Center for Turbulence Research

Annual Research Briefs 2010
275
Large-eddy simulation of stratified combustion
By A. Roux AND H. Pitsch
1. Motivation and objective
Combustion under stratified conditions appears in many systems: the simplest example
is unconfined premixed flames, where fuel mixes with air at the outer shear layer leading
to a reduced equivalence ratio and flame extinction (Boehm et al. 2010), but a more
practical example is gas turbines where evaporating fuel may have not enough time to
be perfectly premixed with air before burning. Stratified burning may occur in a variety
of forms, and the most basic condition used to characterize these forms is the presence
of stoichiometric fuel and air mixtures. If stoichiometric conditions can be observed in a
flow, such as in the classical canonical edge flames (Favier & Vervisch 2001; Domingo et al.
2002), the tribrachial flame structure gathers both premixed and non-premixed regimes.
On the other hand, both the lean-lean combustion, used to reduce NO
X
emissions (Besson
et al. 2000; Zhao et al. 1999), and rich-rich (Barlow et al. 2009) stratification conditions
remain in the premixed regime but might show deviations from a classical flamelet-type
description.
Direct Numerical Simulations (DNS) and Large-Eddy Simulations (LES) have stressed
the changes in heat release and burning velocity occur for instance when a flame prop-
agates through a non-homogeneous mixture. Although current models account for these
effects in a basic way, several questions can be raised regarding whether the accuracy
of current approaches is sufficient (Haworth et al. 2000; Jimenez et al. 2002; Pires Da
Cruz et al. 2000). Anselmo-Filho et al. (2009) have shown that stratification increases
the flame surface density in comparison to a perfectly premixed case. Despite these dif-
ferences and the broad range of applications, until recently, the focus has been on only
canonical perfectly premixed or non-premixed conditions where prior to reactions, re-
actants are either perfectly premixed or completely segregated. Experimentally, only a
few experiments have been dedicated to study stratification effects (Besson et al. 2000;
Anselmo-Filho et al. 2009; Seffrin et al. 2010; Boehm et al. 2010) and validate numerical
models used to simulate turbulent combustion.
The objective of this study is to assess how correctly the multi-regime approach de-
veloped by Knudsen et al. (2010) and, particularly, the G-equation model coupled with a
transported progress variable, capture stratification effects. The focus here will be on ve-
locity and temperature. The chosen burner is relevant for aircraft combustors in that new
low-pollutant engine designs increasingly call for some premixed burning supported by a
more stable pilot flame. In the present work, two different simulations will be compared,
a stratified case and the corresponding fully premixed one with no shear. The paper is
organized as follows. Section 2 presents the configuration of interest, Sec. 3 recalls the
model used to describe premixed combustion and finally, results are analyzed in Sec. 4.
2. Experimental setup
The configuration of interest corresponds to the Darmstadt experimental stratified
burner (TSF) studied by Seffrin et al. (2010) and Boehm et al. (2010). Figure 1 presents
276 A. Roux and H. Pitsch
Figure 1. Simulation setup of the TSF burner. View of the four injection streams.
Figure 2. Central premixed pilot tube with the flame-holder: its position is adjusted so that
the exit velocity profile is as flat as possible and close to the bulk velocity of slot 1 to minimize
shear.
the computational domain. The TSF burner consists of three staged concentric tubes
surrounded by a low-velocity coflow. The inner diameters are 14.8, 37 and 60 mm, re-
spectively. The central tube is used as a premixed pilot. The methane-air flame that
burns in the main combustion chamber is stabilized by a flame-holder positioned 40 mm
upstream of the pilot exit as shown in Fig. 2.
The two outer annular slots are operated independantly, allowing for the investigation
of stratification and/or shear effects. Four specific cases with no shear are studied in the
present work, summarized in Table 1. The first case A i2 corresponds to an iso-thermal
case and is used to validate whether velocity fluctuations are correctly reproduced by
the inflow simulation. The second case, labeled A i1, is non-reacting in the sense that
only the pilot is burning. This case is used to validate the inlet profile used for the pilot
inlet stream. The G r is a fully premixed case where both the bulk velocities and the
equivalence ratios of pipes 1 and 2 are identical. Finally, the equivalence ratio of pipe 2 is
decreased to 0.6 for the case A r to investigate stratification effects. Note that the bulk
velocity given for the pilot corresponds to a velocity taken upstream of the flame-holder,
in unburnt gas. The corresponding bulk velocity at the exit of the pilot is 10 m/s.
Pictures of the mean flame brush studied by Seffrin et al. (2010) are presented in Fig. 3.
The two configurations of interest, A and G, exhibit different behavior: the flame opening
angles differs. Both flames qualitatively have the same appearance at the beginning until
the flame reaches the zone where the mixture fraction is leaner for case A. The angle is
then reduced for the stratified case because of the lower flame speed.
LES of stratified combustion 277
Name Pilot Slot 1 Slot 2 Coflow
U
bulk
φ U
bulk
φ U
bulk
φ U
bulk
φ
A i2 10 m/s 0.0 10 m/s 0.0 10 m/s 0.0 0.1 m/s 0.0
A i1 1 m/s 0.9 10 m/s 0.0 10 m/s 0.0 0.1 m/s 0.0
G r 1 m/s 0.9 10 m/s 0.9 10 m/s 0.9 0.1 m/s 0.0
A r 1 m/s 0.9 10 m/s 0.9 10 m/s 0.6 0.1 m/s 0.0
Table 1. List of flow conditions simulated by LES of the three investigated cases: bulk velocity
and equivalence ratio. Note that the bulk velocity in the pilot is given at its exit and thus may
correspond to burnt gases.
Figure 3. Photographs of the two studied configurations, from Seffrin et al. (2010). Left:
stratified flame and right: corresponding premixed one.
3. Physical and numerical model
3.1. Numerical setup
The configuration is simulated using a structured finite-difference code (Desjardins et al.
2008) with a low Mach number limit and is second order accurate in both space and
time. A third-order accurate WENO scheme is used for scalar advection. The temporal
integration is done using a semi-implicit Crank-Nicolson scheme. A structured grid in
cylindrical coordinates with 408 × 168 × 64 points in the axial, radial and azimuthal
directions, respectively, is used. The mesh is refined in the zone of interest: the cell width
is ∆
x
=1 mm close to the injector (axial location between the exit of the pilot and 200
mm downstream) and is then stretched. The grid is thus composed of 4.4 ×10
6
points.
For practical reasons, the entire set-up is computed by LES using multiple precursor
simulations. First, three turbulent pipe velocity profiles are generated corresponding to
the three slots of the TSF burner. Two of them feed the two outer pipes 1 and 2. The
last profile is used as an inflow condition to generate the velocity profiles at the exit of
278 A. Roux and H. Pitsch
the central pilot tube downstream of the flame-holder where the fine geometric details
cannot be included in the main simulation without increasing the size of the grid. The
domain that is used for the main combustor LES begins 5 mm before the exit of slot 2.
This allows the inflow profiles to adapt to the mesh.
3.2. Combustion modeling
In the thin reaction zone regime of premixed combustion, the premixed flame front is
smaller than the grid size. Depicting chemistry using a transported progress variable ¯c
consequently leads to numerical difficulties at the flame front where the progress vari-
able profile becomes very under-resolved. Additional modeling is therefore needed to
accurately describe the flame propagation speed. This modeling is introduced by solving
an additional equation for a level set function G, defining the evolution of an interface
associated with the flame front as used by Knudsen et al. (2010):

