Combustion and Flame 144 (2006) 461475

www.elsevier.com/locate/combustame
Direct numerical simulation of nonpremixed
amewall interactions
Yi Wang
a
, Arnaud Trouv
b,
a
Department of Mechanical Engineering, University of Maryland, College Park, MD 20742, USA
b
Department of Fire Protection Engineering, University of Maryland, College Park, MD 20742, USA
Received 26 May 2005; received in revised form 4 August 2005; accepted 10 August 2005
Available online 23 September 2005
Abstract
The objective of the present study is to use detailed numerical modeling to obtain basic information on the in-
teraction of nonpremixed ames with cold wall surfaces. The questions of turbulent fuelairtemperature mixing,
ame extinction, and wall-surface heat transfer are studied using direct numerical simulation (DNS). The DNS
conguration corresponds to an ethyleneair diffusion ame stabilized in the near-wall region of a chemically inert
solid surface. Simulations are performed with adiabatic or isothermal wall boundary conditions and with different
turbulence intensities. The simulations feature ame extinction events resulting from excessive wall cooling and
convective heat transfer rates up to 90 kW/m
2
. The structure of the simulated wall ames is studied in terms of a
classical mass-mixing variable, the fuelair based mixture fraction, and a less familiar heat loss variable, the ex-
cess enthalpy variable, introduced to provide a measure of nonadiabatic behavior due to wall cooling. In addition
to the ame structure, extinction events are also studied in detail and a modied ame extinction criterion that
combines the concepts of mixture fraction and excess enthalpy is proposed and then tested against the DNS data.
2005 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
Keywords: Direct numerical simulation; Flamewall interactions; Turbulent wallames; Flame extinction; Excess enthalpy;
Wall heat ux
1. Introduction
Flamewall interactions (FWI) play an important
role in many combustion systems. For instance, in
internal combustion (IC) engines, cooled walls com-
bined with occurrences of short amewall distances
result in ame quenching and FWI has a negative
impact on engine performance, in terms of both ther-
mal efciency and pollution propensity. While the
magnitude of the FWI impact on thermal efciency
*
Corresponding author. Fax: +1 301 405 9383.
E-mail address: atrouve@eng.umd.edu (A. Trouv).
remains small in IC engines, FWI has a more no-
table effect on pollutant emissions and provides one
of the dominant mechanisms for unburnt hydrocarbon
emissions. Similar effects are observed in aeronauti-
cal propulsion and power-generation applications, es-
pecially given the recent trends toward the design of
more compact, smaller (meso- or microscale) com-
bustion chambers; the associated higher surface-to-
volume ratios and shorter amewall distances result
in a larger impact of amewall interactions and heat
transfer on the combustion system performance. En-
closure res are another combustion topic in which
FWI plays an important (albeit different) role. For in-
stance, the burning of a vertical ammable wall is a
0010-2180/$ see front matter 2005 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
doi:10.1016/j.combustame.2005.08.009
462 Y. Wang, A. Trouv / Combustion and Flame 144 (2006) 461475
generic conguration where the fuel is released and
consumed within the buoyancy-driven wall boundary
layer, and the entire combustion process may be con-
sidered as FWI.
The subject of amewall interactions in IC en-
gines has received signicant interest in the past
15 years. Previous studies have focused primarily on
the quenching problem of laminar or turbulent pre-
mixed ames near cold wall boundaries [112]. Re-
sults from laminar ame experiments [1,2] indicate
that (1) quenching events occur near cold wall sur-
faces; (2) these quenching events also correspond to
maximum values of the gassolid heat ux; (3) de-
pending on wall temperature and fuel type, premixed
ames may lose a signicant fraction of their ref-
erence power when driven to quenching. The maxi-
mum heat ux may be as high as 0.51 MW/m
2
for
atmospheric hydrocarbon ames, or approximately
30% of the ame reference power; the reference
power is dened as the heat release rate per unit
ame surface area and is typically on the order of
13 MW/m
2
for stoichiometric ames at atmospheric
pressure.
Detailed numerical modeling has been used in re-
cent years both to reproduce the early experimental
ndings of Refs. [1,2] and to bring further insights
into the controlling factors that determine ame
wall heat transfer in IC engines. Numerical model-
ing has been used in particular to study the effects
of turbulent ow conditions [3,6,9,10], detailed gas-
phase chemical kinetics [4,7,8,1113] and heteroge-
neous gassolid surface chemistry [7,11,12]. Note
that in the treatment of this problem, the ames
are always assumed to be optically thin and ther-
mal radiation is neglected. Consistent with the lam-
inar amelet viewpoint, the dynamics of quenching
events in turbulent ames was found to be similar
to that observed in laminar ame studies: for in-
stance, the magnitude of the peak gassolid heat ux
was approximately the same in laminar and turbulent
ame simulations [3,6]. Furthermore, while numeri-
cal modeling based on simplied (single-step) chem-
istry was found to correctly describe FWI when the
wall temperature was low, detailed descriptions of ho-
mogeneous/heterogeneous chemical reactions were
required for accurate predictions of the gassolid heat
ux when the wall temperature was above 400 K
[4,7,8]. These results combine to draw a complex pic-
ture of FWI in which ame quenching is both ther-
mally and kinetically driven.
References [13,14] also consider the case of a lam-
inar diffusion ame impinging on a cold wall in a
stagnation point ow conguration. Ref. [14] con-
siders the general case of hydrocarbonair ames,
whereas Ref. [13] focuses on the more specic
case of hydrogenoxygen ames. In the impinging
ow/diffusion ame conguration, FWI is a transient
process that must result in ame quenching at the
wall; however, details of the ame-quenching process
depend strongly on the ow strain rate, which is spec-
ied as a free parameter. It is found that at high
strain rates, the diffusion ame comes remarkably
close to the wall surface (prior to quenching) and
the gassolid heat ux takes very large values (for
the hydrocarbonair ames studied in Ref. [14], the
peak value of the gassolid heat ux may be higher
than that obtained in the corresponding premixed stoi-
chiometric conguration). In Ref. [14], the results are
provided in nondimensional units with laminar pre-
mixed ame parameters selected as reference values.
In these units, the minimum amewall distance is
on the order of (D
th
/s
L
), where D
th
is a reference
thermal diffusivity and s
L
the stoichiometric lami-
nar ame speed; the applied strain rate takes values
up to (s
2
L
/D
th
). It is worth noting that these ame
conditions may appear somewhat unrealistic: using
representative values of D
th
=2.2 10
5
m
2
/s and
s
L
=0.5 m/s, one nds that ames are strained up to
more than 10,000 s
1
, and the amewall distance
is decreased to less than 50 m. The high values of
the gassolid heat ux that are obtained under those
conditions may be an artifact of using excessive val-
ues of strain rate and of the specics of the transient
stagnation point ow conguration. The question of
whether these ame-ow conditions are representa-
tive of practical turbulent combustion congurations
remains entirely open.
We now turn to enclosure re applications and
consider three different congurations featuring sub-
stantial amewall interactions [15,16]: a rst con-
guration corresponding to a ammable vertical wall,
in which the ame is fueled by the thermal decom-
position of the wall material; a second conguration
in which the ame is fueled by a separate burner and
where it develops adjacent to an inert vertical wall;
and a third conguration in which the ame is again
fueled by a separate burner and impinges on an in-
ert horizontal (ceiling) wall. These congurations are
representative of the large variety of ame-spread and
heat-transfer mechanisms found in re problems and
correspond to different amewall arrangements. In
the rst conguration, the ame size is determined by
the details of the gassolid heat transfer and in-wall
fuel gasication processes, whereas in the second and
third congurations, the ame size is arbitrary and
simply prescribed by the power output of the pilot
burner. Typical values for the wall heat ux are: up to
50 kW/m
2
in the case of vertical wall res [1518];
up to 120 kW/m
2
in the case of vertical walls ex-
posed to a separate adjacent ame [15,16,19,20]; and
up to 150 kW/m
2
in the case of horizontal (ceiling)
walls exposed to an impinging buoyancy-driven ame
Y. Wang, A. Trouv / Combustion and Flame 144 (2006) 461475 463
[15,16,21,22]. It is worth emphasizing that those esti-
mates correspond to time-averaged values and cannot
be compared directly to the time-resolved, instanta-
neous peak values discussed above in the context of
IC engine applications. The statistical variations of
amewall distances and gassolid heat uxes remain
unknown in re problems; these statistical variations
will depend on the strength of the buoyancy-driven
turbulent motions. The magnitude of the wall heat
ux will also depend on the optical thickness of the
ame gases: previous studies indicate, for instance,
that while the gassolid heat ux in small nonsooty
ames is controlled by convective heat transfer, the
contribution of radiation heat transfer becomes domi-
nant in large ames with signicant smoke production
[15,16].
The present study is a continuation of the lami-
nar FWI study in Ref. [14] and an extension to the
case of turbulent amewall interactions. The ques-
tions of turbulent fuelairtemperature mixing, ame
extinction, and wall-surface heat transfer are studied
using direct numerical simulation (DNS) in a con-
guration corresponding to an ethyleneair diffusion
ame stabilized in the near-wall region of a chem-
ically inert solid surface. While viewed as a ques-
tionable simplication, thermal radiation is neglected
in the present simulations (it will be considered in a
follow-up study). Our objective here is to focus on
turbulent ow effects and to evaluate in particular the
wall-induced modications of the ame structure, the
probability of ame extinction events, and the statisti-
cal distribution of the wall-surface heat ux. The DNS
solver and numerical conguration are presented in
Section 2; results from simulations are discussed af-
terward.
2. DNS solver and numerical conguration
The present study is performed using an advanced
direct numerical simulation solver called S3D. S3D
is a fully compressible NavierStokes solver cou-
pled with an integrator for combustion chemistry and
is based on eighth-order nite differencing [23,24],
fourth-order explicit RungeKutta time integration
[25], characteristic-based boundary conditions treat-
ment [2629], a CHEMKIN-compatible description
of chemical kinetics [30], and a conventional rectan-
gular Cartesian mesh capability. In addition, S3D is
a massively parallel solver based on message pas-
sage interface (MPI) libraries and is currently being
redesigned and enhanced for increased performance
and capability by a consortium of research institu-
tions [31].
The selected ame conguration corresponds to
ethylene burning in ambient air. While S3D features
a detailed chemical kinetics capability as well as new
soot formation and thermal radiation capabilities [31],
the present study takes an intermediate step and fo-
cuses on amewall interactions without soot or radi-
ation. In addition, combustion is here described using
a single-step model proposed in Ref. [32],
(1)
C
2
H
4
+3O
2
2CO
2
+2H
2
O

