Sie sind auf Seite 1von 30

PLEASE SCROLL DOWN FOR ARTICLE

This article was downloaded by: [Universiteit Gent]


On: 25 January 2010
Access details: Access Details: [subscription number 915540789]
Publisher Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-
41 Mortimer Street, London W1T 3JH, UK
Combustion Theory and Modelling
Publication details, including instructions for authors and subscription information:
http://www.informaworld.com/smpp/title~content=t713665226
A generalized flame surface density modelling approach for the auto-
ignition of a turbulent non-premixed system
F. A. Tap
ab
; R. Hilbert
c
; D. Thvenin
c
; D. Veynante
a
a
Laboratoire EM2C, cole Centrale Paris, Grande Voie des Vignes, Chtenay-Malabry, France
b
PSA
Peugeot Citron Automobiles, DRIA/SARA/EMSA/PVMO/C3DM, Vlizy-Villacoublay, France
c
LSS
(Lehrstuhl fr Strmungsmechanik und Strmungstechnik), Otto-von-Guericke-Universitt
Magdeburg, Institute fr Strmungsmechanik und Thermodynamik, Magdeburg, Germany
To cite this Article Tap, F. A., Hilbert, R., Thvenin, D. and Veynante, D.(2004) 'A generalized flame surface density
modelling approach for the auto-ignition of a turbulent non-premixed system', Combustion Theory and Modelling, 8: 1,
165 193
To link to this Article: DOI: 10.1088/1364-7830/8/1/009
URL: http://dx.doi.org/10.1088/1364-7830/8/1/009
Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf
This article may be used for research, teaching and private study purposes. Any substantial or
systematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply or
distribution in any form to anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses
should be independently verified with primary sources. The publisher shall not be liable for any loss,
actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directly
or indirectly in connection with or arising out of the use of this material.
INSTITUTE OF PHYSICS PUBLISHING COMBUSTION THEORY AND MODELLING
Combust. Theory Modelling 8 (2004) 165193 PII: S1364-7830(04)64803-0
A generalized ame surface density modelling
approach for the auto-ignition of a turbulent
non-premixed system
F A Tap
1,3
, R Hilbert
2
, D Th evenin
2
and D Veynante
1,4
1
Laboratoire EM2C,

Ecole Centrale Paris, Grande Voie des Vignes, 92295 Ch atenay-Malabry,
France
2
LSS (Lehrstuhl f ur Str omungsmechanik und Str omungstechnik),
Otto-von-Guericke-Universit at Magdeburg, Institute f ur Str omungsmechanik und
Thermodynamik, Universit atsplatz 2, 39106 Magdeburg, Germany
3
PSA Peugeot Citro en Automobiles, DRIA/SARA/EMSA/PVMO/C3DM, Route de Gisy, 78943
V elizy-Villacoublay, France
E-mail: Denis.Veynante@em2c.ecp.fr
Received 16 June 2003, in nal form 28 November 2003
Published 18 February 2004
Online at stacks.iop.org/CTM/8/165 (DOI: 10.1088/1364-7830/8/1/009)
Abstract
Auto-ignition of turbulent non-premixed systems is encountered in practical
devices such as diesel internal combustion engines. It remains a challenge
for modellers, as it exhibits specic features such as unsteadiness, ame
propagation and combustion far fromstoichiometric conditions. In this paper, a
two-dimensional DNSdatabase of an igniting H
2
/O
2
/N
2
mixing layer, including
detailed chemistry and transport, is extensively post-processed in order to gain
physical insight into the ame structure and dynamics during auto-ignition.
The results are used as a framework for the development of a generalized
ame surface density modelling approach by integrating the equations over
all possible mixture fraction values. The mean reaction rate is split into two
contributions: a generalized ame surface density and a mean reaction rate
per unit generalized ame surface density. The unsteadiness of the ignition
phenomenon is accounted for via a generalized progress variable. Closures
for the generalized surface average of the reaction rate and for the generalized
progress variable are proposed, and the modelling approach is tested a priori
versus the DNS data. The use of a laminar database for the chemistry coupled
to the mean turbulent eld via the generalized progress variable shows very
promising results, capturing the correct ignition delay and the premixed peak
in the turbulent mean heat release rate evolution. This allows condence in
future inclusion and validation of this approach in a RANS-CFD code.
4
Author to whom any correspondence should be addressed.
1364-7830/04/010165+29$30.00 2004 IOP Publishing Ltd Printed in the UK 165
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
166 F A Tap et al
Nomenclature
Latin
c progress variable
c
p
specic heat capacity at constant pressure (J kg
1
K
1
)
D diffusion coefcient (m
2
s
1
)
K constant
n unity vector normal to an iso-level
P(.) probability density function
Re
t
Reynolds number based on the integral length-scale
T temperature (K)
u velocity (ms
1
)
W molar mass (kg mol
1
)
x
i
spatial coordinate in i-direction (m)
Y mass fraction
Z mixture fraction
Z
e
mass fraction of element (e = H, N, O)
Greek
angle (Rad)

asymptotic value of the generalized progress variable


(.) Dirac delta-function
heat conductivity coefcient (Wm
1
K
1
)
integral turbulent length scale (m)
generalized progress variable

Z
ame surface density (m
1
)

Z
generalized ame surface density (m
1
)
density (kg m
3
)

1,2
characteristic timescale (s)
mixture ratio
scalar dissipation rate (s
1
)
reaction rate (kg m
3
s
1
)

T
heat release rate (Wm
3
)

reaction rate per unit surface density (kg m


2
s
1
)

T
heat release rate per unit surface density (Wm
2
)
Indices
mr relative to most-reactive conditions
st relative to stoichiometric conditions
F fuel
O oxidizer
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
Modelling of turbulent non-premixed auto-ignition 167
Exponents
MIX mixing regime
IFC innitely fast chemistry regime
EQ equilibrium regime
conditions in oxidizer and fuel streams
Averages
(.) Reynolds averaged quantity

(.) Favre averaged quantity


(.|.) Conditionally averaged quantity
.