ˆ
G
∂t
+ ¯ v∇
ˆ
G = −
ρ
u
¯ ρ
Vn · ∇
ˆ
G, (3.1)
where t is the time, ¯ v the Favre-filtered velocity, and V the turbulent propagation speed
of a temperature iso-surface in the flame that includes the laminar burning velocity and
the front propagation caused by diffusion in high curvature regions.
The combustion model couples the level set function and a classical transported progress
variable ¯c by evaluating a weighted progress variable such as
¯ c
M
= ¯ c +ξ
_
ˆ
G, ∆, ∆

_
×
_
¯ c
G
_
ˆ
G, ∆
_
− ¯ c
_
. (3.2)
This mixed expression is a model describing the progress variable field in the vicinity
of the flame through the use of ¯ c
G
, and the field outside of this region using ¯c. The weight
ξ reads
ξ
_
ˆ
G, ∆, ∆

_
=
(∆

)
1/4
_
(∆

)
2
+ ∆
2
_
1/8
×exp
_
_
_−6
_
ˆ
G −G
0
_
2
_
(∆

)
2
+ ∆
2
_
1/8
_
_
_. (3.3)
G
0
stands for the value of the level set function in the inner layer of the flame. It is
here assumed that the G = G
0
iso-surface corresponds to half of the progress variable
maximum value, although it is known that the heat release maximum value occurs for
a higher value of the progress variable. ∆

stands for the width where the level-set is
trusted in the simulation. c
G
has to be independent from these two variable as much as
possible. The model for this scalar that describes how the progress variable depends on
a distance coordinate in an unfiltered sense is defined as
c
G
:= c|G = G
0
=
_
¸
¸
_
¸
¸
_
0 ∀G < −δ
l
cmax
δ
l
×G +
cmax
2
∀ −
δ
l
2
< G ≤ G
2
c3−c2
G3−G2
×(G −G
3
) +c
3
∀G
2
< G ≤ G
3
c
max
∀G > G
3
, (3.4)
where
LES of stratified combustion 279
c
max
:= c|
G=∞
G
2
:=
δ
l
2
×
_
c|
G=δ
l
/2

cmax
2
_
C
2
:=
cmax
G2+δ
l
/2
×G
2
+
cmax
2
G
3
:= G|
c=c3
(3.5)
δ
l
is the flame width. The source term reads

