the igni-
tion kinetics. Using this mechanism instead of the one by Peters et al. [67] would
highly reduce the number of partial dierential equations that need to be solved in
the multi-zone model (see Subsec. 6.3.3 below). However, no NO-submechanism
can be integrated into this reduced mechanism, because it does not contain the
necessary chemical species.
6.3.1 Interactively Coupled CFD-Multi-Zone Approach
The interactively coupled CFD-multi-zone approach eciently separates the solu-
tion of the ow eld from the solution of the chemistry. During one time step of
the CFD code, the multi-zone code solves Eqs. (2.1), (2.2), and (2.3) with time
steps that can be much smaller, depending on the chemistry. In this way, the time
scales of the chemistry and the uid dynamics are decoupled. Thus, the approach
does not need to simplify the highly nonlinear chemistry, allowing for any detailed
reaction mechanisms to be used. These mechanisms may also include pollutant
formation and destruction. This procedure yields an approach that includes an ac-
curate description of PCCI chemistry with only a small penalty in computational
cost, since even during auto-ignition, the CFD time step has hardly to be reduced
compared to an inert computation of the chemistry.
Decoupling the time scales of the chemistry and the uid dynamics has proven
successful in predicting conventional Diesel engine combustion in the framework
of Representative Interactive Flamelets (RIF).
47
6.3.2 Stand-alone Multi-Zone Model
In comparison to the interactively coupled CFD-multi-zone approach, the stand-
alone multi-zone model does not solve for the solution of the ow eld. This
yields a signicant advantage in computational costs. However, modeling the CFD
information according to the procedure outlined in Sec. 6.2 reduces the accuracy
of the simulation outcome. This reduction of accuracy is discussed further in
Subsecs. 7.2.2 and 7.2.2.
6.3.3 Run Time Comparison
The computations with both the interactively coupled CFD-multi-zone approach
and the stand-alone multi-zone model were carried out on a Dell PowerEdge 1950
with two Intel Xeon 5160 CPUs (Dual-Core with 3.0 GHz). For this, both ap-
proaches written in FORTRAN 77 were compiled with the Intel Fortran Compiler
9.1.043. They were executed for one high-pressure engine cycle. Subsections 7.1.2
and 7.1.2 below provide a detailed description of the numerical setups. Table 6.1
contains a run time comparison for both approaches, depending on the number
of zones or the number of partial dierential equations, respectively. Thirteen
partial dierential equations are solved in the CFD code (mass, momentum, total
enthalpy, active species stream, mean fuel mixture fraction, and articial scalar
conservation, k-epsilon-equations; see Sec. 3.5). The equations in the CFD code are
evaluated on each grid cell. In the multi-zone model, 40 partial dierential equa-
tions (species mass fractions, temperature, and pressure; see Secs. 2.2 and 2.3) are
solved per zone.
For the interactively coupled CFD-multi-zone approach, the CPU time costs are
approximately 24 hours for the thirteen dierential equations solved in the CFD
code. The CPU time for the multi-zone model depends on the total number of
zones, varying between less than 10 minutes and 12 hours. It roughly scales with
the number of equations cubic, indicating that the stiness of the Jacobian in the
multi-zone model is nearly independent of the number of zones.
It can be seen from Table 6.1 that the stand-alone multi-zone model is signi-
cantly faster in terms of computational time than the interactively coupled CFD-
multi-zone approach. This is due to the fact that the mass exchange rates m
ik
computed within the stand-alone multi-zone model have a smoother progression
throughout the simulations in comparison to the interactively coupled CFD-multi-
zone approach.
48
Number of zones/
number of eqs.
for
CFD-multi-zone
approach or
stand-alone
multi-zone model
CPU time for
interactively
coupled
CFD-multi-zone
approach
CPU time for stand-
alone multi-zone model
5 zones/
13+200 eqs. or 200
eqs., respectively
24.1 hours 0.2 minutes
10 zones/
13+400 eqs. or 400
eqs., respectively
15 zones/
13+600 eqs. or 600
eqs., respectively
20 zones/
13+800 eqs. or 800
eqs., respectively
25 zones/
13+1000 eqs. or
1000 eqs.,
respectively
Table 6.1: Run time comparison for the interactively coupled CFD-multi-zone
approach and the stand-alone multi-zone model. Depending on the number
of zones, the left column contains the number of partial dierential equations
that are solved in the CFD-multi-zone approach or the stand-alone multi-zone
model, respectively. The second and third column contain the corresponding
CPU times for one high-pressure engine cycle simulation
49
7 Validation of the Systematic
Reduction
7.1.1 Experimental Setup
Engine
At the Institut f ur Technische Verbrennung at RWTH Aachen University, Ger-
many, experiments were carried out with a 1.9l GM Fiat Diesel engine. This en-
gine is equipped with a second-generation Bosch Common-Rail injection system.
The mounting of the engine on the test bench is shown in Fig. 7.1.
torque meter
EC - dynamometer
engine speed
AVL
smoke meter
emission analysis
exhaust gas
fuel meter
pressure
sensor
fuel
conditioner
water & oil
conditioner
air
conditioner
airflow meter
cam &
crankshaft
signal
NO
NO
NO
HC
CO
CO
O
2
X
2
2
air
FI2RE
rpm
GM Fiat
Diesel engine
ES1000
IMEP, CA50
SOI, EGR
m
Fuel
EDC16
Figure 7.1: Engine test bench
51
7 Validation of the Systematic Reduction
All relevant engine data are provided in Table 7.1.
Engine type: Four-cylinder
research engine
Displacement volume: 1910 cm
3
Bore: 82.0 mm
Stroke: 90.4 mm
Bore distance: 90.0 mm
Cylinder block height: 236.5 mm
Connecting rod length: 145.0 mm
Compression height: 46.5 mm
Valve diameter I/E: 28.5/25.5 mm
Compression ratio: 17.5:1
Output @ engine speed: 110 kW/4000 rpm
Torque @ engine speed: 315 kW/2000 rpm
Idle speed: 850 rpm
Maximum speed: 5100 rpm
Injection System: 7-hole injector
Included spray angle: 100.0 degs.
Hole diameter: 0.141 mm
Flow number: 440 cm
3
/30s @ 100 bar
Swirl number: 2.5
Fuel: Diesel
Table 7.1: GM Fiat engine specications
A more detailed description regarding the engine, the test cell equipment, and
the injection rate measurements can be found in Vanegas et al. [98].
Operating Conditions
The engine was operated at part-load conditions with a speed of 2000 rpm and
an external EGR rate of 15, 30, or 45 %, respectively. The external EGR rate
was cooled to about 350 K before mixing with the fresh air in the intake section.
Ten variations of start of injection (SOI) were investigated for each external EGR
rate, leading to 30 dierent experiments. The SOI variation started at -45.7 degs.
CA aTDC and lasted until -0.7 degs. CA aTDC, with a time shift of 5.0 degs.
CA between two consecutive SOIs. For all these experiments, the total fuel mass
injected was 10.2 mg/cycle, with an injection duration of 8.8 degs. CA. Two
additional SOI variations were carried out with an external EGR rate of 30 % and
52
a reduced total fuel mass injected of 6.8 mg/cycle, or an increased total fuel mass
injected of 13.6 mg/cycle, respectively, leading to another 20 dierent experiments.
In these additional experiments, the injection duration was 5.9 or 11.7 degs. CA,
depending on the total fuel mass injected. The rail pressure was 700 bar for
all 50 dierent engine operating conditions. The engine operating conditions are
summarized in Table 7.2.
Engine speed: 2000 rpm
Rail pressure: 700 bar
Fuel mass injected: 6.8, 10.2, or 13.6 mg/cycle
Relative air/fuel ratio: between 5.3 and 1.6
External EGR rate: 15, 30, or 45 %
Start of Injection (SOI): -45.7 degs. CA aTDC
-40.7 degs. CA aTDC
-35.7 degs. CA aTDC
-30.7 degs. CA aTDC
-25.7 degs. CA aTDC
-20.7 degs. CA aTDC
-15.7 degs. CA aTDC
-10.7 degs. CA aTDC
-5.7 degs. CA aTDC
-0.7 degs. CA aTDC
Table 7.2: Engine operating conditions
In the remainder of this chapter, the interactively coupled CFD-multi-zone ap-
proach, as well as its reduction to the stand-alone multi-zone model, is exemplarily
validated against ve selected operating conditions. This includes ve variations
of SOI (-35.7, -30.7, -25.7, -20.7, and -15.7 degs. CA aTDC) at an external EGR
rate of 30 % which are referred to as timings 1, 2, 3, 4, and 5 (TS-1, TS-2, TS-3,
TS-4, and TS-5) in the following.
7.1.2 Numerical Setup
Interactively Coupled CFD-Multi-Zone Approach
Computations with the interactively coupled CFD-multi-zone approach were per-
formed for all engine operating conditions presented in Subsec. 7.1.1. All com-
putations started from intake valve closure (IVC) at -165.6 degs. CA aTDC and
ended at exhaust valve opening (EVO) at 149.1 degs. CA aTDC. In the rst part
of the CFD simulation, AC-FluX was run alone. For every operating condition,
53
the coupling to X0D was initialized 7.0 degs. CA after start of injection. The
starting solution at IVC was initialized with pressure and temperature taken from
the experiments. The internal EGR rate was assumed to be 5 %. The velocity
eld was initialized with a swirl number of 2.5.
Two dierent computational meshes were used throughout the simulations to
provide an optimal computational mesh for any piston position. One grid was
used for the compression phase and one for the combustion and expansion phase.
This procedure allowed for rearranging the cells shortly before the beginning of the
combustion event. In the computational mesh for the combustion and expansion
phase, in comparison to the one for the compression phase, the mesh resolution in
the bowl region was rened. For both meshes, a cell layer removal technique was
applied in the cylinder region throughout the simulations. This was done in order
to account for the compression and expansion of the grid cells along the cylinder
axis due to the piston movement. The remap between both grids took place at
-37.0 degs. CA aTDC. There is no sensitivity on the results when advancing or
retarding the remap of the solution.
An outline of the computational mesh for the combustion phase is given in
Fig. 7.2. The simulation used a sector grid representing 1/7th of the combustion
chamber, thereby taking advantage of the axial symmetry with respect to the
placement of the nozzle holes. In Fig. 7.2, the cyclic boundaries are removed for
the sake of a clear insight into the bowl. The mesh size was 52704 cells at top dead
center (TDC). A sensitivity study revealed that this was a sucient resolution for
all cases.
The wall temperatures were set based on experimental experience and held con-
stant during the simulations. The spray model parameters are always subject to
adjustment and were adapted such that the experimental and simulated pressure
curve for TS-3 matched. Otherwise they were held unchanged for all additional
simulations. The parameters used in this work are within the recommended pa-
rameter range proposed by Weber et al. [99].
Stand-alone Multi-Zone Model
For all engine operating conditions presented in Subsec. 7.1.1, computations were
also performed with the stand-alone multi-zone model. For reasons of consistency,
the initial conditions, start and end of simulation, number of zones, etc., were held
identical to the simulations carried out with the interactively coupled CFD-multi-
zone approach.
TS-3 was used as a calibration case for incorporating information from the inter-
actively coupled CFD-multi-zone approach into the stand-alone multi-zone model
(
s
ij
and A
wall,l,i
, see Eqs. (2.1), (2.2), and (2.4)). The information obtained for
TS-3 from the simulation with the interactively coupled CFD-multi-zone approach
54
was used for all experiments investigated with the stand-alone multi-zone model.
This introduces a small error into the other stand-alone multi-zone model simula-
tions (e.g., TS-1, TS-2, TS-4, and TS-5). However, this error is small, since the
calibration conditions of TS-3 do not deviate signicantly from all other engine
operating conditions.
Figure 7.2: Computational grid at -25.0 degs. CA aTDC. Cyclic boundaries are
removed for illustration purposes
In this section, the interactively coupled CFD-multi-zone approach is rst val-
idated against experimental data in terms of pressure curves, IMEP
HP
, CA10,
CA50, CO emissions, combustion eciency, and NO/NO
x
emissions for all ve
aforementioned SOI timings (TS-1, TS-2, TS-3, TS-4, and TS-5). After this, TS-3
is considered in detail. The exchange of information between the CFD code and
the multi-zone model within the interactively coupled CFD-multi-zone approach is
investigated. Following this, the systematic reduction of the interactively coupled
CFD-multi-zone approach towards the stand-alone multi-zone model is exemplar-
ily shown for TS-3. Finally, simulation results obtained with the stand-alone
multi-zone model for all ve SOI timings are compared to experimental data.
55
7.2.1 Interactively Coupled CFD-Multi-Zone Approach
Sensitivity Study
A sensitivity study of the local in-cylinder quantities (e.g., temperature, fuel mass
fraction), the ignition timing, and the overall heat-release rate on the number of
zones was performed rst. Beyond nz = 15, the change of the local in-cylinder
quantities, the ignition timing, and the overall heat-release rate was insignicant,
and thus a number of zones of nz = 15 was chosen for all subsequent simulations.
SOI Variation
Figure 7.3 compares the pressure curves and the corresponding indicated mean
eective pressures for the high-pressure part of the engine cycle (IMEP
HP
) obtained
from the simulations with the interactively coupled CFD-multi-zone approach with
the results from the experiments for the complete SOI variation TS-1 through TS-
5. It becomes evident that with retarded SOI timings the maximum pressure peak
decreases, while the IMEP
HP
increases. For all ve cases, the simulations are in
good agreement with the experiments. There is only a slight overestimation of the
maximum peak pressure in all simulations. The qualitative trends are reproduced
well by the simulations.
-20 -10 0 10 20
Crank Angle [degs.]
20
30
40
50
60
70
80
90
P
r
e
s
s
u
r
e

