Numerical Modeling of Diesel Spray Formation and C

See
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/255568931
Numerical Modeling of Diesel Spray Formation

and Combustion
Article January 2009
CITATIONS READS
10 956
3 authors:
Cemil Bekdemir L.M.T. Somers

TNO Technische Universiteit Eindhoven
19 PUBLICATIONS 144 CITATIONS 101 PUBLICATIONS 876 CITATIONS
SEE PROFILE SEE PROFILE
Philip de Goey
Technische Universiteit Eindhoven
389 PUBLICATIONS 4,671 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
RCCI engine control oriented model development View project
Biomass combustion modeling View project
All content following this page was uploaded by Philip de Goey on 14 August 2014.
The user has requested enhancement of the downloaded file.

.
.
Numerical Modeling of Diesel Spray
Formation and Combustion
Cemil Bekdemir
WVT 2008.15
.
.
Numerical Modeling of Diesel Spray

Formation and Combustion
master thesis
Cemil Bekdemir
Supervisor: dr.ir. L.M.T. Somers
Exam committee: prof.dr. L.P.H. de Goey

dr.ir. L.M.T. Somers
dr.ir. C.C.M. Luijten
dr. C.W.M. van der Geld (Process Technology)
dr.ir. N.A. Beishuizen (University of Twente)
Eindhoven University of Technology

Faculty of Mechanical Engineering
Section Combustion Technology
c 2008 by Cemil Bekdemir

Copyright
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 author.
.
vi
Abstract
Nowadays diesel engine related research, driven by environmental issues and global energy
demand, is becoming more and more important. For that reason, rapid development of high
efficiency and low emission engines in a numerical fashion is needed instead of conventional
prototyping. In modern diesel engines spray formation and combustion are the main processes
that challenge engineers. In this study in the first place diesel spray formation in a constant
volume chamber is modeled accurately and efficiently with a combination of a 1D spray model
and Fluent. Then, ignition and combustion by means of a tabulated chemistry method (FGM)
is applied.
Liquid spray formation concern a lot of physics, starting from breakup of the liquid core into
droplets short after the nozzle exit, called primary breakup. In a second stage the formed
droplets breakup into smaller droplets, called secondary breakup. In automotive applications,
with high ambient pressures and temperatures, the fuel droplets evaporate during their path
until the liquid length is reached. From then on the evaporated fuel penetrates further into
the surrounding gas, and at some point the spray auto-ignites.
Spray modeling can be done with phenomenological and CFD models, from which the last
one is more complex and expensive. But, in order to do a full engine simulation including
combustion 3D CFD is required. Therefore, Fluent and its Euler-Lagrange spray model
is used. However, comparisons with liquid length and spray length measurements on the
EHPC and data from Sandia show unsatisfactory correspondence. Also major numerical
disadvantages exist, like restricted mesh refinement possibilities and lack of parallelization of
the computation.
These problems are circumvented by implementing a 1D Euler-Euler spray model of Versaevel
et al in Matlab. From that, source terms are extracted for 3D calculations in Fluent. The re-
sults are validated with experimental data of IFP and Sandia. The newly created combination
of the 1D spray model with 3D Fluent gives a better overal performance in spray length and
shape compared to the Euler-Lagrange model of Fluent. At the same time the mentioned nu-
merical drawbacks are taken care of, since there is no complex interaction between gas phase
and liquid droplets any more. Improvements for the future should be updated source terms
during the CFD calculation to make simulations with variable volume combustion chambers
possible and sound spray angle prediction methods/relations should be developed.
Finally, an attempt to include ignition and combustion is made. The detailed, though time-
efficient, tabulated chemistry approach called FGM is used for this purpose. This approach
is based on the flamelet concept and the manifold is preprocessed with igniting non-premixed
flamelets. Now, Fluent solves four (mixture fraction, progress variable and their variances
vii
viii
because of turbulence interaction) predefined scalar transport equations that are lookup in-
dices for the 4D table. The spray length and shape are still well predicted, and the spray
auto-ignites. But combustion temperatures stay low, while the flame does not extinguish.
This observation seems to be caused by the PDF integration procedure during the generation
of the FGM, and it should be investigated first by simulating laminar spray combustion. Yet,
the implementation (spray model-CFD interaction, FGM data handling) on its own seems to
work well.
Contents
1 Introduction 1
2 Fundamentals of Liquid Sprays 3

2.1 Spray Regimes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Breakup Regimes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2.1 Primary Breakup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2.2 Secondary Breakup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2.3 Atomization in Diesel Sprays . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Direct Injection Diesel Combustion . . . . . . . . . . . . . . . . . . . . . . . . 10
3 Classification of Models 13
3.1 Thermodynamic Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.2 Phenomenological Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.3 CFD Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4 CFD Spray Model 15

4.1 Fluent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.1.1 Transport Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.1.2 Discrete Phase Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.1.3 Model Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.2 Results and Parametric Dependencies . . . . . . . . . . . . . . . . . . . . . . 24
4.2.1 Spray Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2.2 Liquid Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
ix
x CONTENTS
5 CFD and Phenomenological Spray Model 31

5.1 Phenomenological Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.2 Results; 1D Spray Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.2.1 Spray Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.2.2 Liquid Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.2.3 Sensitivity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.3 Phenomenological Model into CFD . . . . . . . . . . . . . . . . . . . . . . . . 42
5.3.1 Source Terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.3.2 Interaction between Models . . . . . . . . . . . . . . . . . . . . . . . . 44
5.4 Results; 3D Spray Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.5 Achievements and Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6 Modeling Spray Combustion with Tabulated Chemistry 49

6.1 FGM Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
6.1.1 The Flamelet Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
6.1.2 FGM Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
6.1.3 Implementation into CFD . . . . . . . . . . . . . . . . . . . . . . . . . 55
6.2 Results; 3D Spray Model with FGM Combustion Modeling . . . . . . . . . . 56
6.2.1 Spray Length without Combustion . . . . . . . . . . . . . . . . . . . . 57
6.2.2 Ignition Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
6.2.3 Combustion Temperature . . . . . . . . . . . . . . . . . . . . . . . . . 59
6.3 Improvement Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
7 Conclusions and Recommendations 61

7.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
7.2 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Nomenclature 63
A Transport Equations for the Mean and Variance of Z and PV 65
Acknowledgements 67
Bibliography 69
Chapter 1
Introduction
This study contributes to the development of diesel spray formation and combustion models.
The motivation is based on the public need for maintaining, or even improving, current
prosperity, while preserving the environment and the health of mankind. In daily practice
this means, amongst others, that one has to comply with stringent regulations concerning
internal combustion engine emissions. These emissions include pollutants like nitrogen oxides
(NOx ) and soot. More and more also the emission of carbon dioxide (CO2 ) is restricted due
to its involvement with the reinforced greenhouse effect. Another implication of this public
need, together with an increase of the global energy demand, is the approaching depletion of
fossil fuels, which makes the efficient use of organic fuels necessary.
To efficiently deal with fossil sources and to reduce harmful emissions, diesel engine related
research is of upmost importance, since this type of engine is mainly used for heavy duty
transport purposes and also increasingly for passenger cars. Conventional engine design ap-
proaches that rely on prototype development are too time-consuming and expensive, while
development of predictive and efficient computational tools would represent a significant step
forward in the ability to rapidly design high efficiency, low emission engines [FCD+ 07]. Mod-
ern diesel technology consists of direct liquid fuel injection under high pressure, that forms
a non-homogenous mixture leading to relatively high levels of soot. This spray formation
process may seem straightforward, but in reality it is dauntingly complex [flu06b]. And also
the combustion presents especially difficult and complex challenges [RR95].
The aim of this study is in the first place to accurately and efficiently model non-reacting diesel
spray formation. A commercial CFD package (Fluent) with a dedicated spray model (DPM),
and a 1D spray model from literature are used for this purpose. The 1D model of Versaevel et
al [VMW00] is relatively simple compared to DPM, but has some major advantages when used
in 3D simulations. Validation of these spray simulations is done with constant volume, high
pressure cell measurements from several engine research groups (TU/e, Sandia, IFP). When
spray formation results are satisfactory, the second aim is to include ignition and combustion
by means of a detailed, though efficient, tabulated chemistry method called FGM.
First of all, a literary study on spray fundamentals is reviewed in Chapter 2, with special
attention to fuel sprays in engine like conditions. Then, numerical models are classified into
three groups of varying complexity levels in Chapter 3. Furthermore, in Chapter 4 inert
1
2 CHAPTER 1. INTRODUCTION
sprays are modeled in a CFD environment with the standard available DPM model, and the
results are discussed. Subsequently, a 1D spray model is implemented and validated and
then coupled to 3D CFD in Chapter 5, and the results are discussed and compared with the
former spray model results. In Chapter 6 a first attempt to add ignition and combustion to
the transient and turbulent spray model is considered, and improvement points are discussed.
Finally, the conclusions and recommendations are given in Chapter 7.
Chapter 2
Fundamentals of Liquid Sprays
The concept of injecting liquid through a small hole may seem a trivial process, but the
physics of spray formation proves to be extremely complex. Although the analysis of liquid
spray formation is a science discipline on its own, understanding some of its physical aspects
is already valuable for numerical modeling. In this chapter the fundamentals of liquid sprays
in general, like spray regimes, droplet formation and breakup regimes, are presented. These
considerations imply no assumptions about the liquid, so they hold for a liquid spray in
general. But in this study particularly fuel sprays in modern diesel engines are relevant,
therefore in the last section some remarks concerning direct injection diesel spray formation
and combustion are included.
The presented information in this chapter is mostly based on the books of Baumgarten [Bau06]
and Stiesch [Sti03].
2.1 Spray Regimes
Diesel engine sprays are usually of the full-cone type. This means that in the idle mode
the fuel is blocked from the upstream side of the nozzle and during injection the core of the
spray is more dense than the outer regions. In spark-ignition engines however, gasoline is
commonly injected with a hollow-cone type of injector, which will not be discussed in this
study. See Figure 2.1 for a schematic drawing of a full-cone spray. The liquid spray can be
characterized by distinguishing five spatial regimes. Starting from the nozzle exit first there
is an intact liquid core. A few nozzle diameters further downstream in the so-called churning
flow the liquid consists of ligaments (blobs). These liquid parts are like large droplets with
sizes comparable to the nozzle diameter. Then the ligaments breakup into many smaller
droplets in the thick zone where the volume and mass fraction of the liquid phase is high.
Further downstream the breakup process of droplets goes on and in the same time more and
more of the surrounding gas is entrained into the spray area. This results in a diverging
behavior with a characteristic spray angle. The regimes after the thick zone are the thin zone
(low volume but still high mass fraction of liquid) and the dilute zone (negligible volume and
low mass fraction of liquid), respectively.
3
4 CHAPTER 2. FUNDAMENTALS OF LIQUID SPRAYS
Figure 2.1: Full-cone spray example; definition of spray angle, spray length and appearance of different
regimes [Bau06]
In a hot environment the position at which all liquid droplets are evaporated is called the
liquid length. From (automotive) experiments this liquid length is found to be approximately
constant after a short development time. From that point on the evaporated fuel continues
to penetrate the surrounding gas and is denoted as vapor length. In a typical diesel injection
timeframe (few milliseconds) the vapor length does not reach a steady state.
2.2 Breakup Regimes
The disintegration of liquid jets is described by two main mechanisms. The first mechanism
is the breakup of the intact liquid core into droplets and is called primary breakup. This
mechanism is characterized by the droplet size and the breakup length, which is defined as
the length of the intact liquid core. The second mechanism is the breakup of droplets into
smaller ones, which is called secondary breakup. Here the size of the droplets is a characteristic
parameter. Both breakup length and droplet size are dependent on the properties of the liquid
and the surrounding gas. At least as important is the relative velocity between the liquid and
the surrounding gas.
The primary breakup is the most important mechanism in fuel injection systems, because it
determines the size of the droplets that separate from the liquid core, hence therefore also
determines evaporation behavior and it marks the starting point for further breakup into
smaller droplets (secondary breakup). It is also far more difficult to analyse primary breakup
both experimentally and numerically. In the following the breakup regimes are treated in
more detail, but just for clearness the scheme in Figure 2.2 can be kept in mind to have the
big picture right.
2.2. BREAKUP REGIMES 5
Spray
Breakup
Primary Secondary
Breakup Breakup
Aerodynamic
Rayleigh First wind-
droplet breakup
regime induced regime
regimes
Second wind- Atomization

induced regime regime
Turbulence Cavitation
Figure 2.2: Breakup scheme; treated regimes
2.2.1 Primary Breakup
The primary breakup mechanism concerns the breakup of the intact liquid core and can be
divided into four regimes. Namely, the Rayleigh regime, the first and second wind-induced
regimes and last but not least the atomization regime. In order to make a quantitative
classification of the regimes the Ohnesorge number Oh is introduced:

W el
Oh = . (2.1)
Rel
Herein the Weber number W el and the Reynolds number Rel are defined as:
u2 D l
W el = , (2.2)

u D l
Rel = , (2.3)
l
is the density, is the surface tension, is the dynamic viscosity, u is the jet velocity and
D is the diameter of the nozzle. The subscript l denotes the properties of the liquid. The
Weber number is the ratio between inertial (or aerodynamic) and surface tension forces. The
Reynolds number is the ratio between inertial and viscous forces. Substitution of (2.2) and
(2.3) into equation (2.1) gives:
l
Oh = . (2.4)
l D
Thus, the Ohnesorge number is a ratio between viscous forces and surface tension forces.
Now all relevant liquid properties are incorporated, so the various regimes can be classified
in the space Oh as function of the jet velocity, or alternatively Rel . See Figure 2.3 for the
so-called Ohnesorge diagram. In this figure the four regimes and also the relevant zone for
diesel injection applications are indicated.
Figure 2.3: Ohnesorge diagram; four primary breakup regimes, and region for diesel injection applica-
tion indicated [Sch03]
It appears that also the density of the surrounding gas influences the breakup process. By
increasing gas density enhanced atomization is achieved. Therefore the ratio of the gas and
liquid densities is used to span a three dimensional space. A schematic representation of such
a three dimensional Ohnesorge diagram is seen in Figure 2.4.
Figure 2.4: 3D Ohnesorge diagram; primary breakup regimes as function of the Ohnesorge number
(Z = Oh), Reynolds number and gas to liquid density ratio [Bau06]
Now the four regimes are described in more detail by increasing jet velocity. See also the
corresponding illustrations in Figure 2.5.
Rayleigh regime Breakup at low jet velocity due to axisymmetric oscillations initiated by
liquid inertia and surface tension forces. Ddroplet > Dnozzle , the breakup length Ljet is long
and by increasing jet velocity u also Ljet increases.
First wind-induced regime Liquid inertia and surface tension forces are amplified by
aerodynamic forces. The relevant Weber number for this regime is:
u2rel D g
W eg = . (2.5)

Here urel is the relative velocity between liquid and surrounding gas and the subscript g
denotes the gas properties. Ddroplet Dnozzle , Ljet > Dnozzle and by increasing jet velocity
u the breakup length Ljet decreases.
Second wind-induced regime The flow in the nozzle is turbulent. Instable growth of
short wavelength surface waves initiated by the turbulence and amplified by aerodynamic
forces. Ddroplet < Dnozzle and by increasing jet velocity u the breakup length Ljet decreases.
Atomization regime Breakup at surface directly at the nozzle hole, so the intact core
length Ljet goes to zero. Conical spray develops immediately after leaving the nozzle. Ddroplet
Dnozzle , relevant regime for diesel sprays.
Figure 2.5: Schematic representations of the primary breakup regimes: (a) Rayleigh regime, (b) wind-
induced regime, (c) atomization regime [Sti03]
2.2.2 Secondary Breakup
The secondary breakup mechanism concerns the breakup of droplets due to aerodynamic
forces that are induced by the relative velocity between the droplets and the surrounding gas.
On the gas-liquid interface instable growth of waves occur, while in the same time surface
tension counteracts the disintegration process. Similar to the first wind-induced regime for
the liquid core the gas Weber number is the relevant dimensionless quantity to identify the
process, with the only difference that the nozzle diameter D in equation (2.5) is replaced with
the droplet diameter before breakup d :
u2rel d g
W eg = . (2.6)