˙ ω
C
(¯c
M
) =
_
˙ ω
C
(c).δ(c −¯c
M
)dc. (3.6)
Finally, this procedure ensures that the resolved part of the turbulent mixing is well-
described througout the use of the transported progress variable while the tracking of
the flame is ensured at the right speed.
3.3. Heat loss modeling
The effects of radiative as well as wall convective heat transfers are of foremost im-
portance. Different levels of approximation can be used to consider radiation. Thermal
radiation effects can be directly integrated in the combustion model such as in the FPV
model for non-premixed combustion developed by Ihme (2008). Other methods of treat-
ing radiation include approaches such as Discrete Ordinate Methods (DOM), ray tracking
methods or Monte Carlo simulations in a separate code. Note that coupling these meth-
ods to LES solvers is typically very computationally expensive (Desjardin & Frankel
1999; Gonalves dos Santos et al. 2008). A more cost-effective approximation is to use a
thin gray gas model for radiation and to assume that each point has a direct view of
the cold surroundings (Barlow et al. 2001; Schmitt et al. 2007; Ihme 2008). Here this
optically thin approach is employed, and radiative heat transfer is accounted for by a
Stefan-Boltzmann law which depends on local gas composition.
Convective heat transfer is accounted for by a thermal wall-function boundary con-
dition imposing a prescribed temperature. The temperature in the walls of the main
injection port are set to be 300 K so that LES gas temperature matches experimental
measurements at the nozzle exit.
Note that coupling heat transfer with a look-up table would require an additional
dimension: i.e., tabulating along another direction, the enthalpy (Ribert et al. 2006).
However, the size of the different look-up tables is a major issue due to the lack of
memory, especially when dealing with a multi-regime model requiring the use of three
different tables. Different methods such as the ones using the self-similar properties of
turbulent premixed flames have recently emerged but require additional assumptions.
Another way to reduce the computational cost, the result of the dimension of the look-
up tables, has been studied by Singer & Pope (2004) by creating in-situ chemical tables,
i.e., tables that are optimized and extended during the computation. This is the so-called
In-Situ Adaptive Tabulation method (ISAT). Use of neural networks are another way to
improve the size of the look-up tables (Ihme 2008). In this work, another method is used
to lighten the table (Knudsen et al. 2010).
The enthalpy equation is transported but is not a new dimension of the table to spare
memory. Instead, it is here assumed that the progress variable source term ˙ ω
c
can be
modeled such as
˙ ω
c
(z, c) ≈ B(z). exp(−T
a
(z)/T(z, c)), (3.7)
280 A. Roux and H. Pitsch
(a) (b)
Figure 4. Correction of the source term to take into account enthalpy change as defined by
Eq. 3.8. Right: φ = 0.6 and Left: φ = 1.0.
where B(z) is independent of the temperature. Note that this evolution is comparable
to the source term from a one-step chemistry approximation. T
a
is extracted from the
computation of the flamelets and tabulated. An enthalpy scalar is solver for, and can be
then directly used to correct the tabulated source term. This assumed functional form of
the source term can be used to modify a source term from a tabulated chemistry database
in the following manner. First, temperature is calculated from the transported value by
assuming that gas composition is not changed by transfer. A corrected temperature that
matches the value of the transported enthalpy can thus be estimated (Cp are evaluated
using polynomial functions). Then, density is computed by correcting the tabulated one
with the ratio of the exponential functions (corrected over transported). Finally, the
corrected source term reads
˙ ω
corr
c
(z, c) = ˙ ω
tab
c
(z, c) ×
exp
_
−T
tab
a
/T
corr
_
exp (−T
tab
a
/T
tab
)
, (3.8)
where the superscript tab means that the quantity comes from the look-up table. Fig-
ure 4 shows the error introduced using this alternate way to describe heat losses. Three
different flamelets are computed using FlameMaster (Pitsch 1998) with the same mixture
fraction φ = 0.6 or φ = 1.0 but three different enthalpies (assumed to remain constant
throughout the flamelet). The enthalpy changes are induced by decreasing or increasing
the temperature of the cold gases. The enthalpy is either lower (case A), or higher (case
B) than T
0
, with a corresponding inlet temperature of 240, 400 and 320 K, respectively.
The black line T
0
is used as a reference state to tabulate the different quantities of inter-
est. Then, the model from Eq. 3.8 is used to compute the source term at the two other
enthalpies using the temperature of the reference flamelet. The result from FlameMaster
is denoted by the subscript FM and the corrected variable by MODEL. A good agree-
ment is found although the source term peak is slightly shifted toward the burned gases
for lower equivalence ratio.
The laminar burning velocity s
L
appears in the source term V of the G-equation,
Eq. 3.1. It also has to be corrected to take into account heat loss effects and reproducing
the flame lift-off. This correction is performed by noting that s
L
is proportional to the
square root of the pre-exponential factor B of Eq. 3.7. By considering that the correction
of the ratio of the exponential is included in the function B(z), the corrected flame speed
reads
LES of stratified combustion 281
Figure 5. Correction of the flame speed to take into account enthalpy change. Laminar flame
speed versus equivalence ratio for five different initial temperatures (Representing different en-
thalpies). The flamelets described by triangles are used to build the look-up tables. The four oth-
ers are constructed by changing the inlet temperature and keeping the composition unchanged.
Filled symbol: corrected speed from Eq. 3.9, and empty ones, FlameMaster.
s
corr
L
= s
tab
L
×
_
exp
_
−T
tab
a
/T
corr
_
exp (−T
tab
a
/T
tab
)
_
1/2
. (3.9)
Figure 5 shows the evolution of the laminar flame speed versus the equivalence ratio
for five different enthalpies coming from FlameMaster (empty symbols) and the corre-
sponding approximated value (filled symbols and line). Differences arise for high enthalpy
flamelets. However, for the lower enthalpies that are considered in this work, a maximum
error of 15% is found and is considered to be acceptable.
3.4. Chemistry modeling
The progress variable is here defined as the sum of CO, CO
2
, H
2
O and H
2
mass fractions.
The fuel is methane. A flamelet database is generated using FlameMaster by representing
the flame as an unstretched freely propagating premixed flame with non-unity Lewis
numbers. The look-up table has three leading dimensions: the mixture fraction z, the
progress variable c, and the variance z
2
. It is discretized using 200 ×400 ×25 points.
4. Results
In this section, simulation results are presented and assessed against the experiment.
All time-averaged statistics were collected for a period of four flow through times, based
on a bulk velocity of 10 m/s.
4.1. Non-reacting comparisons
Simulation of the isothermal case A i2 and the quasi non-reacting case A i1 are first
performed to validate the applied inflow boundary conditions. Time-averaged velocity
statistics are shown in Fig. 6 where mean (a) and variance (b) values of the axial com-
ponent of the velocity vector are given. A very good agreement is found between the
282 A. Roux and H. Pitsch
40
30
20
10
0
R
a
d
i
u
s

[
m
m
]
10 5 0
Seffrin et al :
A_i1 A_i2
NGA :
A_i1 A_i2
40
30
20
10
0
R
a
d
i
u
s