S
Surface averaged quantity
Acronyms
CFD computational uid dynamics
CFM coherent ame model
CMC conditional moment closure
DNS direct numerical simulation
EGR exhaust gas recirculation
NSCBC NavierStokes characteristic boundary conditions
PDF probability density function
1. Introduction
This work focuses on a new approach to model the auto-ignition of a turbulent non-premixed
combustion system. Understanding and modelling this kind of process is of both great
theoretical and practical interest. This conguration typies the fundamental interaction
between chemical reactions and molecular and thermal diffusion together with turbulent
mixing. It is also a relevant model situation for studying the transition between the pure
mixing and the diffusive burning regimes found in many practical applications, such as
diesel engines (Heywood 1998).
Non-premixed ames correspond to situations where the reactants, fuel and oxidizer are
not initially mixed and originate from two separate streams. The burning rate is controlled by
the speed at which the reactants diffuse towards the reaction zone, where the heat produced by
chemical reactions is released. The physical description of turbulent non-premixed combustion
often relies on the introduction of a conserved scalar, the mixture fraction, Z (Williams 1985),
which takes the value 0 on the oxidizer side and 1 on the fuel side. This approach consists of
assuming that the chemical state and thereby the species mass fractions and the temperature
can be related to the mixture fraction and thus allows a formal decoupling between turbulent
mixing on the one hand and a description of the local ame structure in Z-space on the other
hand. Turbulent mixing is then described by the mixture fraction distribution, and in the
case of well-established non-premixed ames (far from extinction and ignition limits), the
reaction zone is located in the vicinity of the stoichiometric conditions and the ame front can
be identied as the iso-level of Z = Z
st
.
The main characteristics of the auto-ignition process of a non-premixed ame will be
rediscussed in this paper, but to briey summarize the problem, theoretical and numerical
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
168 F A Tap et al
results have provided evidence that the following points may have to be accounted for in
modelling approaches.
It has been proved theoretically (Li n` an and Crespo 1976) and veried from direct
numerical simulations (Mastorakos et al 1997, Sreedhara and Lakshmisha 2000, Hilbert
and Th evenin 2002a) that the mixture formed by the diffusion of heat and species does not
necessarilyignite at stoichiometric conditions but at a so-calledmost-reactive value of the
mixture fraction, denoted as Z
mr
. This most-reactive value of the mixture fraction can be
determined by asymptotic analysis for simple chemistry or by laminar ame calculations
for complex chemistry.
In contrast to well-established diffusion ames, for which the reaction rate increases
(up to quenching) with increasing values of scalar dissipation rate, , a non-premixed
ame ignites preferentially at low values of . In terms of the well-known S-curve
(Williams 1985), auto-ignition processes are represented on the lower branch of the curve.
It has been shown by DNS that ignition processes in non-premixed congurations can
exhibit complex ame structures implying both premixed and non-premixed combustion
regimes (Domingo and Vervisch 1996, Hilbert et al 2001, Echekki and Chen 2002).
When differential diffusion is accounted for in the simulation, focusing of light species
like hydrogen and the simultaneous de-focusing of heat (Law 1988, Takagi and Xu 1994)
may occur. This effect causes the ignition spots to appear on convex (viewed from the
fuel side) parts of the iso-contour of the most-reactive mixture fraction (Im et al 1998,
Hilbert and Th evenin 2002a).
For modelling the ignition and combustion of a non-premixed system, various approaches
may be found in the literature (Veynante and Vervisch 2002). The simplest models are generally
based on an ad hoc coupling between the Arrhenius laws and the Magnussen eddy dissipation
concept (Magnussen and Hjertager 1976). Ignition is treated specically by a generic model
such as the Shell (Halstead et al 1977) or IFP (Pinchon 1989) model. Such an ignition model
accounts for the nite-rate chemistry and is coupled to the mean ow calculation (Natarajan
and Bracco 1984, Sh apert ons and Lee 1985, Theobald and Cheng 1987, Zellat et al 1990, Kong
and Reitz 1993). These ignition models have now been replaced by stiff chemistry solvers
(Chomiak and Karlsson 1996, Agarwal and Assanis 1998, Golovitchev et al 2000, Kong et al
2001, Hong et al 2002).
Approaches based on a presumed PDF method may also be found, such as the MIL
model (Borghi 1988, Demoulin and Borghi 2002), which rely on a description of sudden
chemistry in mixture fraction space by considering either the mixing or the equilibrium
regime and comparing the ignition delay time with a PDF of mixing times. The PDFA/CHI
model (Pires Da Cruz et al 2001) combines an Arrhenius law weighed with a presumed
mixture fraction probability density function for ignition and subsequent ame propagation,
coupled, via a progress variable, to a amelet-type reaction rate for the diffusive combustion.
A CMC-type ignition model was devised by Mastorakos et al (1997), solving for a mean
temperature increment and its variance, conditioned on the most-reactive mixture fraction
value. A transient amelet model for auto-ignition making use of tabulated chemical source
terms has been proposed (Bruel et al 1990, Zhang et al 1995, Chang et al 1996) and applied to
simulate ignition of fuel jets successfully. This model makes use of an ignition time variable,
which acts as a progress variable of the ignition process, and presumes the PDFs for the
mixture fraction, the scalar dissipation rate and the ignition time variable. A amelet library
is generated by solving the unsteady amelet equations in mixture fraction space.
The laminar amelet model (Peters 1984) uses a laminar non-premixed ame as a model
for the turbulent ame. It relies on the expression of the ame structure in the mixing
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
Modelling of turbulent non-premixed auto-ignition 169
space, i.e. on expressions for species and temperature as a function of mixture fraction,
Z. These expressions can be derived from the balance equations for combustion by a
coordinate transformation of the Cartesian equations into a coordinate system attached to the
stoichiometric surface, with the Z-direction perpendicular to the ame sheet (Peters 1984,
Williams 1985). This derivation makes evident the importance of the scalar dissipation
rate = 2D|Z|
2
, which is the inverse of a diffusive time and characterizes the local
mixing rate. This quantity is the parameter used to quantify the departure from the classical
innitely fast chemistry limit. The expression for the mean turbulent reaction rate is
based on a local description of the balance between production in the ame front by the
chemical reactions and diffusion to and away from the reaction zone. The use of tabulated
solutions of the amelet equations (or of laminar counterow ame computations or an
experimental database) is not always justied as the amelets do not respond innitely fast
to changes in the ow eld (Haworth et al 1988); in order to take into account the transient
history of the scalar dissipation rate, an interactive amelet model has been devised (Pitsch
et al 1995, Peters 2000). It solves the unsteady amelet equations interactively with a
CFD code.
The ame surface density concept was rst introduced in the CFM(Marble and Broadwell
1977), in a diffusion ame context. An exact, but unclosed, balance equation for the ame
surface density can be derived (van Kalmthout et al 1996, van Kalmthout and Veynante 1998)
starting from the balance equation for the mixture fraction. The ame surface density,
Z
(the
subscript Z refers to the non-premixed context), measures the available ame surface per unit
volume. The mean burning rate is expressed as the product of the ame surface density and a
laminar reaction rate per unit ame surface density. This denition is based on the identication
of the ame front as the stoichiometric iso-level. With this approach, the expression of the
local mean burning rate can be based on steady, strained laminar ames. The parameter used
for the coupling with turbulent mixing is the strain rate. A library of counterow ames is
calculated a priori, and the values of reaction rates per unit ame surface density are read
from the library during the turbulent CFD calculation. Applications of the CFM approach to
ignition problems may be found in Veynante et al (1991), Dillies et al (1993), Fichot et al
(1994), Musculus and Rutland (1995).
More generally applicable models for non-premixed combustion, such as PDF methods
(Dopazo and OBrien 1974, Pope 1985, Pope 1990) and the CMC(Klimenko and Bilger 1999),
do not need a model for the reaction rate and are thus very promising. The dimensionality of
the problem is large, however, when considering detailed chemistry, and so their application
to an ignition problem in an industrial device (a diesel engine for example) is computationally
very expensive and has not yet been realized.
The goal of this work is to further develop and test a new modelling approach for auto-
ignition (Hilbert et al 2002), based on DNS data. This approach relies on a generalization
of the ame surface density equations and the expression for the local reaction rate via an
unsteady amelet approach. The two stages of this study are rst to prove that the coupling of
these two approaches to simulate ignition is well suited when using a generalized formulation,
and then to validate a strategy for the local description of the ame structure and the coupling
to the mean ow eld.
The DNS database used for physical analysis is presented rst, with some typical results
for the ignition sequence. A detailed investigation of the ame structure during the ignition
process is then proposed for both the laminar and the turbulent cases. Emphasis is put on
the observed distribution of the scalar dissipation rate, (Z). In order to assess qualitatively
the main direction of propagation (with respect to the iso-Z levels) of the ignition kernels
in the DNS simulations, a new parameter, cos , is introduced.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
170 F A Tap et al
A generalized progress variable, , is presented, and the temporal evolution of is
analysed in the laminar DNSsimulations, showing the interest of this quantity for describing the
advancement of the overall ignition process. Flamelet structures in the mixing space, extracted
from the DNS along cuts across the Z iso-levels, are compared with the structure obtained
from a laminar, unsteady and unstrained simulation via the generalized progress variable, ,
in order to observe in particular the transition between the mixing and equilibrium regimes.
Based on these observations, a new modelling approach, relying on the generalization of
the ame surface density equations, is proposed and tested. The different terms of the model
are extracted fromDNS results and discussed. Amodel is proposed for the generalized surface
average of the reaction rate,

, and for the generalized progress variable, . Finally, it is shown
that, when using this modelling approach, the temporal evolution of the mean heat release rate
of the turbulent DNS simulations can be successfully reconstructed.
2. DNS database
In this section, a brief overview of the available literature on DNS of auto-ignition is given, as
well as the details of the DNS database used in this work. Also shown are typical results from
these two-dimensional turbulent DNS simulations.
2.1. Literature overview
Although limited to low Reynolds numbers and academic congurations, DNS is known to be
an interesting tool to help understand the fundamental mechanisms in turbulent combustion
and is often used as a framework for collecting crucial information for modelling (Poinsot
et al 1996, Poinsot 1996, Vervisch and Poinsot 1998, Poinsot and Veynante 2001). Two-
dimensional simulations of auto-ignition of turbulent non-premixed systems were rst carried
out by Mastorakos and co-workers (Mastorakos et al 1995, 1997), using a single-step reaction
model and unity Lewis numbers for the expression of diffusion velocities. The authors
demonstrated the existence of so-called most-reactive conditions during the mixing of the
reactants. These conditions are associated with a given value of the mixture fraction, Z,
denoted as Z
mr
, which can be estimated a priori by maximizing the analytical expression
of the reaction rate. The importance of the scalar dissipation rate = 2D|Z|
2
was also
emphasized in order to predict the localization of the rst appearance of self-ignition. The
main conclusion of this work is that the mixture rst auto-ignites for Z = Z
mr
and at the point
where minimum values of are found. A CMC-type turbulent combustion model was derived
based upon these conclusions (Mastorakos et al 1997, Mastorakos and Bilger 1998).
Im et al (1998) carried out similar two-dimensional turbulent calculations of a globally
rich ame ( = 2.2) using a detailed chemical scheme with nine species and 38 reactions for
the combustion of hydrogen in air. In this study, a constant Lewis number for each species was
considered, without thermal diffusion, and a constant Prandtl number. The authors showed
that for ignition temperatures higher than the crossover temperature, the ignition process can
be described by a chemical non-linear coupling without having to involve thermal feedback.
Also observed was that self-ignition can be slightly delayed, in the turbulent case, at high
turbulence intensities where the turbulent timescale is shorter than the laminar self-ignition
time. But it was concluded that this point requires further research.
Sreedhara and Lakshmisha (2000) presented results of the auto-ignition of an n-heptane/air
ame, where the chemistry was described by a single-step reaction. These DNS results
are used to improve CMC modelling. More recently, the same conguration in a three-
dimensional turbulent mediumwas simulated with a four-step reduced mechanismspecically
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
Modelling of turbulent non-premixed auto-ignition 171
ox,0
Y
fu,0
Y
ox,0
Y
Hot Air
1100 K
Cold Fuel
300 K
Hot Air
1100 K
Periodicity
Periodicity
x
y
P=1 atm
u'
Initial turbulent field