C
2
H
4
=
A
(10
6
)
p+q
M
C
2
H
4

Y
C
2
H
4
M
C
2
H
4

Y
O
2
M
O
2

q
exp(T
a
/T )

,
where
C
2
H
4
is the fuel mass reaction rate (in units
of kg/m
3
/s); A a model coefcient, A =2.0 10
12
(mol/m
3
)
(1pq)
s
1
; p and q the model fuel and
oxygen concentration exponents, p = 0.1, q = 1.65;
the mixture mass density (kg/m
3
); Y
k
the species
k mass fraction; M
k
the species k molecular weight
(kg/mol); T
a
a model activation temperature, T
a
=
15,107 K; and T the uid temperature (K). The cor-
responding heat-release rate is
(2) q
c
=

C
2
H
4

H
c
,
where q
c
is the ame power density (in units of
W/m
3
); and H
c
the heat of combustion (J/kg of
fuel consumed). Note that the present single-step
chemistry version of S3D also assumes a constant
heat capacity, c
p
1006 J/kg K, and that the value of
H
c
has been adjusted accordingly so that the simu-
lated adiabatic ame temperature is 2370 K: we use
H
c
=32.7 MJ/kg.
Additional ame modeling choices include a
temperature-dependent dynamic viscosity ( varies
with T to the power 0.7), a constant Prandtl number,
Pr =0.708, and unity Lewis numbers.
We now turn to a brief discussion of the antici-
pated impact of our ame modeling choices. The ac-
curacy of the simplied chemical kinetics and mole-
cular transport submodels adopted in S3D has been
evaluated in a separate numerical study of strained
laminar diffusion ames. This separate study consid-
ers the generic case of steady, one-dimensional, plane,
counterow ames. To evaluate accuracy, the S3D-
based simulations are compared to more elaborate
calculations performed with the OPPDIF software
[33], using detailed descriptions of chemistry and
molecular transport [34]. In a simple laminar counter-
ow ame conguration, discrepancies between S3D
and OPPDIF predictions are essentially due to differ-
ences in the mathematical formulation of the ame
problem, and may therefore be used to evaluate the
penalty associated with using simplied chemical and
molecular transport models in S3D.
Fig. 1 presents typical results obtained from the
S3D/OPPDIF comparative study. Each data point in
464 Y. Wang, A. Trouv / Combustion and Flame 144 (2006) 461475
Fig. 1. Flame response to changes in the fuelair mixing
rate in a steady, one-dimensional, plane, laminar, diffusion
counterow ame conguration. The top curve (squares)
corresponds to numerical data obtained with OPPDIF using
detailed chemistry [34] and molecular transport; the bot-
tom curve (triangles) corresponds to numerical data obtained
with S3D and using single-step chemistry (Eqs. (1) and (2))
and unity Lewis numbers.
Fig. 1 corresponds to a converged steady ame solu-
tion. The plot corresponds to variations of centerline
combustion intensity with fuelair mixing rate; the
combustion intensity is measured as the heat release
rate per unit surface area; the fuelair mixing rate is
measured as the stoichiometric value
st
of the scalar
dissipation rate , = 2D|Z|
2
, with D the heat
diffusivity and Z the fuelair based mixture fraction.
The inverse of
st
provides an estimate of the mixing
time scale in the vicinity of the reaction zone [3537].
The S3D and OPPDIF curves in Fig. 1 display the
classical ame response to increasing mixing rates,
including the intensication of combustion, observed
for low-to-moderate values of
st
, and the abrupt fall-
off, observed at high values of
st
. This fall off cor-
responds to transition to the supercritical ame ex-
tinction regime. Compared to OPPDIF results, it is
seen that the S3Dcalculations underestimate the com-
bustion intensity in the subcritical regime and predict
extinction for a critical value of the scalar dissipation
rate that is about three times smaller than that ob-
tained with OPPDIF,
ad
st,ext
45 s
1
, where
ad
st,ext
denotes the critical value of
st
obtained under (adia-
batic) extinction conditions.
While the errors documented in Fig. 1 are clearly
a concern, we choose to accept these errors in the
present study and to work with the simplied ame
model described in Eqs. (1) and (2). It is worth em-
phasizing, however, that the present DNS results are
to be interpreted in a qualitative, rather than a quanti-
tative manner. The precomputed database of strained
laminar diffusion ames is used in the following to
help specify inow boundary conditions, as well as
Fig. 2. Numerical conguration corresponding to a turbulent
ethyleneair diffusion ame stabilized near a solid wall. The
computational domain is two-dimensional and features a tur-
bulent inow boundary at x =0 (ow is from left to right);
a solid wall boundary at y =0; and nonreecting boundaries
at x =8 cm and y =4 cm. The turbulent inow boundary is
used for both air and fuel injection. The plot shows a typi-
cal snapshot of temperature isocontours and is taken from a
simulation in which the wall is assumed to be adiabatic.
to provide a point of reference in the analysis of the
more complex ame structures observed under turbu-
lent ow and nonadiabatic combustion conditions.
Next, we turn to a presentation of the FWI numeri-
cal conguration. The FWI conguration corresponds
to a two-dimensional, momentum-driven, chemically
reacting ethyleneair mixing layer developing paral-
lel to an inert solid wall surface (Fig. 2). The com-
bustion region is well ventilated, with excess air sup-
plied from the free stream, while ethylene ow is con-
ned to the near-wall region. As shown in Fig. 2, the
computational domain features an inow boundary at
x =0, a wall boundary at y =0, and two nonreect-
ing boundaries at x = 8 cm and y = 4 cm. The wall
boundary conditions correspond to zero velocity, zero
mass ux, and either zero heat ux (adiabatic wall
case) or prescribed temperature (nonadiabatic wall
case). The inow boundary conditions at x = 0 cor-
respond to prescribed velocity, mixture composition
and temperature. We now discuss those inow condi-
tions in more detail.
The free-stream region of the inow boundary
corresponds to a uniform ow of air under normal
temperature and pressure conditions, and seeded with
turbulent-like perturbations. The mean velocity is
u