Decreasing the droplet diameter d raises the surface tension force . This means that the
critical relative velocity, the relative velocity at which breakup takes place, must be higher.
W eg in equation (2.6) is used to separate the droplet breakup regimes. The values at which
transitions from one regime to another occur, are determined experimentally. A schematic
representation of different droplet breakup processes are depicted in Figure 2.6. In the same
figure the corresponding Weber numbers are indicated according to Wierzba [Wie90], but also
other classifications exist in literature. In engine sprays all droplet breakup regimes occur at
vibrational breakup (We 12)
bag breakup (We < 20)
bag/streamer breakup (We < 50)
stripping breakup (We < 100)
catastrophic breakup (We > 100)
Figure 2.6: Droplet breakup regimes and corresponding transition Weber numbers according to Wierzba
[Wie90][Bau06]
the same time. Near the nozzle the Weber number is high, so most of the breakup takes place
at the nozzle exit. Further downstream the Weber number is lower due to smaller droplet
diameters and lower relative velocities. Therefore the breakup far from the nozzle is much
less.
2.2.3 Atomization in Diesel Sprays
Modern injectors for diesel engines have nozzle diameters of 200 m or less, and the length
of the nozzle hole is approximately 1 mm. Injection pressures up to 200 MPa are used and
therefore the jet velocity u reaches values of 500 m/s and more. These conditions result
in an atomization regime for the primary breakup mechanism. Some possible sources for
atomization are shortly treated in the following.
Aerodynamic shear Aerodynamic shear forces amplify the surface waves created by the
turbulence in the nozzle hole. The waves separate from the jet and form droplets. There
are two reasons why this aerodynamic source is less important. First, this process is time
dependent, but it is known from experiments that jets break immediately at the exit of the
nozzle. Second, aerodynamic breakup is a surface effect, so it cannot explain disintegration
of the inner structure.
Relaxation of velocity profile At the wall inside the nozzle a no-slip boundary conditions
exists, forcing the flow towards a Poiseuille velocity profile. When the liquid exits the nozzle,
the velocity profile will transform into a uniform one. In order to realize that the outer region
of the liquid accelerates, which may cause instabilities and ultimately result in breakup into
droplets. However, in modern diesel engines the length to diameter ratio of the nozzle hole is
L
typically small ([ D ]nozzle = 5), so probably the flow in the nozzle has no time to develop.
Turbulence The presence of radial turbulent velocity fluctuations in the jet results, if strong
enough to overcome the surface tension, in formation of droplets. Turbulence-induced primary
breakup is considered one of the most important mechanisms in high pressure applications.
Cavitation Cavitation is the transition from liquid to gas due to the decrease of static
pressure below the vapor pressure. The curved streamlines at the upstream edge of the noz-
zle result in a radial pressure gradient. So, at places where the pressure is lower than the
vapor pressure, cavitation bubbles are formed. These bubbles in the liquid flow contribute
to primary breakup since they implode when they enter the high pressure environment. Pa-
rameters that influence cavitation are the upstream nozzle edge and the angle between the
injector needle axis and the nozzle hole axis. A sharper edge results in stronger cavitation,
which in turn results in smaller ligaments and a larger cone angle. If the angle between the
needle and the hole is too large, the flow in the nozzle and also the spray is asymmetric
due to the asymmetry of the streamlines. Although cavitation is strongly dependent on the
injector/nozzle geometry, the cavitation number K is an important dimensionless parameter
to predict the inception of cavitation. The cavitation number is defined as follows:
p1 pvap p1
K= . (2.7)
p1 p2 p1 p2
The indices 1 and 2 refer to the upstream and downstream pressures respectively and pvap is
the vapor pressure of the liquid. Since in automotive applications pvap p1 the vapor pressure
may be eliminated from equation (2.7). K is defined such that it decreases with increasing
cavitation intensity. To include the influence of the geometry, an empirical criterium is used
to decide when cavitation occurs. This and more about computational considerations are
treated in the following chapters.
2.3 Direct Injection Diesel Combustion
Until now a qualitative description of spray formation is given without taking combustion
into account. In DI diesel engines however, liquid fuel is injected into an environment at high
pressure and temperature that causes the fuel to evaporate and finally to burn. Here some
general remarks are reproduced from the conceptual model of Dec [Dec97] that is derived
from laser-sheet imaging.
As mentioned before, from the start of injection first the spray penetrates into the combustion
chamber, whereby due to entrainment of ambient gas a spreading angle is observed. Short
after the start of injection the liquid fuel breaks up into ligaments (primary breakup) and
further into smaller droplets (secondary breakup), that in the mean time are heated and
evaporated by the high temperature entrained gas. The point at which all fuel droplets
are evaporated is called the liquid length. This short recapitulation is shown in Figure 2.7
as function of time. The liquid length is depicted with the dark color and has a value of
approximately 20 mm, that can occur for diesel like fuels at typical engine conditions. In the
region direct after the liquid length an excess of fuel exists.
Figure 2.7: Temporal sequence of schematics showing how a DI diesel spray evolves from the start of
injection [Dec97]
In time, somewhere in the spray the equivalence ratio and temperature reach values appro-
priate for ignition. Then the premixed part of the spray auto-ignites and a flame establishes
at the so-called flame lift-off length. This flame is still rich in fuel and produces lots of soot
precursors that move downstream and form soot particles. Finally, most of the soot and its
precursors combust in a diffusion flame surrounding the spray. This hot reaction layer is the
main location of NOx formation [Lui08]. See Figure 2.8 for the combustion evolution. Again
the liquid length and mixed gas region in front of it are shown, but now also the flame lift-off
and soot concentration are indicated.
2.3. DIRECT INJECTION DIESEL COMBUSTION 11
Figure 2.8: Temporal sequence of schematics showing how DI diesel combustion evolves in the early
part of the mixing-controlled burn [Dec97]
Chapter 3
Classification of Models
In-cylinder processes in engines can be modeled on different complexity levels. Some main
model categories are shown here in order to understand the applicability and shortcomings
for modeling sprays.
3.1 Thermodynamic Models
Thermodynamic models originate from the laws of thermodynamics. In these models there
is no flow field present in the computational domain. Instead, the whole domain is ideally
mixed, so the temperature, pressure, species etc. are distributed homogeneously. Due to the
relative simplicity of the models small computational effort is sufficient. Apart from empirical
relations, like for heat release rate, subprocesses cannot be incorporated. This means that
with these simple models important engine parameters like pollutant formation due to local
mixing behavior cannot be predicted. Therefore this class of models is not considered in the
rest of this study.
3.2 Phenomenological Models
In phenomenological models the computational domain is divided into multiple zones which
can have different temperatures, compositions etc. . But also in this model class there is no
flow field present. Spray modeling with phenomenological models is very common, because of
the small computational effort needed, although more than for thermodynamic models. They
are mostly applied to compare spray and liquid lengths for sprays in constant volume vessels
without an initial flow field. As soon as mixing behavior in internal combustion engines is
considered, the (turbulence) interaction of sprays with the surrounding flow is important, so
more advanced models are needed.
Existing phenomenological spray models are roughly distinguished into two modeling schools,
namely the droplet limited evaporation and the mixing limited evaporation [Lui08]. In the
droplet limited case, as the name says, the droplet evaporation process is the limiting factor
for spray formation. Hereby the local conditions around the droplet are important, whereas
13
14 CHAPTER 3. CLASSIFICATION OF MODELS
in the mixing limited case the vapor-liquid (thermodynamic) equilibrium is assumed to settle
immediately, which makes the amount of air entrained a direct limiting factor. The immediate
thermodynamic equilibrium assumption makes the mixing limited models only applicable for
high ambient temperatures and densities, such as in modern diesel engines. The droplet
limited models are valid for any conditions, but are also more complex. In Chapter 4 an
example of the droplet limited approach is used and in Chapter 5 a mixing limited model is
studied in detail.
3.3 CFD Models
In CFD (Computational Fluid Dynamics) models the computational domain is divided into
cells of specific sizes for which standard mass, momentum and energy equations are solved.
Submodels are added to incorporate for example turbulence and spray formation. Interaction
between the flow field and the spray is reached by solving the equations in an iterative fashion.
A possible disadvantage could be that relative small errors in submodels may result in large
errors in the overall result. Also the complexity of the model that makes the necessity of high
CPU power inevitable can be a serious disadvantage. Figure 3.1 gives an indication of the
computational expenses needed for each model class, note the logarithmic scale on the y-axis.
Figure 3.1: Computational effort needed for model classes ranked by complexity [Sti03]
Commercial codes like Fluent and StarCD and open source codes like KIVA supply submodels
that are specially developed for sprays under engine like conditions. In the Combustion
Technology group at the Eindhoven University of Technology, mainly Fluent is used for heat
and flow problems in engine applications, therefore in this study Fluent is the first choice to
model diesel fuel sprays. More about Fluent and its spray submodel is presented in Chapter
4.
Chapter 4
CFD Spray Model
Considering multidisciplinary heat and flow problems, with complex interaction between fluid
phases and a variety of (moving) geometries, the best way to reach numerical solutions is
probably to use a software package that can solve transport equations for all kind of flows
under user specified constraints. Specific geometries/constraints should of course be defined
using a proper user interface and results should be easy to postprocess. Fluent is an example
of a commercial code that more or less meets the mentioned demands, hence the choice for
Fluent to model engine sprays in this study.
This (and also the next) chapter deals with evaporating non-reacting fuel injections only, since
spray formation modeling is best achieved when other modeling features that are necessary
for practical engine simulations with burning sprays and moving pistons are absent. In the
engine community it is very common to investigate diesel injection in a constant volume, high
pressure cell. Through the years experimental techniques are developed to measure important
spray properties like liquid and vapor penetrations and spray angles. Spray penetration and
dispersion are needed to promote fuel-air mixing, but impingement and collection of liquid
fuel on in-cylinder walls (wall wetting) can lead to greater harmful emissions [Sie99]. The
fact that many research groups worldwide (like TU/e, IFP, Sandia) do such experiments with
inert sprays gives the opportunity to use those measured data for validation purposes.
In the following, first the solved equations in Fluent are presented, with special attention
for the spray submodel. Next, the model settings and the resulting solutions are discussed.
Finally, some model dependencies such as mesh size and timestep are investigated.
4.1 Fluent
4.1.1 Transport Equations
Fluent solves a number of transport equations depending on the users specific problem setup.
In this section an overview is given of the (general) continuity, momentum, energy, species
and turbulence equations [flu06b]. Additional models and settings that are required to deal
with sprays are treated in the next sections.
15
16 CHAPTER 4. CFD SPRAY MODEL
Continuity equation The general continuity equation is written as follows:

~
+ (~v ) = Sm , (4.1)
t
where Sm is a mass source from the discrete phase due to evaporation of droplets. It can also
be a user-defined mass source.
Momentum equations The momentum equation that is solved in this study is:
~ (~v~v ) = p
~ +
~ + F~ .
(~v ) + (4.2)
t
Here p is the static pressure, is the stress tensor and F~ is a body force due to interaction of
the discrete phase with the continuous phase and/or a user-defined momentum source. The
gravity term in the momentum equation is neglected because of the minimal contribution
compared to the high momentum injection event.
Energy equation The energy equation in Fluent is written as follows:

X
~ [~v (E + p)] =
(E) + ~ (k + kt )T
~ hj J~j + ( ~v ) + Se , (4.3)
t j
where the term between the brackets on the right hand side consists of energy transfer due to
conduction, species diffusion and viscous dissipation, respectively. Se is a user-defined energy
source. Energy E is defined as follows:
p ~v ~v
E =h + . (4.4)
2
Herein h is the enthalpy for ideal gases, and is written as a summation of mass fractions times
species enthalpy: X
h= Yj hj . (4.5)
j
It is important, especially for Chapter 5, to state that the enthalpy is calculated by integrat-
ing the specific heat from Tref to the instantaneous temperature T , whereby the reference
temperature in Fluent is 298.15 K:
Z T
hj = cp,j dT, Tref = 298.15K. (4.6)
Tref
Species transport equations In spray simulations there are at least two different species,
one species is in the gas phase (oxidizer) and an other one is injected (fuel), which after
evaporation goes into the gas phase where it can mix with the oxidizer. N 1 transport
equations for N species are solved because the sum of fractions must equal one. The transport
equation for the ith species is as follows:
~ (~v Yi ) =
~ J~i + Si .
(Yi ) + (4.7)
t
4.1. FLUENT 17
Si is again a source from the liquid droplet phase that is activated when evaporation occurs.
Also user-defined sources are included in this term. Species transport due to diffusion is
calculated via the diffusion flux J~i . For turbulent flows this flux is:

t ~
J~i = Di,m + Yi , (4.8)
Sct
where Di,m is the diffusion coefficient of the ith species in the mixture. t is the turbulent
dynamic viscosity and Sct is the turbulent Schmidt number:
t
Sct = , (4.9)
Dt
which is equal to 0.7 by default. Dt is the turbulent diffusivity.
Turbulence equations Turbulence is dealt with the transport equations for the turbulent
kinetic energy k and its dissipation rate . Here the realizable k- model is preferred because
it is more suitable for axisymmetric jets than the standard one [flu06b].