[
b
a
r
]
TS-1, Exp.
TS-2, Exp.
TS-3, Exp.
TS-4, Exp.
TS-5, Exp.
TS-1, Sim.
TS-2, Sim.
TS-3, Sim.
TS-4, Sim.
TS-5, Sim.
TS-1 TS-2 TS-3 TS-4 TS-5
SOI Variation
3
3.5
4
4.5
5
I
M
E
P
H
P

[
b
a
r
]
Experiment
Sim. AC-FluX-X0D
Figure 7.3: Average cylinder pressure (left) and IMEP
HP
(right) for the SOI
variation. Comparison between simulation with the interactively coupled CFD-
multi-zone approach and experiment
Figure 7.4 shows the crank angles at which 10 % (CA10) and 50 % (CA50) of the
total injected fuel mass is burned for all simulations and experiments. While CA10
is predicted almost perfectly, the simulated CA50 has a very slight but acceptable
deviation from some of the experimental results.
56
The CO emissions in g/kg fuel in the exhaust gas and the combustion eciency
in percent for the complete SOI variation are summarized in Fig. 7.5. The CO
emissions are rather challenging to predict. They are slightly overestimated in
the simulations. This leads to an underestimation of the combustion eciency,
since the simulated HC emissions (not shown here) are roughly of the same order
of magnitude as the experimentally obtained ones for all cases. All in all, the
qualitative and quantitative agreement over the whole SOI variation is reasonable
for both CO emissions and combustion eciency.
SOI Variation
12.5
-10
-7.5
-5
-2.5
C
A
1
0

[
d
e
g
s
.

C
A
]
Experiment
Sim. AC-FluX-X0D
SOI Variation
12.5
-10
-7.5
-5
-2.5
C
A
5
0

[
d
e
g
s
.

C
A
]
Experiment
Sim. AC-FluX-X0D
Figure 7.4: CA10 (left) and CA50 (right) for the SOI variation. Comparison
between simulation with the interactively coupled CFD-multi-zone approach
and experiment
SOI Variation
0
50
100
150
C
O

E
m
i
s
s
i
o
n
s

[
g

/

k
g

f
u
e
l
]
Experiment
Sim. AC-FluX-X0D
SOI Variation
94
95
96
97
98
99
100
C
o
m
b
u
s
t
i
o
n

E
f
f
i
c
i
e
n
c
y

[
%
]
Experiment
Sim. AC-FluX-X0D
Figure 7.5: CO emissions in g/kg fuel in the exhaust gas (left) and combustion ef-
ciency in percent (right) for the SOI variation. Comparison between simulation
with the interactively coupled CFD-multi-zone approach and experiment
Finally, Fig. 7.6 presents a comparison between numerically and experimentally
obtained NO/NO
x
emissions in g/kg fuel in the exhaust gas for the complete SOI
variation. There is good agreement. Note that the chemical mechanism used in
57
the simulations only accounts for thermal NO (see also Sec. 2.3), while both NO
and NO
2
emissions are measured in the experiments. NO and NO
2
emissions are
usually grouped together as NO
x
emissions, with NO being the predominant oxide
of nitrogen produced inside the engine cylinder [41].
SOI Variation
0
20
40
60
N
O
/
N
O
x

E
m
i
s
s
i
o
n
s

[
g

/

k
g

f
u
e
l
]
Experiment (NO
x
emissions)
Sim. AC-FluX-X0D (NO emissions)
Figure 7.6: NO/NO
x
emissions in g/kg fuel in the exhaust gas for the SOI varia-
tion. Comparison between simulation with the interactively coupled CFD-multi-
zone approach and experiment
Exchange of Information between the CFD Code and the Multi-Zone
Model for TS-3
For TS-3, Fig. 7.7 shows the fuel mixture fraction and the temperature distribution
in ve representative AC-FluX zones and in their corresponding X0D zones around
top dead center ring. At time of X0D initialization (-18.7 degs. CA aTDC), the
mean fuel mixture fractions

Z
fuel,i
vary roughly between 0.0 and 0.125. With
30 % external and 5 % internal EGR, the stoichiometric mixture fraction of Diesel
fuel is approximately Z
Diesel,st
0.053. Thus, some in-cylinder zones are fuel-
rich, whereas most others are fuel-lean. Throughout the simulation, the mean fuel
mixture fraction stratication decreases due to mixing and diusion processes. The
evolution of the fuel mixture fractions in X0D is almost identical to that of AC-
FluX. The mixing model with the individual mass exchange rates again accurately
replicates the mixing between the zones in phase space and keeps the mixing and
distribution of the phase variable consistent between the CFD code and the multi-
zone model at all times. Hence, mixing and diusion processes are captured well
in X0D. The evolution of the temperature in AC-FluX and in X0D for the same
ve representative zones also matches well. The exchange of information between
AC-FluX and X0D as outlined in Chapter 4 and Sec. 6.1 keeps the temperature
evolution in both codes identical. Only small dierences exist. However, these
58
deviations are in an acceptable range. This validates the interactive coupling of
both codes. For TS-3, the zones with the highest peak temperatures are also the
ones with the highest fuel mixture fractions. Figure 7.7 shows that these zones
auto-ignite rst. Subsequently, the other zones auto-ignite or are ignited through
heat and mass transfer from other zones, respectively. The coldest zone contains
the least amount of fuel and has the smallest increase in temperature.
The left hand side of Fig. 7.8 compares the pressure trace obtained from the
experiment with the ones from AC-FluX and X0D. The pressure traces of AC-FluX
and X0D are in perfect agreement over the whole crank angle range, especially
during the heat-release phase. Only negligible dierences exist. The simulated
pressure curves reveal that auto-ignition is captured well in the simulation. The
peak pressures match, too.
-20 -10 0 10 20 30 40 50
Crank Angle [degs.]
0
0.02
0.04
0.06
0.08
0.10
0.12
M
i
x
t
u
r
e

F
r
a
c
t
i
o
n

[

-

]
Sim. (AC-FluX)
Sim. (X0D)
-20 -10 0 10 20 30 40 50
Crank Angle [degs.]
500
750
1000
1250
1500
1750
2000
2250
2500
2750
T
e
m
p
e
r
a
t
u
r
e

[
K
]
Sim. (AC-FluX)
Sim. (X0D)
Figure 7.7: Comparison of fuel mixture fraction (left) and temperature (right)
for X0D and AC-FluX for ve selected zones for TS-3
-20 -10 0 10 20
Crank Angle [degs.]
20
30
40
50
60
70
80
90
P
r
e
s
s
u
r
e

[
b
a
r
]
Experiment
Sim. (AC-FluX)
Sim. (X0D)
-20 -15 -10 -5 0
Crank Angle [degs.]
0
50
100
150
200
H
e
a
t
-
R
e
l
e
a
s
e

R
a
t
e

[
J
/
d
e
g
.
]
Experiment
Sim. (X0D)
Figure 7.8: Average cylinder pressure (left) and average apparent net heat-release
rates (right) for TS-3. Comparison between simulation and experiment. Simu-
lation results are extracted from the CFD code and/or the multi-zone model
59
The right hand side of Fig. 7.8 displays the corresponding average apparent net
heat-release rates. The comparison of the apparent net heat-release rates further
shows that rst and second stage ignition, as well as the subsequent heat release,
are reproduced well by the simulation. Only small discrepancies with respect to
overall burning time exist. The CA50 obtained from the experiment is -9.0 degs.
CA aTDC. The simulation yielded a CA50 of -8.7 degs. CA aTDC (see also
Fig. 7.4).
7.2.2 Stand-alone Multi-Zone Model
Systematic Reduction for TS-3
Subsections 6.2.1, 6.2.2, and 6.2.3 above outline how to systematically model CFD
information within the multi-zone model. Among them, Subsec. 6.2.1 deals with
the treatment of heat transfer to the walls. For TS-3, Fig. 7.9 presents a comparison
of the integral heat transfer to all cylinder walls for the interactively coupled CFD-
multi-zone approach and the stand-alone multi-zone model. Both are in good
agreement.
-20 -10 0 10 20 30 40 50
Crank Angle [degs.]
0
1
2
3
4
H
e
a
t
l
o
s
s