= 5 m/s; the perturbations use a variant of the

random uctuation method of Ref. [38], in which
the velocity uctuations are specied using an aux-
iliary synthetic eld corresponding to homogeneous
isotropic turbulence and a prescribed model kinetic
energy spectrum. In the present study, the turbulent
inow perturbations are characterized by a moderate-
to-high forcing intensity, 1 u

2.5 m/s, and a

small integral length scale, l
t
= 0.17 cm; the cor-
Y. Wang, A. Trouv / Combustion and Flame 144 (2006) 461475 465
responding turbulent ow Reynolds number, Re
t
=
(u

l
t
/), where is the free-stream kinematic viscos-
ity, ranges from 108 to 270.
The near-wall region of the inow boundary cor-
responds to a prescribed velocity prole that satises
the no-slip wall condition and has a certain thick-
ness : we choose =0.15 cm. is also the selected
inlet ame-to-wall distance, with fuel being injected
at y-locations below , i.e., within the velocity bound-
ary layer. Flow temperature and species mass frac-
tions are specied at x = 0 using a separate calcu-
lation corresponding to a low-strain laminar coun-
terow ame solution (see the discussion of Fig. 1
above).
The initial elds correspond to a one-dimensional
laminar ame solution and are constructed from the
inow boundary proles. The computational grid
size is 1216 244. The grid spacing is uniform
in the x-direction, x 66 m, while variable in
the y-direction: the y-grid is uniform in the near
wall/ame region, y 50 m for 0 y 0.8 cm,
and is stretched in the free-stream region. The grid
resolution is chosen based on the precomputed data-
base of laminar diffusion counterow ames (Fig. 1)
and has been determined to be adequate to resolve
the thin ame front up to the extinction strain rate.
The grid resolution is also adequate to resolve the
smallest ow structures: the Kolmogorov-based ow
microscale ranges from 26 to 51 m; the enstrophy-
based ow microscale (a more appropriate scale
to estimate the size of the smallest eddies in two-
dimensional turbulence) ranges from 103 to 163 m.
Time integration is performed at a pace determined
by the acoustic-based CourantFriedrichsLewy sta-
bility condition, t 0.05 s. Computations are per-
formed for a total duration corresponding typically
to three or four free-stream transit times

, where

= (L
x
/u

) = 16 ms, with L
x
the x-size of the
computational domain, L
x
= 8 cm. If we ignore the
initial transient phase of duration, equal to one or two
free-stream transit times, the useful part of the simu-
lations lasts for approximately t
s
2

=32 ms.
S3D is run in a parallel mode, using MPI and
one of the following two supercomputers: 256 proces-
sors on an IBM Power 3 system operated by the Na-
tional Energy Research Scientic Computing Center
(NERSC) at Lawrence Berkeley Laboratory; or 32
processors on an IBM Power 4 system operated by the
National Center for Computational Sciences (CCS) at
Oak Ridge National Laboratory. On these platforms,
the cost of performing a single simulation of total du-
ration 64 ms is approximately 32,000 processor hours
at NERSC and 7,100 processor hours at CCS.
3. DNS results
3.1. Wallame structure
We consider in the following a series of four simu-
lations. The simulations differ by the choice of inow
turbulence intensity and/or thermal boundary condi-
tion applied at the solid wall surface (see Table 1).
The ame structures observed in cases 14 are found
to correspond to different topologies and belong to
one the following three categories: a continuous ame
sheet without extinction; multiple ame sheets with-
out extinction; or multiple ame sheets with extinc-
tion. Figs. 35 provide examples of these three differ-
ent categories.
Figs. 3a and 3b present instantaneous snapshots
of temperature and fuel mass reaction rate, as ob-
tained in case 1 (adiabatic wall, lower turbulence in-
tensity). The gures illustrate the ame response to
incoming ow perturbations and conrm that signi-
cant turbulent mixing and ame wrinkling take place
in the vicinity of the wall. They also show that the
x-domain size is smaller than the wallame length
and burning is observed to continue beyond the out-
ow boundary at x = 8 cm. Fig. 3c presents the lo-
cation of the stoichiometric fuelair interface and a
comparison between Figs. 3b and 3c indicates that
the ame is active along the entire stoichiometric in-
terface: the ame displays the classical downstream
weakening of combustion intensity as the reactants
get depleted and replaced by combustion products,
but it remains free of any aerodynamic ame extinc-
tion event. In the absence of extinction, the tempera-
ture levels are particularly high in the near-wall region
(see Fig. 3a): the wall boundary is almost everywhere
in contact with hot gases at temperatures in excess of
2000 K. Fig. 4 presents similar results, as obtained
in case 2 (adiabatic wall, higher turbulence intensity).
In this case, the turbulence levels are high enough
Table 1
Inow turbulence intensity and wall thermal boundary condition used in cases 14 of the DNS database
Case 1 2 3 4
Turbulence intensity u