[
m
m
]
4 3 2 1 0
Seffrin et al :
A_i1 A_i2
NGA :
A_i1 A_i2
(a) (b)
Figure 6. Mean and variance of axial velocity 1 mm downstream the exit of the pilot for cases
A i1 and A i2.
LES data and the experiment. The simulation captures the velocity fluctuations accu-
rately. However, discrepancies arise at the shear layers between the inflow pipes where
more points could have been used to discretize the lips of the different slots. Another
difference arises in the pilot for the A i2 case on the mean axial velocity: asymmetries
in the pilot flow appear owing to a slight misalignment of the moveable flame holder
in the experiments. This issue does not appear for the other reacting cases according to
the experimentalists. Indeed, hot gases have higher temperature and increased diffusivity
that smoothes the asymmetries. Experimental data sets are available for further axial
locations of z = 50 and z = 100 mm but are not given here. Very good agreement is also
found at these locations.
4.2. Reacting comparisons
Having validated the approach for the non-reacting cases, the LES is extended to the
reacting cases A r and G r. Figure 7 shows an instantaneous iso-surface of progress
variable Y
C
= 0.1. Two different behaviors are seen; as the flow moves downstream, the
hot gases expand radially faster for the G R case than for the stratified case. Figure 8
shows a detailed view of mixture fraction and temperature for the A r case close to the
injector in the region of interest. The maximum temperature is found to be 2090 K which
is 60 K below the adiabatic flame temperature for an equivalence ratio of 0.9. For the G r
case, pockets of hot gas are expelled from the flame core because of Kelvin-Helmholtz
instability at the outer shear layer where a high velocity shear is found. Contours of
mixture fraction allow localizing stratified regions. These show the flame propagating
in regions of gradually leaner mixture fraction from the inner to the outer part of the
jet. Two different zones can be distinguished. Between the pilot exit and around 30 mm
downstream, the shape is similar for both cases as the flame is not yet burning toward
the leaner mixture coming from slot 2. This region particularly shows that the lift-off
is the same for both cases. Farther downstream, two different opening flame angles are
seen. The A r flame angle is reduced compared to that in the fully-premixed case, as seen
in Fig. 3. Interestingly, the gradient of temperature locating the flame front is mostly
perpendicular to the gradient of mixture fraction. However, at some points, a parallel
LES of stratified combustion 283
Figure 7. Instantaneous isosurface of progress variable YC = 10. Top half of image: A r case,
and, bottom half of image: G r case.
Figure 8. Instantaneous fields of iso-levels of mixture fraction and contours of temperature.
Left: G r case, and, right: A r case.
alignment appears that leads to a decreased heat release and temperature because of the
leaner mixture fraction met.
For a more quantitative validation, the differences between the two cases are now
studied and assessed against the experiment. Figures 9 to 12 show the evolution of the
mean and variance of axial and radial components of the velocity vector as a function
of the radial coordinate at six axial locations: z = 1 mm, z = 50 mm, z = 75 mm,
z = 100 mm, z = 125 mm and z = 200 mm. An overall good agreement is found on the
axial velocity. Note that at the outer shear layer, some discrepancies may exist because
of the lack of convergence of the results. Because the coflow velocity (0.1 m/s) is a
hundred times lower than the bulk velocity of the three different pipes, i.e., U
bulk
= 10
m/s, statistics in the outer shear layers converge very slowly relative to the primary jets
statistics. The statistics have been acquired during four flow through times where the
characteristic time-scale is based on the bulk velocity of these three pipes and not on
that of the coflow. Having a fully converged simulation is not adressed in this study.
A bump structure is seen until z = 50 mm because of the segregated inflow. Then the
presence of the flame and the increased viscosity at high temperature smooth out this
effect. The opening of the flame in the radial direction seen in the radial velocity profiles
comes from the gas expansion. Because of non-adiabaticity effects in the pilot, the flame
284 A. Roux and H. Pitsch
Figure 9. Mean axial velocity profiles in [m.s
−1
] at six different axial locations. Comparison
between computation (lines) and experiment (symbols). Top: G r, and, bottom: A r cases.
Figure 10. Variance of axial velocity profiles in [m.s
−1
] at six different axial locations. Com-
parison between computation (lines) and experiment (symbols). Top: G r, and, bottom: A r
cases.
is detached from the exit of the pilot and the radial velocity profiles show a fairly good
prediction of its position in the numerical simulation. The levels of variance are also well
matched. However, a longer simulation seems to be required to have smoother second-
order statistics.
LES of stratified combustion 285
Figure 11. Mean radial velocity profiles in [m.s
−1
] at six different axial locations. Comparison
between computation (lines) and experiment (symbols). Top: G r, and, bottom: A r cases.
Figure 12. Variance of radial velocity profiles in [m.s
−1
] at six different axial locations. Com-
parison between computation (lines) and experiment (symbols). Top: G r, and, bottom: A r
cases.
Temperature predictions are plotted in Figs. 13 and 14. Experimental data sets are
taken from Boehm et al. (2010). The temperature in the pilot (r < 7.4 mm) is below
the adiabatic flame temperature at around 1800 K. Farther downstream, additional heat
release increases the local temperature and forms a bump at the location of the flame
286 A. Roux and H. Pitsch
Figure 13. Mean temperature profiles in [K] at six different axial locations. Comparison
between computation (lines) and experiment (symbols). Top: G r, and, bottom: A r cases.
brush. At the two final downstream locations, increased mean temperature is found for
high r because large flame cusps appear during the propagation of the flame. Regarding
the variance of temperature, a good agreement is found for the magnitude, but the
profiles are slightly shifted toward the inner part of the jet. This could be explained by
the modeled turbulent flame speed, which is responsible for the radial expansion of the
hot gases, being predicted as too small.
5. Conclusions and future work
In this work, the multi-regime flamelet approach was used to perform Large-Eddy
Simulation (LES) of a stratified burner. This study focuses on understanding the limits
of the models used when dealing with stratification. An experimental burner that has
been designed to look at lean premixed stratified combustion is used to validate the
model. Two different reactive cases have been simulated, one a stratified flame and an
equivalent fully premixed one. An overall good agreement is found for the time-averaged
velocity statistics and temperature.
The immediate future work includes NO
x
modeling. Indeed, stratification is a com-
monly used way to reduce NO
x
emissions and comparing profiles of pollutants at the
exit of the burner for different stratified/fully-premixed case would be of great interest.
Acknowledgments
Authors thank Dr. Ing. F. Seffrin and B. Boehm for providing the experimental data.
Support from CTR and funding from SAFRAN is acknowledged. Fruitful discussion with
Dr. Ed. Knudsen, Mr. C. Schmitt and Dr. T. Lederlin are gratefully acknowledged.
LES of stratified combustion 287
Figure 14. Variance of temperature profiles in [K] at six different axial locations. Comparison
between computation (lines) and experiment (symbols). Top: G r, and, bottom: A r cases.
REFERENCES
Anselmo-Filho, P., Hochgreb, S., Barlow, R. & Cant, R. 2009 Experimental
measurements of geometric properties of turbulent stratified flames. Proc. Combust.
Inst. 32 (2), 1763 – 1770.
Barlow, R., Wang, G.-H., Anselmo-Filho, P., Sweeney, M. & Hochgreb, S.
2009 Application of raman/rayleigh/LIF diagnostics in turbulent stratified flames.
Proc. Combust. Inst. 32 (1), 945 – 953.
Barlow, R. S., Karpetis, A. N., Frank, J. H. & Chen, J. Y. 2001 Scalar profiles
and no formation in laminar opposed-flow partially premixed methane/air flames.
Comb. Flame 127 (3), 2102 – 2118.
Besson, M., Bruel, P., Champion, J. & Deshaies, B. 2000 Experimental analysis
of combustion flows developping over a plane-symmetric expansion. J. Thermophys.
Heat Transfer 14 (1), 59 – 67.
Boehm, B., Frank, J. & Dreizler, A. 2010 Temperature and mixing field measure-
ments in stratified lean premixed turbulent flames. Proc. Combust. Inst. In Press,
Corrected Proof, –.
Desjardin, P. E. & Frankel, S. H. 1999 Two-dimensional large eddy simulation of
soot formation in the near-field of a strongly radiating nonpremixed acetylene-air
turbulent jet flame. Comb. Flame 119 (1-2), 121 – 132.
Desjardins, O., Blanquart, G., Balarac, G. & Pitsch, H. 2008 High order con-
servative finite difference scheme for variable density low mach number turbulent
flows. J. Comput. Phys. 227 (15), 7125–7159.
Domingo, P., Vervisch, L. & Bray, K. 2002 Partially premixed flamelets in LES of
nonpremixed turbulent combustion. Combust. Theory Model. 6 (4), 529 – 551.
288 A. Roux and H. Pitsch
Favier, V. & Vervisch, L. 2001 Edge flames and partially premixed combustion in
diffusion flame quenching. Comb. Flame 125 (1-2), 788 – 803.
Gonalves dos Santos, R., Lecanu, M., Ducruix, S., Gicquel, O., Iacona, E. &
Veynante, D. 2008 Coupled large eddy simulations of turbulent combustion and
radiative heat transfer. Comb. Flame 152 (3), 387 – 400.
Haworth, D. C., Blint, R. J., Cuenot, B. & Poinsot, T. J. 2000 Numerical
simulation of turbulent propane-air combustion with nonhomogeneous reactants.
Comb. Flame 121 (3), 395 – 417.
Ihme, M., P. H. 2008 Modeling of radiation and nitric oxide formation in turbulent
nonpremixed flames using a flamelet/progress variable formulation. Phys. Fluids
20 (5).
Jimenez, C., Cuenot, B., Poinsot, T. & Haworth, D. 2002 Numerical simulation
and modeling for lean stratified propane-air flames. Comb. Flame 128 (1-2), 1 – 21.
Knudsen, E., Kim, S. H. & Pitsch, H. 2010 An analysis of premixed flamelet models
for large eddy simulation of turbulent combustion. Phys. Fluid In press.
Pires Da Cruz, A., Dean, A. & Grenda, J. 2000 A numerical study of the laminar
flame speed of stratified methane/air flames. Proc. Combust. Inst. 28 (2), 1925 –
1932.
Pitsch, H. 1998 Flamemaster, a c++ computer program for 0d combustion and 1d
laminar flame calculations Available from http://www.stanford.edu/ hpitsch/.
Ribert, G., Gicquel, O., Darabiha, N. & Veynante, D. 2006 Tabulation of com-
plex chemistry based on self-similar behavior of laminar premixed flames. Comb.
Flame 146 (4), 649 – 664.
Schmitt, P., Poinsot, T., Scghuermans, B. & Geigle, K. 2007 Large-eddy simula-
tion and experimental study of heat transfer, nitric oxide emissions and combustion
instability in a swirled turbulent high-pressure burner. J. Fluid Mech. 570, 17–46.
Seffrin, F., Fuest, F., Geyer, D. & Dreizler, A. 2010 Flow field studies of a new
series of turbulent premixed stratified flames. Comb. Flame 157 (2), 384 – 396.
Singer, M. & Pope, S. 2004 Exploiting isat to solve the reaction-diffusion equation.
Combust. Theory Model. 8 (2), 361 – 383.
Zhao, F., Lai, M. C. & Harrington, D. L. 1999 Automotive spark-ignited direct-
injection gasoline engines. Prog. Energy Combust. Sci. 25 (5), 437 – 562.