~ (k~v ) =
~ t ~ + t S 2 ,
(k) + + k (4.10)
t k

~ (~v ) =
~ t ~ 2
() + + + C1 S C2 . (4.11)
t k +
4.1.2 Discrete Phase Model
Fluent provides a model that is specially developed for spray simulations, or more general
suspended particle trajectory simulations. This is the Discrete Phase Model (DPM) and it
is based on the so-called Euler-Lagrange method. In the computational domain there are
two separate phases present, namely the continuous and the discrete phase (particles). The
transport equations from the previous section are solved for the continuous phase only and
the motion of particles is dealt with particle trajectory calculations. Through an iterative
solution procedure the mass, momentum and energy interaction between both phases can be
realized. Some important aspects of the DPM model are presented in this section. For more
information the reader is referred to the Fluent users guide [flu06b, Chapter 22].
Atomizer In order to simulate spray formation, (discrete) liquid particles have to be in-
troduced to interact with the present (continuous) gas phase. As described in Chapter 2, in
diesel sprays the primary breakup takes place in the atomization regime. So, it is assumed
that there is no liquid core; all the liquid is formed into droplets immediately after the exit of
the nozzle hole. That is where the so-called atomizer model comes into play. The atomizer
creates initial conditions, that depend on the internal nozzle flow, for further particle trajec-
tory calculations by defining initial droplet diameter, velocity and the cone angle of the spray.
Here the procedure to determine the internal nozzle flow state and its consequence for the
calculation of initial quantities is presented without going into the details.
In Fluents Plain-Orifice Atomizer Model three kinds of internal nozzle flows are defined,
namely single-phase, cavitating and flipped flows. Figure 4.1 shows schematic cross-section
drawings of those possible nozzle flows. The upstream radius r, hole diameter d and length
L of the nozzle are geometrical details that are used as parameters in empirical relations.
On the righthand side also the corresponding criteria based on the cavitation number K (see
equation (2.7)) are given. Kincep and Kcrit are the cavitation number at which inception
occurs and the critical cavitation number, respectively. These cavitation numbers can be
obtained with empirical relations based on experimental data. Cavitating nozzle flow is the
main regime that occurs in todays high pressure diesel injectors.
Figure 4.1: Plain-Orifice Atomizer, possible nozzle flows with cavitation number criteria [flu06b]
Once the internal flow state of the nozzle is known, the calculation of the initial droplet
diameter and velocity for the cavitating case proceeds according to the scheme in Figure 4.2.
Given the predetermined cavitation number, the case data (r, d, p1 and p2 ; see Figure 4.1
for definitions), and material properties in the upper left frame, the discharge coefficient Cd
and the initial velocity u0 are calculated. Then, via the effective mass flow rate mef f and the
effective nozzle diameter def f an initial droplet diameter is obtained.
To complete the initialization of the droplets, apart of the size and velocity, the initial direction
(cone angle) should be defined. This is again done by an empirical relation, but now for the
cone angle. The cone angle, from now on also-called spray angle, is twice the angle between
the outer boundary of the spray and the main spray axis. In literature there are several spray
angle relations proposed. In Chapter 5 three of those empirical relations are compared, but
for now only the relation of Reitz & Bracco, that is used in the DPM model, is presented:
s
2 3a
tan = , (4.12)
2 3CA f
4.1. FLUENT 19
K, r, d, p1, p2
m, rl , s, pvap
u0
d0
.
Cd meff deff
Figure 4.2: Atomizer scheme; the way initial velocity and diameter are calculated
with
L
CA = 3 + . (4.13)
3.6d
This angle relation seems complete because it takes into account fluid property as well as
nozzle geometry information. However, later on it will be clear that also other relations with
similar appearances give rather different results.
Particle motion The trajectory calculation of a discrete phase particle is done by inte-
grating the force balance on the droplet. The force balance in vector notation is written as
follows:
d~vp p ~ v ),
= FD (~v ~vp ) + ~g + ~vp (~ (4.14)
dt p p
where the left hand term is the acceleration of the particle in question, the term with FD is
the drag force on the particle. FD is defined as:
18 CD Re
FD = . (4.15)
p d2p 24
The drag coefficient CD is determined from the dynamic drag model that accounts for the
effects of droplet distortion, linearly varying the drag between that of a sphere and a disk
[flu06b]. The term with ~g in equation (4.14) is the contribution of the gravitational accel-
eration. In this study the gravitational effect is neglected because of the very low mass of
the droplets and the short injection times. The last term is an additional force that arises
due to pressure gradients in the fluid. However this contribution is accounted for, in case
of a relative large constant volume pressure cell, gradients of pressure are not that large, in
particular when non-reacting sprays are concerned.
Phase coupling While the discrete particle phase is always influenced by the continuous
phase solution (one-way coupling), the other way around (two-way coupling) is just provided
as an option. In the one-way coupling case the continuous phase is solved first thereafter the
particle trajectory calculation is performed. When two-way coupling is applied an iterative
procedure is followed. Then, after the particle trajectory calculation the continuous flow field
is solved again with updated source terms until convergence is reached. See Figure 4.3 for
a graphical representation of the procedure. Because the discrete phase during an injection
event possesses high momentum, thus affects the continuous phase considerably, the two-way
coupling is turned on.
continuous phase flow field calculation
particle trajectory calculation
update continuous phase source terms
one-way coupling
two-way coupling
Figure 4.3: Solving procedure for one-way and two-way coupling
Source terms The exchange of mass, momentum and energy between the continuous and
discrete phases is computed by the change of the concerning quantity as a particle passes
through a computational cell, see Figure 4.4. These changes act as sources in the continuous
flow calculation. So in case of a non-reacting spray in a hot environment, the mass and
momentum sources are positive. But the energy source is usually negative (energy sink)
because the fuel with a relative low pre-injection temperature has to be heated and possibly
evaporated. For non-reacting droplets Fluent makes a distinction between three modes of
heating/vaporization. The first one is heating without vaporization until the user-defined
vaporization temperature is defined. From then, the droplets can heat up and vaporize at
the same time. The vaporization temperature is of course an artificial boundary between
heating only and vaporization, because liquid can vaporize at any temperature, hence the
concept of vapor pressure. But in this way the mass exchange calculation due to vaporization
at low temperatures can be neglected to save time. Finally, when the user-defined boiling
temperature is reached all added heat to the particles is used for vaporization, so the droplet
temperature does not change any more.
Collision and breakup During the motion of droplets throughout the domain they can
breakup into smaller droplets in several different ways as treated in the chapter on spray
fundamentals. But the droplets can also collide with each other. Both phenomena are covered
in Fluent.
Fluent provides two droplet breakup models, the Taylor Analogy Breakup (TAB) model for
low Weber number injections and the Wave breakup model for high Weber number injections.
In typical fuel injection systems (W e 100) the Wave breakup model of Reitz [Rei87] is
the best option. The Wave breakup model considers the breakup due to the relative velocity
between the gaseous and liquid phases. The shear-off of child droplets from the parent droplet
is induced by the growth of Kelvin-Helmholtz instabilities on the liquid surface. The rate of
change of the droplet radius and the resulting child droplet size are related to the frequency
and wavelength of the fastest growing surface wave [KSR99]. Expressions for and as
function of Weber numbers W e and Ohnesorge number Oh are fitted to numerical stability
analysis solutions. Further details about the stability analysis do not add significant value to
4.1. FLUENT 21
Figure 4.4: Discrete phase particle traveling through a continuous phase cell, exchanging mass, mo-
mentum and energy [flu06b]
the rest of this study, for more details the reader is referred to Reitz [Rei87] or the Fluent users
guide [flu06b]. A schematic representation of the breakup model is depicted in Figure 4.5.
Herein, r is the radius of the newly formed droplet as functions of the wavelength and the
model constant B0 (= 0.61 [Rei87]). is an infinitesimal axisymmetric surface displacement
imposed on the steady motion and acts as a basis for the stability analysis.
Figure 4.5: Sketch of the Wave breakup principle [KSR99]
Collisions between droplets may be important in regions with a high particle density. Looking
at the whole computational domain there are N droplets that result in 12 N 2 possible collision
pairs. This implies the need for huge computing capacity since millions of droplets exist,
or more practical, there is need for a simplified method. The last approach is achieved
through the use of ORourkes algorithm that reduces the computational expense for practical
sprays through the introduction of parcels, considering only in-cell collisions and stochastic
estimation. The so-called parcel is a group of particles that have identical properties like
droplet diameter, temperature and velocity. Now, for example just several thousands of
parcels exist that in collision calculations are treated as they where single droplets, reducing
the amount of possible collision pairs tremendously.
In contrast to breakup, collisions decrease the amount of particles because of coalescence. But
also inelastic collisions (bouncing) are possible that do not change the number of particles.
Once it is determined that two parcels collide, the following step is to decide whether the

result should be coalescence or bouncing. This is done by comparing y(r1 + r2 ), where y is a
random number between 0 and 1 and r is the droplet radius, with a critical value expression
as function of the droplet radii and the relative Weber number. If the value is below critical,
then the collision results in coalescence of the droplets and the new properties are calculated
from basic conservation laws. Otherwise the collision is inelastic and the new properties are
calculated from momentum and kinetic energy conservation, but part of the kinetic energy is
also dissipated.
Limitations So far the DPM model of Fluent seems to contain all necessary modeling
features to capture most of the spray physics present in diesel injection systems, but there
are also some major shortcomings from a computational point of view. Apart from issues
that are described in the next sections, probably the most important drawback of the DPM
model arises from the Euler-Lagrange approach assumption that at most 10 to 12 volume
percent of a cell should contain discrete phase particles. Otherwise the discrete phase would
occupy a significant amount of the continuous phase volume, whereas in the continuous phase
calculation the volume is constant and equal to that of the user-specified size. This would give
erroneous interaction sources between the two phases. In practice this restriction means that
computational cells, especially near the nozzle exit, must be big enough. This is the point
where a tradeoff have to be made between relative large cells in favor of the DPM model on
one hand, and small cells to solve the high velocity flow field as accurate as possible on the
other hand. A direct consequence of large cells is the cell shape and orientation dependency
of the results [Sti03].
Additionally, when cell sizes are decreased to improve the flow field resolution, the statistics
(related to the amount of parcels) would run into convergence problems [SBK+ 04]. This has
to do with the low number of parcels per cell, therefore the total amount of parcels should be
increased, leading to a huge number of parcels and therefore also very high computing times.
Despite the known limitations of the DPM model, it is worthwhile investigating to what
extent these limitations restrict the reach of the ultimate goal; modeling direct injection of
a reactive spray in the variable volume of an auto-ignition engine. Therefore the following
sections show the applied Fluent settings and the resulting solutions.
4.1.3 Model Settings
In the section with the results of the DPM spray model, comparisons with experimental data
gained from high pressure cell setups with a constant volume are done. This approach to
validate spray models is very common, because in a constant volume cell the mean pressure
stays approximately constant even when combusting sprays are considered. Important fea-
tures like spray angle and penetration are then relatively easy to measure due to the controlled
and reproducible conditions. In this study experimental data from several research groups is
4.1. FLUENT 23
used. For more information on their specific experimental layout and measuring techniques
see [Pet07][KDFS+ 06, Eindhoven High Pressure Cell], [VVB98, Institut Francais du Petrole]
and [ECN, Sandia National Laboratories].
First of all a mesh is constructed to define the constant volume environment wherein the spray
is simulated. This is done with the default drawing and mesh creating tool of Fluent, named
Gambit. Two different base meshes are created taking the model related cell size restriction
into account. In typical high pressure fuel injection cases a cell size near the nozzle of 1 mm3
is common and gives the most realistic results [Sti03]. The first base mesh consists of square
cells. In order to reduce computational expense and because of symmetry, only a quarter of
the spray is captured, see Figure 4.6 for a three dimensional view. The base square mesh
Figure 4.6: Square mesh, 14 th piece of the spray
has 90,000 cells which all are 1 mm3 big. The other base mesh is a 18 th slice of a cilinder. For
the same magnitude of mean cell dimensions near the nozzle this mesh contains much less
cells than the square one, namely 6,000. See Figure 4.7 for a three dimensional picture.
Additional variations to the base mesh configurations are only in cell size. Especially the
square mesh is used for cell size variation in different directions because this is much easier
to accomplish than for the slice mesh. Boundary conditions on the mesh surfaces are set as
follows. The cutoff cross-sections are symmetry boundaries. All other surfaces have adiabatic
constraints.
The meshes do not include detailed interior geometry of the experimental constant volume
apparatus. This is allowed because the high pressure cells have cube or cilinder like volumes
that are much bigger than the space occupied by the spray.
In the next section only simulation results for heptane sprays are considered. By choosing a
single-component fuel like heptane, all temperature dependent material properties are defined
Figure 4.7: Slice mesh, 18 th slice of a cilinder
relatively simple. The specific heat, thermal conductivity, viscosity, vapor pressure and surface
tension of the liquid heptane are defined as function of temperature. Also the specific heat
of all gaseous species is set temperature dependent. All these data are gained from the
thermophysical database from DIPPR [DIfPP].
In the DPM model two-way coupling of the phases and droplet collision and breakup are
enabled. The spray origin, spray direction, initial temperature, nozzle diameter and with the
modeled section corresponding mass flow are prescribed.
4.2 Results and Parametric Dependencies
For now only evaporating, non-reacting heptane sprays are considered. Validation of the DPM
model is done with an in-house measurement in the Eindhoven High Pressure Cell (EHPC)
and with a measurement of Sandia National Laboratories. These measurements contain spray
length, liquid length and spray angle information. Well developed experimental techniques
exist to measure these quantities, and they represent easy to compare physical properties of
the spray. Details of the experiments that are necessary for numerical modeling are listed in
Table 4.1. Herein the corner radius is defined as in Figure 4.1. Ideally, things like temperature
field and mass fractions of species should be measured to make more accurate comparisons
with (CFD) spray models. But unfortunately these field quantity measurements are not
common practice at the moment.
In the following successively the spray length and liquid length are investigated, and in the
same time some encountered model dependencies are indicated. The spray angle remains
untreated since it is an input to the DPM model, so validation on the basis of angles has
no meaning. However, in Chapter 5 a closer look is taken at the influence of certain angle
4.2. RESULTS AND PARAMETRIC DEPENDENCIES 25
relations.
Table 4.1: Experimental data of EHPC and Sandia measurements
EHPC Sandia
fuel heptane heptane
nozzle diameter [m] 177 100
nozzle length [mm] 1 0.4
corner radius [m] 0.07 d = 12.4 sharp edge
mass flow per nozzle [kg/s] 0.0088 0.0028
ambient temperature [K] 951 1000
initial fuel temperature [K] 323 373
ambient air composition
[mass fraction CO2 /H2 O/N2 ] 0.130/0.060/0.810 0.065/0.038/0.897
ambient pressure [MPa] 2.45 4.33
4.2.1 Spray Length
Spray lengths (vapor lengths) are experimentally determined with a Schlieren imaging tech-
nique [Pet07][NS96][VVB98]. This is a line of site technique that makes use of the deflection
of light that travels through a medium with density gradients. While this is an appropri-
ate method to measure spray lengths, it makes direct comparison with numerical results not
trivial. In order to make validation possible, a numerical technique is developed by Huijnen
[Hui07]. Using this technique an image is constructed with virtual rays of light that travel
through the 3D density domain that is extracted from the model results. The position on
the resulting image where the rays are most far from the nozzle exit is considered to mark
the end of the spray in the length direction. Due to the imitation of the real Schlieren record
with virtual rays of light, this numerical method is called the virtual Schlieren technique.
Mesh dependency Mainly the EHPC heptane measurement, indicated as a dotted line in
Figure 4.8, is used to investigate the sensitivity of the DPM model for various settings. First,
as expected, the DPM model appears to be highly dependent on cell sizes in the (square)
mesh. The red solid lines are spray lengths as function of time for 1 mm3 cubic cells and the
black dotted lines are for the 110.2 mm3 cells. The last ones are only 0.2 mm in the z-
direction at the nozzle exit, but gradually increase to 1.5 mm to the end of the computational
domain. Also the result for a 2 mm3 mesh is shown, see the star markers. The difference with
the two other configurations is very large, and maybe more important, they do not follow
the same increasing trend. Now imagine the spray in a combustion chamber model with a
moving piston. Then, either a computationally very expensive and sophisticated adaptive
mesh can be generated to align cells near the spray such that the initiated flow enters the
cells perpendicular, or the mesh stays aligned with the liner. In the second case, the spray
axis is chosen to have an angle of 15 degrees compared to the z-axis. The spray length for the
unaligned case is indicated in Figure 4.8 with the diamond markers. Not seen in this figure
is the observed asymmetry of the spray around its own main axis due to the imposed angle
in combination with the relative large cells.
Penetration lengths (EHPC)

100
90
80
70
60
SL [mm]
50
40 Cells t [s] Alignment

1 mm3 106 yes
30 1x1x0.2 mm
3
10
6
yes
1 mm3 105 yes
20 3 5
1x1x0.2 mm 10 yes
3 6
2 mm 10 yes
10
1 mm3 106 no
EHPC measurement
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time [ms]
Figure 4.8: Case: EHPC heptane. Spray length as function of time for various mesh configurations
and solver timesteps
Solver timestep Second, the timestep of the time-dependent solver gives rise to very large
differences in spray length. In Figure 4.8 the circular markers are solutions with a timestep
of 105 s and the square markers indicate the 106 s results. Decreasing the timestep to
5 107 s leads to little improvement while computational expenses increase tremendously.
Even a smaller timestep like 107 s is tried, but it gives alternately flipped and cavitating
nozzle flows, and the solution does not converge at all. This is remarkable because the internal
nozzle flow is determined with empirical relations that does not depend on the solver timestep,
but depend on nozzle geometry and fluid properties. Anyway, even the best result (1 mm3
cells and timestep of 106 s) is still far off from the experimental curve.
Physical properties Spray formation includes thermodynamic interaction between two

phases with large temperature differences in a high pressure environment. Therefore material
properties play an important role in spray modeling, even in the case of inert sprays. The
material properties in Fluent are set as function of temperature with data from the thermo-
physical database of DIPPR [DIfPP]. Especially the specific heat, vapor pressure and boiling
point (see the definition under the caption Source terms in Section 4.1.2) are key properties
that have a big influence on the results. The improved results for the spray length are shown
4.2. RESULTS AND PARAMETRIC DEPENDENCIES 27
in Figure 4.9 with red solid lines, once for the square mesh (square markers) and once for the
slice mesh (triangular markers) configuration. In approximately the first 0.5 ms the model
estimates too large lengths. This is due to the assumed constant massflow in the numerical
case, whereas in (EHPC) practice the massflow takes some time to develop after the injection
starts [SSB05]. To account for this phenomenon the massflow is gradually increased. This is
done manually, because unfortunately there is no other/easier way to do so in Fluent. The
dotted blue lines in Figure 4.9 are the results for the unsteady massflow cases. One can see
that the start of the injection is predicted much better, but thereafter, as expected, the pen-
etration lags behind the measured curve. And once again the square and slice meshes show
substantial differences.
Penetration lengths (EHPC)