[
J
/
d
e
g
.
]
Sim. AC-FluX-X0D
Sim. stand-alone X0D
Figure 7.9: Comparison of the integral heat transfer to all cylinder walls for the
interactively coupled CFD-multi-zone approach and the stand-alone multi-zone
model for TS-3
Subsections 6.2.2 and 6.2.3 deal with the fuel injection and vaporization, as well
as with a simplied mixing model, within the stand-alone multi-zone model. In
Fig. 7.10, the distribution of the vaporized fuel over the zones as obtained from
the simulation with the interactively coupled CFD-multi-zone approach is shown
for TS-3. Further on, Fig. 7.10 shows the distribution of the vaporized fuel over
the zones as incorporated into the corresponding simulation with the stand-alone
multi-zone model. Both are almost identical. In this context, Fig. 7.11 presents
60
the fuel mixture fraction distribution for the interactively coupled CFD-multi-zone
approach and the stand-alone multi-zone model for the same ve representative
zones already shown in Fig. 7.7. The incorporation of information on fuel injection
and vaporization into the stand-alone multi-zone model and the simplied mixing
model lead to a fuel mixture fraction distribution that is qualitatively similar to
the one from the three-dimensional approach.
Figure 7.10: Distribution of the vaporized fuel over all zones as obtained from
the interactively coupled CFD-multi-zone approach and as incorporated into the
stand-alone multi-zone model for TS-3
-20 -10 0 10 20 30 40 50
Crank Angle [degs.]
0
0.02
0.04
0.06
0.08
0.10
0.12
M
i
x
t
u
r
e

F
r
a
c
t
i
o
n

[

-

]
Sim. AC-FluX-X0D
Figure 7.11: Comparison of fuel mixture fraction for the interactively coupled
CFD-multi-zone approach and the stand-alone multi-zone model for ve selected
zones for TS-3
61
The simulation with the interactively coupled CFD-multi-zone approach allows
for evaluating the choice of the parameter r
MB
in Eq. (6.4). The mass exchange
rate m
ik
of zone i with neighbouring zone k is computed according to Eqs. (4.1)
and (4.4). By inserting the mass exchange rate m
ik
into Eq. (6.4), a corresponding
individual mixing parameter r
MB,ik
can be determined which describes the level of
mixing between zones i and k. Averaging the individual mixing parameters r
MB,ik
yields the parameter r
MB
.
Figure 7.12 displays the average parameter r
MB
around top dead center ring for
TS-3, as accordingly evaluated from the simulation with the interactively coupled
CFD-multi-zone approach and as incorporated into the stand-alone multi-zone
model. For incorporation, r
MB
was set to 75.0 during compression and expansion,
with a linear increase up to 500.0 during injection and a linear decrease back
again to 75.0 after end of injection. Qualitatively, both evolutions are similar.
Altogether, r
MB
decreases after EOI with decaying turbulence intensity after EOI.
The evolution of r
MB
as evaluated from the interactively coupled CFD-multi-zone
approach reveals two peaks. These two peaks are due to rst and second stage
ignition. The corresponding integral average of the whole simulation is equal 169.7.
-10 0 10 20 30 40 50
Crank Angle [degs.]
0
100
200
300
400
500
600
r
M
B

[

-

]
Sim. AC-FluX-X0D
Figure 7.12: Average parameter r
MB
as evaluated from the interactively coupled
CFD-multi-zone approach and as incorporated into the stand-alone multi-zone
model for TS-3
The SOI variation TS-1 through TS-5 was also investigated in a previous work
by Felsch et al. [27], with just one constant value of 100.0 being chosen for r
MB
,
which is of the same order of magnitude as 169.7. In [27], the overall simulation
results for the stand-alone multi-zone model in terms of pressure curves, IMEP
HP
,
CA10, CA50, CO emissions, combustion eciency, and NO/NO
x
emissions were
not as accurate in predicting the experimental results as the ones presented in this
thesis, where the evolution of r
MB
was qualitatively adjusted to the one from the
three-dimensional approach.
62
It should be emphasized, too, that the order of magnitude of the parameter r
MB
strongly depends on the level of premixing. In [39], where more advanced injection
timings were investigated, a constant value of 6.0 was sucient for r
MB
.
Finally, the left hand side of Fig. 7.13 compares the pressure trace obtained from
the stand-alone multi-zone model with the one obtained from the interactively
coupled CFD-multi-zone approach and the experimental data. All three pressure
curves match very well. This also holds for the comparison of the corresponding
average apparent net heat-release rates shown on the right hand side of Fig. 7.13.
-20 -10 0 10 20
Crank Angle [degs.]
20
30
40
50
60
70
80
90
P
r
e
s
s
u
r
e

[
b
a
r
]
Experiment
Sim. AC-FluX-X0D
-20 -15 -10 -5 0
Crank Angle [degs.]
0
50
100
150
200
H
e
a
t
-
R
e
l
e
a
s
e

R
a
t
e

[
J
/
d
e
g
.
]
Experiment
Sim. AC-FluX-X0D
Figure 7.13: Average cylinder pressure comparison (left) and average apparent
net heat-release rate comparison (right) for stand-alone multi-zone model to the
interactively coupled CFD-multi-zone approach and the experiment for TS-3
Figures 7.9, 7.10, 7.11, 7.12, and 7.13 reveal that systematically reducing the in-
teractively coupled CFD-multi-zone approach according to the procedure outlined
in Sec. 6.2 leads to a reliable stand-alone multi-zone model that can be used to
eciently model PCCI combustion. This is further shown in the following.
63
SOI Variation
Similar to Subsec. 7.2.1, Fig. 7.14 rst compares the pressure curves and the cor-
responding IMEP
HP
s obtained from the stand-alone multi-zone model simulations
with the results from the experiments for the complete SOI variation TS-1 through
TS-5. Again, for all cases the experimental pressure curves and the experimental
IMEP
HP
s are reproduced well with the stand-alone multi-zone model. There is
only a slight overestimation of the maximum peak pressure and a slight underesti-
mation of the IMEP
HP
in all simulations. Figure 7.15 shows CA10 and CA50 for
all experiments and simulations. Once more, CA10 and CA50 are in very good
agreement with the experimental results. Figure 7.16 summarizes the CO emis-
sions in g/kg fuel in the exhaust gas and the combustion eciency in percent for
the complete SOI variation. The simulations overestimate the CO emissions and
underestimate the combustion eciency. The qualitative agreement for the CO
emissions and the combustion eciency over the whole SOI variation is good. The
HC emissions (not shown here) are captured reasonably. In Fig. 7.17, a compari-
son between numerically and experimentally obtained NO/NO
x
emissions in g/kg
fuel in the exhaust gas is presented for the complete SOI variation. There is good
agreement.
64
-20 -10 0 10 20
Crank Angle [degs.]
20
30
40
50
60
70
80
90
P
r
e
s
s
u
r
e

[
b
a
r
]
TS-1, Exp.
TS-2, Exp.
TS-3, Exp.
TS-4, Exp.
TS-5, Exp.
TS-1, Sim.
TS-2, Sim.
TS-3, Sim.
TS-4, Sim.
TS-5, Sim.
SOI Variation
3
3.5
4
4.5
5
I
M
E
P
H
P

[
b
a
r
]
Experiment
Figure 7.14: Average cylinder pressure (left) and IMEP
HP
(right) for the SOI vari-
ation. Comparison between simulation with the stand-alone multi-zone model
and experiment
SOI Variation
-12.5
-10
-7.5
-5
-2.5
C
A
1
0

[
d
e
g
s
.

C
A
]
Experiment
SOI Variation
-12.5
-10
-7.5
-5
-2.5
C
A
5
0

[
d
e
g
s
.

C
A
]
Experiment
Figure 7.15: CA10 (left) and CA50 (right) for the SOI variation. Comparison
between simulation with the stand-alone multi-zone model and experiment
SOI Variation
0
50
100
150
C
O

E
m
i
s
s
i
o
n
s

[
g

/

k
g

f
u
e
l
]
Experiment
SOI Variation
94
95
96
97
98
99
100
C
o
m
b
u
s
t
i
o
n

E
f
f
i
c
i
e
n
c
y

[
%
]
Experiment
Figure 7.16: CO emissions in g/kg fuel in the exhaust gas (left) and combus-
tion eciency in percent (right) for the SOI variation. Comparison between
simulation with the stand-alone multi-zone model and experiment
65
SOI Variation
0
20
40
60
N
O
/
N
O
x