S
u
b
s
o
n
i
c

n
o
n
-
r
e
f
l
e
c
t
i
n
g

o
u
t
l
e
t
S
u
b
s
o
n
i
c

n
o
n
-
r
e
f
l
e
c
t
i
n
g

o
u
t
l
e
t
Figure 1. Turbulent double-ame calculation layout. The physical domain is 1 cm 1 cm.
derived for auto-ignition of n-heptane/air mixtures (Sreedhara and Lakshmisha 2002a,b).
The authors studied in particular the effects of three-dimensional turbulence on auto-ignition
characteristics (Sreedhara and Lakshmisha 2002a). This work particularly shows the interest
in a three-dimensional description of turbulence: a comparison between two-dimensional and
three-dimensional simulations reveals differences that can be attributed to the effect of vortex
stretching. Parametric studies show that the ignition time decreases with turbulence intensity,
which is consistent with the results presented by Hilbert and Th evenin (2002a).
2.2. DNS conguration
The DNS database used in this paper has been presented in detail in previous work (Hilbert
and Th evenin 2002a). The DNS code parcomb (Th evenin et al 1996, 1997) has been used
with a detailed chemical scheme of nine species (H
2
, O
2
, H
2
O, OH, H, O, HO
2
, H
2
O
2
and N
2
)
and 37 elementary reactions to describe the combustion of H
2
in air (Maas and Warnatz 1988)
and taking into account multi-component diffusion velocities and thermo-diffusion effects
(Hirschfelder et al 1954, Maas and Warnatz 1989). The importance of a detailed description
of transport properties has been emphasized in previous work (Hilbert and Th evenin 2002a).
The simulation is initialized with uniform proles on the left and right sides and in the middle,
with steps for temperature and species (gure 1). On the left and right sides, the initial
conditions T = 1100 K, Y
O
2
= 0.233 and Y
H
2
= 0 are imposed. In the middle, the values
T = 300 K, Y
O
2
= 0 and Y
H
2
= 0.023 are imposed. An appropriate nitrogen complement is
added everywhere. These initial compositions result in a global equivalent ratio of = 0.8.
A 501 501 regularly spaced grid is used, with a spatial resolution of 20 m, to simulate a
1 cm 1 cm domain. Extended NSCBCs (Poinsot and Lele 1992, Baum et al 1995) are used,
and a pseudo-turbulent velocity eld is initiated in the associated Fourier space using a von
K arm an spectrum with Pao correction following Hinze (1975), with a random-seed number
for the velocity uctuations. In this conguration, the two uids, the hot oxidizer and the
cold diluted fuel, are initially separated, and the hot oxidizer serves as an energy source, such
that the temperature of the mixture formed by turbulent diffusion of heat and reactants will be
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
172 F A Tap et al
high enough to promote self-ignition of a non-premixed ame. Calculations are performed at
atmospheric pressure, and the integral-scale turbulent Reynolds number is Re
t
= 297. The
calculations have been repeated in order to compensate for the randominitial conditions, which
are due to the method used to generate the initial synthetic turbulent velocity eld. The DNS
database nally consists of six independent realizations of the turbulent auto-ignition process.
The simulations were stopped at the time when the ame leaves the computational domain,
which is not the same time in each case; the complete statistics of the database are 3000 points
in the y-direction up to t = 0.28 ms, 2000 points up to t = 0.35 ms and 1000 points up to
t = 0.38 ms. This database still allows us to extract statistically meaningful mean proles
as most of the post-processed data have been taken from times around the rst appearance of
ignition kernels, around t = 0.27 ms.
2.3. Typical results
When dealing with multi-step chemistry and detailed transport, there is no standard denition
for the mixture fraction, and many denitions can be found that do not necessarily correspond
to each other (Hilbert and Th evenin 2002b). In this study, the mixture fraction is calculated
using the denition proposed by Bilger (1988):
Z =
1/2 Z
H
/W
H
+ (Z

O
Z
O
)/W
O
1/2 Z

H
/W
H
+ Z

O
/W
O
, (1)
where Z
H
and Z
O
are elemental mass fractions, Z

H
, Z

O
represent the respective values on
the fuel and oxidizer sides and W
H
, W
O
are the molar masses of elements H and O. With this
denition, the stoichiometric mixture corresponds to Z = Z
st
= 0.56.
Atypical result for turbulent simulations is presented in gure 2. Before ignition, reactants
behave as if evolving in a perfect mixing regime. The initial increase of heat release indicates
the rst position of ignition. It has been shown and explained in detail in previous publications
(Mastorakos et al 1997, Hilbert and Th evenin 2002a) that auto-ignition appears at the point
where |Z
mr
(i.e. conditioned on the Z = Z
mr
iso-contour) is minimal. After ignition, the
ignition kernels propagate towards the stoichiometric iso-level, implying both premixed and
non-premixed combustion regimes. The importance of partially premixed combustion during
this process has been emphasized in previous work (Hilbert et al 2001), which is consistent
with the results from Echekki and Chen (2002) in a slightly different conguration. In order
to allow comparisons between laminar and turbulent auto-ignition, a laminar simulation has
been carried out in a one-dimensional domain with exactly the same parameters.
The DNS database is used in this paper to study local ame structures during auto-ignition
and to test the newmodelling approach based on generalized formulations for the ame surface
density and the progress variable.
3. Flame structure analysis
The transient ame structure of the igniting non-premixed system described above will be
analysed in this section, starting with the laminar case. These results are compared with the
turbulent case by extracting local amelet structures following the mixture fraction gradient.
In order to gain condence in the use of laminar diffusion ames as a sub-model for an
igniting, turbulent, non-premixed system, a new criterion is introduced, cos , that compares
the transfers of fuel in the direction across iso-Z levels with the transfers along iso-Z levels.
Section 3.4 introduces a generalized progress variable, , that allows a description of the degree
of progress of the non-premixed ignition phenomenon. This variable is used to compare local
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
Modelling of turbulent non-premixed auto-ignition 173
7.2
6.1
5.0
3.9
2.8
1.7
0.6
10
9
W.m
-3
Figure 2. Self-ignition of a non-premixed, turbulent, diluted hydrogen/air ame, computed with
DNS using a complete reaction scheme and zeroth-order approximation of the multi-component
diffusion velocities including thermal diffusion. Instantaneous vorticity elds (thin dashed lines),
heat release (thin solid lines) and two specic iso-levels (thick solid lines) of the most-reactive
(left line) and stoichiometric (right line) mixture fractions at four successive times (from left to
right t = 0.20, 0.25, 0.30, 0.40 ms). The hot air is located on the left-hand side of the domain, the
cold fuel on the right-hand side. The iso-levels are identical on all gures to facilitate comparisons.
Reproduced from Hilbert and Th evenin (2002a).
amelet structures from the two-dimensional turbulent DNS calculations with the laminar
proles.
3.1. Laminar ame structure
This rst paragraph discusses the self-ignition of a laminar non-premixed ame (the laminar
case corresponding to gure 1), which is a useful exercise as the laminar ame exhibits
several very interesting features. The ame structure is represented as a function of mixture
fraction. In the cases considered in this work, the mixture fraction is dened by equation (1).
Figure 3 shows instantaneous proles of Y
F
(Z), Y
MIX
F
(Z) and Y
EQ
F
(Z) at three different times.
Y
MIX
F
(Z) is the mixing prole, i.e. the fuel mass fraction prole if no reaction takes place.
Here, the initial prole of Y
F
(Z) is taken, which is not a straight line because of differential
diffusion. Y
EQ
F
(Z) is the equilibrium prole of the fuel mass fraction that would be reached
if there were no strain present. As detailed chemistry is used, Y
EQ
F
(Z) must be locally
computed, which is done with the code EQUIL (Lutz et al 1996), from the CHEMKIN package
(Kee et al 1989), which calculates the equilibriumcomposition and temperature by minimizing
the free energy of a perfect-gas mixture at constant pressure. Also shown in gure 3 is the
normalized prole of the heat release as an indicator of the instantaneous location of the
reaction zone.
The rst graph in gure 3 corresponds to a time just after ignition has occurred at
Z
mr
= 0.11. The second graph shows the ame structure a few instants later, corresponding
to t = 0.40 ms. The premixed ame is propagating from Z
mr
towards Z
st
= 0.56. The third
graph shows the ame structure when the diffusion ame is established at the stoichiometric
Z-level; the premixed front quenches in the rich ammability limit. Figure 3 clearly reveals
the propagative (across iso-Z levels) nature of the ignition phenomenon. Under stoichiometric
conditions, the fuel mass fraction does not exactly go to zero because of the strain in the ame.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
174 F A Tap et al
0 0.2 0.4 0.6 0.8 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Z
N
o
r
m
a
l
i
z
e
d

v
a
r
i
a
b
l
e
s
t =0.30 ms
0 0.2 0.4 0.6 0.8 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Z
N
o
r
m
a
l
i
z
e
d

v
a
r
i
a
b
l
e
s
t =0.40 ms
0 0.2 0.4 0.6 0.8 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Z
N
o
r
m
a
l
i
z
e
d

v
a
r
i
a
b
l
e
s
t =0.55 ms
Figure 3. Results from the laminar one-dimensional simulation: normalized instantaneous proles
of Y
F
(Z) (), Y
MIX
F
(Z) (- - - -), Y
EQ
F
(Z) ( ) and heat release rate ( ) at three different
times.
One can also see that the premixed form of combustion releases much more heat than the
resulting diffusion ame, which in this case is a diffusion ame between the excess fuel and
oxidizer in the burnt gases.
In most studies of auto-ignition, the scalar dissipation rate, , is used as the controlling
parameter of ignition location. It is dened as
= 2D
Z
_
Z
x
i
_
2
, (2)
where D
Z
is the diffusivity of the mixture fraction, which is not dened since the mixture
fraction is reconstructed from elemental mass fractions (see equation (1)). Here, D
Z
is set to
be equal to the diffusivity of heat:
D
Z
=

c
p
. (3)
Physically, the scalar dissipation rate is the inverse of a diffusion time, characterizing the speed
at which the reactants move towards each other on a molecular level, and also the speed at which
heat diffuses to and away from the reaction zone. Therefore, it is an important characteristic
parameter in ignition problems. It has been shown that also in the turbulent simulations
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
Modelling of turbulent non-premixed auto-ignition 175
0 0.2 0.4 0.6 0.8 1
0
50
100
150
200
250
300
350
400
450
Z


(
s

1
)


(
s

1
)