The two outer annular slots are operated independantly. The second case. Both flames qualitatively have the same appearance at the beginning until the flame reaches the zone where the mixture fraction is leaner for case A. The inner diameters are 14. A and G.8. Figure 2. The TSF burner consists of three staged concentric tubes surrounded by a low-velocity coflow. The angle is then reduced for the stratified case because of the lower flame speed. The corresponding bulk velocity at the exit of the pilot is 10 m/s. Four specific cases with no shear are studied in the present work. labeled A i1. Pitsch Figure 1. . 3. The first case A i2 corresponds to an iso-thermal case and is used to validate whether velocity fluctuations are correctly reproduced by the inflow simulation. respectively. Note that the bulk velocity given for the pilot corresponds to a velocity taken upstream of the flame-holder. Finally. in unburnt gas. summarized in Table 1. This case is used to validate the inlet profile used for the pilot inlet stream. The G r is a fully premixed case where both the bulk velocities and the equivalence ratios of pipes 1 and 2 are identical. the equivalence ratio of pipe 2 is decreased to 0. The methane-air flame that burns in the main combustion chamber is stabilized by a flame-holder positioned 40 mm upstream of the pilot exit as shown in Fig.276 A. View of the four injection streams.6 for the case A r to investigate stratification effects. Simulation setup of the TSF burner. The two configurations of interest. Central premixed pilot tube with the flame-holder: its position is adjusted so that the exit velocity profile is as flat as possible and close to the bulk velocity of slot 1 to minimize shear. the computational domain. exhibit different behavior: the flame opening angles differs. Roux and H. 2. Pictures of the mean flame brush studied by Seffrin et al. is non-reacting in the sense that only the pilot is burning. allowing for the investigation of stratification and/or shear effects. 37 and 60 mm. (2010) are presented in Fig. The central tube is used as a premixed pilot.

is used. For practical reasons. three turbulent pipe velocity profiles are generated corresponding to the three slots of the TSF burner.1 m/s m/s m/s m/s 0. Figure 3.0 0.1.9 10 G r 1 m/s 0. 2008) with a low Mach number limit and is second order accurate in both space and time. Left: stratified flame and right: corresponding premixed one. Two of them feed the two outer pipes 1 and 2.9 10 Table 1.9 10 A r 1 m/s 0.9 Slot 2 Ubulk φ 10 10 10 10 m/s m/s m/s m/s 0. radial and azimuthal directions.9 0.9 0. Note that the bulk velocity in the pilot is given at its exit and thus may correspond to burnt gases. Numerical setup The configuration is simulated using a structured finite-difference code (Desjardins et al.1 0.0 0. Photographs of the two studied configurations. The last profile is used as an inflow condition to generate the velocity profiles at the exit of .0 0.1 0. The grid is thus composed of 4.0 0. First.4 × 106 points. List of flow conditions simulated by LES of the three investigated cases: bulk velocity and equivalence ratio.0 10 A i1 1 m/s 0. the entire set-up is computed by LES using multiple precursor simulations. from Seffrin et al.0 0. respectively.0 0.0 A i2 10 m/s 0. (2010). The temporal integration is done using a semi-implicit Crank-Nicolson scheme. The mesh is refined in the zone of interest: the cell width is ∆x =1 mm close to the injector (axial location between the exit of the pilot and 200 mm downstream) and is then stretched.6 Coflow Ubulk φ 0.0 0. 3. A third-order accurate WENO scheme is used for scalar advection. A structured grid in cylindrical coordinates with 408 × 168 × 64 points in the axial. Physical and numerical model 3.LES of stratified combustion 277 Name Pilot Ubulk φ Slot 1 Ubulk φ m/s m/s m/s m/s 0.1 0.