E
m
i
s
s
i
o
n
s

[
g

/

k
g

f
u
e
l
]
Experiment (NO
x
emissions)
Sim. stand-alone X0D (NO emissions)
Figure 7.17: NO/NO
x
emissions in g/kg fuel in the exhaust gas for the SOI vari-
ation. Comparison between simulation with the stand-alone multi-zone model
and experiment
7.3 Concluding Remarks
Starting from an interactively coupled CFD-multi-zone approach, a computation-
ally ecient stand-alone multi-zone model was derived. Both the interactively
coupled CFD-multi-zone approach, as well as the stand-alone multi-zone model,
were applied for ve cases of a PCCI operating strategy in a 1.9l FIAT GM Diesel
engine. Overall good agreement between experiments and simulations was ob-
tained for all cases for the interactively coupled CFD-multi-zone approach. These
simulations allowed for the systematic reduction of the interactively coupled CFD-
multi-zone approach towards the stand-alone multi-zone model. The stand-alone
multi-zone model also proved successful in predicting the ve investigated PCCI
combustion cases.
66
8 Gas Exchange
The stationary validation described in the previous chapter is restricted to the
high-pressure part of the engine cycle. The usage of the stand-alone multi-zone
model within a closed-loop controller for PCCI combustion requires that it is
capable of predicting the dependency of the controlled variables on the actuators.
As already mentioned above, the actuators are the SOI, the external EGR rate,
and the total fuel mass injected. The controlled variables are the IMEP and CA50.
The latter is directly obtained from a simulation with the stand-alone multi-zone
model. The former, however, can only be predicted for the time frame from closing
of the intake until opening of the exhaust valves. For this reason, the calculation
of the IMEP
HP
is extended to the gas exchange part of the engine cycle. This
extension is described in the present chapter.
The IMEP of the whole engine cycle is composed of the high-pressure part and
the part describing the losses caused by the gas exchange. From the viewpoint
of automatic control, a model predicting the accumulated losses occurring from
exhaust valve opening until intake valve closing is sucient. Furthermore, as one
aim of this thesis is model reduction, the lowest acceptable approach in terms
of detail level and calculation time is desired. Thus, a mean value model for
the gas exchange is presented in the following. Nevertheless, if desirable, many
commercially available software packages exist for a more detailed calculation of
the gas exchange part of the engine cycle (e.g., WAVE [78] or GT-POWER [33]).
The calculation of the indicated mean eective pressure throughout the gas
exchange (IMEP
GE
) is physically inspired by pumping losses. This approach is
derived from the one proposed in [79]. It was introduced in Homann et al. [44].
The formulation is valid for a constant engine speed. However, for the test case
investigated, this restriction related to the engine speed can be disregarded, be-
cause the model validation is carried out at a constant engine speed of 2000 rpm.
The IMEP
GE
is given by
IMEP
GE
= x
1
(p
ae
p
be
) + x
2
p
ae
p
be
+ p
be
x
3
p
be
T
be
+ x
4
p
be
T
be
+ x
5
T
ae
T
be
+ x
6
. (8.1)
67
8 Gas Exchange
Equation (8.1) includes terms of the volume ow through an orice and through a
throttle, as well as the dependency on the volumetric eciency or on the density
at the intake, respectively. The latter is represented by the fraction p
be
/T
be
, which
is related to the ideal gas law = p/RT = const. p/T. The engine-dependent
constants x
1
, x
2
, x
3
, x
4
, x
5
, and x
6
need to be determined by a tting against
experimental data.
The mean value model for the gas exchange depends on the state directly after
and before the combustion process or the manifolds after and before the engine,
respectively. In Eq. (8.1), these states are denoted with the subscripts
ae
and
be
. Inputs to the model from the exhaust manifold (i.e., p
ae
and T
ae
) can be
taken from the high-pressure calculation with the stand-alone multi-zone model,
while the inputs to the model from the intake manifold (i.e., p
be
and T
be
) become
inuencing variables to the plant model. These inuencing variables can serve as
a disturbance input within the closed-loop control.
After appropriately tting the abovementioned constants x
1
through x
6
, the
approach is capable of predicting the IMEP
GE
within 5 % accuracy for most of
the 50 operating conditions presented in Subsec. 7.1.1. This is shown in Figs. 8.1
and 8.2. Combining the stand-alone multi-zone model X0D with this gas exchange
model therefore leads to a static model, which can be used to determine the static
dependency of the controlled variables IMEP and CA50 on the actuated variables
SOI, external EGR rate, and total fuel mass injected. Inuencing inputs are the
pressure and temperature in the intake manifold.
68
0 10 20 30 40 50
0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
Static Measurement Point
I
M
E
P
G
E

[
b
a
r
]

Measurement
Mean Value Model
Figure 8.1: IMEP
GE
of the gas exchange for all 50 operating conditions pre-
sented in Subsec. 7.1.1. Comparison between mean value model calculation and
experiment
0 10 20 30 40 50
20
15
10
5
0
5
10
15
Static Measurement Point
P
e
r
c
e
n
t
a
g
e
d

D
e
v
i
a
t
i
o
n

[
%
]
Figure 8.2: Percentaged deviation of the calculated IMEP
GE
of the gas exchange
from all 50 operating conditions presented in Subsec. 7.1.1
69
9 System Identication
9.1 Wiener Model
The combination of stand-alone multi-zone model and mean value model for the
IMEP
GE
of the gas exchange does so far not include any dynamics of the con-
trolled variables. Hence, a structure has to be chosen which is suitable for adding
the dynamic aspect to the static model. In control theory, two types of models or
a combination of both have been established for this purpose: Wiener and Ham-
merstein models [59, 63]. Both have in common a separation of a dynamic transfer
function from the systems nonlinear static transfer behavior. With Hammerstein
models, the systems static nonlinearity is modeled before, with Wiener models
after the systems dynamics. As the dynamic physical behavior of the real engine
shall be enforced on the whole static model, a Wiener-type dynamics is desired [44].
With this choice, the inputs to the static model are overlaid with time attributes,
which enforces the dynamic behavior on the combustion simulation. Thus, IMEP
and CA50 are consequently aected by the dynamics to be identied.
Figure 9.1 depicts the implemented Wiener-type dynamics. Here, the static non-
linear behavior is described by the combined integrated static model, or alterna-
tively the gains K
1
through K
6
and the two oset values A
1
and A
2
, respectively.
For the system identication described in this chapter, the latter alternative is
used. Thereby, the gains K
1
through K
6
and the two oset values A
1
and A
2
are
dependent on the operating point due to the systems nonlinearity.
As only load-transient conditions with a static engine speed are regarded as a
rst step, the identied dynamics is assumed to be given by a PT
1
. If speed-
transient conditions would additionally apply, this approach would either have
to be extended by time constants tabulated over engine speed or by a physical
model for the mechanical dynamics of pistons, conrods, crank, and additionally
gas exchange, respectively. The total model would still have to be tted to mea-
surements as well. With the latter approach, the unknown masses and inertias of
the drivetrain, as well as parameters of the gas dynamics need to be identied. An
example for such an approach is the work by Schulze et al. [85].
71
IMEP
K
1
K
2
K
3
K
4
A
1
CA5O
A
2
G
SOI
G
EGR
Stand-alone multi-zone
model and gas
exchange model
SOI
EGR rate
K
5
K
6
G
FMI
FMI
Figure 9.1: Model structure used for identifying the systems dynamic time-
discrete transfer functions. The dashed line shows the part that may be substi-
tuted by the complete integrated static model (see Sec. 10.1)
9.2 Step Response Experiments
A description of the dynamics from the systems inputs SOI, external EGR rate,
and total fuel mass injected (FMI) towards the outputs IMEP and CA50 is needed.
For this purpose, several approaches for arranging experiments exist in the litera-
ture. In this regard, step responses are the most common way for the identication
of a systems dynamics. The experimental setup (see also Subsec. 7.1.1) realizes
this by using an ES1000 system by ETAS as the central unit for impressing the
steps on the actuated variables, as well as for simultaneously collecting the data of
the step responses in the controlled variables. More details regarding this Rapid
Control Prototyping system can be found in [23]. The steps in the actors were
enabled by an ETK-connection of the ES1000 system to the engines electronic
control unit (ECU, type EDC16). The physical actor for the SOI and the to-
tal fuel mass injected obviously is the Common-Rail direct injector, while for the
EGR rate there is no explicit acting device. For this, the EGR rate was changed
by the combination of the actors EGR valve and variable geometry turbine (VGT)
72
9.3 Dynamic Transfer Functions
position. It is impossible to make the engine step in EGR rate. Nevertheless,
the system shows a dynamic response to a change in the EGR rate caused by
a simultaneous step in EGR valve and VGT position. For the operating points
to step between, the corresponding values for EGR throttle and VGT position
were determined for a steady state operation. Furthermore, the test bed oers no
direct possibility to measure the dynamics of the EGR rate, which is the input
and actuating variable for the static model. Commonly, in modern ECUs in se-
ries applications, the mass ow of fresh air aspirated by the engine is measured
instead [35]. The EGR rate can be calculated by
EGR rate =
m
EGR
m
trapp. mass
=
m
trapp. mass
m
air
m
trapp. mass
, (9.1)
where m
air
denotes the mass ow of fresh air. The total trapped gas mass
m
trapp. mass
mainly depends on the engine speed, as well as on the pressure and
temperature in the intake manifold. The engine was run at a constant speed. In
addition, the intake manifold conditions did not vary signicantly during the step
response experiments. The total trapped gas mass m
trapp. mass
is therefore con-
sidered approximately constant and, for this reason, the main dynamics aecting
the EGR rate in this case is the same but negative dynamics as with the fresh air
mass m
air
. This was measured by the engines ECU, but also handed over to the
measuring device ES1000 via ETK. Because the steady state value for the EGR
rate is known before and after applying the step in VGT position and EGR valve,
the dynamic EGR rate during the tests can be recalculated and used as an input
to the model.
Another prerequisite for the system identication is the measurement of the
target values or outputs of the model, respectively. Both IMEP and CA50 are
calculated from the pressure progression measurement. This has to be realized by
an additional device, as in series applications no pressure analysis is implemented
in the ECU. In this work, a FI2RE system by IAV was used [95]. It analyzes the
pressure signal in real time and hands it over to the measurement system ES1000
via CAN.
The step response experiments provide a dataset containing the temporally re-
solved signals for SOI, external EGR rate, and FMI, as well as the corresponding
response in IMEP and CA50. The aim of the extraction of the systems dynamics is
the abovementioned Wiener-type dynamics. The advantage of this approach is also
its disadvantage. By assigning the systems dynamics to the inputs, a relatively
simple model structure concerning the dynamic transfer functions is achieved, as
their number is reduced to the number of inputs (see Fig. 9.1). But this also is
73
the main drawback of the approach, because all three inputs inuence both out-
puts. This results in six dynamic dependencies, which have to be described by
three transfer functions. The dynamic eects on IMEP and CA50 excited by the
changes in SOI thus have to be represented by a transfer function G
SOI
, the dy-
namic inuence on both outputs caused by changes in the external EGR rate or
the total fuel mass injected by a transfer function G
EGR
or G
FMI
, respectively.
The transfer function G
SOI
can be determined from the evaluation of the re-
sponses of IMEP and CA50 to steps in SOI. The same sampling rate as used
during the measurements with the ES1000 system was used for the time-discrete
transfer function. The sampling rate T
s
of the calculation of IMEP and CA50 is
preset by the engines speed and is given by
T
s
=
2 60 s/min
n
eng.
[rpm]
. (9.2)
With n
eng.
= 2000 rpm, a new value for IMEP and CA50 is calculated every 0.06
seconds. To ensure compliance with Shannons theorem, the measurements were
sampled with 0.01 seconds, and consequently the same sample time was chosen for
the time-discrete transfer function.
In mathematics and signal processing, the Z-transform converts a discrete time-
domain signal, which is a sequence of real or complex numbers, into a complex
frequency-domain representation [53]. It is the discrete equivalent of the Laplace
transform. The discrete time-domain function f(k T
spl
) of sample time T
spl
is
transformed to a z-domain transfer function F(z)
(T
spl
)
valid for the specied sample
time. In this representation, z can be understood as a time shift operator according
to x(k T
spl
n T
spl
)
d ...........
t
z
n
X(z)
(T
spl
)
. The measurements indicate the use of
a PT
1
-transfer function. With a sample time of 0.01 seconds the identied PT
1
is
given by
G(z)
SOI,(0.01)
=
SOI
SOI
=
0.1
z 0.9
. (9.3)
Here, the superscript indicates the temporal attribute of the signal. Figure 9.2
shows the results obtained with this transfer function. Shown is the response of
IMEP and CA50 in measurement and system identication to various steps in SOI.
For the system identication, the corresponding static gains K
1
through K
6
and
the two oset values A
1
and A
2
were adjusted to the operating point to substitute
the complete integrated static model. The results indicate that the time constant of
the chosen transfer function is suitable to describe the engines temporal attribute
regarding a change in SOI.
The sampling rate used here is obviously not consistent with the engine speed.
The static integrated model is executed every engine cycle, demanding a sample
74
time of 0.06 seconds for 2000 rpm. To remedy this problem and to enable the im-
plementation of the transfer function into the integrated model, the time-discrete
transfer function is resampled with the new sample time of 0.06 seconds to
G(z)
SOI,(0.06)
=
SOI
SOI
=
0.4686
z 0.5314
. (9.4)
In an analogue manner, the transfer functions G(z)
EGR
and G(z)
FMI
can be
determined from the evaluation of the responses of IMEP and CA50 to the dynamic
EGR rate or dynamic total fuel mass injected signal, respectively. Thereby, the
EGR rate signal is recalculated as described above. The corresponding gains and
oset values in Fig. 9.1 are again adjusted such that the static nonlinear transfer
behavior is met. The tting of the systems dynamics is evaluated with a measuring
sample time of 0.01 seconds and resampled to 0.06 seconds afterwards.
The inuence of the external EGR rate is given by
G(z)
EGR,(0.01)
=
EGR rate
EGR rate
=
0.015
z 0.985
(9.5)
and can be resampled with T
s
= 0.06 seconds to
G(z)
EGR,(0.06)
=
EGR rate
EGR rate
=
0.08669
z 0.9133
. (9.6)
Using G(z)
EGR,(0.01)
, the results shown in Fig. 9.3 were achieved.
For the dynamic dependency of the models outputs on the total fuel mass
injected the transfer function
G(z)
FMI,(0.01)
=
FMI
FMI
=
0.071
z 0.9290
(9.7)
was determined, which leads to the resampled transfer function
G(z)
FMI,(0.06)
=
FMI
FMI
=
0.3572
z 0.6428
. (9.8)
Figure 9.4 demonstrates the achieved result for various steps in FMI modeled by
Eq. (9.7). Note that in this case there are two active inuences, as it is impossible
to only change the value for FMI and keep the EGR rate constant without adapting
the settings of EGR valve and VGT position. Therefore, the dependencies of IMEP
and CA50 on FMI and on the EGR rate have to be tted simultaneously.
75
42 44 46 48 50 52
2.6
2.8
3
3.2
3.4
I
M
E
P