=1 m/s u

=2.5 m/s u

=1 m/s u

=2.5 m/s
Wall thermal boundary Adiabatic Adiabatic Isothermal Isothermal
condition (T
w
=300 K)
a
(T
w
=300 K)
a
a
T
w
is the gassolid wall temperature.
466 Y. Wang, A. Trouv / Combustion and Flame 144 (2006) 461475
(a)
(b)
(c)
Fig. 3. Instantaneous spatial variations of (a) uid tempera-
ture (in units of K), and (b) fuel mass reaction rate (in units
of kg/m
3
/s); (c) the location of the corresponding stoichio-
metric isocontour of the fuelair-based mixture fraction Z.
Case 1 solution.
to break the ame sheet into several segments; these
levels remain subcritical, however, and a comparison
between Figs. 4b and 4c indicates that the ame re-
mains extinction free. The temperature levels in the
near-wall region (see Fig. 4a) exhibit larger uctua-
tions than that found in case 1 due to the intermittent
penetration of cold free stream gases down to the wall
surface.
Fig. 5 presents the results obtained in case 3 (cold
wall temperature, lower turbulence intensity). A com-
parison between Figs. 5b and 5c indicates that case 3
features several ame extinction events (similar ob-
servations are made in case 4). As seen in Fig. 5a,
the structure of the thermal boundary layer is quite
different from that observed in Figs. 3a and 4a: a thin
cold subregion develops near the wall and the ame is
now exposed to wall-induced heat losses. These heat
(a)
(b)
(c)
Fig. 4. See caption of Fig. 3. Case 2 solution.
losses occasionally become supercritical and lead to
the extinction events seen in Fig. 5b.
We continue our discussion of the ame struc-
ture below, but choose to adopt hereafter a more
theoretical perspective. We start from the classical
BurkeSchumann description of nonpremixed com-
bustion, in which equilibrium chemistry (i.e., fast
chemistry) is assumed and the ame is controlled by
the fuelair mixing process [3537]. In the Burke
Schumann solution, the full reactive mixture compo-
sition is uniquely mapped as a function of the fuelair
based mixture fraction Z (Appendix A). While use-
ful as a starting point and a valuable reference, the
BurkeSchumann theory is also often too simple and
deviations from chemical equilibrium are frequently
observed in practical applications; these deviations
are particularly large in the presence of ame ex-
tinction. In the classical amelet theory of turbulent
nonpremixed combustion, deviations from fast chem-
istry behavior are monitored by the introduction of
Y. Wang, A. Trouv / Combustion and Flame 144 (2006) 461475 467
(a)
(b)
(c)
Fig. 5. See caption of Fig. 3. Case 3 solution. Two wall-
induced ame extinction events are observed in (b), near
(x = 0.8 cm; y = 0.2 cm) and (x = 5.5 cm; y = 0.2 cm);
(c) also presents the spatial variations of the excess enthalpy
variable H.
a second mixing variable
st
: the deviations are in-
signicant for
st
much smaller than
ad
st,ext
, and large
otherwise. This point of viewis similar to that adopted
in Fig. 1. Thus, in the amelet theory, the ame chem-
ical structure is described in terms of the two variables
Z and
st
[3537].
We now extend the previous discussion to the
BurkeSchumann and amelet descriptions of uid
temperature variations. We consider the adiabatic
case rst (cases 1 and 2 in Table 1). The extension is in
that case trivial. For instance, in the BurkeSchumann
theory, T is simply described in terms of Z (Appen-
dix A), while in the amelet theory, T is a function
of both Z and
st
. Fig. 6 presents a scatterplot of
uid temperature versus mixture fraction, as obtained
in case 2. The adiabatic BurkeSchumann solution
Fig. 6. Scatterplot of temperature versus fuelair-based mix-
ture fraction. Case 2 solution. The DNS data correspond
to an arbitrarily chosen time in the simulation and are
compared to two reference curves: the adiabatic Burke
Schumann solution (squares, Eq. (A.4)) and one laminar
ame prole (circles). The laminar prole corresponds to a
steady, one-dimensional, plane, laminar counterow ame,
with
st
=36.3 s
1
.
Fig. 7. Scatterplot of ame temperature versus fuelair mix-
ing rate. Case 2 solution. The DNS data correspond to an
arbitrarily chosen time period of 10 ms in the simulation
and are compared to reference data points (circles). The ref-
erence data points correspond to steady, one-dimensional,
plane, laminar counterow ames.
(Eq. (A.4) in Appendix A) and one laminar ame pro-
le are also plotted for reference; the ame prole is
taken from the laminar diffusion ame database de-
scribed in Section 2 and corresponds to a high (albeit
subcritical) value of the mixing rate
st
. Fig. 6 shows
that the temperature variations in the present ame
wall conguration are laminar-like, well-correlated
with Z, and essentially similar to those observed in
traditional wall-free congurations [3537].
A slightly different perspective is adopted in Fig. 7
where the variations of the ame temperature T
st
468 Y. Wang, A. Trouv / Combustion and Flame 144 (2006) 461475
are plotted as a function of the mixing rate
st
.
Flame temperatures are dened as the values of T
conditioned on being on the stoichiometric isocon-
tour of mixture fraction, Z = Z
st
. Data points from
our precomputed database of strained laminar diffu-
sion ames are also plotted for reference. Fig. 7 re-
veals that in case 2, the amewall conguration fea-
tures occasional transcritical ame-ow conditions,
characterized by large values of
st
, up to
ad
st,ext

45 s
1
, and by signicant reductions in ame tem-
perature, down to less than 1200 K. However, as illus-
trated in Fig. 4, the transcritical events are not strong
enough and the turbulent ame remains extinction-
free. These results are consistent with observations
made in previous studies of unsteady (oscillating)
strained laminar diffusion ames, in which it is shown
that the instantaneous occurrence of supercritical con-
ditions is a necessary but not a sufcient condition to
lead to ame extinction (see, for instance, Ref. [39]).
It is shown in Ref. [39] that supercritical conditions
have to be sustained over a sufciently long period of
time for extinction to occur.
We now turn to the isothermal cold wall case
(cases 3 and 4 in Table 1). Cases 3 and 4 exhibit
two new features: ame extinction and nonadiabatic
combustion conditions (Fig. 5). Both features con-
tribute to make the description of uid temperature
variations signicantly more complex: the occur-
rence of ame extinction results in large deviations
from chemical equilibrium; and the basic analogy
between mass and heat mixing is lost under nona-
diabatic conditions. As explained in Appendix A, the
BurkeSchumann coupling relations between reactive
species mass and thermal energy are no longer valid
in nonadiabatic ames, and the several denitions
of mixture fraction, as a fuelair, fueltemperature,
or oxygentemperature mixing variable, are all dif-
ferent. The developments presented in Appendix A
suggest that even in the framework of fast chemistry,
more than one variable is now needed to describe the
thermal ame structure.
Fig. 8 presents a scatterplot of uid temperature
versus the fuelair-based mixture fraction. In con-
trast to Fig. 6, where T and Z exhibit a high level
of correlation, the same variables in Fig. 8 now ap-
pear uncorrelated. Low temperature levels on the fuel
side of the ame, Z 0.15, result from (direct) wall
cooling effects, while low temperature levels in the
vicinity of the ame zone, Z Z
st
, or on the air
side of the ame, Z Z
st
, are associated with ame
extinction events. The increased complexity in the de-
scription of temperature variations is also apparent
in Fig. 9, where the ame temperatures are plotted
versus the stoichiometric fuelair mixing rate. Fig. 9
conrms that in case 4, the amewall conguration
features a signicant probability of ame extinction,
Fig. 8. Scatterplot of temperature versus fuelair-based mix-
ture fraction. Case 4 solution. The DNS data correspond to
an arbitrarily chosen time in the simulation, and are com-
pared to the adiabatic BurkeSchumann solution (squares,
Eq. (A.4)).
Fig. 9. Scatterplot of ame temperature versus fuelair mix-
ing rate. Case 4 solution. The DNS data correspond to
an arbitrarily chosen time period of 10 ms in the simula-
tion, and are compared to reference data points (circles).
The reference data points correspond to (adiabatic) steady,
one-dimensional, plane, laminar counterow ames.
as characterized by large reductions in ame temper-
ature, down to less than 500 K. In Fig. 9, intermediate
temperature levels, 600 K T 1600 K, may be in-
terpreted as transient extinction or reignition events.
To gain further insight into the wallame tem-
perature variations, we now consider case 3, in which
ame extinction is observed (Fig. 5) but is less fre-
quent than in case 4. We also limit our discussion to
an instantaneous snapshot from the case 3 solution
and select a time in the simulation that is extinction-
free, t = t
ef
. Under ame-extinction-free conditions,
we expect the temperature variations to be fairly well
approximated by the extended BurkeSchumann so-
lution discussed in Appendix A. In Appendix A, the
Y. Wang, A. Trouv / Combustion and Flame 144 (2006) 461475 469
ame response to wall cooling is described in terms
of the mixture fraction Z and a normalized heat
loss variable H called the excess enthalpy variable
[40,41],
H =
c
p
(T T