90
80
70
60
50
SL [mm]
40
30
20
Square mesh, massflow constant
Slice mesh, massflow constant
10 Square mesh, massflow(t)
Slice mesh, massflow(t)
EHPC measurement
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time [ms]
Figure 4.9: Case: EHPC heptane. Spray length as function of time for the square and slice meshes,
and with timedependent massflow
From many simulations of the Sandia case (see Table 4.1) similar trends as for the EHPC
are found, therefore only the best practice result is shown in Figure 4.10. From the former
considerations best practice means the 1 mm3 square mesh with a solver timestep of 106 s,
and of course also this time temperature dependent material properties are used. Now, the
massflow is kept constant because the development time of the Sandia injection is just less
than 0.1 ms [ECN]. Overall one can conclude that also for the Sandia spray the results are
not satisfactory.
Penetration lengths (Sandia)
70
60
50
SL [mm]
40
30
20
10
Square mesh
Sandia measurement
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time [ms]
Figure 4.10: Case: Sandia heptane. Spray length as function of time for the square mesh
4.2.2 Liquid Length
Liquid length (LL) is defined as the position at which all fuel is just evaporated. In an
experimental setup the liquid length is measured using the scattering of light caused by small
droplets. In Fluent it is determined simply by the position of the droplets-parcel that is most
far from the nozzle exit. The amount of parcels is a key issue here, recall the definition of
a parcel from Section 4.1.2 under the caption Collision and breakup. Suppose that too few
parcels exist in the domain at an arbitrary moment in time. The parcel furthest in the domain
prescribes the liquid length at that time. Now suppose that one timestep later that parcel does
not exist anymore because it is evaporated. Because of the little amount of parcels present,
one can imagine that the next furthest parcel is still far upstream compared to the evaporated
one. This causes the liquid length to vary in time as depicted with square markers in Figure
4.11, where 10 new particle streams (or parcels) are introduced every timestep. Increasing
this amount finally makes the liquid length to stay at a steady value, see the 100 and 1000
particle stream solutions marked with circles and stars, respectively.
Figure 4.11 is only shown to state the influence of the chosen amount of particle streams. The
EHPC heptane measurement does not include liquid length information, so the comparison
can only be done for the Sandia experiment. But also the IFP heptane case, which is consid-
ered in the next chapter (see Table 5.1), is already compared concerning the liquid length in
order to give a more complete picture of the reliability of the DPM model. Sandia reports a
steady liquid length of 9.2 mm, whereas the DPM model calculates a length of approximately
4.3. DISCUSSION 29
2.5 mm. This large difference comes from the predicted droplet diameter in Fluent which
is too small, in the order of 0.01 m whereas in practice the droplets have diameters in the
order of 10 m. In the first place the diameter after injection is too small, and secondly after
breakup the droplets become even smaller. The IFP heptane measurement is considerably
different from Sandia. This results in a cavitating nozzle flow instead of a flipped nozzle flow
as in the Sandia simulation and therefore one order of magnitude larger droplets. One would
also expect a smaller liquid length this time, because the droplets are still relatively small,
but remarkably the liquid length reaches values larger than 16 mm whereas the experimental
value is 10.8 mm.
Liquid Lengths (EHPC)
1
LL
particle stream: 10
particle stream: 100
particle stream: 1000
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time [ms]
Figure 4.11: Case: EHPC heptane. Liquid length as function of time for various particle stream
(amount of parcels per timestep) settings
4.3 Discussion
Fluents DPM model is extensively used to model evaporating, but inert heptane sprays. This
is done with special attention for temperature dependent material properties and for many
different setups, including various meshes, solver timesteps and amount of parcels. The results
are compared with a measurement on the EHPC and experimental data from Sandia. From
these comparisons it is found that the DPM model gives unsatisfactory results concerning
spray and liquid lengths. Nevertheless, some best practice setups resulted from this study,
which are at least valid for heptane sprays in engine like conditions. That are 1 mm3 cells
aligned with the spray axis and solved with a solver timestep of 106 s, and injection of 1000
parcels per timestep. Besides these numerical features, setting material properties as function
of temperature is probably the most important.
From a numerical point of view there are also major disadvantages. One of them is the
imposed limitation to mesh refinement which is far from desirable when detailed in-cylinder
mixture formation and combustion are to be modeled. The second disadvantage is that the
discrete phase part of the calculations cannot be parallelized, while those detailed investiga-
tions require fine resolution, thus expensive simulations.
Chapter 5
CFD and Phenomenological Spray

Model
From the previous chapter, about the 3D spray model of Fluent, called DPM, clearly follows
that practical engine sprays are difficult to solve accurately with an Euler-Lagrange method.
The results are highly mesh and timestep dependent and often convergence problems may
occur. Also the statistical approach with parcels is a source of problems due to large com-
puting times when parcels accumulate in the domain. Many authors tried to overcome these
problems by fine tuning the submodels for specific cases, but this is obviously not the way
to go due to the fundamental discrepancy between, limitation to cell sizes and lack of paral-
lelization possibilities, and solving in-cylinder velocity fields and turbulence with increasingly
finer meshes.
However, others also developed (semi) Euler-Euler methods to model spray formation. One of
them is the so-called ICAS (Interactive Cross section Averaged Spray) model [WP97][SBK+ 04]
that combines a 1D (Eulerian) spray model with the existing Langrangian CFD model. The
region near the nozzle exit is then covered with the 1D spray model and somewhere further
downstream droplet source terms are introduced to incorporate 3D spray formation in the
CFD environment. By introducing parcels anyway, the influence of in-cylinder flow to spray
formation would still be captured [Sti03], but to do so the Euler-Euler region should be not
more than a few millimeters from the nozzle exit, because further downstream (behind LL)
there are no droplets anymore. This makes complex, adaptive meshes necessary, and the
DPM model of Fluent would still be needed. However, with the results of the preceding
chapter in mind, the CFD code should ideally be used for gas phase calculations only in order
to get rid of the complex discrete phase interaction. Therefore a 1D model that covers the
complete spray region is presented in this chapter. This model is coupled to Fluent with
appropriate source terms for mass (fuel vapor), momentum and energy. Initially, just like
the DPM model, the continuous phase exists of solely ambient gas. When fuel vaporizes it is
introduced into the computational domain through a fuel vapor source term. Whereas in the
DPM model real droplets and their trajectories are calculated to determine source terms, in
the 1D model this is done relatively simple.
In the following, first the phenomenological spray model proposed by Versaevel et al [VMW00]
is treated extensively. The model is then implemented in Matlab and validated with mea-
31
32 CHAPTER 5. CFD AND PHENOMENOLOGICAL SPRAY MODEL
surements of IFP [VVB98] and Sandia [ECN][Sie99]. Last but not least, source terms are
extracted from the 1D model and put into Fluent via UDFs (User-Defined Function), and the
resulting 3D solutions are also compared with measurements.
5.1 Phenomenological Model
The 1D quasi steady spray model of Versaevel et al [VMW00] is an extension of the earlier
efforts of Naber et al [NS96] and Siebers [Sie99]. Naber and Siebers developed a 1D model
for non-vaporizing spray penetration first, and later Siebers added some thermodynamics to
distinct liquid penetration from vapor penetration. But Siebers contribution is based on
the assumption that only at the steady liquid length position thermodynamic equilibrium
exists. This approach implies that no temperature information is available, except at the
liquid length position. Also the composition of the spray volume between the nozzle exit and
liquid length is unknown. Versaevel et al overcame this shortcoming by introducing a void
fraction m that couples the mass, momentum and energy equations.
The model is based on five basic assumptions from which the first four are the same as in the
work of Naber et al [NS96] and Siebers [Sie99].
no velocity slip between gas and liquid phases

whole system at constant pressure
uniform velocity, density and temperature profiles
constant spray angle
whole system at thermodynamic equilibrium
The last assumption is necessary to gain information about the temperature and composition
of the spray in the entire 1D domain. The mass, momentum and energy equations are
derived by considering a control volume as defined in Figure 5.1. The spray is described in
one direction due to the constant angle and the axisymmetry. The figure shows that from
fuel injection (mf l,0 ) into the x-direction the spray diverges due to air entrainment (ma ) into
the spray volume. Air entrainment is controlled by the prescribed spray angle ( 2 ). For this
purpose an angle relation is chosen, like is done with the DPM model in the previous chapter.
At the liquid length just enough hot air is entrained into the spray to evaporate all liquid fuel,
so from that point on the fuel penetrates the surrounding
gas as a vapor. def f in Figure 5.1 is
the effective orifice diameter defined as: def f = Ca d, with d the geometric orifice diameter
and Ca the area contraction coefficient.
Now the mass, momentum and enthalpy balances as defined by Versaevel et al [VMW00] are
shown, preceded by the description in words.
Mass balance Liquid fuel mass flow rate at the injector hole = remaining liquid fuel mass
flow rate at x + gaseous evaporated fuel mass flow rate at x
f l,0 A0 vf l,0 = f l [1 m(x)]A(x)v(x) + Yf g (x)g (x)m(x)A(x)v(x). (5.1)
5.1. PHENOMENOLOGICAL MODEL 33
. control volume
ma
q /2
.
mfl,0 x
deff df
. gas and liquid gas only

ma
Figure 5.1: Definition of the control volume used in the derivation of the 1D model [VMW00]
Here, is the density, A represents the cross-section area of the spray, v is the spray velocity,
m is the void fraction and Y stands for mass fraction. The subscripts f l, 0, f g and g refer to
the liquid fuel, the nozzle exit, the gaseous fuel and the whole gaseous phase, respectively.
Momentum balance Liquid fuel x-momentum flow rate at the injector hole = remaining
liquid fuel x-momentum flow rate at x + gaseous mixture x-momentum flow rate at x
f l,0 A0 vf2l,0 = f l [1 m(x)]A(x)v(x)2 + g (x)m(x)A(x)v(x)2 (5.2)
Enthalpy balance Liquid fuel enthalpy flow rate at the injector hole + entrained air en-
thalpy flow rate = liquid fuel enthalpy flow rate at x + fuel vapor enthalpy flow rate at x +
air enthalpy flow rate at x
mf l,0 hf l (Tf l,0 ) + ma (x)ha (Ta ) = f l [1 m(x)]A(x)v(x)hf l (T (x))+ (5.3)
Yf g (x)g (x)m(x)A(x)v(x)hf g (T (x)) + ma (x)ha (T (x))
Here, m is the mass flow rate, h represents the enthalpy and T is the temperature. The
subscript a refers to the ambient gas.
The quantities that depend on the spatial coordinate x are indicated explicitly. Only the
spray cross-section area A(x) is known on beforehand through the applied correlation for the
spray angle :
2
def f
A(x) = + x tan . (5.4)
2 2
Now the empirical angle relation of Siebers is used, like Versaevel et al, to compare the results.
Siebers experimental fit is as follows:
!0.19
0.5
a fl
tan = ac 0.0043 , (5.5)
2 f l a
where the factor ac is a fitting parameter. It must be said that the parameter ac is the largest
point of concern of this phenomenological model for conventional diesel sprays. The choice of
a proper coefficient is explained later on. In the next section some common angle relations are
compared to investigate the predictive capacities and give recommendations for improvement.
The coupled non-linear balance equations (5.1), (5.2) and (5.3) are written in an other form
to make it suitable for solving with a non-linear solver. With the nozzle exit velocity of the
fuel:
mf l,0
vf l,0 = , (5.6)
f l,0 A0
and with the velocity at any axial position:
ma (x)
v(x) = , (5.7)
g (x)[1 Yf g (x)]A(x)m(x)
the mass balance equation is written as:

ma (x) m(x)[1 Yf g (x)]g (x)
(x) = = . (5.8)
mf l,0 f l [1 m(x)] + m(x)Yf g (x)g (x)
In the same manner with equations (5.6) and (5.7) the momentum balance is written as:
m(x)2 [1 Yf g (x)]2 g (x)2 A(x)

(x)2 = . (5.9)
A0 f l,0 ([1 m(x)]f l + m(x)g (x))
Also the enthalpy equation is rewritten, using equation (5.7) and (5.8). Additionally an
enthalpy relation that relates liquid and gaseous enthalpies through the heat of vaporization
Lv is applied:
hf l (T (x)) = hf g (T (x)) Lv (T (x)), (5.10)
then the enthalpy equation is written as:
!
Yf g (x)(x)
(x)[ha (Ta ) ha (T (x))] = 1 Lv (T (x)) + Lv (Tf l,0 ) + [hf g (T (x)) hf g (Tf l,0 )].
1 Yf g (x)
(5.11)
Together with a constant pressure ideal gas relation for the density of the total gaseous phase:
Ta 1
g (x) = a (T (x)) Ma
, (5.12)
T (x) Yf g (x) M + [1 Yf g (x)]
f
wherein M is the molecular mass, the coupled non-linear equations (5.8), (5.9) and (5.11)
form the ingredients to describe the 1D spray.
The solution procedure is split into two spatial parts, the liquid length part and the vapor
length part. First the part from the nozzle exit to the liquid length is considered. Here
5.1. PHENOMENOLOGICAL MODEL 35
are three unknowns, namely the temperature T (x), the massflow ratio (x) and the void
fraction m(x). Yf g (x) is not considered as an unknown because of the imposed thermodynamic
equilibrium, that states that the fuel partial pressure is equal to the fuel vapor saturation
pressure:
psat (T (x)) = pa Xf g (x). (5.13)
Mg
Then using the relation between molar and mass fractions Xf g (x) = Mf Yf g (x), Yf g becomes:
1
Yf g (x) = , (5.14)
pa Ma
psat (T (x)) 1 Mf +1
thus Yf g (x) is calculated when a solution for T (x) is found. And then also g (x) is defined
via equation (5.12).
Once the void fraction reaches unity in x-space, the rest of the spray in increasing x-direction
is solved for the fuel vapor with the same three equations (5.8), (5.9) and (5.11) as for the
liquid length part. Now the unknowns are also the temperature T (x) and massflow ratio
(x), but instead of the void fraction m(x) this time the gaseous fuel mass fraction Yf g (x) is
unknown. Downstream of the liquid length position the void fraction is by definition constant
and equal to unity, provided that the fuel does not condensate. g (x) is again determined
afterwards via equation (5.12).
So far the 1D spray model is steady, thus the coupled equations can be solved for any chosen
x-position. Once all quantities are solved for the whole x-domain the next step is to introduce
timedependence, in order to mimic the unsteady behavior of a practical spray. This is accom-
plished by following the assumption of Naber et al [NS96] that the velocity v(x), determined
with equation (5.7), is equal to the velocity of the spray tip. In this way one can construct
an unsteady penetration length SL(t) of the spray from the steady velocity profile v(x):
dSL
= v(x). (5.15)
dt
The phenomenological spray model is completely implemented in Matlab. The standard non-
linear solver of Matlab is used for this purpose. Material properties, except the liquid fuel
density, are temperature dependent, and are obtained from the thermophysical database of
DIPPR [DIfPP]. The current Matlab model works with a run-file with input fields that are
required to be filled in, in order to start a well defined simulation. These inputs are the
case setup data as is shown in Table 5.1 and model related inputs like the preferred spatial
resolution and initial guesses of the three unknowns. Furthermore one can choose to solve
for the steady liquid length or transient spray length only, or both at the same time. And
finally, the user can turn on the plot during calculation option. If plotting is turned on, six
quantities (m, T , , v, Yf g and g ) are shown as function of x during the whole simulation,
otherwise the progress of the simulation is indicated with a progress bar and the simulation
time is much shorter. An example output to the screen during calculation is shown in Figure
5.2. For each defined x-position the quantities are computed and added to the figure until
the loop over all meshpoints is finished.
The curves in Figure 5.2 give some understanding about what happens from x = 0 at the
nozzle exit up to the point that the void fraction reaches a value of 1. For example the
ac = 0.2537 dx = 0.1 mm exitflag = 1 status: FINISHED