(
s

1
)
t =0.30 ms
0 0.2 0.4 0.6 0.8 1
0
50
100
150
200
250
300
Z
t =0.40 ms
0 0.2 0.4 0.6 0.8 1
0
20
40
60
80
100
120
140
160
180
Z
t =0.55 ms
Figure 4. Results from the laminar one-dimensional simulation: instantaneous proles of (Z)
at three different times. The dashed line represents the location of maximum heat release in the
mixture fraction space.
considered in this work the ignition locations correspond to lowest values of scalar dissipation
rate along the most-reactive mixture fraction iso-level (Hilbert and Th evenin 2002a).
Figure 4 shows scalar dissipation rate proles as a function of Z, extracted fromthe laminar
DNS results, for different times during the auto-ignition process. The proles exhibit a local
minimum of , in the vicinity of the position in the mixture fraction space where maximum
heat release is observed (represented by the dashed line). This particular distribution of in
the Z-space has also been observed in a ame/vortex interaction, in cuts across the ame front
where partially premixed phenomena are of importance (Hilbert 2002), and it is not equivalent
to the one classically used in amelet approaches, i.e. the inverse of the complementary error
function (Peters 2000).
3.2. Local amelet structures in the turbulent case
In order to compare the laminar ame structure with the local ame structure in a turbulent
ow eld, a post-processing procedure has been devised (Hilbert 2002) to extract amelet
information from the turbulent DNS database. This procedure relies on non-linear cuts in
mixture fraction space (across the ame), starting from a point of maximum heat release and
following the steepest mixture fraction gradient.
Figure 5 shows the proles of the fuel mass fraction and scalar dissipation rate extracted
from the turbulent simulation (t = 0.28 ms), at three different locations, following the local
mixture fraction gradient. Cut A is across the rst-appeared ignition kernel (on the Z
mr
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
176 F A Tap et al
1
1.5
2
2.5
3
3.5
4
4.5
5
5.5
6
x 10
9
0 1 2 3 4 5
x 10
0
1
2
3
4
5
6
7
8
9
x 10
3
X (m)
Y

(
m
)
Heat release
rate (Wm
3
)
0 0.5 1
0
0.005
0.01
0.015
0.02
Z
Y
H
2
A
B
C
0 0.5 1
0
200
400
600
800
1000
Z
A
B
C
A
C
B


(
s

1
)
Figure 5. Left: results from the two-dimensional turbulent simulation: contours of heat release
with the Z
st
iso-contour () and the different cuts following mixture fraction gradient (- - - -).
The hot air is located on the left-hand side of the domain, the cold fuel on the right-hand side.
Upper right: proles of Y
H
2
in mixture fraction space at the three locations considered; bottom
right: scalar dissipation rate distribution for the different cuts. Time corresponds to t = 0.28 ms.
iso-level), which is now propagating towards the stoichiometric mixture fraction contour. Its
fuel mass fraction prole corresponds more or less to the rst graph of gure 3 but appears
slightly ahead in time. The observation that the turbulent ame may locally ignite faster than
the laminar ame was discussed by Hilbert and Th evenin (2002a) and is attributed to the
enhanced diffusion of the light and reactive H-species. Cut B is across an igniting kernel,
where combustion has just started, as can be seen from its fuel mass fraction prole. Cut C
nally is across a still non-reacting part of the mixing layer, and its fuel mass fraction prole
is the same as Y
MIX
F
in gure 3. One can note that the structures exhibited here look similar to
those described by Vervisch and Poinsot (1998).
The local scalar dissipation rate distributions during auto-ignition showas original a shape
in the mixing space as the laminar ones from the rst graph of gure 4, with a local minimum
value corresponding to the instantaneous location of the ame front. The maximum values
are much higher though, resulting in higher heat release rates. Thus it appears, at least in
this weakly turbulent case, that the igniting turbulent mixing layer locally has the same ame
structure as the corresponding laminar diffusion ame, with a shift in combustion timing and
intensity. This is an encouraging argument for using laminar diffusion ames to model the
turbulent ame.
3.3. Assessment of the amelet approach
The amelet concept identies the ame structure normal to the ame front (the Z
st
iso-level)
as a laminar ame. In fact, the derivation of the amelet equations implicitly neglects all terms
representing transfers along mixture fraction iso-surfaces (Peters 2000).
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
Modelling of turbulent non-premixed auto-ignition 177
0.5
1
1.5
2
2.5
3
3.5
4
4.5
x 10
9
0 1 2 3 4
x 10
3
0
1
2
3
4
5
6
7
8
9
x 10
3
X (m)
Y

(
m
)
Heat release
rate (Wm
3
)
0.7 0.8 0.9 1
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
x 10
9
cos
Figure 6. Results from the two-dimensional turbulent simulation. Left: instantaneous iso-levels of
heat release rate at t = 0.28 ms and most reactive iso-level (Z = Z
mr
, - - - -); the hot air is located
on the left-hand side of the domain, the cold fuel on the right-hand side. Right: averaged values
of heat release rate conditioned on the values of cos , extracted from the most-reactive mixture
fraction iso-level.
In order to assess the importance of transfers occurring across iso-levels of the mixture
fraction in the case of auto-ignition, a new criterion is introduced here (Hilbert 2002),
considering the angle between the unity vectors normal to Z iso-levels and to fuel mass fraction
iso-levels. Dening these vectors respectively as
n
Z
=
Z
|Z|
(4)
and
n
H
2
=
Y
H
2
|Y
H
2
|
, (5)
one can express the cosine of the angle between n
Z
and n
H
2
as
cos = n
Z
n
H
2
=
Y
H
2
|Y
H
2
|

Z
|Z|
. (6)
If cos 1, the direction of the mass fraction of fuel is mainly the same as the direction
given by the gradient of the mixture fraction, meaning that transfers along the iso-levels of
Z are negligible and that a description of the local ame structure as a function of Z only is
possible. On the other hand, if cos 0, transfers along Z iso-levels are of importance and
ame propagation along the mixture fraction iso-contours needs to be taken into account.
Figure 6 shows the conditional average of heat release rate on cos , extracted from the
most-reactive mixture fraction iso-contour in the turbulent simulation at t = 0.28 ms. It can
be seen clearly that most of the transfers occur in the direction following the mixture fraction
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
178 F A Tap et al
gradient as the highest values of ( | cos ) occur for values of cos > 0.8. This also holds for
several followingtime steps, duringwhichthe different ignitionkernels become interconnected,
whereafter it becomes impossible to use this post-processing method to evaluate ( | cos )
because of pocket formation in the ame front. This result supports the use of the amelet
concept for the ignition problem by considering heat and mass transfer only in the direction
given by the gradient of the mixture fraction. This also implies the conceptual idea of the
global ignition of this mixing layer being a sequential process of local ignitions, as opposed
to being locally propagating ames along mixture fraction iso-levels.
3.4. Generalized progress variable
The fact that the auto-ignition process is controlled by both mixing and chemistry processes
suggests that, contrarily to an established non-premixed or premixed ame, at least two param-
eters are necessary to describe the ame structure. The mixture fraction is used as an indicator
of the degree of mixing between the reactants, and a second parameter has to be introduced to
describe the degree of progress of the chemical reaction. Several approaches for this question
may be found in the literature on turbulent auto-ignition and partially premixed problems. For
example, Echekki and Chen (1996) use the temperature as the second independent variable
to parametrize the structure of a triple ame; Mastorakos et al (1997) introduce a reactedness
to describe auto-ignition, and Haworth et al (2000) dene a progress variable by the mean of
elemental mass fractions to account for multi-step chemistry and differential diffusion.
Most often, the way to describe the progress of the reaction is to introduce a progress
variable as done for perfectly premixed problems, where the so-called progress variable acts
as a descriptive variable of the ame structure (Poinsot and Veynante 2001). When dealing with
a single-step reaction mechanism and unity Lewis numbers, it is possible to build a progress
variable c(Z) in terms of the local value of the mixture fraction (Poinsot et al 1996). For a
given value of Z, c(Z) can be written as a measure of the departure to the fast chemistry limit,
normalized by the distance between the pure mixing and innitely fast chemistry limits.
This variable can be dened as
c(Z) =
Y
MIX
F
(Z) Y
F
(Z)
Y
MIX
F
(Z) Y
IFC
F
(Z)
, (7)
where Y
MIX
F
(Z) and Y
IFC
F
(Z) represent the mass fraction of fuel proles in the mixing
space for the two limiting regimes known as the mixing regime and the innitely fast
chemistry regime. Its theoretical basis is sketched in gure 7 for a single-step, irreversible
chemistry with unity Lewis numbers. The gure represents an ignition at Z
mr
, the most-
reactive mixture fraction. The value of Z
mr
depends on the initial conditions of temperature
gradient, composition and pressure of the system considered (Mastorakos et al 1997, Hilbert
and Th evenin 2002a). In the results presented hereafter, ignition takes place on the lean (and
hot) side; for heavy hydrocarbon ignition, a rich ignition is more likely (Pitsch and Peters
1998). Equation (7) thus is the ratio of distances AB and AC when the most-reactive mixture
fraction is considered. The so-built progress variable possesses all the properties needed for
modelling, i.e. it varies between 0 and 1 and may obey a convectivediffusive balance equation
with a source term. It has been used to couple ignition models to diffusive combustion models
(Pires Da Cruz et al 2001, Hong et al 2002).
There are several phenomena that occur during the ignition phase that unfortunately
complicate the direct use of equation (7) for the description of the global ignition problem.
First, ignition does not happen for a xed value of the mixture fraction in each case, but
this most-reactive value of the mixture fraction depends on initial conditions. Second, after
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
Modelling of turbulent non-premixed auto-ignition 179
Figure 7. Schematic representation of the ame structure in the mixture fraction space of an
igniting non-premixed system, represented by the dotted line. Ignition location is denoted by Z
mr
.
Figure 8. Schematic representation of equation (8) of the generalized progress variable, , for an
igniting non-premixed system. Lines are the same as in gure 7.
ignition, the ame propagates through the mixture fraction space, meaning that all values of the
mixture fraction should be considered in this transitional phase frommixing to equilibrium.
Equation (7) then is very difcult to model in a RANS context as it is local information that is
not available on the grid scale.
In order to characterize the degree of advancement of the overall ignition process, a
generalized progress variable is introduced at this point. The denition adopted here is based
on the evolution of the surfaces under the curves of gure 7:
=
_
1
0
Y
MIX
F
(Z) dZ
_
1
0
Y
F
(Z) dZ
_
1
0
Y
MIX
F
(Z) dZ
_
1
0
Y
IFC
F
(Z) dZ
. (8)
This denition allows us to describe the evolution of the system from the mixing to the
innitely fast chemistry regime, regardless of the ignition location in the mixture fraction
space and the instantaneous ame location in Z-space during the propagation phase. This
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
180 F A Tap et al
0 0.2 0.4 0.6 0.8 1
x 10
3 time (s)