)
H
c

1
Y

C
2
H
4
+
1
Y

O
2
/r
s

(3) +
Y
C
2
H
4
Y

C
2
H
4
+
Y
O
2
Y

O
2
1,
where Y

k
designates the value of the species k mass
fraction in the k-supply stream; r
s
the stoichiometric
oxygen-to-fuel mass ratio; and T

the free-stream
temperature. H is a nondimensional quantity that
varies between (1) and 0. Under adiabatic condi-
tions, H = 0; under nonadiabatic conditions, H = 0
and the deviations of H from 0 give a local mea-
sure of the amount of heat loss resulting from the
wall cooling process. In Appendix A, the tempera-
ture variations are expressed as a function of Z and
H (Eqs. (A.4) and (A.12)):
T (Z, H) =T

C
2
H
4
H
c
c
p

(Z +HZ
st
),
0 Z Z
st
,
T (Z, H) =T

O
2
H
c
r
s
c
p

(1 Z) +H(1 Z
st
)

,
(4) Z
st
Z 1.
The validity of the extended BurkeSchumann so-
lution given in Eq. (4) is tested in Fig. 10. Fig. 10
presents three scatter plots of temperature versus mix-
ture fraction; the plots correspond to the preselected
time t = t
ef
and to three different x-locations. The
adiabatic BurkeSchumann solution is also plotted
for reference. Fig. 10 shows the expected downstream
decrease in values of Z as gases from the fuel supply
stream mix with, and get diluted by, the surround-
ing gases from the air stream. Fig. 10 also shows
the downstream increase in values of H as the near-
wall cold sublayer grows in size and interacts with the
ame more deeply.
These results are consistent with the plot shown in
Fig. 5c (with the difference that Fig. 5 features two
ame-extinction events). Fig. 5c compares the loca-
tion of the wall-cooled region (i.e., the region that ex-
hibits negative values of H) to that of the ame (i.e.,
the stoichiometric fuelair interface) and thereby pro-
vides a helpful graphic representation of occurrences
of thermal amewall interactions: ame elements
that lie in the white or light gray regions in Fig. 5c
are quasi-adiabatic, whereas ame elements that lie
in the dark gray region are strongly nonadiabatic and
susceptible to wall-induced extinction events.
(a)
(b)
(c)
Fig. 10. Scatterplots of ame temperature versus fuelair-
based mixture fraction. Case 3 solution. The DNS data corre-
spond to a time in the simulation that is ame-extinction-free
and to three different x-locations: (a) x = 0; (b) x = 1.97;
(c) x =7.9 cm. The DNS raw data (diamonds) are also com-
pared to the adiabatic BurkeSchumann solution (solid line,
Eq. (A.4)) and to reference data (squares) processed accord-
ing to the extended BurkeSchumann solution in Eq. (4).
470 Y. Wang, A. Trouv / Combustion and Flame 144 (2006) 461475
We now return to Fig. 10. Each plot in Fig. 10
also compares DNS raw data with processed data.
Processed data correspond to estimates of uid tem-
perature using Eq. (4) and the local values of Z
and H. Figs. 10a and 10b reveal some discrepancies
between raw and processed data near Z = Z
st
; these
discrepancies may be explained by nite-rate chem-
istry effects (i.e., effects of
st
in a amelet descrip-
tion), which are unaccounted for in Eq. (4). Despite
these discrepancies, the overall agreement between
raw and processed data is good, and the compari-
son lends support to the theoretical developments pre-
sented in Appendix A.
Thus, we may conclude that in the absence of
ame extinction, nonadiabatic temperature variations
may be described as a function of mixture fraction
Z and excess enthalpy H. In Section 3.2, we ex-
tend this discussion to cases with ame extinction.
We suggest that the classical amelet description may
be adapted readily to treat this case, using Z, H, and

st
as controlling variables, with the difference that
ame extinction is then predicted with a modied cri-
terion.
3.2. Wallame extinction
As discussed above, the adiabatic cases 1 and 2
remain extinction-free (Figs. 3 and 4). Case 2 does
feature occasional occurrences of transcritical ame-
ow conditions (Fig. 7), but these occurrences lack
adequate strength and fail to fully overpower the
ame. In contrast, the nonadiabatic cases 3 and 4 fea-
ture a number of local ame extinction events (Fig. 5).
Thus, a simple comparison between cases 1 and 2 and
3 and 4 indicates that in our particular DNS congu-
ration, ame extinction is the direct consequence of
wall-cooling effects.
We focus our attention in this section on the
conditions that lead to ame extinction. We start
from the classical theory of (adiabatic) laminar dif-
fusion ames in which ame extinction is described
as a consequence of excessive values of the fuel
air mixing rate:
st

ad
st,ext
[3537]. The criti-
cal value
ad
st,ext
may be evaluated from theoretical
expressions [42], from detailed numerical calcula-
tions of one-dimensional strained laminar diffusion
ames (Fig. 1), or from experiments. A useful pre-
diction from high-activation-energy asymptotic the-
ory is that the critical value
ad
st,ext
is an exponen-
tially increasing function of the ame temperature,

ad
st,ext
exp(T
a
/T
ad
st
), where T
a
is a characteris-
tic activation temperature and T
ad
st
the adiabatic ame
temperature [42]. We now use this prediction to es-
tablish a correspondence between the different ame
responses to stretch under adiabatic and nonadiabatic
conditions.
In the presence of wall-induced heat losses, the
ame temperature T
st
is decreased, T
st
T
ad
st
, and
following the previous discussion, we expect the ex-
tinction value of the fuelair mixing rate to be lowered
accordingly:
st,ext

ad
st,ext
. Assuming an exponen-
tial relationship between
st,ext
and T
st
, we have the
relation

st,ext
=
ad
st,ext
exp(T
a
/T
st
)
exp(T
a
/T
ad
st
)
(5) =
ad
st,ext
exp

H
st
1 +H
st
(T
ad
st
T

)/T
ad
st

or
(6)
st,ext

ad
st,ext
exp(H
st
),
where is a characteristic Zeldovich number, =
T
a
(T
ad
st
T

)/(T
ad
st
)
2
( 5.57 in the present S3D
ame model, as described in Eqs. (1) and (2)), and
H
st
is the value of excess enthalpy at the ame loca-
tion, H
st
= H(Z
st
) (keep in mind that wall cooling
results in negative values of H). Using Eq. (6), ame
extinction is predicted when
(7)
st