1.4 700
1.2 650
1 600
d = 200 mm
0.8 Cd = 0.8 550
T [K]
m [-]
T = 373 K
0.6 0 500
T = 800 K
a
0.4 P = 80 MPa 450
inj
Pa = 6.2783 MPa
0.2 400
= 25 kg/m3
a
0 350
0 5 10 15 20 25 30 0 5 10 15 20 25 30
x [mm] x [mm]
10 500
8 400
6 300
v [m/s]
G[-]
4 200
2 100
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
x [mm] x [mm]
0.5 70
60
0.4
50
0.3
g [kg/m ]
3
40
Y [-]
fg
30
0.2
20
0.1
10
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
x [mm] x [mm]
Figure 5.2: Example output to screen of the 1D spray model during calculation
mass fraction of gaseous fuel increases due to evaporation and has its maximum at the liquid
length. From then on, only gaseous fuel is present while air entrainment keeps on going, so
Yf g decreases. A discontinuity at the liquid length does not appear in the velocity curve, it
declines asymptotically from a maximum velocity at the nozzle exit to zero. At the end of the
simulation also curves of the transient vapor length (SL) and liquid length (LL) are plotted
in a separate window, see the next section for the results.
5.2 Results; 1D Spray Model
In this section IFP [VVB98] and Sandia [Sie99][ECN] measurements are used for validation
purposes. These are all for single component fuels that are well documented, so thermo-
physical data needed for the numerical model is found in literature. The most important
experimental setup information, needed as input for the spray model, is included in Table
5.1. Also now, penetration lengths form the basis for validation for the same reasons as men-
tioned in Chapter 4. But this time more attention is paid to the effect of different spray angle
relations.
5.2. RESULTS; 1D SPRAY MODEL 37
Next, the spray length is shortly viewed. The rest of the comparison is mostly done for
the steady liquid length because in literature only very few measurements of transient spray
lengths for single component fuels is available. Finally, the sensitivity of the model to mea-
surement errors and physical property prediction errors is analyzed.
Table 5.1: Experimental data of IFP and Sandia measurements
IFP Sandia
fuel heptane, dodecane cetane
nozzle diameter d [m] 200 246
contraction coefficients
Cv /CA /Cd 0.8/1.0/- 0.81/-/0.78
initial fuel temperature Tf l,0 [K] 373 438
ambient temperature Ta [K] 800 850/1000/1150
ambient density a [kg/m3 ] 25.0 7.3/14.8/30.2
injection pressure pinj [MPa] 80 136
ambient air composition
mass fraction CO2 /H2 O/N2 0.017/0.114/0.869 0.065/0.038/0.897
5.2.1 Spray Length
Regarding the fitting coefficient ac in equation (5.5), validation of the model by Versaevel et
al [VMW00] with experimental data from IFP [VVB98], results in the choice of two different
values. One of them gives good agreement with the experiments concerning the unsteady
vapor length, and has a value of 0.2537. Versaevel et al argue that this value is in good
correspondence with the findings of Naber and Siebers [NS96]. But for predicting the steady
liquid length this value is too large. To obtain the best fit for the experimental data, Versaevel
et al propose an ac of 0.104 for steady liquid length calculations. Therefore the model is run
two times, once for each ac, and from the ac = 0.2537 solution the timedependent spray
length is constructed as shown in equation (5.15). Subsequently, from the philosophy that
the unsteady paths of the vapor and liquid phases should match until the steady liquid length
is reached, the ac = 0.104 solution is only used to determine the steady value of the liquid
length.
In Figure 5.3 the results for an IFP case with dodecane is shown. The vapor length as is
measured by IFP is indicated with the dotted line, which is a fit of Verhoeven et al [VVB98]
through measured points. The 1D model results are indicated with the black and red line for
ac = 0.25 and ac = 0.29 respectively. Versaevel et al choose ac = 0.25 as a best fit coefficient,
but from the figure it is seen that ac = 0.29 fits better. Therefore in the rest of this study
the angle relation coefficient ac for the vapor length calculations is set to 0.29. The unsteady
liquid length, indicated with the dotted blue line, follows the same curve as the unsteady
vapor length as mentioned before. In correspondence with practice, when the steady value of
20.7 mm is reached the length remains constant. More vapor length comparisons, for heptane
and different ambient densities, between the 1D model and measurements are not shown but
can be found in the paper of Versaevel et al [VMW00]. There it is shown that overall good
agreement is achieved except for low ambient temperatures because of the mixing limited
assumption of this model.
Penetration lengths (dodecane)

50
45
40
35
SL and LL [mm]
30
25
20
15
10
SL with ac = 0.29
LL
5
SL with ac = 0.25
fit through IFP measurements
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time [ms]
Figure 5.3: Case: IFP dodecane. Spray length and liquid length as function of time
5.2.2 Liquid Length
Now the focus is on the liquid length prediction, which is for this study more important than
the vapor length because the liquid length part is used to couple spray formation with CFD
calculations. Figure 5.4 contains steady liquid lengths for various ambient densities. The
experiment points (black stars) are again from IFPs dodecane measurements. The results of
the 1D model of Versaevel et al (red squares) are within 10 % of the experiments, while the
1D model of Siebers (blue circles) [Sie99] predicts liquid lengths that are up to 50 % too large.
This would make no sense since the findings of Siebers in his paper show that his model is
also within 10 % accurate, but in his case compared with Sandia data. Sandias experimental
data for cetane are found in Figure 5.5 (stars). These liquid length measurements are plotted
as function of the ambient temperature and for three different ambient densities. Also the
numerical results of the Versaevel and Siebers 1D models are shown, with squares and circles
respectively. Now Siebers model deviates just up to 10 %, while Versaevels model is up to 20
% off. Both models are based on basic conservation laws and most assumptions are common,
so the striking differences could only have to do with the capability to predict an appropriate
spray angle for a certain case.
In literature there are three commonly used angle relations for vaporizing sprays. One is
Liquid lengths (dodecane)

40
35
30
LL [mm]
25
20
Versaevel (Matlab model)

IFP measurement
Siebers (Matlab model)
15
10 12 14 16 18 20 22 24 26 28 30 32
3
a [kg/m ]
Figure 5.4: Case: IFP dodecane. Liquid length comparison between Versaevel and Siebers models for
various ambient densities
from Siebers as is used in this chapter until now. And the others are from Reitz & Bracco
and Hiroyasu & Arai. These relations are conveniently arranged in Table 5.2. The thing
that attracts attention from this table is the dissimilarity of quantity dependencies. Siebers
empirical relation is a function of fuel and ambient densities only, whereas Reitz & Bracco
also include nozzle geometry information. Hiroyasu & Arais relation is a function of the
nozzle diameter, ambient properties and the pressure drop over the nozzle. The impact of
these differences on the spray angle is made visible in Figure 5.6. Herein the full spray angles
Table 5.2: Three commonly used angle relations

a 0.19 f l 0.5
Siebers tan( 2 ) = 0.26 f l 0.0043 a

3a 0.5
Reitz & Bracco tan( 2 ) = 2
3CA f l , CA = 3 + L
3.6d
h i
a (Pinj Pa )d2 0.25
Hiroyasu & Arai = 0.05 a2
Liquid lengths (cetane)

70
= 7.3 kg/m3
a
= 14.8 kg/m3
a
60 = 30.2 kg/m3
a
Siebers (Matlab model)
Sandia measurement
Versaevel (Matlab model)
50
LL [mm]
40
30
20
10
800 850 900 950 1000 1050 1100 1150 1200
Ta [K]
Figure 5.5: Case: Sandia cetane. Liquid length comparison between Versaevel and Siebers models for
various ambient temperatures and densities
are plotted as function of several ambient temperatures and densities. By creating this
figure special attention is paid to the definition of the spray angle, in the sense that some
authors use the half spray angle whereas others prefer the full spray angle. So the large
variations in the angles is solely due to the lack of predictive ability of the relations in cases
for which their are not fitted on. Also Jung et al [JA01] observed this discrepancy (they only
looked at Reitz & Bracco and Hiroyasu & Arai). They choose to work with Reitz & Braccos
relation because their compared experimental data was a bit closer to that model than that
of Hiroyasu & Arai and the measured angles did not show any dependency on the injection
pressure. Obviously, a good functioning angle relation is important, therefore Siebers formula
fitted to IFP measurements is continued to use in the rest of this study. But for future spray
studies better and generic angle predictions may be important, therefore it is recommended
to do an extensive (experimental) study to improve existing empirical relations or come with
other solutions. In the short term it may be more important to modify an existing formula
to fit experiments on the EHPC. Subsequently any proposed angle prediction model can be
put into the spray model in a straightforward fashion.
5.2.3 Sensitivity Analysis
The last 1D analysis before the phenomenological model is used to model 3D sprays in a
CFD environment is a sensitivity analysis. This is very relevant considering the high temper-
Comparison between angle relations

35
Siebers
Reitz & Bracco
Hiroyasu & Arai
30 a = 14.8 kg/m3
3
a = 30.2 kg/m
Full spray angle [degrees]
25
20
15
Sandia cetane measurement conditions

10
800 850 900 950 1000 1050 1100 1150 1200
Ta [K]
Figure 5.6: Three commonly used angle relations applied to the Sandia cetane case
ature and pressure levels at which spray measurements are done and material properties are
modeled. The question is to what extent the computed temperature in the spray, the liquid
length and the vapor length are influenced by varying measured quantities and thermody-
namic properties. The measured quantities Ta and a (pa actually) typically have an error of
10 %, and the ambient gas composition can vary more than that because it is not measured
directly. And also the spray angle is varied 10 % to investigate the influence of possibly
malfunctioning angle relations. The results in Table 5.3 are for the indicated percentages of
the reference values for the IFP dodecane case. The deviations from the reference tempera-
ture, liquid length (LL) and vapor length (SL) are positive or negative referring to larger or
smaller values, respectively. The same holds for the material properties, but these are varied
5 %. Sensitivity to material properties at elevated temperatures and pressures is important
because often there is simply not sufficient temperature dependent fuel information present
in literature, especially for practical (multi-component) fuels. Most remarkable values in the
table are found at the sensitivity of LL for variations in the ambient temperature. This is due
to the importance of the energy equation for the liquid length part of the spray where evapo-
ration of liquid fuel is a crucial process. Concerning the spray angle, or indirectly the amount
of air entrainment, one can see that the liquid length varies approximately with the same
percentage. The sensitivity to thermodynamic properties corresponds well to the findings of
van Erp [vE07], who did similar work with Naber and Siebers spray model.
Table 5.3: Sensitivity analysis
[%] T [%] LL [%] SL [%]

Measurement errors
Ta 90 9.54 +28.50 +1.45
110 +9.51 18.84 1.28
a 90 1.87 +4.83 +3.90
110 +1.80 4.35 3.42
YCO2 , YH2 O 50 1.09 +2.90 0.17
200 +2.01 5.31 +0.20
90 2.47 +11.11 +5.02
110 +2.34 9.18 4.35
Property errors
f l 95 +1.03 3.86 0.38
105 1.01 +3.38 +0.36
cp,f g 95 +0.73 1.93 +0.14
105 0.65 +1.45 0.14
psat 95 +0.21 +1.45 +0.01
105 0.21 1.45 0.01
Lv 95 +0.57 1.45 +0.08
105 0.58 +0.97 0.08
5.3 Phenomenological Model into CFD
The 1D phenomenological spray model discussed in the previous section is, in contrary to
the earlier model of Naber and Siebers, suitable to apply in combination with a 3D CFD
code. To accomplish such an interaction, source terms are extracted from the 1D model and
are assigned to the corresponding transport equations in Fluent. Subsequently the combined
model is validated through spray length comparison with experimental data.
5.3.1 Source Terms
Three source terms are needed to describe spray formation, namely mass, momentum and
energy sources. The coupling via source terms is realized only in the domain between the
nozzle exit and the liquid length (x < LL). Further downstream (x > LL) all liquid fuel is
evaporated, so there is only gaseous phase involved in the flow that already is covered with
the transport equations in Fluent. In the following, each source term is treated one after the
other.
Mass source The mass source is related to the evaporation of fuel in the x < LL part.
Because of evaporation, gaseous heptane mass is introduced in the computational domain.
5.3. PHENOMENOLOGICAL MODEL INTO CFD 43
Following Versaevel et al, the liquid fuel mass flow rate Mf l [kg/s] is:
Mf l (x) = f l [1 m(x)]A(x)v(x), (5.16)
differentiation to x and division by A(x) gives the mass source needed in [kg/m3 s]:
1 dMf l
Msource (x) = . (5.17)
A(x) dx
This equation is based on the change of liquid mass, which is the opposite of gaseous mass
change, hence the minus sign.
Momentum sources Momentum consists of three components, one in each spatial direc-
tion. Only one component is provided by the 1D model, and as long as the spray in Fluent is
oriented along a major axis (x, y or z) no extra measures have to be taken. The x-momentum
source from the 1D model is then set as the source for one of the other directions in 3D, as is
done in this study. But unfortunately when real in-cylinder sprays are simulated, the spray
axis generally does not match with the main axes. Some mathematics is included to make
the necessary rigid body rotation possible, see the next section for details. Now for one di-
rection the same approach as for the mass source gives the liquid fuel momentum flow rate
Tf l [kgm/s2 ]:
Tf l (x) = Mf l (x)v(x), (5.18)
again, differentiation and division gives the momentum source [N/m3 ] in x-direction:
1 dTf l
Tsource (x) = . (5.19)
A(x) dx
Energy source The derivation of a source term for the energy equation is not that straight-
forward as for the mass equation. In the first place there is a difference in the definition of
energy/enthalpy between the 1D model and Fluent. The 1D model works on an enthalpy
basis, but in Fluent energy is used and for liquid fuel it is defined as follows:
pa v(x)2
Ef l (x) = hf l (x) + . (5.20)
f l 2
The same definition also holds for the gaseous fuel:
pa v(x)2
Ef g (x) = hf g (x) + . (5.21)
f g (x) 2
With equations (5.16) and (5.20) the liquid fuel energy flow rate [J/s] is determined:
Ef l (x) = Mf l (x)Ef l (x) (5.22)
The second difficulty is due to the added gaseous fuel mass into the computational domain.
Whenever this happens, the mass enters the domain without thermal energy, so the energy
source must have an additional term to account for the temperature at which the liquid fuel
has evaporated. Then, the full energy source [J/m3 s] reads as follows:
1 dEf l
Esource (x) = + Msource (x)Ef g (x) (5.23)
A(x) dx
5.3.2 Interaction between Models
In this study the approach is limited to a constant pressure assumption, just like with the
derivation of the 1D model. Such an assumption is allowed as long as the (non-reacting)
spray is analyzed in a relatively large constant volume environment. From a numerical point
of view this is beneficial because the 1D model in Matlab is ran, source terms for the 3D
simulation are calculated and saved in a data-file as plain text. This procedure is followed
just one time for each different case setup. From then on, everything is done with Fluent.
Fluent is a commercial CFD package from which the source code is not accessible. Custom
settings that are not adjustable via the graphical or text user interface are only accessed
by means of user-defined functions (UDF) [flu06a]. UDFs are C language scripts that are
compiled in Fluent before the initialization of the case. Even with the use of UDFs the
possibilities are restricted to special formats, called macros, that are predefined by Fluent.
Hence, UDFs are the only way to provide sources of mass, momentum and energy to Fluent.
In the header part of the c-file the work directory, spray origin, spray direction and the
initial ambient conditions are defined. The rest of the c-file is universal, thus does not need
adaptations for each case. The interaction between files and models is schematically depicted
in Figure 5.7.
m-file 1D spray model data

file
Run Matlab ***
manual input
c-file
case setup*
UDF
model settings**
* d, Cd, T0, Ta, Pinj, Pa or a, Yi