norm. heat release rate


0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Figure 9. Time evolution of the generalized progress variable, (), and normalized mean
heat release rate (- - - -) in the laminar case.
generalized progress variable is very similar to the progress variable introduced in Zhang et al
(1995), based on combustion products. The different terms are represented in the schematic
in gure 8. By denition, = 0 corresponds to a mixing regime and = 1 to the innitely
fast chemistry regime, and the evolution between 0 and 1 quanties the degree of progress
of the global ignition phenomenon. As the theoretical innitely fast chemistry limit is never
reached, the asymptotic value of can also characterize strain effects in the ame. This is
a main difference from the progress variable of Zhang et al (1995), dened with respect to
the equilibrium value of the product species considered. Such a progress variable will always
reach unity, whereas the asymptotic value, , may vary as a function of the amelet boundary
conditions. Equation (8) also has the benet of an analytical solutions once the prole Y
F
(Z)
is known, which is not the case when considering the equilibrium limit as opposed to the
innitely fast chemistry limit.
The time evolution of the quantity dened by equation (8) is shown in gure 9 for the
laminar case. After the ignition delay, evolves rapidly towards 1 during the ame propagation
phase. It shows asymptotic behaviour towards 1 because the near-equilibrium chemistry in
the ame never reaches the theoretical limit of innitely fast chemistry. Also shown is the
normalized time evolution of the heat release rate, demonstrating that the premixed peak,
inherent to the mixture formation during the ignition delay, is captured when is evolving
from zero to unity. Moreover, is a global variable that can be resolved on a RANS-type grid
scale, using a balance equation for example.
3.5. Local ame structure reconstruction using the generalized progress variable,
As demonstrated in the foregoing paragraphs, a laminar non-premixed ame could be used to
represent the local ame structure of an igniting turbulent diffusion ame, and the generalized
progress variable, , characterizes the advancement of the ignition process. To investigate this
point in more detail, local turbulent ame structures are extracted from the DNS database and
compared with the laminar ame structures, in gures 1012. The extracted paths in gure 10
start at a point of maximum heat release rate and follow the local mixture fraction gradient.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
Modelling of turbulent non-premixed auto-ignition 181
1
1.5
2
2.5
3
3.5
4
4.5
5
x 10
9
2 0 2 4 6
x 10
3
0
1
2
3
4
5
6
7
8
9
x 10
3
X (m)
Y

(
m
)
Heat release rate (Wm
3
)
1
1.5
2
2.5
3
3.5
4
4.5
5
5.5
x 10
9
2 0 2 4 6
x 10
3
0
1
2
3
4
5
6
7
8
9
x 10
3
X (m)
Y

(
m
)
Heat release rate (Wm
3
)
Figure 10. Contours of heat release, Z
st
iso-contour () and the extracted cut following the
local mixture fraction gradient (- - - -), for two different values (along the cut) of the generalized
progress variable, . Left: t = 0.29 ms, = 0.08. Right: t = 0.33 ms, = 0.40. The hot air is
located on the left-hand side of the domain, the cold fuel on the right-hand side.
0 0.2 0.4 0.6 0.8 1
0
0.005
0.01
0.015
0.02
Z
Y
H
2
= 0.08
TURBULENT
LAMINAR
0 0.2 0.4 0.6 0.8 1
0
0.005
0.01
0.015
0.02
Z
Y
H
2
= 0.40
TURBULENT
LAMINAR
0 0.2 0.4 0.6 0.8 1
0
200
400
600
800
1000
1200
1400
Z
T
e
m
p
e
r
a
t
u
r
e

(
K
)
= 0.08
TURBULENT
LAMINAR
0 0.2 0.4 0.6 0.8 1
0
200
400
600
800
1000
1200
1400
1600
Z
T
e
m
p
e
r
a
t
u
r
e

(
K
)
= 0.40
TURBULENT
LAMINAR
Figure 11. Extracted ame structure in the mixture fraction space from the turbulent case and
comparison with the laminar ame for two different values (along the cut) of the generalized
progress variable, . Upper row: fuel mass fraction as a function of mixture fraction. Lower row:
temperature as a function of mixture fraction.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
182 F A Tap et al
0 0.2 0.4 0.6 0.8 1
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
x 10
9
Z
H
e
a
t

r
e
l
e
a
s
e

r
a
t
e

(
W
m

3
)
= 0.08
TURBULENT
LAMINAR
0 0.2 0.4 0.6 0.8 1
0
1
2
3
4
5
6
x 10
9
Z
H
e
a
t

r
e
l
e
a
s
e

r
a
t
e

(
W
m

3
)
= 0.40
TURBULENT
LAMINAR
0 0.2 0.4 0.6 0.8 1
0
0.5
1
1.5
2
x 10
3
Z
Y
O
H
= 0.08
TURBULENT
LAMINAR
0 0.2 0.4 0.6 0.8 1
0
1
2
3
4
5
x 10
3
Z
Y
O
H
= 0.40
TURBULENT
LAMINAR
Figure 12. Extracted ame structure in the mixture fraction space from the turbulent case and
comparison with the laminar ame for two different values (along the cut) of the generalized
progress variable, . Upper row: the heat release rate as a function of mixture fraction. Lower
row: OH mass fraction as a function of mixture fraction.
Different variables can be extracted along these paths, and the extracted fuel mass
fraction and mixture fraction are used to calculate the local generalized progress variable, ;
the extracted ame structure is then compared with the laminar ame at the corresponding value
of . A simplied version of equation (8) is retained here because the equilibrium and mixing
proles are not known. The differential diffusion effects on the fuel mass fraction distribution
in the mixture fraction space are neglected by assuming a piecewise-linear relationship between
the two quantities, as sketched in gures 7 and 8. With this assumption, one can write
Y
MIX
F
(Z) = Y

F
Z and Y
IFC
F
= max
_
0, Y

F

Z Z
st
1 Z
st
_
, (9)
and the generalized progress variable becomes
=
_
1
0
Y
MIX
F
(Z) dZ
_
1
0
Y
F
(Z) dZ
_
1
0
Y
MIX
F
(Z) dZ
_
1
0
Y
IFC
F
(Z) dZ

(1/2)Y

F

_
1
0
Y
F
(Z) dZ
(1/2)Y

F
(1/2)Y

F
(1 Z
st
)
=
1
Z
st
_
1
2
Y

F
_
1
0
Y
F
(Z) dZ
_
. (10)
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
Modelling of turbulent non-premixed auto-ignition 183
Figure 11 shows the fuel mass fraction and temperature distributions in the mixture fraction
space at two different instants during the ignition sequence, characterized by the generalized
progress variable, , extracted from the paths shown in gure 10. A very good agreement
is seen in both cases, especially for the fuel mass fraction. The temperature proles show a
discrepancy at the onset of ignition, but the results are in better agreement for the value of
= 0.40. However, for the intermediate species, OH, and the heat release rate, the agreement
is less good, as shown in the gure 12, although the location of the ame in the mixture fraction
space is well represented. The effects of curvature and higher scalar dissipation rates need to
be accounted for in order to reconstruct correctly the heat release rate.
4. Modelling approach and a priori tests on the DNS database
The aim of this section is to use the previously described physical insights to propose a new
modelling approach for the global ignition of a turbulent non-premixed system. The concept
of a generalized ame surface density is rst introduced, and the use of the generalized
progress variable coupled with the laminar, unsteady and unstrained simulation for the local
reaction rate sub-model is explained. The modelling approach is tested against the turbulent
DNS results.
4.1. Generalized ame surface density model formulation
The extension of the ame surface density concept to ignition problems is investigated in this
work. During auto-ignition, occurring on the Z
mr
iso-level, the ame cannot be identied
as the Z
st
iso-level as usually done for a well-established non-premixed ame. In order to
overcome this problem, the reaction rate and the ame surface density can be integrated over
all possible values of mixture fraction (Hilbert et al 2002).
Several denitions are introduced at this point. The surface density (i.e. the surface per
unit volume) of the Z = Z

iso-level is given classically by (Veynante and Vervisch 2002)

Z
= |Z|(Z = Z

) = (|Z||Z = Z

)P(Z

), (11)
where (Z Z

) is the Dirac delta function, (|Z||Z = Z

) the conditional average of |Z|


for Z = Z

and P(Z

) the probability of having Z = Z

.
A surface-averaged quantity along the Z = Z

iso-surface, Q

S
, is dened as (Veynante
and Vervisch 2002)
Q

S
=
1

(Q|Z|(Z Z

)) =
(Q|Z||Z = Z

)
(|Z||Z = Z

)
. (12)
In the ame surface density approach, the mean reaction rate is split into two contributions:
a reaction rate per unit of ame surface density and a mean turbulent ame surface density
(Marble and Broadwell 1977). The reaction rate per unit of ame surface density is usually
tabulated from laminar model ame calculations, and the ame surface density is modelled by
a balance equation.
Ageneralized surface density,

Z
, is nowintroduced by integrating over the entire domain
of Z (Veynante and Vervisch 2002):

Z
=
_
1
0
(|Z||Z = Z

)P(Z

) dZ

= |Z|, (13)
which is a global variable that could be solved via a balance equation. This surface density
is affected by turbulent motions, taking into account the straining effects of turbulence. The
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
184 F A Tap et al
mean reaction rate is here again split into two contributions and reads (Veynante and Vervisch
2002):
=
_

|Z|
_
S

Z
=

Z
, (14)
where, with the use of equation (12),

=
1
|Z|
_
1
0
_

|Z|
_

S

Z
dZ

=
1
|Z|
_
1
0
(( /|Z|)|Z||Z = Z

)
(|Z||Z = Z

)
(|Z||Z = Z

) P(Z

) dZ

=
1
|Z|
_
1
0
( |Z = Z

) P(Z

) dZ

=

|Z|
(15)
is the generalized surface average of the reaction rate, i.e the mean reaction rate per unit of
generalized surface density. Note that /|Z|