st,ext

ad
st,ext
exp(H
st
).
Equations (6) and (7) provide a simple correction to
the adiabatic ame extinction criterion established in
the classical diffusion ame theory. The correction
takes the form of a multiplicative factor exp(H
st
);
this factor depends on the magnitude of the ame heat
losses, through H
st
, as well as on the sensitivity of the
ame chemistry to temperature variations, through .
Equation (7) implies that ame extinction conditions
are more readily achieved as the ame moves closer to
the wall and experiences larger negative values of H
st
.
We now turn to a DNS-based evaluation of the
wall-modied ame extinction criterion proposed in
Eq. (7). We consider case 4, in which frequent ame
extinction events are observed, and case 2, in which
occasional transcritical events are observed, but with-
out ame extinction. The evaluation of ame extinc-
tion conditions in case 4 is not straightforward since,
as seen in Fig. 9, the simulated ame exhibits a range
of combustion conditions that correspond to different
states: a burning state; a nonburning state; and two
transitional states, from burning to nonburning (ex-
tinction) and from nonburning to burning (reignition).
Therefore, to differentiate between all possible states
and to achieve our objective of focusing on extinc-
tion conditions, we need rst to develop a detection
scheme that can extract the incipient ame extinction
events from the DNS raw data.
Fig. 11 presents our ame extinction detection
scheme. The scheme may be applied to any instan-
taneous snapshot of the DNS solutions; it is based on
Y. Wang, A. Trouv / Combustion and Flame 144 (2006) 461475 471
Fig. 11. Flame extinction detection scheme. The scheme
consists in monitoring the variations of the ame weakness
factor, R = (
st
/
st,ext
) (
st
/
ad
st,ext
) exp(H
st
). The
upper plot shows the ame location in physical space, while
the lower plot shows the corresponding variations of R as
a function of arc length along the ame contour. The ame
weakest spots are readily identied as peak values of R in
the lower plot, and ame extinction is predicted to occur for
R above a critical value (i.e., R 1 in Eq. (7)).
(A) the identication of the instantaneous ame sur-
face as the stoichiometric isocontour of the fuelair-
based mixture fraction, Z =Z
st
; (B) the evaluation of
the local fuelair mixing rate
st
and excess enthalpy
H
st
along the entire ame surface; (C) the subsequent
evaluation of a ame weakness factor, dened as R =
(
st
/
st,ext
) (
st
/
ad
st,ext
) exp(H
st
); (D) and -
nally the identication of the ame weakest spots as
local peak values of R.
Once potential ame extinction events are identi-
ed as R-maxima, we can proceed to track their time
evolution by simply repeating the detection scheme at
different closely spaced time intervals. Note that the
proposed diagnostic remains Eulerian-based and is
different from, and much simpler than, a Lagrangian
tracking technique that would accurately follow a par-
ticular ame element; the Eulerian treatment presents
the advantage of tracking the ame locations that fea-
ture the most critical conditions, as measured by the
weakness factor R, and it remains unclear whether a
Lagrangian treatment would provide more valuable
insight. The weakest spots are tracked as they are
convected downstream and out of the computational
domain; the data processing activated during tracking
consists in recording the local values of temperature
at the different marked ame locations.
Consistent with the criterion proposed in Eq. (7),
we nd that in case 4, observations of ame extinction
events are well-correlated with occurrences of large
peak values of the ame weakness factor R (similar
observations are made in case 3). In Eq. (7), extinc-
tion is predicted to occur for a critical value of R
equal to 1. This prediction is now tested in Fig. 12.
Fig. 12. Variations of the ame temperature T
st
with
the ame weakness factor R at selected ame locations.
Case 2 (triangles) and case 4 (lines) solutions. The DNS
data correspond to several ame weak spots, as identied
using the detection algorithm presented in Fig. 11. The
DNS data are compared to reference data points (circles).
The reference data points correspond to (adiabatic) steady,
one-dimensional, plane, laminar counterow ames.
Fig. 12 presents the time variations of temperature,
as recorded at several ame weak spot locations, and
using the detection algorithm of Fig. 11. To facilitate
the comparison with Eq. (7), the time variations of
T
st
are plotted as a function of R, instead of time.
Fig. 12 includes one transcritical event from case 2
and three supercritical events from case 4; data points
from our precomputed database of strained laminar
diffusion ames are also plotted for reference. The
plot displays the classical evolution of temperature
as the ame experiences a change from sub to super-
critical conditions: the ame temperature exhibits a
gradual decrease at rst, for low-to-moderate values
of R, followed by an abrupt fall-off (down to ambient
temperatures), as R is further increased beyond a crit-
ical value denoted R
c
and the ame undergoes sudden
extinction.
In agreement with Eq. (7), Fig. 12 shows that the
trans- and supercritical ame events from cases 2
and 4 are characterized by large values of R close to
unity. In contrast to Eq. (7), however, ame extinction
is observed for a range of values of R
c
, rather than a
unique distinct value: we nd that 0.3 R
c
1.2.
Also, there is some level of disagreement between re-
sults from cases 2 and 4: while in Fig. 12, the ame
event from case 2 features large values of R, up to
more than 1, and yet remains chemically active, one
of the ame events in case 4 undergoes complete ex-
tinction at a signicantly lower value of the ame
weakness factor, R
c
0.30.5.
Despite these discrepancies, we nd that the DNS
results lend support to the modied ame extinction
criterion proposed in Eq. (7). In our view, this crite-
472 Y. Wang, A. Trouv / Combustion and Flame 144 (2006) 461475
rion provides a useful quantication of the magnitude
of the ame weakening resulting from the wall cool-
ing process.
3.3. Wall surface heat ux
We now turn to an evaluation of the wall surface
heat ux, as obtained in cases 3 and 4 of the DNS
database. As discussed in previous sections, thermal
radiation transport is neglected in the present study
and gas-to-solid heat exchanges are limited to con-
vective heat transfer. Also, the DNS simulations are
well resolved and the wall heat ux is simply ob-
tained from its classical Fourier-law expression and
direct differentiation of the temperature eld.
Fig. 13 presents the statistical distribution of the
gassolid heat ux, as obtained in cases 3 and 4. The
probability density functions in Fig. 13 correspond to
data collected on the entire wall surface, and over a
time period of 20 ms. In both cases 3 and 4, the aver-
aged and peak values of the wall heat ux are approx-
imately 40 and 90 kW/m
2
, respectively. These values
are much smaller than those reported in Ref. [14]; the
peak heat ux in Ref. [14] is as high as the ame
reference power, i.e., up to 13 MW/m
2
for typical
hydrocarbon ames at atmospheric pressure.
Discrepancies between the present results and
those reported in Ref. [14] may be due to differences
in the respective ame-ow congurations, as well as
differences in the adopted combustion models. For in-
stance, our present choice of a simplied ame model
results in an underestimate of the ame resistance
to stretch (Fig. 1) which may lead in turn to over-
predictions in minimum amewall distances and
underpredictions in the associated wall heat uxes.
Fig. 13. Probability density function (PDF) of the wall sur-
face heat ux. Case 3 (solid line) and case 4 (dashed line) so-
lutions. The PDF data are accumulated both in space, along
the entire wall surface at y =0, and in time, over a time pe-
riod of 20 ms.
We plan to address this issue in future work by per-
forming DNS simulations with a more realistic ame
chemistry model.
4. Conclusions
Direct numerical simulation is used in this study
to obtain basic information on the interactions of non-
premixed ames with adiabatic or cold wall surfaces.
The simulations correspond to momentum-driven
ethyleneair turbulent wallames and feature sub-
critical ame dynamics, in the case of an adiabatic
wall, and supercritical ame dynamics, in the case of
a cold wall.
The different simulated wall ames are analyzed
in the present study with a amelet view point. In the
adiabatic wall case, the ame chemical and thermal
structure is described in terms of classical variables,
i.e., mixture fraction and scalar dissipation rate. In the
cold wall case, the description of the ame thermal
structure requires the introduction of an additional
variable, called excess enthalpy. The excess enthalpy
concept was apparently rst applied to the descrip-
tion of nonadiabatic diffusion ames in Refs. [40,41].
In agreement with Refs. [40,41], we nd that in our
amewall conguration, the excess enthalpy vari-
able provides a convenient description of the amount
of heat loss resulting from the wall cooling process.
Using mixture fraction and excess enthalpy as princi-
pal variables, an extended BurkeSchumann descrip-
tion of the ame structure, valid in the limit of equi-
librium chemistry, is presented in Appendix A.
A modied ame extinction criterion is then pro-
posed, which provides a correction factor to the clas-
sical scalar dissipation rate criterion obtained in adia-
batic ame theory. This correction factor depends on
the magnitude of the wall-induced ame heat losses,
through the excess enthalpy, as well as on the sensitiv-
ity of the ame chemistry to temperature variations,
through a Zeldovich number. The modied ame ex-
tinction criterion is tested against the DNS data using
an innovative detection algorithm designed to identify
the locations of the ame weakest spots. We nd that
the results are encouraging and lend support to an ex-
tended amelet framework enhanced by the concept
of excess enthalpy.
Additional results include a quantication of the
mean and peak values of the convective wall heat ux
resulting from the short amewall distances. These
values are in the range of a few tens of kW/m
2
and
are much smaller than the maximum heat ux re-
ported in a similar study, but for a highly strained
laminar diffusion counterow ame [14]. Differences
between our results and those of Ref. [14] remain
unexplained and will be the focus of follow-up stud-
Y. Wang, A. Trouv / Combustion and Flame 144 (2006) 461475 473
ies. It is worth emphasizing again that the present
numerical formulation includes a number of simpli-
cations/limitations: a two-dimensional computational
domain; a single-step ame chemistry model; no het-
erogeneous gassolid surface chemistry; no soot for-
mation; and no thermal radiation transport. Future
work will be aimed at removing those limitations,
with the goal of achieving more quantitative simula-
tions.
Acknowledgments
This work was supported by the Department of
Energy, Ofce of Basic Energy Sciences, SciDAC
Computational Chemistry Program (Grant DE-FG02-
01ER15227). Interactions with other participants in
our SciDAC project (see Ref. [31]) are gratefully ac-
knowledged.
Appendix A. An extension of the
BurkeSchumann description of diffusion ame
structure to the case of nonadiabatic combustion
The diffusion ame structure near a solid wall sur-
face may be signicantly different from that found in
traditional wall-free congurations. We discuss in this
section some of the complications found in the de-
scription of diffusion ames evolving near cold wall
boundaries. We limit our discussion to the classi-
cal BurkeSchumann theoretical framework for non-
premixed combustion, in which equilibrium chem-
istry is assumed and the ame is controlled by the
fuelair mixing process [3537]. Furthermore, we
limit our discussion to the amewall conguration
of Fig. 2, in which deviations from adiabatic behav-
ior are due to convective heat transfer at the wall, with
the wall located at the y =0 boundary. Equal molecu-
lar diffusion coefcients for mass and thermal energy
are also assumed, and following standard practice, a
nondimensional fuelair mixing variable, known as
the mixture fraction and denoted Z, is used as prin-
cipal variable,
(A.1) Z =
Y
C
2
H
4
(Y
O
2
/r
s
) +(Y