[
b
a
r
]
Time [s]

Measurement
System Ident.
42 44 46 48 50 52
13.5
13
12.5
12
11.5
11
10.5
10
C
A
5
0

[
d
e
g
s
.

C
A
]
Time [s]

Measurement
System Ident.
(a) IMEP (left) and CA50 (right) for an SOI step from -40.7 to -30.7 degs. CA aTDC and back
with an external EGR rate of 30 % and an injected fuel mass of 10.2 mg/cycle
34 36 38 40 42 44
2.8
2.9
3
3.1
3.2
3.3
3.4
I
M
E
P

[
b
a
r
]
Time [s]

Measurement
System Ident.
34 36 38 40 42 44
13
12
11
10
9
8
7
6
5
4
C
A
5
0

[
d
e
g
s
.

C
A
]
Time [s]

Measurement
System Ident.
(b) IMEP (left) and CA50 (right) for an SOI step from -20.7 to -30.7 degs. CA aTDC and back
22 24 26 28 30 32
2.8
2.9
3
3.1
3.2
3.3
I
M
E
P

[
b
a
r
]
Time [s]

Measurement
System Ident.
22 24 26 28 30 32
8
7
6
5
4
3
2
1
C
A
5
0

[
d
e
g
s
.

C
A
]
Time [s]

Measurement
System Ident.
(c) IMEP (left) and CA50 (right) for an SOI step from -30.7 to -20.7 degs. CA aTDC and back
Figure 9.2: Comparison between experiment and system identication with iden-
tied discrete transfer function G(z)
SOI,(0.01)
76
10 20 30 40 50 60 70 80
2.4
2.6
2.8
3
3.2
I
M
E
P

[
b
a
r
]
Time [s]

Measurement
System Ident.
10 20 30 40 50 60 70 80
14
12
10
8
6
4
2
C
A
5
0

[
d
e
g
s
.

C
A
]
Time [s]

Measurement
System Ident.
(a) IMEP (left) and CA50 (right) for an EGR step from 30 to 45 % and back with an SOI of
-40.7 degs. CA aTDC and an injected fuel mass of 10.2 mg/cycle
30 40 50 60 70 80
2.7
2.8
2.9
3
3.1
3.2
3.3
3.4
3.5
I
M
E
P

[
b
a
r
]
Time [s]

Measurement
System Ident.
30 40 50 60 70 80
12
10
8
6
4
2
0
2
C
A
5
0

[
d
e
g
s
.

C
A
]
Time [s]

Measurement
System Ident.
(b) IMEP (left) and CA50 (right) for an EGR step from 30 to 45 % and back with an SOI of
10 20 30 40 50
2.9
3
3.1
3.2
3.3
3.4
I
M
E
P

[
b
a
r
]
Time [s]

Measurement
System Ident.
10 20 30 40 50
8
7
6
5
4
3
2
1
0
1
C
A
5
0

[
d
e
g
s
.

C
A
]
Time [s]

Measurement
System Ident.
(c) IMEP (left) and CA50 (right) for an EGR step from 30 to 45 % and back with an SOI of
EGR,(0.01)
77
35 40 45 50
2
2.5
3
3.5
4
4.5
I
M
E
P

[
b
a
r
]
Time [s]

Measurement
System Ident.
35 40 45 50
14.5
14
13.5
13
12.5
12
C
A
5
0

[
d
e
g
s
.

C
A
]
Time [s]

Measurement
System Ident.
(a) IMEP (left) and CA50 (right) for an injected fuel mass step from 10.2 to 13.6 mg/cycle and
back with an SOI of -35.7 degs. CA aTDC and an external EGR rate of 45 %
75 80 85 90
2
2.5
3
3.5
4
4.5
5
5.5
I
M
E
P

[
b
a
r
]
Time [s]

Measurement
System Ident.
75 80 85 90
14.5
14
13.5
13
12.5
12
11.5
11
C
A
5
0

[
d
e
g
s
.

C
A
]
Time [s]

Measurement
System Ident.
(b) IMEP (left) and CA50 (right) for an injected fuel mass step from 10.2 to 17.0 mg/cycle and
back with an SOI of -35.7 degs. CA aTDC and an external EGR rate of 45 %
FMI,(0.01)
78
10 Integrated Model
In the previous three chapters, all discussed model parts (stand-alone multi-zone
model, gas exchange model, and dynamic time response) were each validated sep-
arately. In this chapter, the realization of an integrated model composed of these
three parts within a suitable test environment for control applications is described.
Afterwards, the integrated model is validated against transient experimental data.
This chapter closes with a discussion of this models potential in the framework of
closed-loop control development.
10.1 Suitable Test Environment for Control
Applications
For application within a closed-loop control simulation, the stand-alone multi-zone
model was transferred to an environment suitable for the conception and testing of
controllers. For this study, the Matlab/Simulink environment of The MathWorks,
Inc. was chosen as an appropriate platform [92]. The stand-alone multi-zone
model X0D written in FORTRAN 77 was embedded into a Matlab/Simulink FOR-
TRAN s-function enabling the simulation of the stand-alone multi-zone model from
within Matlab/Simulink. Moreover, the gas exchange model was implemented into
this FORTRAN s-function. The three dynamic transfer functions were added by
means of time-discrete PT
1
-dynamics within appropriate function blocks. The
Matlab/Simulink setup is visualized in Fig. 10.1.
The simulation environment is realized such that the initial in-cylinder con-
ditions reect the engine-out conditions of the prior cycle. This establishes the
cycle-to-cycle connectivity. However, as the external EGR was cooled (see also
Subsec. 7.1.1), the measurements did not show a strong dependency of the intake
conditions on the engine-out conditions, except for the external EGR composi-
tion when varying the total fuel mass injected. Aside from the initial in-cylinder
conditions, the numerical setup for the multi-zone model (e.g., start and end of
simulation, number of zones, incorporated information, etc.) within the Mat-
lab/Simulink environment was kept unchanged in comparison to the stand-alone
multi-zone model simulation setup described in Subsec. 7.1.2.
79
10 Integrated Model
The computational time for the integrated model amounts to approximately six
minutes per engine cycle, using a Dell Latitude D630 notebook with an Intel Core
2 Duo T7500 CPU (Dual-Core with 2.2 GHz). Obviously, inserting the stand-
alone multi-zone model into the Matlab/Simulink environment does not increase
the overall computational time.
Figure 10.1: Matlab/Simulink setup for transient simulations
10.2 Transient Simulations
Figure 10.2 shows the IMEP (left) and the CA50 (right) obtained from the step
response experiment, the system identication with the identied discrete trans-
fer function G(z)
SOI,(0.01)
, and the transient simulation with the integrated model
using G(z)
SOI,(0.06)
for an SOI step from -20.7 to -30.7 degs. CA aTDC and back
with an external EGR rate of 30 % and an injected fuel mass of 10.2 mg/cycle.
The results obtained from the step response experiment and the system identi-
cation are already shown in Fig. 9.2(b) above. The simulation results are in very
good agreement with the measurements. In particular, the dynamic step responses
are reproduced well. Similar simulation results (not shown here for brevity) were
achieved for all other step response experiments presented in Sec. 9.3. This vali-
dates the integrated model composed of the stand-alone multi-zone model, the gas
exchange model, and the identied system dynamics.
80
10.3 Potential within Closed-Loop Control Development
34 36 38 40 42 44
2.8
2.9
3
3.1
3.2
3.3
3.4
3.5
I
M
E
P

[
b
a
r
]
Time [s]

Measurement
System Ident.
Integrated Model
34 36 38 40 42 44
12
11
10
9
8
7
6
5
4
C
A
5
0

[
d
e
g
s
.