S
is the mean reaction rate per unit of Z = Z

surface density. This formulation thus directly extends the ame surface density concept
to non-innitely thin ame surfaces. The challenge now lies in the closure of these terms,
especially

, on which the remainder of this work will be focused.
Post-processing of the two-dimensional DNS results is done by assuming a one-
dimensional mean ame structure, and the quantities of interest are averaged along the
y-direction (gure 1); the x-direction thus corresponds to a variation of

Z between 0 and 1.
The rst step consists of extracting the three terms in equation (14) from the turbulent DNS
data, in the homogenous y-direction, where all six available cases have been added in order to
have statistically representative mean values. These results have already been presented and
discussed by Hilbert et al (2002) and are again shown in gure 13 for different instants around
the ignition time. The generalized ame surface density appears to be almost constant in time
and reaches its maximum value in the middle of the ame brush. Because of the increase in
the mean reaction rate per unit of generalized surface density, the averaged heat release rate
increases with time. Assuming that the generalized surface density may be described by an
algebraic formulation or a balance equation, the challenge is now to model

= /|Z|
S
to describe the auto-ignition process.
To illustrate the benet of the ame surface density approach, the mean values of heat
release rate and the generalized surface average of the heat release rate are plotted against time
for both the laminar and turbulent cases in gure 14 (these are total mean values of
T
(x, y)
and

T
(x)). The peak value of the turbulent mean heat release rate is much higher than the
laminar one as expected because of the higher strain and occurs later. It must also be noted
that in this case the laminar and turbulent ignition delays are nearly the same when looking
at mean values. This might be attributed to the fact that the turbulence levels in the DNS
calculations are quite low. The second graph shows that the generalized surface averages of
the heat release,

T
, have the same peak values and thus are independent of strain effects, as in
both cases the heat release rate has been normalized by the generalized ame surface density

Z
= |Z|, thus accounting for the straining effects. The evolution in the laminar case is
slightly faster than the turbulent one. A possible explanation for this is that in the turbulent
case the generalized ame surface density decreases faster than in the laminar case because
of the wrinkling of the mixing layer. This decomposition of the mean reaction rate into two
contributions, according to equation (14), allows us to model, at least in this low-turbulence
case, the mean reaction rate per unit of generalized ame surface density,

, froman unstrained
laminar diffusion ame.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
Modelling of turbulent non-premixed auto-ignition 185
X (cm)
0 0.1
1
0.2 0.3 0.4 0.5
0.0E+00
1.0E+08
2.0E+08
3.0E+08
4.0E+08
5.0E+08
6.0E+08

DNS
X (cm)
0 0.1 0.2 0.3 0.4 0.5
100
200
300
400
500
600
700

X (cm)
0 0.1 0.2 0.3 0.4 0.5
0.0E+00
2.0E+05
4.0E+05
6.0E+05
8.0E+05
1.0E+06
1.2E+06
T
.
Figure 13. Spatial distribution of turbulent mean heat release rate,


T
(top left), generalized
ame surface density,

Z
(top right), and generalized surface average of the heat release rate,

T
(bottom), at different times during the ignition sequence. Times correspond to t = 0.25 ms,
; t = 0.26 ms, ; t = 0.27 ms, ; t = 0.28 ms, . Reproduced from
Hilbert et al (2002).
4.2. Model for the generalized progress variable,
In order to be able to reconstruct froma RANSmean oweld calculation, one must knowthe
instantaneous prole of fuel mass fraction in mixture fraction space (see equation (8)), which
is not available on the grid scale. The strategy adopted here to is try and forecast the mean
evolution of instead of calculating it fromthe consumption of the fuel by chemical reactions.
This seems feasible as, during the ignition process, the evolution of will qualitatively be the
same in all cases, i.e. an asymptotic behaviour to unity after a certain ignition delay. Such an
evolution is parametrized using different characteristic timescales, as shown in gure 15, where

1
represents an ignition delay time,
2
a characteristic time for the speed of the reaction and an
asymptotic value,

, that accounts for the non-equilibrium effects. All these three parameters
will most likely depend on the scalar dissipation rate, , but this is only considered for
2
for
now. Such an evolution may be simulated with a hyperbolic tangent function, for example
(t ) =

1 + tanh(
1
/
2
(t ))
_
tanh
_
t
1

2
(t )
_
+ tanh
_

1

2
(t )
__
. (16)
This function is zero until the delay time,
1
, is reached and shows asymptotic behaviour
towards

, on a timescale characterized by
2
. These parameters are determined fromlaminar
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
186 F A Tap et al
0 0.5 1 1.5 2 2.5 3 3.5 4
x 10
4
0
1
2
3
4
5
6
7
x 10
8
time (s)
m
e
a
n

h
e
a
t

r
e
l
e
a
s
e

r
a
t
e

(
W
m

3
)
TURBULENT
LAMINAR
0 0.5 1 1.5 2 2.5 3 3.5 4
x 10
4
0
0.5
1
1.5
2
2.5
x 10
6
time (s) g
e
n
e
r
a
l
i
z
e
d

s
u
r
f
a
c
e

a
v
e
r
a
g
e

o
f

h
e
a
t

r
e
l
e
a
s
e

r
a
t
e

(
W
m

2
)
TURBULENT
LAMINAR
Figure 14. Temporal evolution of the laminar and turbulent mean heat release rates,


T
(left), and
generalized surface averages of the heat release rates,

T
(right).
0 0.2 0.4 0.6 0.8 1
x 10
3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
time (s)

LAMINAR
MODEL

Figure 15. Temporal evolution of from the laminar simulation () with the different
characteristic parameters used for the modelled -equation (equation (16)), represented by - - - -.
ame computations as

1
= 0.38 ms,

2
(t ) =
K
(t )
, K = 0.0044,

= 0.966,
where the constant K has been determined froma least-squares t. Fromequation (8), ignition
occurs when > 0, but here
1
is taken to be equal to the time where = 0.5, which allows
a better adjustment of the hyperbolic-tangent function.
2
is taken to be inversely proportional
to the average scalar dissipation rate , which can be estimated in a RANS context.
With these parameters, the evolution of , calculated through equation (8), can be
modelled with equation (16), also shown in gure 15. The agreement is quite good, but
some discrepancies are found at the beginning of combustion ( > 0) and at the end of the
propagation phase (

). The slope of the curve during the ame propagation phase is


D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
Modelling of turbulent non-premixed auto-ignition 187
Figure 16. Schematic representation of the ingredients required for the modelling approach
proposed in this work. In an actual RANS computation, the information issued from the box
two-dimensional turbulent DNS will be replaced by information from the CFD code.
well captured by the inverse dependence on . The current values for the parameters
1
, K and

are retained as they are a best t to the laminar data. The coupling between the modelled
equation for and the generalized ame surface density model is now elaborated.
4.3. Modelling of the mean heat release rate from the turbulent DNS database
The modelled equation for , equation (16), is used to describe the mean heat release rate
from the two-dimensional turbulent DNS calculations. The following ingredients are required
(see gure 16).
From the two-dimensional turbulent DNS (or the turbulent ow eld in an actual RANS
simulation): mean scalar dissipation rate as a function of time, (t ), and the mean
generalized ame surface density as a function of time,

Z
(t ).
From the one-dimensional laminar ame: generalized surface average of the heat release
rate as a function of the generalized progress variable,

T
(), and the parameters in
equation (16),
1
= 0.38 ms, K = 0.0044 and

. The last is set to unity here because


the simulation is too short for to reach its asymptotic value and this parameter has no
effect on the initial part of the prole given by equation (16).
As mentioned before, the calculations end before the ignition process was completed, but the
onset of combustion as well as the premixed peak are captured, as seen in gure 17. This
graph compares the mean value of

T
() extracted from the two-dimensional turbulent DNS
database and

T
() from the laminar model. The graph on the right shows the same variables
but as a function of time. Although the mean scalar dissipation rate in the turbulent ame is
about two times higher than in the laminar case, the mean value of the generalized surface
average of heat release rate agrees well with the laminar model. This is also shown in the
direct comparison of gure 14. The discrepancy at the onset of combustion seems inherent
to the observed error in gure 15, i.e. an ignition that happens too early. The overprediction
of the premixed peak may be due to the fact that not all the points in the mixing layer have
ignited and thus reduce the turbulent mean value of
T
(x, y) with respect to the laminar case.
The last step consists of multiplying the extracted value of

T
() from the laminar
database with the generalized ame surface density,

Z
, fromthe turbulent database, following
equation (14) and gure 16. This is shown in gure 18, where the mean value of the heat
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
188 F A Tap et al
0 0.2 0.4 0.6 0.8 1
0
0.5
1
1.5
2
x 10
6
DNS
MODEL
0 0.5 1 1.5 2 2.5 3 3.5 4
x 10
4
0
0.5
1
1.5
2
x 10
6
DNS
MODEL
g
e
n
e
r
a
l
i
z
e
d

s
u
r
f
a
c
e

a
v
e
r
a
g
e

o
f

h
e
a
t

r
e
l
e
a
s
e

r
a
t
e

(
W
m

2
)
g
e
n
e
r
a
l
i
z
e
d

s
u
r
f
a
c
e

a
v
e
r
a
g
e

o
f

h
e
a
t

r
e
l
e
a
s
e

r
a
t
e

(
W
m

2
)

time (s)
Figure 17. Left: mean value of the generalized surface average of the heat release rate,

T
,
extracted from the two-dimensional turbulent database () and

T
from the one-dimensional
laminar ame (- - - -) as a function of . Right: same variables as a function of time.
0 0.5 1 1.5 2 2.5 3 3.5 4
x 10
4
0
1
2
3
4
5
6
7
x 10
8
time (s)
h
e
a
t

r
e
l
e
a
s
e

r
a
t
e

(
W
m

3
)
DNS
MODEL
Figure 18. Temporal evolution of the mean heat release rate,


T
, for the two-dimensional turbulent
DNS calculation () and modelled with the one-dimensional laminar ame (- - - -).
release rate extracted from the two-dimensional turbulent DNS database is compared with the
computed value, using the laminar ame as a sub-model. The agreement is very satisfactory.
The model only uses global turbulent variables ( and