O
2
/r
s
)
Y

C
2
H
4
+(Y

O
2
/r
s
)
,
where Y

C
2
H
4
is the fuel mass fraction in the fuel sup-
ply stream; Y

O
2
the oxygen mass fraction in the air
free stream; and r
s
the stoichiometric oxygen-to-fuel
mass ratio. Z is a passive scalar with well-dened
boundary conditions: Z = 0 in the air stream; Z = 1
in the fuel stream; and Z/y =0 at the wall. In the
BurkeSchumann framework, the full reactive mix-
ture composition is uniquely described in terms of Z;
for instance, the fuel and oxygen mass fractions are
given by
Y
C
2
H
4
=0, Y
O
2
=Y

O
2
(1 Z/Z
st
),
0 Z Z
st
,
Y
C
2
H
4
=Y

C
2
H
4
(Z Z
st
)
(1 Z
st
)
, Y
O
2
=0,
(A.2) Z
st
Z 1,
where Z
st
is the stoichiometric value of the mixture
fraction, Z
st
=Y

O
2
/(r
s
Y

C
2
H
4
+Y

O
2
).
We now turn to the problem of describing the
uid temperature variations. Following the classi-
cal BurkeSchumann theory, we consider some al-
ternative denitions of mixture fraction using fuel
temperature or oxygentemperature mixing variables,
Z
2
=
Y
C
2
H
4
Y

C
2
H
4
+
c
p
(T T

)
Y

C
2
H
4
H
c
,
(A.3) Z
3
=

1
Y
O
2
Y

O
2

c
p
(T T

)
(Y

O
2
/r
s
)H
c
,
where T

is the air free-stream temperature (as-

sumed equal to the fuel-supply stream temperature)
and c
p
the specic heat of the reactive mixture at con-
stant pressure (assumed constant). Z
2
and Z
3
satisfy
the same convectiondiffusion equation that Z does;
these variables also satisfy the same well-dened
boundary conditions in the reactant supply streams:
Z
2
= Z
3
= 0 in the air stream, Z
2
= Z
3
= 1 in the
fuel stream. The wall boundary conditions, however,
may or may not be the same as those for Z.
Let us rst consider the adiabatic wall case. In this
case, the wall boundary conditions for Z
2
and Z
3
are
identical to that for Z, Z
2
/y = Z
3
/y = 0, and
we have the trivial result Z = Z
2
= Z
3
. This result
leads in turn to the classical BurkeSchumann tem-
perature solution, in which T is a piecewise linear
function of Z [3537]:
T
ad
=T

C
2
H
4
H
c
c
p

Z, 0 Z Z
st
,
T
ad
=T

O
2
H
c
r
s
c
p

(1 Z),
(A.4) Z
st
Z 1.
In Eq. (A.4) and in what follows, we use the ad super-
script to refer to this adiabatic BurkeSchumann solu-
tion, T
ad
(Z). Equations (A.2) and (A.4) show that in
the adiabatic case, the Z-structures of the wallame
and wall-free ame are identical.
Next we consider the case of a cold wall sur-
face with a prescribed temperature, T (x, 0, t ) = T
w
.
474 Y. Wang, A. Trouv / Combustion and Flame 144 (2006) 461475
In that case, Z
2
and Z
3
are found to satisfy more com-
plex wall boundary conditions,
Z
2
y
=
(c
p
/)
Y

C
2
H
4
H
c
q

w
,
(A.5)
Z
3
y
=
(c
p
/)
(Y

O
2
/r
s
)H
c
q

w
,
where is the mixture thermal conductivity and q

w
the solid-to-gas heat ux ( q

w
is negative in the present
problem). The boundary conditions in Eq. (A.5) dif-
fer from the zero normal-gradient condition satised
by Z. The implications are then two fold: rst, Z, Z
2
,
and Z
3
are not identical variables; and second, the
temperature is not a function of Z only, as was the
case in Eq. (A.4).
The deviations of Z
2
and Z
3
from Z may in fact
be interpreted as a direct measure of the amount of
heat loss resulting from the wall-cooling process. Be-
fore we make that point, it is worth noting that Z, Z
2
,
and Z
3
are related by
(A.6) Z =Z
3
Z
st
+Z
2
(1 Z
st
).
In the following, we stop discussing Z
2
and use Z and
Z
3
as principal variables.
Combining the denition of Z
3
in Eq. (A.3) with
the Z-variations of oxygen mass fraction in Eq. (A.2),
the temperature variations under nonadiabatic com-
bustion conditions may be expressed as
T =T

O
2
H
c
r
s
c
p

Z
Z
st
Z
3

,
0 Z Z
st
,
T =T

O
2
H
c
r
s
c
p

(1 Z
3
),
(A.7) Z
st
Z 1,
where T is now described as a function of Z and Z
3
.
Equation (A.7) may be recast in yet another form, in
which temperature is expressed as the sum of the adi-
abatic BurkeSchumann solution plus a perturbation:
(A.8) T =T
ad
(Z)

O
2
H
c
r
s
c
p

(Z
3
Z).
In this expression, temperature is now described as a
function of both the fuelair-based mixture fraction
Z and the difference (Z
3
Z). The role of the lat-
ter quantity is now more readily understood: under
adiabatic conditions, (Z
3
Z) =0 and T =T
ad
(Z),
whereas under nonadiabatic conditions, (Z
3
Z) =0
and T T
ad
(Z).
An alternative form of Eq. (A.8) is
(A.9) (Z
3
Z) =
T
ad
(Z) T
T
ad
st
T