1) where t is the time. The model for this scalar that describes how the progress variable depends on a distance coordinate in an unfiltered sense is defined as   0  cmax  cmax δl × G + 2 = c3 −c2   G3 −G2 × (G − G3 ) + c3  cmax ∀G < −δl ∀ − δl < G ≤ G 2 2 . and the field outside of this region using c.4) where . ∆.2) This mixed expression is a model describing the progress variable field in the vicinity of the flame through the use of cG .3) G0 stands for the value of the level set function in the inner layer of the flame.2. The combustion model couples the level set function and a classical transported progress variable c by evaluating a weighted progress variable such as ˆ cM = c + ξ G. (2010): ˆ ρu ∂G ˆ + v G = − Vn · ∂t ρ ¯ ˆ G. defining the evolution of an interface associated with the flame front as used by Knudsen et al. ∆ − c . The domain that is used for the main combustor LES begins 5 mm before the exit of slot 2. The weight ¯ ξ reads  ˆ ξ G. This allows the inflow profiles to adapt to the mesh. It is here assumed that the G = G0 iso-surface corresponds to half of the progress variable maximum value. and V the turbulent propagation speed of a temperature iso-surface in the flame that includes the laminar burning velocity and the front propagation caused by diffusion in high curvature regions. ∆ = (∆ ) 2 1/4 1/8 ˆ G − G0 (∆ ) + ∆2 2 2  1/8  (∆ ) + ∆2  × exp −6  . Roux and H. v the Favre-filtered velocity. Pitsch the central pilot tube downstream of the flame-holder where the fine geometric details cannot be included in the main simulation without increasing the size of the grid. cG has to be independent from these two variable as much as possible.278 A. Additional modeling is therefore needed to accurately describe the flame propagation speed. ∀G2 < G ≤ G3 ∀G > G3 cG := c|G = G0 (3. ∆ × cG G. 3. Depicting chemistry using a transported progress variable c consequently leads to numerical difficulties at the flame front where the progress variable profile becomes very under-resolved. ¯ ˆ (3. the premixed flame front is smaller than the grid size. (3. This modeling is introduced by solving an additional equation for a level set function G. although it is known that the heat release maximum value occurs for a higher value of the progress variable. Combustion modeling In the thin reaction zone regime of premixed combustion. ∆ stands for the width where the level-set is trusted in the simulation. ∆. (3.

exp(−Ta (z)/T (z. i. it is here assumed that the progress variable source term ωc can be ˙ modeled such as ωc (z. Different levels of approximation can be used to consider radiation. this procedure ensures that the resolved part of the turbulent mixing is welldescribed througout the use of the transported progress variable while the tracking of the flame is ensured at the right speed. and radiative heat transfer is accounted for by a Stefan-Boltzmann law which depends on local gas composition. Heat loss modeling The effects of radiative as well as wall convective heat transfers are of foremost importance.7) . Schmitt et al. The enthalpy equation is transported but is not a new dimension of the table to spare memory.LES of stratified combustion cmax G2 C2 G3 := c|G=∞ := δl × c|G=δl /2 − cmax 2 2 cmax := G2 +δl /2 × G2 + cmax 2 := G|c=c3 279 (3.δ(c − cM )dc.e. the enthalpy (Ribert et al. 2010).e. Another way to reduce the computational cost. c)). Different methods such as the ones using the self-similar properties of turbulent premixed flames have recently emerged but require additional assumptions. has been studied by Singer & Pope (2004) by creating in-situ chemical tables. ˙ (3. Thermal radiation effects can be directly integrated in the combustion model such as in the FPV model for non-premixed combustion developed by Ihme (2008). 2008). Convective heat transfer is accounted for by a thermal wall-function boundary condition imposing a prescribed temperature. especially when dealing with a multi-regime model requiring the use of three different tables. tables that are optimized and extended during the computation. ray tracking methods or Monte Carlo simulations in a separate code.6) Finally. Instead. ˙ (3. Other methods of treating radiation include approaches such as Discrete Ordinate Methods (DOM). 2001. the size of the different look-up tables is a major issue due to the lack of memory. 2007. Ihme 2008). This is the so-called In-Situ Adaptive Tabulation method (ISAT). Note that coupling heat transfer with a look-up table would require an additional dimension: i. Use of neural networks are another way to improve the size of the look-up tables (Ihme 2008).5) δl is the flame width. tabulating along another direction. 2006). 3. The source term reads ωC (cM ) = ˙ ωC (c)..3. another method is used to lighten the table (Knudsen et al. Note that coupling these methods to LES solvers is typically very computationally expensive (Desjardin & Frankel 1999. c) ≈ B(z). A more cost-effective approximation is to use a thin gray gas model for radiation and to assume that each point has a direct view of the cold surroundings (Barlow et al. However. Gonalves dos Santos et al. The temperature in the walls of the main injection port are set to be 300 K so that LES gas temperature matches experimental measurements at the nozzle exit.. Here this optically thin approach is employed. the result of the dimension of the lookup tables. In this work.

The enthalpy is either lower (case A). This correction is performed by noting that sL is proportional to the square root of the pre-exponential factor B of Eq. 3. Ta is extracted from the computation of the flamelets and tabulated. Correction of the source term to take into account enthalpy change as defined by Eq. The black line T0 is used as a reference state to tabulate the different quantities of interest. Three different flamelets are computed using FlameMaster (Pitsch 1998) with the same mixture fraction φ = 0.8) where the superscript tab means that the quantity comes from the look-up table. 3. Roux and H. The laminar burning velocity sL appears in the source term V of the G-equation.7. A corrected temperature that matches the value of the transported enthalpy can thus be estimated (Cp are evaluated using polynomial functions). the corrected flame speed reads . Finally. the model from Eq. and can be then directly used to correct the tabulated source term. Note that this evolution is comparable to the source term from a one-step chemistry approximation. or higher (case B) than T0 . Pitsch (a) (b) Figure 4. the corrected source term reads ωc (z. The enthalpy changes are induced by decreasing or increasing the temperature of the cold gases. Figure 4 shows the error introduced using this alternate way to describe heat losses.8 is used to compute the source term at the two other enthalpies using the temperature of the reference flamelet. density is computed by correcting the tabulated one with the ratio of the exponential functions (corrected over transported). respectively. c) × ˙ corr ˙ tab tab exp −Ta /T corr . Right: φ = 0. Then.0 but three different enthalpies (assumed to remain constant throughout the flamelet). Then.0.1. An enthalpy scalar is solver for. Eq. It also has to be corrected to take into account heat loss effects and reproducing the flame lift-off. 3. The result from FlameMaster is denoted by the subscript F M and the corrected variable by M ODEL.6 or φ = 1. This assumed functional form of the source term can be used to modify a source term from a tabulated chemistry database in the following manner.280 A. where B(z) is independent of the temperature. 400 and 320 K. tab exp (−Ta /T tab ) (3. 3.8.6 and Left: φ = 1. c) = ωc (z. By considering that the correction of the ratio of the exponential is included in the function B(z). with a corresponding inlet temperature of 240. First. A good agreement is found although the source term peak is slightly shifted toward the burned gases for lower equivalence ratio. temperature is calculated from the transported value by assuming that gas composition is not changed by transfer.