Z
) and laminar ame data (
1
, K and

) as input. The local ignition phenomenon in the two-dimensional turbulent case remains a
blackboxas noinformationis neededonthe instantaneous ame locationinthe mixture fraction
space or any other conditioned variable on the mixture fraction (such as scalar dissipation rate).
This contributes greatly to the simplicity of the approach.
4.4. Implementation into a RANSCFD code
The generalized progress variable, , may be described by a convectivediffusive balance
equation:

t
+ u = (D) +

, (17)
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
Modelling of turbulent non-premixed auto-ignition 189
where the source term,

, is required to ensure that evolves from 0 towards its asymptotic


value. Such a source term may be derived from equation (16) by taking its time derivative and
then using equation (16) to eliminate the time variable, t , from the formulation, yielding

2
(1 + k
1
)
_
1
_

_
1 + k
1

_
k
1
_
2
_
(18)
with
k
1
= tanh
_

2
_
. (19)
The parameters
1
,
2
and

for the hyperbolic tangent function may be estimated from


laminar, unsteady and strained diffusion ame computations.
The simplest implementation of the proposed modelling approach into a RANSCFDcode
for diesel combustion is to add the following equations to the usual set of balance equations:
Balance equations for mixture fraction and its variance, including spray source terms
(R eveillon and Vervisch 2000, Demoulin and Borghi 2002).
An algebraic closure (Jones and Whitelaw1982) or a balance equation (Mantel and Borghi
1994) for the scalar dissipation rate.
A model to link the mean scalar dissipation rate to the instantaneous value used in the
laminar ame computations. Different models can be found in Peters (2000), based on a
presumed -distribution for the mixture fraction PDF.
A balance equation for the generalized ame surface density, which could be adapted
from the original formulation by Marble and Broadwell (1977) or may be derived from
the mixture fraction balance equation (van Kalmthout and Veynante 1998, Veynante and
Vervisch 2002). Note that it is possible to relate
Z
and directly, as demonstrated by
Veynante and Vervisch (2002).
A balance equation for the generalized progress variable, which can be obtained by
averaging and closing equation (17).
A database or correlations for the parameters
1
,
2
and

, needed in equation (18),


depending at least on scalar dissipation rate and two variables to determine the
thermodynamic state of the uid, oxidizer temperature and pressure, for example.
A database of unsteady, strained, laminar diffusion ame computations, where the
generalized surface average of the heat release rate and the reaction rate for the
species of interest are stored as a function of the generalized progress variable and
two thermodynamic variables. Additional entries might be of interest as well, such as
percentage of EGR or scalar dissipation rate, as described below. If scalar dissipation rate
is not an entry in the look-up table, assuming that straining effects are accounted for by

Z
, an approach as in the MIL model (Borghi 1988) can be used to inhibit ignition in
regions where the scalar dissipation rate is higher than its critical value.
Such an implementation into a RANSCFD code is currently in progress and thus no
computations are presented in this work. The results presented by Chang et al (1996), who have
developed a similar transient amelet model using tabulated chemistry (but using a presumed
PDF approach instead of a ame surface density approach), allows condence in capturing
the main characteristics of igniting fuel jets using a transient amelet approach together with
tabulated chemistry. The main advantage of the generalized ame surface density approach
presented here as opposed to the presumed PDF method by Chang et al (1996) is that by
integrating the reaction rate across the ame, under the hypothesis of a thin ame front,
there is no need for the mean mixture fraction and its variance as coordinates in the look-up
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
190 F A Tap et al
table, reducing its dimensionality and thus the computation time and the implementation
complexity.
4.5. Further extensions
It would be of great interest to introduce the effects of the scalar dissipation rate, , in the local
mean reaction rate sub-model, writing

=
_

|Z|
_
S
=
1
|Z|
_
1
0
_
+
0
(Z, ) P(|Z) P(Z) d dZ, (20)
where P(|Z) is the conditional probability of the scalar dissipation rate, , for a given value
of the mixture fraction, Z. For a given value of , identifying the reaction rate as the one of a
one-dimensional laminar ame, extending (15), leads to

=
_
+
0
__
+

(, ) d
_
P() d =
_
+
0

() P() d, (21)
where

() is the mean reaction rate per unit of ame surface density of a one-dimensional
laminar ame submitted to the scalar dissipation rate, , and is the spatial coordinate across
the laminar ame. Of course,

and P() may also depend on time. The idea would thus be
in the future to build a library of unsteady strained laminar ames to be able to determine the
value of

=

(, ) and to use the results as a local model for the mean reaction rate per
unit surface area.
5. Conclusions and further work
A DNS database including detailed chemistry and multi-component transport phenomena of
the auto-ignition of a turbulent H
2
/O
2
/N
2
non-premixed system has been post-processed in
detail to gain insight in the interaction between chemistry, molecular diffusion and turbulent
mixing during the ignition process. A generalized ame surface density modelling approach
is proposed for the calculation of the mean reaction rate in a RANS context.
A comparison of igniting laminar diffusion ame structures with local turbulent amelet
structures, extracted from the DNS database, shows very similar behaviour during the ignition
sequence. In order to assess the preferred propagation direction of the ame during ignition,
a new parameter is introduced, cos , showing that the transfers occur mainly in the direction
perpendicular to the Z iso-levels. The degree of advancement of the ignition process is
quantied using a generalized progress variable, , which is used to compare the laminar
and turbulent data. These ndings support the use of the laminar diffusion amelet approach
for modelling the non-premixed ignition phenomenon and the use of to account for the
unsteadiness.
A generalized ame surface density formulation is proposed to model the auto-ignition
of a turbulent non-premixed ame. This approach extends the usual ame surface density
models to situations where several mixture fraction levels should be taken into account. The
mean reaction rate,

, is written as the product of the generalized ame surface density,

Z
,
by the generalized surface average of the reaction rate,

. These quantities both need to be
modelled; preliminary results show that

is mainly controlled by chemistry effects and could
be modelled from one-dimensional laminar ame calculations, whereas

Z
takes into account
the inuence of turbulence and could be described using a balance equation.
An a priori test based on the DNS database provided evidence that the mean reaction
rate per unit of generalized ame surface density,

, may be estimated from a laminar ame
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
Modelling of turbulent non-premixed auto-ignition 191
database, via the generalized progress variable, . A model for is proposed that makes
use of laminar ame parameters and mean scalar dissipation rate, , from the turbulent DNS
database.
Due to the actual limitations of DNS, in terms of kinetics and of Reynolds numbers, further
work still needs to be performed to simulate auto-ignition under conditions more representative
of those in a practical combustion system and to answer some remaining questions. In
particular, the applicability of the amelet approach for the evaluation of the mean reaction
rate per unit of generalized surface density in congurations in which the reaction zone can be
fragmented and/or for very high turbulence intensities is still an open question. Nevertheless,
results given by the reconstruction of the turbulent mean heat release rate,


T
, using the
generalized ame surface density approach and the generalized progress variable, are very
promising and allow us to be condent of the future inclusion of this approach in a RANS
CFD code for validation in an industrial conguration. An extension of the model, to account
in more detail for scalar dissipation rate effects, is proposed and will be tested in future work.
Acknowledgments
This collaboration was initiated at the EM2C laboratory and is supported nancially by PSA
Peugeot Citroen Automobiles. The support of the French national supercomputer centre IDRIS
is gratefully acknowledged.
References
Agarwal A and Assanis D 1998 Multi-dimensional modeling of natural gas ignition under compression ignition
conditions using detailed chemistry SAE Paper 980136
BaumM, Poinsot Tand Th evenin D1995 Accurate boundary conditions for multicomponent reactive ows J. Comput.
Phys. 116 24761
Bilger R W 1988 The structure of turbulent nonpremixed ames Proc. Combust. Inst. 22 47588
Borghi R 1988 Turbulent combustion modeling Prog. Energy Combust. Sci. 14 24592
Bruel P, Rogg B and Bray K 1990 On auto-ignition in laminar and turbulent non-premixed systems Proc. Combust.
Inst. 23 75966
Chang C-S, Zhang Y, Bray Kand Rogg B1996 Modelling and simulation of autoignition under simulated diesel-engine
conditions Combust. Sci. Technol. 113114 20519
Chomiak J and Karlsson A 1996 Flame lift-off in diesel sprays Proc. Combust. Inst. 26 255764
Demoulin F X and Borghi R 2002 Modeling of turbulent spray combustion with application to diesel like experiment
Combust. Flame 129 28193
Dillies B, Marx K, Dec J and Espey C 1993 Diesel engine combustion modeling using the coherent ame model in
kiva-ii SAE Paper 930074
Domingo P and Vervisch L 1996 Triple ame and partially premixed combustion in autoignition on non-premixed
turbulent mixtures Proc. Combust. Inst. 26 23340
Dopazo C and OBrien E E 1974 A probabilistic approach to the autoignition of reacting turbulent mixtures Acta
Astronaut. 1 123955
Echekki T and Chen J 1996 The effects of complex chemistry on triple ames Proc. Summer Program Center for
Turbulent Research, Stanford University/NASA-Ames, pp 21734
Echekki T and Chen J H2002 High temperature combustion in autoigniting non-homogeneous hydrogenair mixtures
Proc. Combust. Inst. 29 20618
Fichot F, Delhaye B, Veynante D and Candel S 1994 Strain rate modelling for a ame surface density equation with
application to non-premixed turbulent combustion Proc. Combust. Inst. 25 127381
Golovitchev V I, Nordin N, Jarnicki R and Chomiak J 2000 Three-dimensional diesel spray simulations using a new
detailed chemistry turbulent combustion model SAE Paper 2000-01-1891
Halstead M P, Kirch L J and Quinn C P 1977 The autoignition of hydrocarbon fuels at high temperatures and
pressurestting of a mathematical model Combust. Flame 30 4560
Haworth D, Blint R J, Cuenot B and Poinsot T J 2000 Numerical simulation of turbulent propaneair combustion with
nonhomogeneous reactants Combust. Flame 121 395417
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
192 F A Tap et al
Haworth D C, Drake M C, Pope S B and Blint R J 1988 The importance of time-dependent ame structure in stretched
laminar amelet models for turbulent jet diffusion ames Proc. Combust. Inst. 22 58997
Heywood J B 1988 Internal Combustion Engine Fundamentals (New York: McGraw-Hill)
Hilbert R 2002 Etude de la combustion turbulente non pr em elang ee et partiellement pr em elang ee par simulations
num eriques directes PhD Thesis