** dx, N, initial guess of , m and T at nozzle exit
ac for spray angle relation 3D spray model
*** mass, momentum and enthalpy sources
spray geometry information
(spray angle and unsteady LL)
Fluent
Figure 5.7: Followed procedure from manual input to 3D model in Fluent
As is mentioned in the previous section about momentum sources, the spray axis is generally
not the same as one of the main axes. Therefore rotations are performed to virtually coincide
de 1D spray axis with the real spray axis in the 3D domain. The z-axis in Fluent is chosen to
be the default spray direction. Other directions are defined by elementary rotations around
the x-axis first and then around the new, after the first rotation obtained, y-axis, such that
the z-axis after two rotations points in the desired spray direction.
5.4 Results; 3D Spray Model
For simulations with the combined 1D spray model and Fluent, the same slice mesh config-
uration and boundary conditions as shown in Figure 4.7 in Section 4.1.3 is used, but now
with smaller cells. In radial direction the cell length is approximately 0.09 mm, because from
Abrahams work [Abr97] follows that at least 2 cells should be present in the nozzle hole
geometry in order to have mesh independent results near the nozzle. In axial direction the
cells also have lengths of 0.09 mm at the nozzle exit. To save simulation time (domain is
largest in axial direction) these cells gradually increase to at most 0.5 mm at the end of the
domain. But according to Versaevel et al further refinement to half of the axial and radial
cell size barely (approximately 1 %) influences the results.
Another numerical consideration is the solver timestep. The results shown in this section
are for a timestep of 105 s. Smaller steps like 106 and 5 107 s give only little (1-2 %)
improvement, except for the first part (around 0.1 ms) where the spray length is predicted up
to 10 % better. Nevertheless a timestep of 105 s is preferred at this stage because of, again,
computational expenses.
The results in Figure 5.8 represent the spray lengths for the IFP dodecane case. The IFP
measurement is indicated with the dotted black line and the 1D model result is indicated
with the red line. The circular markers are spray lengths from the combined 3D model, that
proves to be consistent with the experimental values. Leaving the timedependent massflow
at the start of injection out of consideration, the maximal deviation is 6.3 %.
case: IFP dodecane

50
45
40
35
30
SL [mm]
25
20
15
10
3D model
5
1D model
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time [ms]
Figure 5.8: Spray length as function of time with the 3D model, compared to 1D and measurement
A similar picture is observed for the IFP heptane case in Figure 5.9. But now, also Fluent
DPM results are shown, indicated with the stars. The DPM model is run with the square
mesh configuration (1 mm3 cells) and 106 s as the solver timestep. The liquid length is in
this case far off from the experimental value. The spray length however corresponds better to
the IFP measurements. Still, the DPM model gives up to 18.3 % deviation whereas the 3D
model is 8.6 % off. And, as mentioned earlier, the error of the 3D model prediction decreases
to about 6.7 % for smaller timesteps and decreases maybe even slightly more when a higher
resolution mesh is applied.
case: IFP heptane

50
45
40
35
30
SL [mm]
25
20
15
10
3D model
Fluent DPM
5
1D model
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time [ms]
Figure 5.9: Spray length as function of time with the 3D model, compared to DPM, 1D and measure-
ment
The correctness of the 3D model prediction is much better visualized with the contours of
fuel mass fraction at the spray cross-section shown in Figure 5.10. The upper spray is a DPM
simulation result and the other one is gained with the 3D model, both at 1 ms. Apart from the
obvious spray length difference, the shape/width of the sprays are also dissimilar. It is clearly
seen that the DPM spray is much wider than the more realistic 3D model spray, especially
near the nozzle, because there are 1 mm3 large cells present in the DPM case. The result is
that evaporated heptane is introduced in a larger cell, and therefore spreads the results out,
hence the lower absolute mass fraction values.
5.5. ACHIEVEMENTS AND FUTURE WORK 47
Figure 5.10: Case: IFP heptane. Contours of fuel mass fraction gained with Fluents DPM model and
the implemented 3D spray model. Note the minimum/maximum values between the brackets at the
right hand side, the colorbar is scaled for each separately
5.5 Achievements and Future Work
Summarizing the achievements in this chapter, the 1D mixing limited spray model proposed
by Versaevel et al is successfully implemented in Matlab. The results for spray and liquid
lengths correspond well with the measurements of IFP. Next, source terms are extracted from
the 1D results and are via UDFs used in 3D simulations with Fluent. This leads to similar
spray lengths as for the 1D model and the overall performance (spray length, spray shape)
is much better than the solution of the DPM model from Chapter 4. Furthermore, together
with the proper mesh resolution behavior (higher resolution gives better solutions), the ability
to parallelize is a major advantage compared with Fluents DPM model, which uses only one
CPU to do all discrete phase calculations no matter how many CPUs are available. The
consequence is large computation times despite the relatively small amount of cells.
Possible improvements to the 1D/3D spray model for the future may be the following, in
order of decreasing importance:
Make 3D spray model pressure dependent in order to simulate spray formation in a

variabele volume combustion chamber. The best way to do so, is probably by converting
the 1D model script (m-file) into C language. Then the 1D model, which generates
sources for the 3D calculation (matter of seconds), can update the sources online at
each timestep by, for example, looking at the average pressure in the volume at that
moment.
Sound spray angle prediction. Not really an improvement to the model itself, but
spray angle prediction certainly has a large effect on the final results as extensively
discussed in Section 5.2. Therefore the quest for proper and generic angle prediction
methods/relations should be promoted.
Include radial distribution of velocity, density and temperature profiles. For example the
use of a Gaussian radial distribution under the assumption of a fully-developed turbulent
flow as is done by Pastor et al [PLGP08]. This issue becomes particularly important
when other than conventional, circular nozzle hole, injectors would be modeled with the
developed 1D to 3D infrastructure.
Chapter 6
Modeling Spray Combustion with

Tabulated Chemistry
All non-premixed combustion phenomena go hand in hand with mixing processes. A chal-
lenging example from a numerical point of view is diesel spray formation and combustion.
In the preceding chapters a 3D Euler-Euler spray model is implemented and validated for
evaporating inert fuel sprays. This model is mesh and solver timestep independent, and is
suitable for parallel simulations. So, regarding the status of modeling the mixing process,
additional modeling features, which may require fine spatial and time resolutions, can be
included. In this chapter an attempt is made to add combustion, more specific, the emphasis
is on the application of FGMs (Flamelet Generated Manifolds) in modeling of the turbulent
combustion of a transient spray.
In the following, first the principle of flamelets and its use for modeling combustion with
tabulated chemistry is mentioned. Then, the procedure of 4D FGM generation is described.
Subsequently, its implementation into Fluent and from that arising necessary adaption to the
way of implementation of the spray model is shown. Finally, some results and emerged issues
that still have to be solved are discussed.
6.1 FGM Approach
Many combustion models exist, varying from heat release rate functions to detailed chemical
kinetics with finite-rate Arrhenius reaction sources. These examples are two extremes of sim-
ple and detailed combustion modeling, respectively. Simple models are known for their short
computing times but also for their lack of accuracy. Detailed models however, can be very
accurate, but unfortunately also computationally very expensive. To overcome the imprac-
tical computing times, while solving the combustion process still with high detail (depends
on used reaction scheme), an approach with tabulated chemistry is applied. This so-called
FGM approach is developed by van Oijen [vO02] for laminar premixed flames, and makes use
of 1D laminar flamelet data to tabulate composition, density, temperature etc. as function
of local control variables. Later, Ramaekers [Ram05] extended the application to turbu-
lent partially-premixed combustion by choosing one control variable describing non-premixed
49
50CHAPTER 6. MODELING SPRAY COMBUSTION WITH TABULATED CHEMISTRY
(mixture fraction Z) and one describing premixed (reaction progress variable P V ) combus-
tion, and by PDF (Probability Density Function) integration to account for turbulence.
6.1.1 The Flamelet Concept
The flamelet concept views the turbulent flame as an ensemble of thin, laminar, locally 1D
flames, called flamelets, embedded within the turbulent flow field [flu06b]. Furthermore,
the concept is based on the assumption that the smallest turbulent time and length scales
are much larger than the chemical ones, and there exists a locally undisturbed sheet where
chemical reactions occur [SRM06].
A counterflow diffusion flame is a common laminar flame that is used to represent a flamelet in
a turbulent flow. The counterflow geometry consists of two opposed flows, with in this study
pure fuel at one side and oxidizer (mixture of ambient air) at the other side. Increasing flow
velocities and/or decreasing the distance between the flows results in straining of the flame and
increasingly departs from chemical equilibrium until the flame extinguishes at some point. In
non-premixed combustion it is common practice to introduce the mixture fraction Z defined
as:
sYF u YO2 + YO2 ,2
Z= , (6.1)
sYF u,1 + YO2 ,2
where s is the stoichiometric mass fraction and the subscripts 1 and 2 refer to the constant
mass fraction in the original fuel and oxidizer streams, respectively. In the fuel stream the
mixture fraction is equal to unity and monotonically decreases to zero at the oxidizer stream.
In the 1D laminar flamelet concept this property of the mixture fraction is used to study the
concerned scalars as function of Z instead of the spatial coordinate x. See reference [dGSB+ 07]
for a descriptive derivation or van Oijen [vO02] for a complete derivation. An additional
control variable, called the reaction progress variable P V , is introduced to parameterize the
progress of the irreversible combustion process. In this study a combination of CO2 , CO and
CH2 O mass fractions is chosen as a reaction progress variable:
YCO2 YCO YCH2 O
PV = + + . (6.2)
MCO2 MCO MCH2 O
This decision is based on successful auto-ignition modeling efforts at the University of Karl-
sruhe [SHS+ 08], and on common hydrocarbon chemistry knowledge that formaldehyde (CH2 O)
is an intermediate that has a driving force on reactions in the early stages of combustion at
relative low temperatures. Later in the combustion process CO becomes more important,
ultimately (ideally) all carbon atoms end in CO2 molecules.
The succes of this concept is related to the fact that all occurring compositions tend to
have a common, low-dimensional, attractor in composition space, a so-called intrinsic low-
dimensional manifold (ILDM) [MP92]. Hence, the complex chemistry is reduced and com-
pletely described by the mixture fraction Z and the reaction progress variable P V .
6.1.2 FGM Generation
The low-dimensional manifold is now tabulated as function of the two spanning variables Z
and P V . This reduction allows the flamelet calculation to be preprocessed and all species
6.1. FGM APPROACH 51
mass fractions Yi , temperature T , density , source of the progress variable sP V etc. to be

tabulated as function of the two control variables in a 2D look-up table. By preprocessing
the chemistry, computational costs are reduced considerably [flu06b]. The flamelet equations
with counterflow boundary conditions are solved with CHEM1D [che], which is a specialized
one-dimensional laminar flame code developed at the Eindhoven University of Technology.
The turbulence-chemistry interaction is accounted for by integrating the quantities in the
2D table with a -PDF function as follows:
Z 1Z 1
= (Z, P V ) P (Z)P (P V ) dZdP V. (6.3)
0 0
Note that this explicit formulation assumes that Z and P V are statistically independent.
The overtilde stands for Favre (mass) averaged. Both control variables are from then on
described with a mean value (Z, PgV ) and a variance (Z2 , P V 2 ), so a quantity is defined
by the probability of occurrence for several states instead of one fixed state. Details on the
application of the -PDF can be found in [Ram05]. The chemistry is in this way extended to a
4D look-up table with the means and variances of the two control variables as the parameters
(look-up indices).
The way(s) a FGM is constructed in this study is depicted schematically in Figure 6.1. First a
heptane flamelet database at constant pressure is calculated with CHEM1D, making use of a
reduced TRF (Toluene Reference Fuels) mechanism from which the original C1-C3 chemistry
is replaced by that of the GRI 3.0 mechanism [SGF+ ] to reduce the total number of species,
ultimately leading to 248 elementary reactions between 48 species. For more information
about reduced TRF mechanisms the reader is referred to for instance [ABCK07].
FGMs can be generated in many ways. For stationary flames, there is a classical way with
steady flamelets only, where a loop over strainrates is performed until the flame extinguishes.
An illustrative example is shown in Figure 6.2, see the gray area between the solution for
the lowest strainrate and the solution at wich the strainrate reached its maximum before
extinction. For the stationary TRF mechanism solutions the quenching part of the well
known S-curve is plotted in Figure 6.3; maximum flame temperature as function of one over
the strainrate. Temperatures up to 2400 K are observed for small strainrates and at a value of
approximately 14600 s 1 the maximum temperature drops beneath 1700 K prior to extinction.
But now the situation is unsteady and initially non-reacting, so to cover the ignition pro-
cess the table should also contain information in the area beneath the quenching strainrate
solution. Several ways exist to fill this gap in the Z-P V plane. One way is to solve a timede-
pendent flamelet with a higher strainrate than the highest possible non-quenching strainrate,
in this way forcing the flame to extinguish and in the mean time sampling data to fill the gap.
Another approach, that is more appropriate for this study, is solving timedependent flamelets
from a mixed, but non-reacting initial state in a counterflow setup. The ignition behavior is
followed in time until a steady flame is reached. This process for heptane ignition is shown
in Figure 6.4 for P V as function of Z at several moments in time. It is seen that the steady
solution (the one with the widest base) has a lower peak than few solutions earlier in time.
So, the mixture ignites, reaches high values for the progress variable in Z-space and before a
steady solution is reached, the progress variable slightly decreases. The third possibility is to
reproduce ignition of mixtures covering the entire Z-space with PSR (Perfectly Stirred Re-
actor) auto-ignition calculations as Michel et al [MCV07] applied to laminar diffusion flames

recently. All three methods to fill the Z-P V gap are depicted schematically in Figure 6.2.
In this study however, only the quenching and igniting flamelet approaches, as is shown in
Figure 6.1, are used.
TRF mechanism
48 species
248 reactions
CHEM1D
solves:
flamelet equations
(constant pressure )
quenching flame igniting flame
Z-PV domain filled with Z-PV domain filled with , from

stationary loop over strainrates , initial pure mixing solution ,
and with timedependent igniting flame , using the
quenching flame timedependent solver
2D FGM
Z, PV table
(laminar)
interpolated laminar flamelet data ,

sPV, Yi and T as function of
Z and PV
4D FGM
"2 ,PV table
Z% , Z , PV "2
(turbulence included )
2D data integrated with PDF

m,sPV
functions. , sPV
sPV , sPV"2v, Yi and T
%
as function of mmmmmmmm ,PV "2
Z, Z "2 , PV
Figure 6.1: FGM construction scheme

Flamelet database generation

PV []
extinguishing flamelet
igniting flamelet
igniting PSRs
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Z []
Lowest strainrate
Steady solutions region
Highest nonquenching strainrate
Timedependently extinguishing or igniting flamelet
Perfectly Stirred Reactors (PSRs) before ignition
Figure 6.2: Ways to generate a full flamelet database