(1 Z
st
),
where T
ad
st
is the adiabatic ame temperature, T
ad
st
=
T
ad
(Z
st
). Equation (A.8) shows that the maximum
value of (Z
3
Z) is (1 Z
st
), and therefore suggests
using the renormalization
(A.10) H
m
=
(Z
3
Z)
(1 Z
st
)
=
T
ad
(Z) T
T
ad
st
T

,
where H
m
is a nondimensional heat-loss variable that
varies between 0 and 1, is equal to 0 under adiabatic
conditions, and is positive otherwise.
A similar quantity was previously introduced in
Refs. [40,41] in two different studies of nonadia-
batic diffusion ames. The ame conguration stud-
ied in Ref. [40] includes radiative heat losses, while
the conguration studied in Ref. [41] features con-
vective wall cooling. The quantity proposed in Refs.
[40,41] is called the excess enthalpy variable and is
denoted H. H is dened as
H =
c
p
(T T

)
H
c

1
Y

C
2
H
4
+
1
(Y

O
2
/r
s
)

(A.11) +
Y
C
2
H
4
Y

C
2
H
4
+
Y
O
2
Y

O
2
1.
Using Eqs. (A.2), (A.4), and (A.10), it is straightfor-
ward to show that H =H
m
. H is therefore a nondi-
mensional quantity that varies between (1) and 0, is
equal to 0 under adiabatic conditions, and is negative
otherwise. In the main body of the text, we follow the
choice made in Refs. [40,41] and use H to describe
the modications in ame structure due to the pres-
ence of a cold wall surface.
In conclusion, the BurkeSchumann temperature
variations are given by the two-parameter expression
(A.12) T (Z, H) =T
ad
(Z) +H

T
ad
st
T

,
where (1) H 0, and where T
ad
(Z) is given by
Eq. (A.4) and T
ad
st
= T
ad
(Z
st
). Equations (A.2) and
(A.12) provide a complete description of the ame
structure under nonadiabatic conditions, in the limit
of innitely fast chemistry.
References
[1] J.H. Lu, O. Ezekoye, R. Greif, R.F. Sawyer, Proc. Com-
bust. Inst. 23 (1990) 441446.
[2] O. Ezekoye, R. Greif, R.F. Sawyer, Proc. Combust.
Inst. 24 (1992) 14651472.
[3] T.J. Poinsot, D.C. Haworth, G. Bruneaux, Combust.
Flame 95 (1993) 118132.
[4] P. Popp, M. Baum, Heat Transfer and Pollutant Forma-
tion Mechanisms in Insulated Combustion Chambers,
SAE Technical Paper 952387, 1995.
[5] I. Wichman, G. Bruneaux, Combust. Flame 103 (1995)
296310.
Y. Wang, A. Trouv / Combustion and Flame 144 (2006) 461475 475
[6] G. Bruneaux, K. Akselvoll, T. Poinsot, J.H. Ferziger,
Combust. Flame 107 (1996) 2744.
[7] P. Popp, M. Smooke, M. Baum, Proc. Combust. Inst. 26
(1996) 26932700.
[8] P. Popp, M. Baum, Combust. Flame 108 (1997) 327
348.
[9] G. Bruneaux, T. Poinsot, J.H. Ferziger, J. Fluid Mech.
349 (1997) 191219.
[10] T.M. Alshaalan, C.J. Rutland, Proc. Combust. Inst. 27
(1998) 793799.
[11] P. Aghalayam, P.-A. Bui, D.G. Vlachos, Combust. The-
ory Modelling 2 (1998) 515530.
[12] S. Raimondeau, D. Norton, D.G. Vlachos, R.I. Masel,
Proc. Combust. Inst. 29 (2002) 901907.
[13] F. Dabireau, B. Cuenot, O. Vermorel, T. Poinsot, Com-
bust. Flame 135 (2003) 123133.
[14] A. de Lataillade, F. Dabireau, B. Cuenot, T. Poinsot,
Proc. Combust. Inst. 29 (2002) 775779.
[15] V. Babrauskas, in: Proc. Interam, WileyInterscience,
New York, 1993, pp. 5774.
[16] B.Y. Lattimer, in: SFPE Handbook Fire Prot. Eng.,
third ed., NFPA, 2003, pp. 2269.
[17] T. Ahmad, G.M. Faeth, J. Heat Transfer 100 (1978)
112119.
[18] T. Ahmad, G.M. Faeth, Proc. Combust. Inst. 17 (1979)
11491160.
[19] J.G. Quintiere, T.G. Cleary, Fire Technol. 30 (1994)
209231.
[20] G. Back, C. Beyler, P. DiNenno, P. Tatem, in: Proc.
Fourth Int. Symp. Fire Safety Sci., IAFSS, 1994,
pp. 241252.
[21] H.Z. You, G.M. Faeth, Fire Mater. 3 (1979) 140147.
[22] M.A. Kokkala, in: Proc. Third Int. Symp. Fire Safety
Sci., Elsevier, Amsterdam, 1991, pp. 261270.
[23] S.K. Lele, J. Comput. Phys. 103 (1992) 1642.
[24] C.A. Kennedy, M.H. Carpenter, Appl. Numer. Math.
14 (4) (1994) 397433.
[25] C.A. Kennedy, Appl. Numer. Math. 35 (3) (2000) 177
219.
[26] T. Poinsot, S.K. Lele, J. Comput. Phys. 101 (1992)
104129.
[27] M. Baum, T. Poinsot, D. Thevenin, J. Comput. Phys.
116 (1995) 247261.
[28] J.C. Sutherland, C.A. Kennedy, J. Comput. Phys. 191
(2003) 502524.
[29] C.S. Yoo, Y. Wang, A. Trouv, H.G. Im, submitted for
publication.
[30] R.J. Kee, F.M. Rupley, E. Meeks, J.A. Miller,
CHEMKIN-III: A Fortran Chemical Kinetics Package
for the Analysis of Gas-Phase Chemical and Plasma
Kinetics, Report SAND96-8216, Sandia National Lab-
oratories, 1996.
[31] Terascale High-Fidelity Simulations of Turbulent Com-
bustion with Detailed Chemistry, SciDAC Computa-
tional Chemistry Program, DOE, Ofce of Basic En-
ergy Sciences, available at: http://purl.org/net/tstc/.
[32] C.K. Westbrook, F.L. Dryer, Combust. Sci. Technol. 27
(1981) 3143.
[33] A.E. Lutz, R.J. Kee, J.F. Grcar, F.M. Rupley, OPPDIF:
A Fortran Program for Computing Opposed-Flow Dif-
fusion Flames, Report SAND96-8243, Sandia National
Laboratories, 1996.
[34] San Diego Mech., available at: http://maeweb.ucsd.edu/
combustion/cermech/.
[35] F.A. Williams, Combustion Theory, second ed., Benja-
minCummings, Redwood City, CA, 1985.
[36] T. Poinsot, D. Veynante, Theoretical and Numerical
Combustion, Edwards, Ann Arbor, MI, 2001.
[37] D. Veynante, L. Vervisch, Prog. Energy Combust. Sci.
28 (2002) 193266.
[38] S. Lee, S.K. Lele, P. Moin, Phys. Fluids A 4 (1992)
1521.
[39] A.F. Ghoniem, M.C. Soteriou, O.M. Knio, B. Cetegen,
Proc. Combust. Inst. 24 (1992) 223230.
[40] A. Ray, I.S. Wichman, Int. J. Heat Transfer 41 (1998)
30753085.
[41] R. Vance, I.S. Wichman, Int. J. Heat Transfer 43 (2000)
921933.
[42] N. Peters, Combust. Sci. Technol. 30 (1983) 117.