C
A
]
Time [s]

Measurement
System Ident.
Integrated Model
Figure 10.2: IMEP (left) and CA50 (right) for an SOI step from -20.7 to -30.7
degs. CA aTDC and back with an external EGR rate of 30 % and an injected fuel
mass of 10.2 mg/cycle. Comparison between experiment, system identication
with identied discrete transfer function G(z)
SOI,(0.01)
, and integrated model
10.3 Potential within Closed-Loop Control
Development
The strict derivation of the stand-alone multi-zone model from a detailed three-
dimensional CFD approach integrates the detailed knowledge from combustion
simulation tools into closed-loop control. This novel procedure establishes a broad
eld of possibilities for testing completely new controlled process variables. In this
context, Fig. 10.3 displays further outcomes of the transient simulation for an SOI
step from -20.7 to -30.7 degs. CA aTDC and back with an external EGR rate of
30 % and an injected fuel mass of 10.2 mg/cycle. Exemplary, the NO emissions,
the maximum pressure, the burning time from CA10 to CA90, and the ringing
intensity are shown.
The combustion noise level is quantied applying the ringing intensity correla-
tion developed by Eng [22]. This correlation relates the ringing intensity to the
maximum pressure rise rate, the maximum pressure, and the speed of sound. The
ringing intensity is expressed as
Ringing Intensity
1
2
dp
dt max
2
p
max
RT
max
. (10.1)
In Eq. (10.1), dp/dt
max
denotes the maximum pressure rise rate, p
max
the maximum
pressure, and T
max
the maximum temperature. is a scaling factor determined
81
10 Integrated Model
from the experimental data. In this work, it was set to 0.005 ms. In [22], a value
of 0.05 ms was applied.
34 36 38 40 42 44
28
30
32
34
36
38
40
42
44
N
O

[
g
/
k
g

f
u
e
l
]
Time [s]

Integrated Model
34 36 38 40 42 44
81
81.5
82
82.5
83
83.5
M
a
x
i
m
u
m

P
r
e
s
s
u
r
e

[
b
a
r
]
Time [s]

Integrated Model
(a) NO emissions (left) and maximum pressure (right) for an SOI step from -20.7 to -30.7 degs. CA
aTDC and back with an external EGR rate of 30 % and an injected fuel mass of 10.2 mg/cycle
34 36 38 40 42 44
8.5
9
9.5
10
10.5
C
A
1
0

C
A
9
0

[
d
e
g
s
.

C
A
]
Time [s]

Integrated Model
34 36 38 40 42 44
7.8
8
8.2
8.4
8.6
8.8
x 10
6
R
i
n
g
i
n
g

I
n
t
e
n
s
i
t
y

[
W
/
m
2
]
Time [s]

Integrated Model
(b) Burning time from CA10 to CA90 (left) and ringing intensity (right) for an SOI step from -20.7
to -30.7 degs. CA aTDC and back with an external EGR rate of 30 % and an injected fuel mass of
10.2 mg/cycle
Figure 10.3: Additional process variables obtained from a transient simulation
with the integrated model
Figure 10.3 reveals that the NO emissions increase with advanced SOI timings
due to an increasing maximum pressure or maximum temperature, respectively.
Advancing the SOI also increases the burning time. This reduces the pressure
rise rate, leading to a decreasing ringing intensity. These results indicate the
potential of the integrated model in the framework of closed-loop control develop-
ment. Beyond the controlled variables IMEP and CA50, the integrated model is
capable of predicting various additional process variables, with four selected ones
82
10.3 Potential within Closed-Loop Control Development
being briey discussed in this section. These additional process variables allow
for dening further process constraints or may even replace the original controlled
variables, depending on the specic control task.
83
11 Summary and Conclusion
This thesis proposes a new approach for the integration of detailed physical com-
bustion knowledge into closed-loop simulations. Closed-loop simulations are a
necessary and common tool in the development process of controllers. A detailed
interactively coupled CFD-multi-zone approach was derived for modeling com-
pression ignition combustion. This detailed interactively coupled CFD-multi-zone
approach was validated against a test case of HCCI combustion in a gasoline
single-cylinder research engine. Addressing PCCI combustion, the approach was
then reduced to a stand-alone multi-zone model such that the resulting calcula-
tion time could be signicantly lowered with only a small penalty in accuracy.
The stand-alone multi-zone model is capable of describing the PCCI combustion
characteristics for the high-pressure part of the engine cycle. The controller to
be developed shall actuate SOI, external EGR rate, and total fuel mass injected
to control the IMEP of the whole engine cycle and the CA50. The IMEP
HP
of
the high-pressure engine cycle and the CA50 can be extracted from simulations
with the stand-alone multi-zone model. The former was combined with a newly
proposed mean value model for the losses of the gas exchange to calculate the
IMEP of the whole engine cycle. The combination of these models leads to an
accurate static model, which was further extended to a Wiener model for captur-
ing temporal cycle-to-cycle dependencies. For every input of the model, a transfer
function was determined, forcing the whole model to follow the engines dynamics.
All model parts (stand-alone multi-zone model, gas exchange model, and dynamic
time response) were each validated separately. Afterwards, the integrated model
composed of all three model parts was validated against transient experimental
engine data.
85
12 Future Work
The dynamic model will build the framework in which a controller for the PCCI
combustion process will be developed. With the developed plant model, this con-
troller will be laid out and tested.
The actuated variable external EGR rate is theoretical as long as there is no
corresponding actor to it. Because of this, the overall goal of this research project
(carried out within the Collaborative Research Centre SFB 686 - Modellbasierte
Regelung der homogenisierten Niedertemperatur-Verbrennung at RWTH Aachen
University, Germany, and Bielefeld University, Germany [83]) is to develop not
only a controller for the PCCI combustion process, but also an additional con-
troller enabling the fastest possible changes in external EGR rate by simultane-
ously adjusting EGR valve and VGT position [18]. This controller, in combination
with its actuated variables EGR valve and VGT position, will serve as the virtual
actor for the external EGR rate.
87
Bibliography
[1] S. M. Aceves, D. L. Flowers, F. Espinoza-Loza, A. Babajimopoulos, and
D. N. Assanis. Analysis of Premixed Charge Compression Ignition Combus-
tion with a Sequential Fluid Mechanics-Multizone Chemical Kinetics Model.
Paper No. SAE 2005-01-0115, 2005. 1
[2] S. M. Aceves, D. L. Flowers, J. Martinez-Frias, J. R. Smith, C. K. Westbrook,
W. Pitz, R. Dibble, M. Christensen, and B. Johansson. A Multi-Zone Model
for Prediction of HCCI Combustion and Emissions. Paper No. SAE 2000-
01-0327, 2000. 2.1
[3] S. M. Aceves, D. L. Flowers, J. Martinez-Frias, J. R. Smith, C. K. West-
brook, W. Pitz, R. Dibble, J. F. Wright, W. C. Akinyemi, and R. P. Hes-
sel. A Sequential Fluid-Mechanic Chemical Kinetic Model of Propane HCCI
Combustion. Paper No. SAE 2001-01-1027, 2001. 2.1
[4] A. Albrecht, O. Grondin, F. Le Berr, and G. Le Solliec. Towards a Stronger
Simulation Support for Engine Control Design: a Methodological Point of
View. Oil & Gas Science and Technology - Rev. IFP, 62:437456, 2007. 1
[5] A. Babajimopoulos, D. N. Assanis, D. L. Flowers, S. M. Aceves, and R. P.
Hessel. A fully coupled computational uid dynamics and multi-zone model
with detailed chemical kinetics for the simulation of premixed charge com-
pression ignition engines. International Journal of Engine Research, 6:497
512, 2005. 2.1
[6] A. Babajimopoulos, G. A. Lavoie, and D. N. Assanis. Modelling HCCI
Combustion with High Levels of Residual Gas Fraction - A Comparison of
Two VVA Strategies. Paper No. SAE 2003-01-3220, 2003. 2.1
[7] H. D. Baehr and K. Stephan. Warme- und Sto ubertragung. Springer, Berlin,
2006. 6.2.3
[8] H. Barths, C. Hasse, and N. Peters. Computational uid dynamics modelling
of non-premixed combustion in direct injection diesel engines. International
Journal of Engine Research, 1:249267, 2000. 1, 4.2.1
89
Bibliography
[9] J. Bengtsson, P. Strandh, R. Johansson, P. Tunestal, and B. Johansson.
Hybrid control of homogeneous charge compression ignition (HCCI) engine
dynamics. International Journal of Control, 79:422448, 2006. 1
[10] J. Bengtsson, P. Strandh, R. Johansson, P. Tunestal, and B. Johansson. Hy-
brid modelling of homogeneous charge compression ignition (HCCI) engine
dynamics - a survey. International Journal of Control, 80:18141847, 2007.
1
[11] J. Boussinesq. Theorie de l
Ecoulement Tourbillant. Mem. Presentes par