Maximum temperature as function of strainrate

2500
2400
2300
2200
Temperature [K]
2100
2000
1900
1800
1700
Tmax
1600
5 4 3 2 1 0 1
10 10 10 10 10 10 10
1
Strainrate , 1/a [s]
Figure 6.3: S-curve, quenching part
Progress variable as function of mixture fraction
6 Z = (s*YFu YO + YO ,2) / (s*YFu,1 + YO ,2)

2 2 2
PV = YCO /MCO + YCO/MCO + YCH O/MCH O

2 2 2 2
4
PV []
ignition
3
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Z []
Figure 6.4: Ignition from an initially mixing-only solution in a counterflow setup

6.1.3 Implementation into CFD
The implementation of the 4D FGM in Fluent, in order to do turbulent spray combustion sim-
ulations, is once again done with UDFs. Remember from Section 5.3.2 that UDFs are the only
way to do serious user-defined adjustment in Fluent. This time there are much more UDFs
involved than for the implementation of the 1D spray model, because the FGM approach
g
requires four variables to be solved, namely: Z, P V , Z2 and P V 2 . These four scalars are
solved with user-defined scalar transport equations, in addition to the standard continuity,
momentum and turbulence equations. Compared to inert spray formation calculations, there
are no species transport equations and energy equation solved any more. All species concen-
trations and corresponding temperatures are in principle known from the flamelet database
for any mixture fraction and progress variable combination. But only important species, like
those defining the reaction progress variable, are tabulated in the 2D and 4D FGM together
with essential quantities like mean densities and temperatures, to save FGM generation time
and memory.
The transport equations for Z, P g V , Z2 and P V 2 are derived by Ramaekers [Ram05], but
these are for stationary flames without spray formation in contrary to the diesel injection case.
Therefore some modifications are needed. In the first place all four equations are extended
with an unsteady term. And secondly a source term for the mean mixture fraction is added
since the mixture fraction is not a conserved scalar anymore. Here, a short derivation is given.
The transport equations for fuel and oxidizer are:
~ (~v YF u )
~ (DY
~ F u ) = F u + SF u ,
(YF u ) + (6.4)
t
~ (~v YO )
~ (DY
~ O ) = O + 0 = sF u ,
(YO2 ) + 2 2 2 (6.5)
t
respectively. s is the stoichiometric mass fraction, is a source due to reactions and SF u is
an injection mass source. Substraction of equation (6.5) from equation (6.4) multiplied with
s, gives:
~ (~v [sYF u YO ])
~ (D[sY
~
([sYF u YO2 ]) + 2 F u YO2 ]) = sSF u . (6.6)
t
Together with the definition of the mixture fraction:
sYF u YO2 + YO2 ,2 sYF u YO2
Z= = + Zst , (6.7)
sYF u,1 + YO2 ,2 sYF u,1 + YO2 ,2
with Zst the constant mixture fraction at the location where stoichiometric conditions exist,
the transport equation for Z becomes:
~ (~v Z)
~ (DZ)
~ s
(Z) + = SF u . (6.8)
t sYF u,1 + YO2 ,2
From this equation with a source term on the right hand side, transport equations for Z and
Z2 can be derived like Ramaekers did. However, to evaluate the terms SF u and ZSF u that
appear in those equations, PDF integration is required. Remember from equation (6.3) that
the PDFs are functions of either Z or P V , but the mass source from the 1D spray model is not.
The mass source is preprocessed and tabulated as function of position and time only, so there
is no coupling with, for example, variations in ambient temperature and density. Therefore,
the term on the right hand side is considered to be the source for the mean mixture fraction,
and the variance of the mixture fraction is assumed to be unaffected by the fuel injection.
See Appendix A for an overview of the solved transport equations.
Furthermore, the same momentum sources are set as before. The 4D FGM and the spray
model sources are read into Fluent prior to initialization of the problem. From then on the
simulation is ready to start. During computations, at each timestep, the four user-defined
scalars are solved and used to look-up the corresponding density , source of the mean progress
variable (sPgV ) and the source of its variance (sP V 2 ). These are passed to the Fluent solver
in order to update the corresponding equations for the new timestep. In Figure 6.5 a scheme
of this CFD-FGM interaction is shown. In the mean time all preferred quantities like the
temperature and mass fractions of certain species are saved to allocated memory in Fluent,
so they can be postprocessed to view contour plots for instance. However, these quantities
are not used for any purpose during the simulation itself.
CFD % , Z"2v , PV
Zm m, PV"2v
4D FGM
Z, Z , PV ,PV
solves : look up tabulated quantities at :
u, p, Z%m, Z
"2
, Zv,,PV
PV,PV
"2
m, PVv m,PV
% ,, ZZ"2v,, PV
ZZm PV , PV"2 v
,sPV
sPV,
, sPV sPV v
"2
Figure 6.5: CFD-FGM interaction
6.2 Results; 3D Spray Model with FGM Combustion Model-

ing
A reacting spray, modeled with the FGM approach as discussed in the preceding section, is
applied to simulate the IFP heptane experiment. The details of this experiment are shown
in Table 5.1, however, the initial ambient gas composition is different now. There is also 0.21
(mass fraction) O2 present, while the fractions of CO2 and H2 O are the same. In the paper of
Verhoeven et al [VVB98] heptane ignition and combustion results are not presented, mainly
the dodecane case is treated elaborately. But for now, the lack of detailed reaction mechanisms
for relative heavy hydrocarbons like dodecane prevents the combustion simulation of such
fuels with FGMs. Although the simulation results with heptane, for which detailed reaction
mechanisms are available, are not compared directly with measurements (often ignition delay
and flame lift-off length) at the moment, they are still very important to monitor the FGM
spray combustion modeling status in a quantitative fashion. This monitoring is done first by
checking the spray length without combustion and then by looking at the ignition behavior
and the combustion temperatures.
6.2. RESULTS; 3D SPRAY MODEL WITH FGM COMBUSTION MODELING 57
6.2.1 Spray Length without Combustion
Now the FGM approach is implemented, and with that also the way the spray model originally
was implemented, is changed. An important first result to check is therefore the, by now
familiar, spray length as function of time for a not yet ignited fuel spray. Figure 6.6 is similar
to Figure 5.9, actually it should be the same if the implementation is passed successfully. The
black circular marks indicate the spray length gained from the 3D spray model with FGM.
It is seen that now, the length corresponds even better with the original 1D model and the
maximum deviation from the experimental values is just 4.8 %, whereas with the original 3D
model this was 6.7 %. So one can say that concerning the spray formation the model is still
functioning satisfactory.
case: IFP heptane

50
45
40
35
30
SL [mm]
25
20
15
10
3D model with FGM (no combustion)
Fluent DPM
5
1D model
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time [ms]
Figure 6.6: Spray length as function of time with the 3D model with FGM, compared to DPM, 1D and
measurement
6.2.2 Ignition Behavior
Due to the unsteady nature of a diesel injection event, ignition modeling is at least as im-
portant as combustion modeling. Following the FGM approach, besides combustion, ignition
should be covered inherently. But not surprisingly this depends on the way the FGM is gen-
erated. The extinguishing flamelet approach, as presented in Section 6.1.2, is applied and
does not lead to ignition of the spray. Instead, only local temperatures slightly above the
initial ambient temperature are found, and the source of the reaction progress variable is not
big enough to end in total ignition within a few milliseconds. A FGM constructed with an
igniting flamelet database is also applied, and this time the result is auto-ignition of the whole
spray in very short time.
The evolution from the early stage of ignition to further combustion of the spray is shown
at four moments in time in Figure 6.7. The upper half of the plots represent the values of
the progress variable and the lower parts are contours of temperature. Several interesting
observations are done from this figure. First, at 0.025 ms, the outer edge of the spray shows
activity by means of an increased (and still increasing) progress variable and a modest increase
in temperature. This activity is particularly present close to the spray axis and at the tip of
the spray. Further in time, at 0.05 ms, the outer contour of the igniting spray is clearly seen
due to high values of P V and T , but reactions are still located at the diffusive edge of the
spray. In the next plot, at 0.1 ms, the combustion region is expanded to the inner volume
and the maximum temperature continues to rise. From 0.2 ms on the general picture does
not change much.
Figure 6.7: Case: IFP heptane. Temporal sequence of progress variable and temperature contours
showing the early ignition process resulting in total combustion
The diffusion flame at the spray edge stays the hottest region, as can be expected on the
basis of the used non-premixed flamelet database. But in Section 2.3 on the conceptual
diesel combustion model it is stated that in DI diesel injection, regions with premixed and
non-premixed combustion can be distinguished. The next step may be the creation of a
partially-premixed database in order to capture the presence of both regimes at the same
time. Again, there are no measurements for this case to compare, but until now there are no
6.2. RESULTS; 3D SPRAY MODEL WITH FGM COMBUSTION MODELING 59
strange things happening.
6.2.3 Combustion Temperature
Despite the good looking ignition and flame geometry, temperatures do not reach realistic
values above 2000 K. Therefore also a much simpler combustion model is used that is available
in Fluent, called eddy-dissipation model [flu06b]. This model is, like the flamelet approach,
a mixing-limited combustion model, with the differences that only the global reaction from
fuel and O2 to CO2 and H2 O is considered and immediate reaction is assumed. So, it is
not expected that this simple model can be as accurate as the much more detailed FGM
method, but it gives nice material to compare regarding for example temperature profiles.
Such a comparative picture is given in Figure 6.8, taken at 1 ms. In contrast to the eddy-
dissipation model, a flame lift-off settles automatically using tabulated chemistry. It is seen
that the eddy-dissipation (upper spray) solution gives temperatures up to 2400 K whereas
the FGM solution stay around 1300-1400 K. This outcome is remarkable when remembering
the laminar quenching temperature of about 1650 K shown in Figure 6.3 and knowing that
the FGM flame does not extinguish at all but burns at those low temperatures.
Figure 6.8: Case: IFP heptane. Contours of temperature gained with the eddy-dissipation model and
the implemented FGM model. Note the minimum/maximum values between the brackets at the right
hand side, the colorbar is scaled for each separately
It is thought that this occurrence is caused by the PDF integration procedure. In order to
confirm this view a similar heptane spray at 1 atmosphere (atm) instead of the original high
pressure of 60 atm is simulated, because the reaction layer thickness will increase at lower
pressures and therefore high peak temperatures will spread out over a larger region resulting
in lower peak temperatures. In this way the averaging PDF integration should have less
influence. From new CHEM1D calculations indeed follows that the reaction layer thickness
increases from tens of micrometers to the order of millimeters. Since flamelets at 373 K fuel
and 800 K oxidizer temperature does not auto-ignite, the quenching method is performed.
With the obtained quenching flamelet database a FGM is generated and used in manually
ignited spray simulations. From simulations with the original case it is found that manual
ignition does not influence the steady combustion behavior and corresponding temperatures
after few milliseconds. The temperatures at 1 atm ambient pressure stay at values of 1400 K,
where the maximum possible temperature according to the FGM is 1810 K. At the original 60
atm case the maximum in the table is 2230 K. The temperatures from the 1 atm simulation
are closer to the tabulated maximum temperatures, so relatively the FGM combustion model
performs better now. This is also supported by the fact that only 0.052 % of the Z, P gV,
2 2
Z and P V combinations in the FGM give temperatures above 2000 K. Concluding, the
simulations at 1 atmosphere give an affirmative answer to the notion that PDF integration is
a source for the low temperatures.
6.3 Improvement Points
So far, the attempted application of a FGM to model the combustion of a transient and
turbulent fuel spray is partially succeeded. Spray formation is still predicted accurately and
ignition follows automatically from the igniting flamelet database. However, the spray auto-
ignites too early and (therefore) also too close to the nozzle. Combustion temperatures are
too low, (partly) due to PDF integration of the laminar 2D table. Yet, the implementation
(spray model-CFD interaction, FGM data handling) on its own seems to work well.
In the future, first the PDF integration process should be avoided by simulating 3D sprays
with a laminar flamelet database. When this approach gives promising results, one can focus
to the averaging procedure in more detail. If not, then it should be considered to go back to
a 1D laminar case to get ignition and combustion right, without the difficulties introduced
by working in a not entirely transparant 3D CFD environment with complex data handling.
Ultimately, when serious improvements are achieved, direct validation with measurements
like ignition delay time and flame lift-off length can be performed.
Chapter 7
Conclusions and Recommendations
7.1 Conclusions
In this study first Fluents Euler-Lagrange DPM model is used to model evaporating, inert
heptane sprays, since heptane is considered as a surrogate for diesel from which all material
properties are well known. Special attention is paid to model the fuel properties as func-
tion of temperature, making use of the thermophysical database of DIPPR. The results are
compared with spray length and liquid length measurements on the EHPC and from Sandia,
because these are common spray characterization quantities. Numerical parameters like mesh
configurations, solver timestep and amount of parcels are varied to investigate the influence
on spray length and liquid length. From the comparisons it is found that the DPM model
gives erroneous results, but more significant are the numerical inadequacies. One of them
is the imposed limitation to mesh refinement which is far from desirable when detailed in-
cylinder mixture formation and combustion are to be modeled. The second disadvantage is
that the discrete phase part of the calculations cannot be parallelized, while those detailed
investigations require fine resolution, thus expensive simulations. Nevertheless, some best
practice setups resulted from this study, which are at least valid for heptane sprays in engine
like conditions. That are 1 mm3 cells aligned with the spray axis and solved with a solver
timestep of 106 s, and injection of 1000 parcels per timestep.
In the second stage, recognizing the limitations of the DPM model, a 1D mixing limited spray
model proposed by Versaevel et al is successfully implemented in Matlab. The results for spray
and liquid lengths correspond well with the measurements of IFP. Source terms are extracted
from this 1D results in order to use in 3D Euler-Euler simulations with Fluent. This leads to
similar spray lengths as for the 1D model and the overall performance (spray length, spray
shape) is much better than the solution of Fluents Euler-Lagrange approach. Furthermore,
together with the proper mesh resolution behavior (higher resolution gives better solutions),
the ability to parallelize is a major advantage compared with the DPM model, which uses
only one CPU to do all discrete phase calculations no matter how many CPUs are available.
The consequence is large computation times despite the relatively small amount of cells.
Finally, having achieved sound mixture formation modeling, ignition and combustion of the
fuel spray is considered. The flamelet concept is applied for this purpose. A laminar flamelet
61
62 CHAPTER 7. CONCLUSIONS AND RECOMMENDATIONS
database for heptane is created and PDF integrated, resulting in a 4D FGM. This so-called
tabulated chemistry approach is known for its time-efficiency and at the same time high
accuracy. With this method the implementation of spray sources and solved equations are
modified, but spray formation is still predicted accurately. So far, this first attempt to apply
FGMs to model the combustion of a transient and turbulent fuel spray is partially succeeded.
Ignition follows automatically from the igniting flamelet database and the whole spray contin-
ues to burn. However, the spray auto-ignites too early and combustion temperatures are too
low, (partly) due to PDF integration of the laminar 2D table. Yet, the created infrastructure
(spray model-CFD interaction, FGM data handling) on its own seems to work well.
7.2 Recommendations
Directions for future research, taking this study as a starting point, are recapitulated, both
for spray formation as for spray combustion modeling, respectively.
Possible improvements to the 1D/3D spray model may be the following, in order of decreasing
importance:
Make 3D spray model pressure dependent in order to simulate spray formation in a