A very good agreement is found between the . CO2 . and empty ones. the progress variable c. (3. a maximum error of 15% is found and is considered to be acceptable. Laminar flame speed versus equivalence ratio for five different initial temperatures (Representing different enthalpies). However. It is discretized using 200 × 400 × 25 points. based on a bulk velocity of 10 m/s. for the lower enthalpies that are considered in this work. H2 O and H2 mass fractions. 3. simulation results are presented and assessed against the experiment. 4. The flamelets described by triangles are used to build the look-up tables.LES of stratified combustion 281 Figure 5. A flamelet database is generated using FlameMaster by representing the flame as an unstretched freely propagating premixed flame with non-unity Lewis numbers. All time-averaged statistics were collected for a period of four flow through times.9) Figure 5 shows the evolution of the laminar flame speed versus the equivalence ratio for five different enthalpies coming from FlameMaster (empty symbols) and the corresponding approximated value (filled symbols and line). Non-reacting comparisons Simulation of the isothermal case A i2 and the quasi non-reacting case A i1 are first performed to validate the applied inflow boundary conditions. 4. Correction of the flame speed to take into account enthalpy change. Results In this section. The look-up table has three leading dimensions: the mixture fraction z. FlameMaster.1. Time-averaged velocity statistics are shown in Fig. The fuel is methane. 3. and the variance z 2 . The four others are constructed by changing the inlet temperature and keeping the composition unchanged.9.4. scorr = stab × L L tab exp −Ta /T corr tab exp (−Ta /T tab ) 1/2 . Chemistry modeling The progress variable is here defined as the sum of CO. Differences arise for high enthalpy flamelets. 6 where mean (a) and variance (b) values of the axial component of the velocity vector are given. Filled symbol: corrected speed from Eq.

Between the pilot exit and around 30 mm downstream. Roux and H. The maximum temperature is found to be 2090 K which is 60 K below the adiabatic flame temperature for an equivalence ratio of 0. Figure 7 shows an instantaneous iso-surface of progress variable YC = 0. the LES is extended to the reacting cases A r and G r.9. Experimental data sets are available for further axial locations of z = 50 and z = 100 mm but are not given here. Two different zones can be distinguished.1. two different opening flame angles are seen. The simulation captures the velocity fluctuations accurately. hot gases have higher temperature and increased diffusivity that smoothes the asymmetries. For the G r case. The A r flame angle is reduced compared to that in the fully-premixed case. This region particularly shows that the lift-off is the same for both cases. However. the hot gases expand radially faster for the G R case than for the stratified case. Interestingly.282 40 Seffrin et al : A_i1 NGA : A_i1 A. as seen in Fig. pockets of hot gas are expelled from the flame core because of Kelvin-Helmholtz instability at the outer shear layer where a high velocity shear is found. However. LES data and the experiment. Reacting comparisons Having validated the approach for the non-reacting cases. Pitsch A_i2 A_i2 30 Radius [mm] 40 Seffrin et al : A_i1 NGA : A_i1 A_i2 A_i2 30 Radius [mm] 20 20 10 10 0 0 5 10 0 0 1 2 3 4 (a) (b) Figure 6. the gradient of temperature locating the flame front is mostly perpendicular to the gradient of mixture fraction. discrepancies arise at the shear layers between the inflow pipes where more points could have been used to discretize the lips of the different slots. Two different behaviors are seen. This issue does not appear for the other reacting cases according to the experimentalists. Indeed. the shape is similar for both cases as the flame is not yet burning toward the leaner mixture coming from slot 2.2. Very good agreement is also found at these locations. Another difference arises in the pilot for the A i2 case on the mean axial velocity: asymmetries in the pilot flow appear owing to a slight misalignment of the moveable flame holder in the experiments. 3. Contours of mixture fraction allow localizing stratified regions. Mean and variance of axial velocity 1 mm downstream the exit of the pilot for cases A i1 and A i2. at some points. as the flow moves downstream. 4. Farther downstream. a parallel . These show the flame propagating in regions of gradually leaner mixture fraction from the inner to the outer part of the jet. Figure 8 shows a detailed view of mixture fraction and temperature for the A r case close to the injector in the region of interest.

z = 125 mm and z = 200 mm. Top half of image: A r case. the differences between the two cases are now studied and assessed against the experiment. z = 100 mm. right: A r case. An overall good agreement is found on the axial velocity.LES of stratified combustion 283 Figure 7. Because the coflow velocity (0. The opening of the flame in the radial direction seen in the radial velocity profiles comes from the gas expansion. z = 50 mm. Figures 9 to 12 show the evolution of the mean and variance of axial and radial components of the velocity vector as a function of the radial coordinate at six axial locations: z = 1 mm. z = 75 mm. Because of non-adiabaticity effects in the pilot. Instantaneous isosurface of progress variable YC = 10. Having a fully converged simulation is not adressed in this study. For a more quantitative validation. Ubulk = 10 m/s. Note that at the outer shear layer. i. Then the presence of the flame and the increased viscosity at high temperature smooth out this effect. and. Figure 8. and. statistics in the outer shear layers converge very slowly relative to the primary jets statistics. bottom half of image: G r case.e. Instantaneous fields of iso-levels of mixture fraction and contours of temperature.1 m/s) is a hundred times lower than the bulk velocity of the three different pipes. Left: G r case. A bump structure is seen until z = 50 mm because of the segregated inflow. some discrepancies may exist because of the lack of convergence of the results. The statistics have been acquired during four flow through times where the characteristic time-scale is based on the bulk velocity of these three pipes and not on that of the coflow.. alignment appears that leads to a decreased heat release and temperature because of the leaner mixture fraction met. the flame .

Pitsch Figure 9.s−1 ] at six different axial locations. The levels of variance are also well matched. Roux and H. Top: G r. Figure 10. Mean axial velocity profiles in [m.284 A. . and. Variance of axial velocity profiles in [m. However. bottom: A r cases. Top: G r. a longer simulation seems to be required to have smoother secondorder statistics. and. bottom: A r cases.s−1 ] at six different axial locations. Comparison between computation (lines) and experiment (symbols). is detached from the exit of the pilot and the radial velocity profiles show a fairly good prediction of its position in the numerical simulation. Comparison between computation (lines) and experiment (symbols).

bottom: A r cases. Temperature predictions are plotted in Figs. Mean radial velocity profiles in [m. Farther downstream. Comparison between computation (lines) and experiment (symbols). Comparison between computation (lines) and experiment (symbols). and.LES of stratified combustion 285 Figure 11. 13 and 14. Variance of radial velocity profiles in [m. The temperature in the pilot (r < 7. (2010). Figure 12. and. Top: G r.s−1 ] at six different axial locations. additional heat release increases the local temperature and forms a bump at the location of the flame . Experimental data sets are taken from Boehm et al. Top: G r.s−1 ] at six different axial locations. bottom: A r cases.4 mm) is below the adiabatic flame temperature at around 1800 K.