Divers Savants Acad. Sci. Inst. Fr., 23:4650, 1877. 3.4
[12] M. Christensen, B. Johansson, and P. Einewall. Homogeneous Charge Com-
pression Ignition (HCCI) using Isooctane, Ethanol and Natural Gas - a Com-
parison with Spark Ignition Operation. Paper No. SAE 972874, 1997. 5.2.2
[13] O. Colin and A. Benkenida. The 3-Zones Extended Coherent Flame Model
(ECFM3Z) for Computing Premixed/Diusion Combustion. Oil & Gas Sci-
ence and Technology - Rev. IFP, 59:593609, 2004. 1
[14] O. Colin, A. Pires da Cruz, and S. Jay. Detailed chemistry-based auto-
ignition model including low temperature phenomena applied to 3-D engine
calculations. Proc. Combust. Inst., 30:26492656, 2005. 1
[15] D. J. Cook, H. Pitsch, J. H. Chen, and R. E. Hawkes. Flamelet-based
modeling of auto-ignition with thermal inhomogeneities for application to
HCCI engines. Proc. Combust. Inst., 31:29032911, 2007. 1
[16] C. T. Crowe, M. P. Sharma, and D. E. Stock. The particle-source-in cell (PSI-
Cell) model for gas-droplet ows. Journal of Fluids Engineering, 99:325332,
1977. 3.6
[17] J. E. Dec. Advanced compression-ignition engines Understanding the in-
cylinder processes. Proc. Combust. Inst., 32:27272742, 2009. 1
[18] P. Drews, K. Homann, R. Beck, R. Gasper, A. Vanegas, C. Felsch,
N. Peters, and D. Abel. Fast Model Predictive Control for the Air Path
of a Turbocharged Diesel Engine. In European Control Conference 2009,
Budapest, Hungary, 2009. 12
[19] J. K. Dukowicz. A particle-uid model numerical model for liquid sprays. J.
Comput. Phys., 35:229253, 1980. 3.6
90
Bibliography
[20] P. Duret and J. Lavy. Gasoline Controlled Auto-Ignition (CAI) - Potential
and Prospects for the Future Automotive Application. In IMechE Interna-
tional Conference: 21st Century Emissions Technologies, London, UK, 2000.
5.1.1
[21] E. Easley, A. Agarwal, and G. A. Lavoie. Modeling of HCCI Combustion and
Emissions Using Detailed Chemistry. Paper No. SAE 2001-01-1029, 2001. 1
[22] J. A. Eng. Characterization of Pressure Waves in HCCI Combustion. Paper
No. SAE 2002-01-2859, 2002. 10.3, 10.3
[23] ETAS GmbH. ES 1000.3 Benutzerhandbuch. Technical report, ETAS GmbH,
2008. 9.2
[24] J. Ewald, F. Freikamp, G. Paczko, J. Weber, D. C. Haworth, and N. Peters.
GMTEC: GMTEC Developers Manual. Technical report, Advanced Com-
bustion GmbH, 2003. 3.5
[25] C. Felsch, M. Gauding, C. Hasse, S. Vogel, and N. Peters. An extended
amelet model for multiple injections in DI Diesel engines. Proc. Combust.
Inst., 32:27752783, 2009. 4.2.1
[26] C. Felsch, T. Sloane, J. Han, H. Barths, and A. M. Lippert. Numerical
Investigation of Recompression and Fuel Reforming in a SIDI-HCCI Engine.
Paper No. SAE 2007-01-1878, 2007. 2.2
[27] C. Felsch, A. Vanegas, B. Kerschgens, N. Peters, K. Homann, P. Drews,
D. Abel, and H. Barths. Systematic Reduction of an Interactively Coupled
CFD-Multi-Zone Approach Towards a Stand-alone Multi-Zone Model for
PCCI Combustion. In Thiesel 2008: Thermo-and uid dynamic processes in
Diesel Engines, Universidad Politecnica de Valencia, Valencia, Spain, 2008.
7.2.2
[28] J. H. Ferziger. Higher level simulations of turbulent ows. In J.-A. Essers
(Ed.), Computational Methods for Turbulent, Transonic, and Viscous Flows.
Springer, Berlin, 1983. 3.4
[29] J. H. Ferziger and M. Peric. Computational Methods for Fluid Dynamics.
[30] D. L. Flowers, S. M. Aceves, and A. Babajimopoulos. Eects of Charge Non-
Uniformity on Heat Release and Emissions in PCCI Engine Combustion.
Paper No. SAE 2006-01-1363, 2006. 2.1
91
Bibliography
[31] D. L. Flowers, S. M. Aceves, J. Martinez-Frias, R. P. Hessel, and R. Dibble.
Eects of Mixing on Hydrocarbon and Carbon Monoxide Emissions Predic-
tion for Isooctane HCCI Engine Combustion Using a Multi-Zone Detailed
Kinetics Solver. Paper No. SAE 2003-01-1821, 2003. 1
[32] F. Freikamp. Towards Multi-Cycle Simulation of In-Cylinder Flow and Com-
bustion. PhD thesis, RWTH Aachen University, Germany, 2008. 3.5, 3.1
[33] Gamma Technologies. GT-POWER Users Manual and Tutorial, GT-SUITE
Version 7.0. Technical report, Gamma Technologies, 2008. 8
[34] O. Gicquel, N. Darabiha, and D. Thevenin. Laminar premixed hydrogen/air
counterow ame simulations using ame prolongation of ILDM with dier-
ential diusion. Proc. Combust. Inst., 28:19011908, 2000. 1
[35] L. Guzzella and C. H. Onder. Introduction to Modeling and Control of In-
ternal Combustion Engine Systems. Springer, Berlin, 2004. 9.2
[36] V. Hamosfakidis, H. G. Im, and D. N. Assanis. A regenerative multiple zone
model for HCCI combustion. Combust. Flame, 156:928941, 2009. 1
[37] A. Harada, N. Shimazaki, S. Sasaki, T. Miyamoto, H. Akagawa, and K. Tsu-
jimura. The Eects of Mixture Formation on Premixed Lean Diesel Com-
bustion. Paper No. SAE 980533, 1998. 1
[38] C. Hasse and N. Peters. A two mixture fraction amelet model applied to
split injections in a DI Diesel engine. Proc. Combust. Inst., 30:27552762,
2005. 4.2.1
[39] C.-A. Hergart, H. Barths, and R. M. Siewert. Modeling Approaches for
Partially Premixed Compression Ignition Combustion. Paper No. SAE 2005-
01-0218, 2005. 1, 2.2, 4.6, 6.2.3, 7.2.2
[40] R. P. Hessel, D. E. Foster, S. M. Aceves, M. Davisson, F. J. Espinosa-Loza,
D. L. Flowers, W. Pitz, J. E. Dec, M. Sjoberg, and A. Babajimopoulos.
Modeling Iso-octane HCCI using CFD with Multi-Zone Detailed Chemistry;
Comparison to Detailed Specication Data over a Range of Lean Equivalence
Ratios. Paper No. SAE 2008-01-0047, 2008. 2.1
[41] J. B. Heywood. Internal Engine Combustion Fundamentals. McGraw-Hill,
New York, 1988. 2.3, 4.4, 7.2.1
[42] M. Hillion, H. Buhlbuck, J. Chauvin, and N. Petit. Combustion Control of
Diesel Engines Using Injection Timing. Paper No. SAE 2009-01-0367, 2009.
1
92
Bibliography
[43] M. Hillion, J. Chauvin, O. Grondin, and N. Petit. Active Combustion Con-
trol of Diesel HCCI Engine: Combustion Timing. Paper No. SAE 2008-01-
0984, 2008. 1
[44] K. Homann, P. Drews, D. Abel, C. Felsch, A. Vanegas, and N. Peters.
A Cycle-Based Multi-Zone Simulation Approach Including Cycle-to-Cycle
Dynamics for the Development of a Controller for PCCI Combustion. Paper
No. SAE 2009-01-0671, 2009. 8, 9.1
[45] D. T. Hountalas, D. A. Kouremenos, G. C. Mavropoulos, K. B. Binder,
and V. Schwarz. Multi-Zone Combustion Modelling as a Tool for DI Diesel
Engine Development-Application for the Eects of Injection Pressure. Paper
No. SAE 2004-01-0115, 2004. 1
[46] R. I. Issa. Solution of the implicitly discretised uid ow equations by
operator-splitting. J. Comput. Phys., 62:4065, 1986. 3.5
[47] Y. Iwabuchi, K. Kawai, T. Shoji, and Y. Takeda. Trial of New Concept
Diesel Combustion System - Premixed Compression Ignited Combustion.
Paper No. SAE 1999-01-0185, 1999. 1
[48] W. P. Jones and B. E. Launder. The prediction of laminarization with a
two-equation model of turbulence. Int. J. Heat Mass Transfer, 15:301314,
1972. 3.4
[49] R. J. Kee, F. M. Rupley, and J. A. Miller. CHEMKIN-II: A Fortran Chemical
Kinetics Package for the Analysis of Gas-Phase Chemical Kinetics. Technical
report, Sandia Report SAND89-8009, Sandia National Laboratories, 1989.
2.3
[50] B. Khalighi, S. H. El Thary, D. C. Haworth, and M. S. Huebler. Computation
and Measurement of Flow and Combustion in a Four-Valve Engine with
Intake Variations. Paper No. SAE 950287, 1995. 3.5
[51] B. E. Launder and B. I. Sharma. Application of the energy-dissipation model
of turbulence to the calculation of ow near a spinning disc. Lett. Heat Mass
Transfer, 1:131138, 1974. 3.4, 3.4
[52] D. Law, D. Kemp, J. Allen, G. Kirkpatrick, and T. Copland. Controlled
Combustion in an IC-Engine with a Fully Variable Valve Train. Paper No.
SAE 2001-01-0251, 2001. 5.1.1
[53] J. Lunze. Regelungstechnik 2: Mehrgroensysteme, Digitale Regelung.
93
Bibliography
[54] U. C. M uller, N. Peters, and A. Li nan. Global kinetics for n-heptane ignition
at high pressures. Symposium (International) on Combustion, 24:777784,
1992. 6.2.4
[55] P. M. Najt and D. E. Foster. Compression-Ignited Homogeneous Charge
Combustion. Paper No. SAE 830264, 1983. 5.2.2
[56] S. Nakayama, T. Ibuki, H. Hosaki, and H. Tominaga. An Application of a
Model Based Combustion Control to Transient Cycle-by-Cycle Diesel Com-
bustion. Paper No. SAE 2008-01-1311, 2008. 1
[57] K. Narayanaswamy, R. P. Hessel, and C. J. Rutland. A New Approach to
Model DI-Diesel HCCI Combustion for Use in Cycle Simulation Studies.
Paper No. SAE 2005-01-3743, 2005. 1, 1
[58] K. Narayanaswamy and C. J. Rutland. A Modeling Investigation of Com-
bustion Control Variables During DI-Diesel HCCI Engine Transients. Paper
No. SAE 2006-01-1084, 2006. 1
[59] O. Nelles. Nonlinear System Identication: From Classical Approaches to
Neural Networks and Fuzzy Models. Springer, Berlin, 2000. 9.1
[60] T. Noda and D. E. Foster. A Numerical Study to Control Combustion
Duration of Hydrogen-fueled HCCI by Using Multi-Zone Chemical Kinetics
Simulation. Paper No. SAE 2001-01-0250, 2001. 1
[61] A. Oakley, H. Zhao, N. Ladommatos, and T. Ma. Experimental Studies
on Controlled Auto-Ignition (CAI) Combustion of Gasoline in a 4-Stroke
Engine. Paper No. SAE 2001-01-1030, 2001. 5.1.1
[62] S. V. Patankar. Numerical Heat Transfer and Fluid Flow. Taylor & Francis,
London, 1980. 3.5
[63] E. Perez, X. Blasco, S. Garca-Nieto, and J. Sanchis. Diesel Engine Identi-
cation and Predictive Control using Wiener and Hammerstein Models. In
Proceedings of the 2006 IEEE International Conference on Control Applica-
tions, Munich, Germany, 2006. 9.1
[64] N. Peters. Laminar diusion amelet models in non-premixed turbulent
combustion. Prog. Energy Combust. Sci., 10:319339, 1984. 4.2.1
[65] N. Peters. Laminar amelet concepts in turbulent combustion. Symposium
(International) on Combustion, 21:12311250, 1986. 4.2.1
94
Bibliography
[66] N. Peters. Turbulent Combustion. Cambridge Univ. Press, Cambridge, 2000.
1, 3.1, 4.2.1
[67] N. Peters, G. Paczko, R. Seiser, and K. Seshadri. Temperature cross-over and
non-thermal runaway at two-stage ignition of n-heptane. Combust. Flame,
28:3859, 2002. 2.3, 6.2.4
[68] L. R. Petzold. A Description of DASSL: A Dierential/Algebraic System
Solver. Technical report, Sandia Report SAND82-8637, Sandia National
Laboratories, 1982. 2.2
[69] H. Pitsch, H. Barths, and N. Peters. Three-Dimensional Modeling of NO
x
and Soot Formation in DI-Diesel Engines Using Detailed Chemistry Based
on the Interactive Flamelet Approach. Paper No. SAE 962057, 1996. 4.2.1
[70] H. Pitsch and N. Peters. Private Communication, 2001. 2.3
[71] S. B. Pope. PDF methods for turbulent reactive ows. Prog. Energy Com-
bust. Sci., 11:119192, 1985. 1
[72] S. B. Pope. Computationally ecient implementation of combustion chem-
istry using in situ adaptive tabulation. Combust. Theory & Modelling, 1:41
63, 1997. 1
[73] S. B. Pope. Turbulent Flows. Cambridge Univ. Press, Cambridge, 2000. 3.2,
3.4
[74] L. Prandtl.