variabele volume combustion chamber. The best way to do so, is probably by converting
the 1D model script (m-file) into C language. Then the 1D model, which generates
sources for the 3D calculation (matter of seconds), can update the sources online at
each timestep by, for example, looking at the average pressure in the volume at that
moment.
Sound spray angle prediction. Not really an improvement to the model itself, but
spray angle prediction certainly has a large effect on the final results as extensively
discussed in Section 5.2. Therefore the quest for proper and generic angle prediction
methods/relations should be promoted.
Include radial distribution of velocity, density and temperature profiles. For example the
use of a Gaussian radial distribution under the assumption of a fully-developed turbulent
flow as is done by Pastor et al [PLGP08]. This issue becomes particularly important
when other than conventional, circular nozzle hole, injectors would be modeled with the
developed 1D to 3D infrastructure.
Considering spray combustion, first the PDF integration process should be avoided by sim-
ulating 3D sprays with a laminar flamelet database. When this approach gives promising
results, one can focus to the averaging procedure in more detail. If not, then it should be
considered to go back to a 1D laminar case to get ignition and combustion right, without the
difficulties introduced by working in a not entirely transparant 3D CFD environment with
complex data handling.
Ultimately, when serious improvements are achieved, direct validation with measurements
like ignition delay time and flame lift-off length can be performed.
Nomenclature
Greek symbols
ratio of mass flow rates [-]
turbulent kinetic energy dissipation rate []
full spray angle [degrees]
wavelength [m]
dynamic viscosity [kg/ms]
density [kg/m3 ]
mass source [kg/m3 s]
surface tension [kg/s2 ]
frequency [Hz]
Sub-/superscripts
0 initial and/or at nozzle exit
a ambient gas
ef f effective
f fuel
fg gaseous fuel
fl liquid fuel
g whole gaseous phase
l liquid
t turbulent
e Favre (mass) average
63
Latin symbols
A cross-section area [m2 ]
Cd discharge coefficient [-]
cp specific heat [J/kgK]
d, D diameter [m]
D diffusion coefficient [m2 /s]
E energy [J/kg]
E energy flow rate [J/s]
F force [N]
~g gravitational acceleration [m/s2 ]
h enthalpy [J/kg]
J~ diffusion flux [kg/ms]
k turbulent kinetic energy []
K cavitation number [-]
L length [m]
LL liquid length [m]
Lv latent heat of vaporization [J/kg]
m void fraction [-]
m, M mass flow rate [kg/s]
M molecular mass [kg/mol]
N amount of droplets [-]
Oh Ohnesorge number [-]
p pressure [Pa]
P Probability Density Function [-]
PV reaction progress variable [-]
r radius [m]
Re Reynolds number [-]
s stoichiometric mass fraction [-]
Sc Schmidt number [-]
SL spray/vapor length [m]
sP V source of progress variable [kg/m3 s]
t time [s]
T temperature [K]
T momentum flow rate [kgm/s2 ]
u, v velocity [m/s]
~v velocity vector [m/s]
x coordinate on spray axis [-]
X mole fraction [-]
Y mass fraction [-]
We Weber number [-]
Z mixture fraction [-]
64
Appendix A
Transport Equations for the Mean

and Variance of Z and PV
The steady terms of the following equations are adopted from Ramaekers [Ram05], except
for the mean mixture fraction equation. The mean mixture fraction equation is derived in
Section 6.1.3.
Mean mixture fraction Z:
~ (~vZ)
~ ([D + DT ]
~ Z) = s
(Z) + SF u . (A.1)
t sYF u,1 + YO2 ,2
Variance of mixture fraction Z2 :

!
2
~ (~vZ2 )
(Z2 ) + ~ ([D + DT ]Z ~ 2 ]2 Z
~ 2 ) = 2 DT [Z . (A.2)
t k
g
Mean reaction progress variable P V:
g ~ (~vP
g ~ ([D + DT ]
~Pg g
(P V ) + V) V ) = P V = sP V. (A.3)
t
Variance of reaction progress variable P V 2 :

!
~ ~vP V 2 )([D+D
~ ~ ~ 2 2
P V 2
(P V 2 )+( 2
T ]P V ) = 2 DT [P V ] +2P V P V ,
t k
(A.4)
where the last term represents the tabulated source of the progress variable variance
sP V 2 = P V P V .
65
66APPENDIX A. TRANSPORT EQUATIONS FOR THE MEAN AND VARIANCE OF Z AND PV
Acknowledgements
Many people contributed to the realization of this Masters thesis. I would like to thank
everyone who helped me in one way or another, and a few people in particular.
First of all, my supervisor Bart Somers for his coaching and for the freedom he gave me in
doing my research. Especially his enthusiasm during discussions about combustion modeling
were motivating. Concerning spray formation modeling, the enthusiasm and help of Carlo
Luijten was motivating in the same way. And Philip de Goey for his sincere interest in the
project and encouragement. The informal discussions at the Spray Meetings I had with
Philip, Bart and Carlo resulted in better understanding from both sides. Thank you all for
that.
Thanks to Giel Ramaekers for his support by implementing the FGM approach into Flu-
ent. For several weeks/months I worked together with Justin Gerritzen (spray modeling in
Fluent) and Jos Reijnders (phenomenological spray model), thank you both for the smooth
cooperation.
Last but not least, I would like to thank the other master students in the hallway for the
pleasant times and excellent atmosphere. Once again special thanks to Jos Reijnders and
Dennis Verwaaijen, who were my roommates for the whole and half year respectively, for the
daily help and above all the many amusing moments.
67
68
Bibliography
[ABCK07] J.C.G. Andrae, P. Bjornbom, R.F. Cracknell, and G.T. Kalghatgi. Autoigni-
tion of toluene reference fuels at high pressures modeled with detailed chemical
kinetics. Combustion and Flame, 149(2007):224, February 2007.
[Abr97] John Abraham. What is adequate resolution in the numerical computations of

transient jets? SAE paper, (SAE 970051), February 1997.
[Bau06] C. Baumgarten. Mixture formation in internal combustion engines. Springer,

Heidelberg, 2006.
[che] CHEM1D, A one-dimensional laminar flame code, Eindhoven University of Tech-

nology, http://www.combustion.tue.nl/chem1d.
[Dec97] John E. Dec. A conceptual model of di diesel combustion based on laser-sheet

imaging. SAE paper, (SAE 970873), February 1997.
[dGSB+ 07] L.P.H. de Goey, L.M.T. Somers, H. Bongers, M.F.G. Cremers, M.J. Remie, and
S. Delhaye. Chemie en Transport in Energie Conversie Processen. Eindhoven
University of Technology, 2007. Lecture notes.
[DIfPP] DIPPR Design Institute for Physical Properties. http://dippr.byu.edu/.
[ECN] ECN Engine Combustion Network. http://public.ca.sandia.gov/ecn/index.php.
[FCD+ 07] J.T. Farrell, N.P. Cernansky, F.L. Dryer, D.G. Friend, C.A. Hergart, C.K. Law,
R.M. McDavid, C.J. Mueller, A.K. Patel, and H. Pitsch. Development of an
experimental database and kinetic models for surrogate diesel fuels. SAE paper,
(SAE 2007-01-0201), 2007.
[flu06a] Fluent 6.3 UDF Manuel, September 2006.
[flu06b] Fluent 6.3 Users Guide, September 2006.
[Hui07] Vincent Huijnen. On the Application of Large-Eddy Simulations in Engine-related

Problems. PhD thesis, Eindhoven University of Technology, Combustion Tech-
nology, 2007.
[JA01] Dohoy Jung and Dennis N. Assanis. Multi-zone di diesel spray combustion model
for cycle simulation studies of engine performance and emissions. SAE paper,
(SAE 2001-01-1246), March 2001.
69
70 BIBLIOGRAPHY
[KDFS+ 06] R.J.H. Klein-Douwel, P.J.M. Frijters, L.M.T. Somers, W.A. de Boer, and R.S.G.
Baert. Macroscopic diesel fuel spray shadowgraphy using high speed digital imag-
ing in a high pressure cell. Fuel, 86(2007):19942007, November 2006.
[KSR99] S.C. Kong, P.K. Senecal, and R.D. Reitz. Developments in spray modeling in
diesel and direct-injection gasoline engines. Oil & Gas Science and Technology,
IFP, 54(2):197204, 1999.
[Lui08] C.C.M. Luijten. Fuels and lubes (4n850). Lecture, Master Mechanical Engineer-
ing, Eindhoven University of Technology, February 2008.
[MCV07] Jean-Baptiste Michel, Olivier Colin, and Denis Veynante. Modeling ignition
and chemical structure of partially premixed turbulent flames using tabulated
chemistry. Combustion and Flame, 152(2008):8099, September 2007.
[MP92] U. Maas and S.B. Pope. Simplifying chemical kinetics: Intrinsic low-dimensional
manifolds in composition space. Combustion and Flame, 88(1992):239264, 1992.
[NS96] Jeffrey D. Naber and Dennis L. Siebers. Effects of gas density and vaporization on
penetration and dispersion of diesel sprays. SAE paper, (SAE 960034), February
1996.
[Pet07] Roel Peters. Penetration and dispersion research of non-reacting evaporating

diesel sprays. Masters thesis, Eindhoven University of Technology, March 2007.
[PLGP08] Jose V. Pastor, J. Javier Lopez, Jose M. Garcia, and Jose M. Pastor. A 1d model
for the description of mixing-controlled inert diesel sprays. Fuel, 87(2008):2871
2885, 2008.
[Ram05] W.J.S. Ramaekers. The application of flamelet generated manifolds in modelling

of turbulent partially-premixed flames. Masters thesis, Eindhoven University of
Technology, Combustion Technology, 2005.
[Rei87] R.D. Reitz. Mechanisms of atomization processes in high-pressure vaporizing

sprays. Atomization and Spray Technology, 3:309337, 1987.
[RR95] R.D. Reitz and C.J. Rutland. Development and testing of diesel engine cfd
models. Prog. Energy Combust. Sci., 21:173196, 1995.
[SBK+ 04] R. Steiner, C. Bauer, C. Kruger, F. Otto, and U. Maas. 3d-simulation of di-
diesel combustion applying a progress variable approach accounting for complex
chemistry. SAE paper, (SAE 2004-01-0106), March 2004.
[Sch03] B. Schneider. Experimentelle Untersuchungen zur Spraystruktur in transienten,

verdampfenden und nicht verdampfenden Brennstoffstrahlen unter Hochdruck.
PhD thesis, Technische Wissenschaften, Eidgenossische Technische Hochschule
Zurich, 2003.
[SGF+ ] Gregory P. Smith, David M. Golden, Michael Frenklach, Nigel W. Moriarty, Boris
Eiteneer, Mikhail Goldenberg, C. Thomas Bowman, Ronald K. Hanson, Soonho
Song, William C. Gardiner Jr., Vitali V. Lissianski, and Zhiwei Qin. GRImech
3.0 reaction mechanism, Berkeley, http://www.me.berkeley.edu/gri mech/.
View publication stats
BIBLIOGRAPHY 71
[SHS+ 08] W. Sauter, S. Hensel, U. Spicher, A. Schubert, R. Schiessl, and U. Maas. Effects
of mixture quality on controlled auto-ignition. ATZ autotechnology, 8, March
2008.
[Sie99] Dennis L. Siebers. Scaling liquid-phase fuel penetration in diesel sprays based on
mixing-limited vaporization. SAE paper, (SAE 1999-01-0528), March 1999.
[SRM06] Satbir Singh, Rolf D. Reitz, and Mark P.B. Musculus. Comparison of the charac-
teristic time (ctc), representative interactive flamelet (rif), and direct integration
with detailed chemistry combustion models against optical diagnostic data for
multi-mode di diesel engine. SAE paper, (SAE 2006-01-0055), April 2006.
[SSB05] X.L.J. Seykens, L.M.T. Somers, and R.S.G. Baert. Detailed modeling of common
rail fuel injection process. MECCA, III(2+3):3039, 2005.
[Sti03] G. Stiesch. Modeling engine spray and combustion processes. Springer, Berlin,
2003.
[vE07] Dennis van Erp. Acquiring accurate thermodynamic properties for in-house
diesel spray models. Internship assignment, Eindhoven University of Technol-
ogy, November 2007.
[VMW00] Philippe Versaevel, Paul Motte, and Karl Wieser. A new 3d model for vaporizing
diesel sprays based on mixing-limited vaporization. SAE paper, (SAE 2000-01-
0949), March 2000.
[vO02] J.A. van Oijen. Flamelet-Generated Manifolds: Development and Application

to Premixed Laminar Flames. PhD thesis, Eindhoven University of Technology,
Combustion Technology, 2002.
[VVB98] Dean Verhoeven, Jean-Luc Vanhemelryck, and Thierry Baritaud. Macroscopic

and ignition characteristics of high-pressure sprays of single-component fuels.
SAE paper, (SAE 981069), February 1998.
[Wie90] A. Wierzba. Deformation and breakup of liquid drops in a gas stream at nearly
critical weber numbers. Experiments in Fluids, 9(1990):5964, 1990.
[WP97] Y.P. Wan and N. Peters. Application of the cross-sectional average method
to calculations of the dense spray region in a diesel engine. SAE paper, (SAE
972866), 1997.

Numerical Modeling of Diesel Spray Formation and C

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Numerical Modeling of Diesel Spray Formation and C

Uploaded by

Copyright:

Available Formats

See

Numerical Modeling of Diesel Spray Formation

Article January 2009

Cemil Bekdemir L.M.T. Somers

SEE PROFILE SEE PROFILE

RCCI engine control oriented model development View project

Biomass combustion modeling View project

The user has requested enhancement of the downloaded file.

Numerical Modeling of Diesel Spray

Exam committee: prof.dr. L.P.H. de Goey

Eindhoven University of Technology

c 2008 by Cemil Bekdemir

2 Fundamentals of Liquid Sprays 3

4 CFD Spray Model 15

5 CFD and Phenomenological Spray Model 31

6 Modeling Spray Combustion with Tabulated Chemistry 49

7 Conclusions and Recommendations 61

A Transport Equations for the Mean and Variance of Z and PV 65

Fundamentals of Liquid Sprays

2.1 Spray Regimes

2.2 Breakup Regimes

Second wind- Atomization

Figure 2.2: Breakup scheme; treated regimes

2.2.1 Primary Breakup

2.2.2 Secondary Breakup

vibrational breakup (We 12)

bag breakup (We < 20)

bag/streamer breakup (We < 50)

stripping breakup (We < 100)

catastrophic breakup (We > 100)

2.2.3 Atomization in Diesel Sprays

2.3 Direct Injection Diesel Combustion

3.1 Thermodynamic Models

3.2 Phenomenological Models

3.3 CFD Models

CFD Spray Model

4.1.1 Transport Equations

Continuity equation The general continuity equation is written as follows:

Energy equation The energy equation in Fluent is written as follows:

4.1.2 Discrete Phase Model

continuous phase flow field calculation

particle trajectory calculation

update continuous phase source terms

Figure 4.3: Solving procedure for one-way and two-way coupling

Figure 4.5: Sketch of the Wave breakup principle [KSR99]

4.1.3 Model Settings

Figure 4.6: Square mesh, 14 th piece of the spray

Figure 4.7: Slice mesh, 18 th slice of a cilinder

4.2 Results and Parametric Dependencies

Table 4.1: Experimental data of EHPC and Sandia measurements

4.2.1 Spray Length

Penetration lengths (EHPC)

40 Cells t [s] Alignment

Physical properties Spray formation includes thermodynamic interaction between two

Penetration lengths (EHPC)

Penetration lengths (Sandia)

4.2.2 Liquid Length

Liquid Lengths (EHPC)

CFD and Phenomenological Spray

5.1 Phenomenological Model

no velocity slip between gas and liquid phases

. gas and liquid gas only

f l,0 A0 vf2l,0 = f l [1 m(x)]A(x)v(x)2 + g (x)m(x)A(x)v(x)2 (5.2)

mf l,0 hf l (Tf l,0 ) + ma (x)ha (Ta ) = f l [1 m(x)]A(x)v(x)hf l (T (x))+ (5.3)

6 Z = (sYFu YO + YO ,2) / (sYFu,1 + YO ,2)