COMBUSTION

AND

FLAME

96: 1-21 (1994)

Dynamics of a Strongly Radiating Unsteady Ethylene Jet

Diffusion Flame
CAROLYN R. KAPLAN*
Chemistry Division, Naval Research Laboratory, Washington, D.C. 20375

SEUNG W. BAEK
Department of Aerospace Engineering, Korea Advanced Institute of Science and Technology, 373-1 Gusung-dong,
Yusung-ku, Taejon, Korea

ELAINE S. ORAN
Laboratory for Computational Physics and Fluid Dynamics, Naval Research Laboratory, Washington, D.C. 20375

JANET L. ELLZEY
Department of Mechanical Engineering, Universityof Texas, Austin, TX 78712
Time-dependent numerical simulations of an axisymmetric ethylene-air diffusion flame are used to quantify
the way in which radiation transport affects the development, structure, and dynamics of the flame. The
numerical model solves the time-dependent Navier-Stokes equations coupled to submodels for chemical
reaction and heat release (ethylene combustion), soot formation, and radiation transport. The soot formation
model includes algorithms for soot nucleation, surface growth, coagulation, thermophoresis, and oxidation.
The radiative heat flux is found by solving the radiative transfer equation using the Discrete Ordinates
Method and includes radiative effects from soot, CO 2 and H20. The model is tested by comparing
simulation results with previously published experimental data for a coflowing laminar ethylene-air flame.
Simulations of a higher-speed jet at 5 m / s show that radiative heat losses reduce the flame temperature,
which decreases the chemical heat release rate. The reduction in heat release rate decreases the volumetric
expansion, causing the flame to shrink considerably, and hence changes the overall temperature, species
concentration, and soot volume fraction distributions in the flame. The computations show that radiative
intensity is attenuated significantly within the heavily sooting region. Radiative heat flux vectors are primariy
directed in the radial direction; however, there is a significant axial component that follows the curvature of
the sooting region. The computations for an undiluted fuel jet show that heat transfer by radiation dominates
transfer by conduction and convection in the heavily sooting regions of the flame.

NOMENCLATURE
a

A
Cv
e
E

L
G
hk

absorption coefficient
surface area of control volume
heat capacity at constant volume
specific internal energy density
fluid energy density
soot volume fraction
gravitational acceleration
enthalpy of species k
heat of combustion

I
im

Ib
k
kc
n
nk
nd
No

P
Po2

Q
* Corresponding author: Dr. Carolyn Kaplan, Code 6183,
Naval Research Laboratory, Washington, D.C. 20375
Copyright 1994 by The Combustion Institute
Published by Elsevier Science Inc.

qc
qr

directional intensity
intensity in discrete ordinates
blackbody intensity
Boltzmann constant
thermal conductivity
total species number density
number density of species k
soot number density
Avogadros number
pressure
partial pressure of oxygen
energy released from chemical
reaction
thermal conductive heat flux
radiative heat flux
0010-2180/94/$6.00

2
r
R
Rox
s
S
t
T
w m

AV
V
Uk

~t
Xk
z

C . R . KAPLAN ET AL.
radial direction
universal gas constant
rate of soot oxidation
scattering coefficient
optical path
time
temperature
Gaussian quadrature weight
volume of control volume
fluid velocity
diffusion velocity of species k
thermophoretic velocity
mole fraction of species k
axial direction

Greek Symbols
a

intensity flow term in curved coordinate

/3
extinction (absorption + scattering) coefficient
II
direction of radiation
oJ~
production/loss of species k due
to chemical reaction
~on~
production/loss of soot number
density
toL
production/loss of soot volume
fraction
/~, ~:, r/
radial, axial, and azimuthal direction cosines
~"
interpolation factor used in
DOM
fluid density
P
density of soot particle
Psoot
7"
viscous stress tensor
( f V --, f~) scattering phase function

azimuthal angle
wavelength
A
E~
spectral emissivity
emissivity at boundary
ew
P
kinematic viscosity
opacity of layer of thickness S
K~(s)
INTRODUCTION
Numerical simulation of unsteady diffusion
flames is a challenge due to the difficulty of
resolving the very disparate time and space
scales of the controlling physical and chemical
processes, and insufficient knowledge of the

input data. The physical and chemical processes can cover time scales ranging over nine
orders of magnitude and space scales ranging
over five orders of magnitude. It is not practical to develop a direct numerical simulation
that can resolve the full range of relevant time
and space scales applicable for unsteady jet
diffusion flames. However, by taking advantage
of what is known about the physics and chemistry of diffusion flames, one can choose appropriate optimized algorithms with adaptive or
variable gridding techniques to develop a numerical model that is computationally feasible.
There have been a number of numerical
studies of steady-state laminar diffusion flames.
In some steady-state cases, the flame interface
is constant in space and time and the fuel and
oxidizer mix through diffusion of the reactants
into the flame zone. Most flames are, however,
unsteady or fluctuating and the mixing process
is more complex. First, buoyancy-driven lowfrequency (10-20 Hz) structures [1-5] form
outside the flame zone and result in flickering.
Also, when the jet velocity is high enough,
there are smaller, high-frequency structures
(200 Hz) at the interface between the highvelocity and low-velocity fluid that result from
Kelvin-Helmholtz instabilities [2-5]. Due to
these unsteady convective processes, fuel and
oxidizer mix as they are entrained by the
large-scale structures and then are convected
into the high-temperature region. Diffusive
processes then mix the reactants at the molecular scale where chemical reactions can occur.
Recently time-dependent axisymmetric numerical simulations of unsteady hydrogen-air
[6-8] and propane-air [9] diffusion flames have
been reported. Studies [7, 8] of the effects of
heat release, viscosity, and gravity on the dynamics of the hydrogen-air flame showed that
heat release and viscosity damp the highfrequency Kelvin-Helmholtz instabilities while
gravity (buoyancy) is responsible for the formation of the low-frequency outer structures
(flame flicker). Combined numerical and experimental investigations [3, 9] of propane-air
jet diffusion flames also demonstrated the importance of buoyancy in the formation of the
low-frequency oscillations.
In hydrocarbon flames, soot usually forms