Uber die ausgebildete Turbulenz. ZAMM, 5:136139, 1925. 3.4
[75] L. Prandtl.

Uber ein neues Formelsystem f ur die ausgebildete Turbulenz.
Nachr. Akad. Wiss. Gottingen, Math.-Phys. Klasse, pages 619, 1945. 3.4
[76] C. D. Rakopoulos and D. T. Hountalas. Development and Validation of a
3-D Multi-Zone Combustion Model for the Prediction of DI Diesel Engines
Performance and Pollutants Emissions. Paper No. SAE 981021, 1998. 1
[77] R. Reitz and C. J. Rutland. Development and testing of diesel engine CFD
models. Prog. Energy Combust. Sci., 21:173196, 1995. 1
[78] Ricardo Software. WAVE Basic Manual, Documentation/Users Manual,
Version 5.0. Technical report, Ricardo Software, 2003. 5.1.2, 8
[79] F. Richert. Objektorientierte Modellbildung und Nichtlineare Pradiktive
Regelung von Dieselmotoren. PhD thesis, RWTH Aachen University, Ger-
many, 2005. 8
95
Bibliography
[80] W. Rodi and D. B. Spalding. A two-parameter model of turbulence and its
application to free jets. Warme- und Sto ubertragung, 3:8595, 1970. 3.4
[81] J. C. Rotta. Statistische Theorie nichthomogener Turbulenz. Zeitschrift f uer
Physik, 129:547572, 1951. 3.4
[82] J. C. Rotta. Turbulente Str omungen. Teubner Verlag, Stuttgart, 1972. 3.4
[83] RWTH Aachen University and Bielefeld University. SFB 686 - Mod-
ellbasierte Regelung der homogenisierten Niedertemperatur-Verbrennung.
http://www.sfb686.rwth-aachen.de. 12
[84] H. Santoso, J. Matthews, and W. Cheng. Characteristics of HCCI Engine
Operating in the Negative-Valve-Overlap Mode. Paper No. SAE 2005-01-
2133, 2005. 5.1.1
[85] T. Schulze, M. Wiedemeier, and H. Schuette. Crank Angle - Based Diesel
Engine Modeling for Hardware-in-the-Loop Applications with In-Cylinder
Pressure Sensors. Paper No. SAE 2007-01-1303, 2007. 9.1
[86] G. M. Shaver, J. C. Gerdes, M. J. Roelle, P. A. Caton, and C. F. Edwards.
Dynamic Modeling of Residual-Aected Homogeneous Charge Compression
Ignition Engines with Variable Valve Actuation. Journal of Dynamic Sys-
tems, Measurement, and Control, 127:374381, 2005. 1
[87] G. M. Shaver, M. J. Roelle, P. A. Caton, N. B. Kaahaaina, N. Ravi, J.-P.
Hathout, J. Ahmed, A. Kojic, S. Park, C. F. Edwards, and J. C. Gerdes.
A physics-based approach to the control of homogeneous charge compres-
sion ignition engines with variable valve actuation. International Journal of
Engine Research, 6:361375, 2005. 1
[88] G. M. Shaver, M. J. Roelle, and J. C. Gerdes. Modeling cycle-to-cycle dy-
namics and mode transition in HCCI engines with variable valve actuation.
Control Engineering Practice, 14:213222, 2006. 1
[89] P. Spiekermann, S. Jerzembeck, C. Felsch, S. Vogel, M. Gauding, and
N. Peters. Experimental Data and Numerical Simulation of Common-Rail
Ethanol Sprays at Diesel Engine-Like Conditions. Atomization and Sprays,
19:357386, 2009. 3.6
[90] Y. Takeda and N. Keiichi. Emission Characteristics of Premixed Lean Diesel
Combustion with Extremely Early Staged Fuel Injection. Paper No. SAE
961163, 1996. 1
96
Bibliography
[91] H. Tennekes and J. L. Lumley. A First Course in Turbulence. The MIT
Press, Cambridge, 1972. 3.2
[92] The MathWorks, Inc.. Matlab 7.6 Manual (R2008a). Technical report, The
MathWorks, Inc., 2008. 10.1
[93] R. H. Thring. Homogeneous Charge Compression Ignition (HCCI) Engines.
Paper No. SAE 892068, 1989. 5.2.2
[94] R. Turin, R. Zhang, and M.-F. Chang. Systematic Model-Based Engine
Control Design. Paper No. SAE 2008-01-0994, 2008. 1
[95] IAV GmbH Ingenieursgesellschaft Auto und Verkehr. FI2RE
- Flexible Injection and Ignition for Rapid Engineering.
http://www.iav.com/de/1 engineering/powertrainmechatronik/messtechnik/2re.php,
2008. 9.2
[96] T. Urushihara, K. Hiraya, A. Kakuhou, and T. Itoh. Expansion of HCCI
Operating Region by the Combination of Direct Fuel Injection, Negative
Valve Overlap, and Internal Fuel Reformation. Paper No. SAE 2003-01-
0749, 2003. 5.1.1
[97] J. A. van Oijen, F. A. Lammers, and L. P. H. de Goey. Modeling of complex
premixed burner systems by using amelet-generated manifolds. Combust.
Flame, 127:21242134, 2001. 1
[98] A. Vanegas, H. Won, C. Felsch, M. Gauding, and N. Peters. Experimental
Investigation of the Eect of Multiple Injections on Pollutant Formation in
a Common-Rail DI Diesel Engine. Paper No. SAE 2008-01-1191, 2008. 7.1.1
[99] J. Weber, P. Spiekermann, and N. Peters. Model Calibration for Spray
Penetration and Mixture Formation in a High-Pressure Fuel Spray Using a
Micro-Genetic Algorithm and Optical Data. Paper No. SAE 2005-01-2099,
2005. 7.1.2
[100] J. Willand, R.-G. Nieberding, G. Vent, and C. Enderle. The Knocking
Syndrome: Its Cure and Potential. Paper No. SAE 982483, 1998. 5.1.1
[101] F. A. Williams. Spray Combustion and Atomization. Phys. Fluids, 1:541
545, 1958. 3.6
[102] G. Woschni. A Universally Applicable Equation for Instantaneous Heat
Transfer Coecients in the Internal Combustion Engine. Paper No. SAE
670931, 1967. 6.2.1
97
Bibliography
[103] H. Xu. Modelling of HCCI Engines: Comparison of Single-Zone, Multi-Zone
and Test Data. Paper No. SAE 2005-01-2123, 2005. 1
[104] M. Yao, Z. Zheng, and H. Liu. Progress and recent trends in homogeneous
charge compression ignition (HCCI) engines. Prog. Energy Combust. Sci.,
35:398437, 2009. 1
[105] Y. Z. Zhang, E. H. Kung, and D. C. Haworth. A PDF method for
multidimensional modeling of HCCI engine combustion: eects of turbu-
lence/chemistry interactions on ignition timing and emissions. Proc. Com-
bust. Inst., 30:27632771, 2005. 1
98
Denitions, Acronyms, Abbreviations
aTDC after Top Dead Center
CA Crank Angle
CA10 Crank Angle of 10 % burnt mass
CA50 Crank Angle of 50 % burnt mass
CFD Computational Fluid Dynamics
CO Carbon Monoxide
DDM Discrete Droplet Model
ECU Electronic Control Unit
EGR Exhaust Gas Recirculated
EOI End of Injection
EVO Exhaust Valve Opening
FMI Total Fuel Mass Injected
HCCI Homogeneous Charge Compression Ignition
IMEP Indicated Mean Eective Pressure
IMEP
GE
Indicated Mean Eective Pressure of the Gas Exchange
IMEP
HP
Indicated Mean Eective Pressure of the High-Pressure Cycle
IVC Intake Valve Closure
LTC Low-temperature Combustion
NO
x
Nitrogen Oxides
p
ae
Pressure in the Exhaust Manifold (after engine)
p
be
Pressure in the Intake Manifold (before engine)
PCCI Premixed Charge Compression Ignition
PCI Premixed Charge Ignition
rpm Revolutions per Minute
RWTH Rheinisch Westfalische Technische Hochschule Aachen
SOI Start of Injection
T
ae
Temperature in the Exhaust Manifold
99
T
be
Temperature in the Intake Manifold
TDC Top Dead Center
TS Timing Sweep
VGT Variable Geometry Turbine
100
Lebenslauf
Personliche Daten
Name Christian Felsch
Geburtsdatum 25. Dezember 1977
Geburtsort Neuss
Familienstand ledig
Nationalitat deutsch
Schulbildung
08/198806/1997 Stadtisches Meerbusch-Gymnasium, Meerbusch;
Abschluss: Abitur
Grundwehrdienst
07/199704/1998 Luftwae, Goslar und Geilenkirchen
Studium
10/199809/2000 Grundstudium Maschinenbau an der Gerhard-Mercator-
Universitat Duisburg;
Abschluss: Diplom-Vorpr ufung
10/200007/2003 Hauptstudium Maschinenbau an der RWTH Aachen
University, Fachrichtung Luftfahrttechnik;
Abschluss: Diplom
Externe Diplomarbeit
Investigation of Boundary Conditions and Pressure
Waves in Numerical Modelling of Compressible Turbu-
lent Flow with Combustion
Department of Energy and Process Engineering, Norwe-
gian University of Science and Technology, Trondheim,
Norwegen (im Rahmen eines DAAD-Stipendiums)
Beruicher Werdegang
seit 10/2003 Wissenschaftlicher Angestellter am Institut f ur Tech-
nische Verbrennung (ehemals Institut f ur Technische
Mechanik) der RWTH Aachen University

Combustion Modeling ForDiesel Engine Control

Transféré par

Informations du document

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Combustion Modeling ForDiesel Engine Control

Transféré par

Droits d'auteur :

Formats disponibles

Combustion Modeling for

Diesel Engine Control Design

is the symmetric stress

denoting the Kronecker delta. is the laminar viscosity.

denotes the viscous heating term. This term is small

. It can be seen from Eqs. (3.17), (3.18), and (3.19) that

with the density (rather than f

appear. They represent the classical closure problem of turbu-

. These triple correlations are also unclosed.

is similar to the Reynolds stresses in Eq. (3.26).

k the mean turbulent kinetic energy

was already proposed by Launder and Sharma [51]. It

Ecoulement Tourbillant. Mem. Presentes par

Vous aimerez peut-être aussi