C. Ing. bottom: A r cases. The immediate future work includes NOx modeling. Fruitful discussion with Dr. stratification is a commonly used way to reduce NOx emissions and comparing profiles of pollutants at the exit of the burner for different stratified/fully-premixed case would be of great interest. An overall good agreement is found for the time-averaged velocity statistics and temperature. the multi-regime flamelet approach was used to perform Large-Eddy Simulation (LES) of a stratified burner. being predicted as too small. Acknowledgments Authors thank Dr. Two different reactive cases have been simulated. Knudsen. Comparison between computation (lines) and experiment (symbols). but the profiles are slightly shifted toward the inner part of the jet. Schmitt and Dr. . Seffrin and B. a good agreement is found for the magnitude. Roux and H. Boehm for providing the experimental data. Indeed. T. Regarding the variance of temperature. This could be explained by the modeled turbulent flame speed. This study focuses on understanding the limits of the models used when dealing with stratification. which is responsible for the radial expansion of the hot gases. Conclusions and future work In this work. Top: G r. Lederlin are gratefully acknowledged. increased mean temperature is found for high r because large flame cusps appear during the propagation of the flame. and. Pitsch Figure 13. brush. one a stratified flame and an equivalent fully premixed one. Support from CTR and funding from SAFRAN is acknowledged. Mean temperature profiles in [K] at six different axial locations. Mr. An experimental burner that has been designed to look at lean premixed stratified combustion is used to validate the model. Ed. 5.286 A. At the two final downstream locations. F.

& Pitsch. Inst. 2000 Experimental analysis of combustion flows developping over a plane-symmetric expansion. R. G. Besson. Bruel. Comparison between computation (lines) and experiment (symbols). Combust. S. 32 (2). G. J. 2001 Scalar profiles and no formation in laminar opposed-flow partially premixed methane/air flames. Phys. H. P. 2010 Temperature and mixing field measurements in stratified lean premixed turbulent flames. Barlow. P. S. J. Flame 127 (3). Desjardins. Frank. A. M. Boehm. Balarac. Inst. M. O. Comb. 2009 Application of raman/rayleigh/LIF diagnostics in turbulent stratified flames. J.. Karpetis. Variance of temperature profiles in [K] at six different axial locations. P. Domingo. J. Top: G r. J. Theory Model. P.. In Press. Barlow. Corrected Proof. & Hochgreb. Inst. Hochgreb. Comput. S. 32 (1). Proc. L. . Barlow. 6 (4).. Proc. N. Blanquart. 945 – 953. Comb. 1763 – 1770.. 1999 Two-dimensional large eddy simulation of soot formation in the near-field of a strongly radiating nonpremixed acetylene-air turbulent jet flame.. S. & Cant. Combust. J.. and. Y..LES of stratified combustion 287 Figure 14. 2102 – 2118. 59 – 67. Anselmo-Filho.. G. –. 2002 Partially premixed flamelets in LES of nonpremixed turbulent combustion. REFERENCES Anselmo-Filho. H. B. Sweeney. & Deshaies. Heat Transfer 14 (1). A. 227 (15). & Frankel. Combust. bottom: A r cases.. 2009 Experimental measurements of geometric properties of turbulent stratified flames.. B. Flame 119 (1-2). & Chen. Champion. H. 121 – 132.. Thermophys. Desjardin. Vervisch. R. Proc. E. R. 529 – 551.-H. K. & Dreizler. Frank. P.. Combust. Wang. R. 2008 High order conservative finite difference scheme for variable density low mach number turbulent flows. & Bray.. 7125–7159.

Seffrin. J. Pitsch. 788 – 803. & Dreizler. 1999 Automotive spark-ignited directinjection gasoline engines. 2000 A numerical study of the laminar flame speed of stratified methane/air flames. Comb. Prog. Zhao. 570. B. Phys.. Gicquel. Ihme.. J. Combust. H. H. P. M. J. Comb. Poinsot. M. nitric oxide emissions and combustion instability in a swirled turbulent high-pressure burner. H. Scghuermans. Haworth. J. 437 – 562. Energy Combust. A. N.. 8 (2). M. F. 25 (5).. A... 2010 Flow field studies of a new series of turbulent premixed stratified flames. & Veynante. 2010 An analysis of premixed flamelet models for large eddy simulation of turbulent combustion. Jimenez.. D. 1998 Flamemaster. Inst. D. K. R. F. Poinsot. Gonalves dos Santos. Flame 125 (1-2).. Roux and H. V. Cuenot. C. M. T. 361 – 383. Phys. Geyer. 387 – 400. C. Sci. Fluids 20 (5). G. Comb.288 A. Ducruix. H. Lai. Ribert. S.. D.. T. Flame 121 (3). Fluid Mech. 2001 Edge flames and partially premixed combustion in diffusion flame quenching. Flame 157 (2). E. Singer. D. 2008 Coupled large eddy simulations of turbulent combustion and radiative heat transfer. Flame 152 (3). A. R. 2004 Exploiting isat to solve the reaction-diffusion equation. . & Pitsch. Gicquel. Fluid In press. 17–46. L.stanford.. & Poinsot. Kim. S. Pitsch Favier. B. Dean. & Veynante. 2008 Modeling of radiation and nitric oxide formation in turbulent nonpremixed flames using a flamelet/progress variable formulation. & Geigle. 2007 Large-eddy simulation and experimental study of heat transfer. Flame 128 (1-2).. 2002 Numerical simulation and modeling for lean stratified propane-air flames. 28 (2). 395 – 417. & Grenda. D.. 1 – 21. 2000 Numerical simulation of turbulent propane-air combustion with nonhomogeneous reactants. P. Comb. C. Darabiha.. 384 – 396. Cuenot. L.. Knudsen. 649 – 664. Comb.. Theory Model. & Harrington. S. O. E. B. O. 1925 – 1932. a c++ computer program for 0d combustion and 1d laminar flame calculations Available from http://www. Lecanu.. & Vervisch. T. 2006 Tabulation of complex chemistry based on self-similar behavior of laminar premixed flames. Comb. Schmitt. Blint. Fuest. Combust. F.edu/ hpitsch/. Proc. D. Iacona. Pires Da Cruz. & Haworth.. & Pope. Flame 146 (4).