Ecole Centrale Paris, France
Hilbert R, Tap F, Th evenin D and Veynante D 2002 A new modeling approach for the autoignition of a nonpremixed
ame using dns Proc. Combust. Inst. 29 207985
Hilbert R and Th evenin D 2002a Autoignition of turbulent non-premixed ames investigated using direct numerical
simulations Combust. Flame 128 2237
Hilbert R and Th evenin D 2002b Turbulent non-premixed ames investigated using direct simulations with detailed
chemistry IUTAM Symp. on Turbulent Mixing and Combustion ed A Pollard and S Candel (Dordrecht: Kluwer)
pp 2518
Hilbert R, Th evenin D and Vervisch L 2001 Partially-premixed combustion during autoignition of a turbulent
nonpremixed ame Direct and Large Eddy Simulation vol IV, ed B J Geurts et al (Dordrecht: Kluwer) pp 1218
Hinze J O 1975 Turbulence in Fluids (New York: McGraw-Hill)
Hirschfelder J O, Curtiss C F and Bird R B 1954 Molecular Theory of Gases and Liquids (New York: Wiley)
Hong S, Woolridge M S and Assanis D N 2002 Modeling of chemical and mixing effects on methane autoignition
under direct injection stratied charged conditions Proc. Combust. Inst. 29 7118
Im H G, Chen J H and Law C K 1998 Ignition of hydrogen/air mixing layer in turbulent ows Proc. Combust. Inst. 27
104756
Jones W and Whitelaw J 1982 Calculation methods for reacting turbulent ows: a review Combust. Flame 48
126
Kee R J, Rupley F M and Miller J A 1989 CHEMKIN-II: a FORTRAN chemical kinetics package for the analysis of
gas-phase chemical kinetics Technical Report SAND80-8003 Sandia National Laboratories
Klimenko A Y and Bilger R W 1999 Conditional moment closure for turbulent combustion Prog. Energy Combust.
Sci. 25 595687
Kong S C, Marriot C D, Reitz R D and Christensen M 2001 Modeling and experiments of HCCI engine combustion
using detailed chemical kinetics with multidimensional CFD SAE Paper 2001-01-1026
Kong S Cand Reitz RD1993 Multidimensional modeling of diesel ignition and combustion using a multistep kinetics
model J. Eng. Gas Turbines Power 115 7819
Law C K 1988 Dynamics of stretched ames Proc. Combust. Inst. 22 1381402
Li n` an Aand Crespo A1976 An asymptotic analysis of unsteady diffusion ames for large activation energies Combust.
Sci. Tech. 14 95117
Lutz A E, Rupley M and Kee R J 1996 Equil: A CHEMKIN implementation of STANJAN, for computing chemical
equilibria Technical Report Sandia National Laboratories, http://www.ca.sandia.gov/chemkin/docs/
Maas U and Warnatz J 1988 Ignition processes in carbon-monoxidehydrogenoxygen mixtures Proc. Combust.
Inst. 22 1695704
Maas U and Warnatz J 1989 Simulation of chemically reacting ows in two-dimensional geometries Impact Comput.
Sci. Eng. 1 394420
Magnussen B F and Hjertager B H 1976 On mathematical modeling of turbulent combustion with special emphasis
on soot formation and combustion Proc. Combust. Inst. 16 719
Mantel T and Borghi R 1994 A new model of premixed wrinkled ame propagation based on a scalar dissipation
equation Combust. Flame 96 44357
Marble F E and Broadwell J E 1977 The coherent ame model of non-premixed turbulent combustion Technical
Report Project Squid TRW-9-PU Purdue University
Mastorakos E, Baritaud T A, Cuenot B and Poinsot T J 1995 Numerical simulations of autoignition in turbulent ows
with non-premixed reactants 10th Symp. on Turbulent Shear Flows vol 2, The Pennsylvania State University,
pp 2530
Mastorakos E, Baritaud T A and Poinsot T J 1997 Numerical simulations of autoignition in turbulent mixing ows
Combust. Flame 109 198223
Mastorakos E and Bilger R W 1998 Second-order conditional moment closure for the autoignition of turbulent ows
Phys. Fluids 10 12468
Mastorakos E, da Cruz A P, Baritaud T A and Poinsot T J 1997 A model for the effects of mixing on the autoignition
of turbulent ows Combust. Sci. Technol. 125 24382
Musculus M and Rutland C 1995 Coherent ame modeling of diesel engine combustion Combust. Sci. Technol. 104
295337
Natarajan B and Bracco F 1984 On multidimensional modelling of auto-ignition in spark-ignition engines Combust.
Flame 57 17997
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0
Modelling of turbulent non-premixed auto-ignition 193
Peters N 1984 Laminar diffusion amelet models in non-premixed turbulent combustion Prog. Energy Combust.
Sci. 10 31939
Peters N 2000 Turbulent Combustion (Cambridge: Cambridge University Press)
Pinchon P 1989 Three dimensional modelling of combustion in a prechamber diesel engine SAE Paper 890666
Pires Da Cruz A, Baritaud T A and Poinsot T J 2001 Self-ignition and combustion modeling of initially nonpremixed
turbulent systems Combust. Flame 124 6581
Pitsch Hand Peters N1998 Investigation of the ignition process of sprays under diesel engine conditions using reduced
n-heptane chemistry SAE Paper 982464
Pitsch H, Wan Y P and Peters N 1995 Numerical investigation of soot formation and oxidation under diesel engine
conditions SAE Paper 952357
Poinsot T 1996 Using direct numerical simulations to understand premixed turbulent combustion Proc. Combust.
Inst. 26 21932
Poinsot T, Candel S and Trouv e A1996 Applications of direct numerical simulation to premixed turbulent combustion
Prog. Energy Combust. Sci. 21 53176
Poinsot T and Lele S 1992 Boundary conditions for direct simulations of compressible viscous ows J. Comput.
Phys. 101 10429
Poinsot T and Veynante D 2001 Theoretical and Numerical Combustion ed R T Edwards
Poinsot T, Veynante D, Trouv e A and Ruetsch G 1996 Turbulent ame propagation in partially premixed ames Proc.
Summer Program Center for Turbulent Research, Stanford University/NASA-Ames, pp 11141
Pope S B 1985 Pdf methods for turbulent reacting ows Prog. Energy Combust. Sci. 19 11992
Pope S B 1990 Computations of turbulent combustion: progresses and challenges Proc. Combust. Inst. 23 591612
R eveillon J and Vervisch L 2000 Spray vaporization in nonpremixed turbulent combustion modeling: a single droplet
model Combust. Flame 121 7590
Sh apert ons H and Lee W 1985 Multidimensional modelling of knocking combustion in SI engines SAE Paper 850502
Sreedhara S and Lakshmisha K N 2000 Direct numerical simulation of autoignition in a non-premixed, turbulent
medium Proc. Combust. Inst. 28 2534
Sreedhara S and Lakshmisha K N 2002a Assessment of conditional moment closure models of turbulent autoignition
using DNS data Proc. Combust. Inst. 29 206977
Sreedhara S and Lakshmisha K N 2002b Autoignition in a non-premixed medium: DNS studies on the effects of
3-dimensional turbulence Proc. Combust. Inst. 29 20519
Takagi T and Xu Z 1994 Numerical analysis of laminar diffusion ames-effects of preferential diffusion of heat and
species Combust. Flame 96 509
Theobald M A and Cheng W K 1987 A numerical study of diesel ignition ASME Paper 87-FE-2
Th evenin D, Behrendt F, Maas U, Przywara B and Warnatz J 1996 Development of a parallel direct simulation code
to investigate reactive ows Comput. Fluids 25 48596
Th evenin D, van Kalmthout E and Candel S 1997 Two-dimensional direct numerical simulations of turbulent diffusion
ames using detailed chemistry Direct and Large Eddy Simulation vol II, ed J Chollet et al (Dordrecht: Kluwer)
pp 34354
van Kalmthout E and Veynante D 1998 Direct numerical simulation analysis of ame surface density models for non
premixed turbulent combustion Phys. Fluids 10 234768
van Kalmthout E, Veynante D and Candel S 1996 Direct numerical simulation analysis of ame surface density
equation in non-premixed turbulent combustion Proc. Combust. Inst. 26
Vervisch Land Poinsot T1998 Direct numerical simulation of non premixed turbulent ames Ann. Rev. Fluid Mech. 30
65592
Veynante D, Lacas F and Candel S 1991 Numerical simulation of the transient ignition regime of a turbulent diffusion
ame AIAA J. 29 84851
Veynante D and Vervisch L 2002 Turbulent combustion modeling Prog. Energy Combust. Sci. 28 193266
Williams F A 1985 Combustion Theory 2nd edn (Reading, MA: Addison-Wesley)
Zellat M, Rolland T and Poplow F 1990 Three-dimensional modeling of combustion and soot formation in an indirect
injection diesel engine SAE Paper 900254
Zhang Y, Rogg B and Bray K 1995 Two-dimensional simulation of turbulent autoignition with transient laminar
amelet source term closure Combust. Sci. Technol. 105 21127
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t
e
i
t

G
e
n
t
]

A
t
:

1
6
:
3
6

2
5

J
a
n
u
a
r
y

2
0
1
0

Das könnte Ihnen auch gefallen