STRONGLY RADIATING ETHYLENE DIFFUSION FLAME

and radiation becomes important to the dynamical and chemical processes. Radiative
properties of weakly radiating (nonluminous)
diffusion flames can be calculated with reasonable accuracy [10-12]. The laminar flamelet
concept [13, 14] for nonluminous flames is
based on the observation that scalar properties
in laminar diffusion flames are nearly universal
functions (state relationships) of mixture fraction. Hence, knowledge of the transient behavior of the mixture fraction is sufficient to obtain estimates of all transient scalar (gas species
concentrations, temperature, density) properties. This concept has been extended to nonluminous turbulent diffusion flames by viewing
them as wrinkled laminar flames having
the same properties [11, 15]. Therefore, direct
measurements of scalar properties in nonluminous laminar flames can provide the necessary
state relationships for nonluminous turbulent
flames.
Prediction of the properties of strongly radiating luminous flames remains a greater challenge. For these strongly radiating flames,
radiative heat transfer is dominated by continuum radiation from soot particles [11]. An important study of both the structure and radiation properties of vertically-upflowing laminar
and turbulent ethylene-air diffusion flames [16]
included experimental measurements of mean
and fluctuating velocities, mean concentrations
of major gas species, soot volume fraction,
monochromatic absorption, spectral emission,
and total radiative heat flux distributions. The
results showed that the major gas species followed nearly universal state relationships for
both the laminar and turbulent cases, but that
soot volume fraction only roughly followed such
a universal state relationship due to the effects
of hydrodynamics. Predictions of flame structure (using a Favre-averaged turbulence model
and the laminar flamelet approximation) compared favorably with measured flame structure.
However, significant differences between mean
property and stochastic radiation emission predictions (using a narrow-band model) indicated
strong turbulence-radiation interactions. The
internal redistribution of energy by radiation
is appreciable in these luminous flames [16].
Other experimental investigations [17, 18] have

studied the effects of flow rate, fuel type, and

temperature on soot formation of ethylene,
ethane, and methane flames. More recent experimental and theoretical studies [19-21] have
correlated local soot formation rates with mixture fraction and temperature in the sooting
regions of laminar ethylene and ethane diffusion flames. In addition, detailed measurements of mixture fraction, temperature, and
soot volume fraction have been used to develop a two-equation model of soot formation
for two-dimensional turbulent nonpremixed
ethylene-air [22] and methane-air [23] flames.
Other investigations [24-26] have stressed
the necessity of including accurate radiation
models in luminous flames. A four-flux model
with the gray gas assumption, used to compute
distributions of temperature and radiant heat
transfer in a furnace [24], showed that the
model accurately predicted these quantities at
furnace walls, but provided poor predictions
within the fluid flow. Another study [25] evaluated two models for predictings flame radiation in turbulent wall fires: the first model
assumed that the radiated power is a constant
fraction of the energy liberated per unit time
by chemical reaction, while the second model
assumed that radiation is emitted by a thin,
constant-temperature (1400 K) layer of soot
particles at the flame front. Comparison with
experimental data for PMMA showed that the
soot-band model was a more accurate predictor of pyrolysis rate and flame radiance. A
more recent study investigated the effect of
fully coupling the radiation calculation to the
conservation equation of mean total enthalpy
[26], including the radiative heat loss/gain
term. This study showed that only coupled
calculations provided good estimates of emission temperatures and radiation intensities for
luminous flames.
This article describes numerical simulations
of the nonsteady behavior of a strongly radiating axisymmetric jet diffusion flame formed
between undiluted ethylene and a coflowing
stream of air. The numerical model is based on
one originally developed for hydrogen jet diffusion flames [6-8], but now includes a chemical reaction and energy release model for ethylene oxidation and models for soot formation

C . R . KAPLAN ET AL.

and radiation transport. The computations are

carried out to study the effects of radiation
transport on the development, structure, and
dynamics of the flame.

q- V" ( n k V ) = - - V " (nkU k) q- Ca)k,

(4)

+ t7. ( n d V ) = --17. ('~tnd) + tO~d,

(5)

diffusive transport (thermal conduction, molecular diffusion, viscosity) terms. However, we

consider only the radial component of the
thermophoretic term in Eqs. 5 and 6, and only
the axial component of the gravitational acceleration term in Eq. 2.
These equations are then rewritten in terms
of finite-volume approximations on an Eulerian mesh and solved numerically for specified
boundary and initial conditions. A complete
solution to these governing equations requires
solving the terms for each of the individual
processes, as well as accounting for the interaction among the processes. The model consists of separate algorithms for each of the
individual processes, which are then coupled
together by the method of timestep splitting
[27]. The algorithms for convection, thermal
conduction, molecular diffusion, viscosity and
the coupling of the individual processes have
been previously discussed in detail [7], and
therefore are only briefly described here. The
new additions to the model, namely the chemical reaction, soot formation, and radiation
transport algorithms, are more thoroughly described in this article. In addition, a detailed
description of the implementation of the thermal radiation algorithm for this problem is
presented in an Appendix.

(6)

Convection

NUMERICAL METHOD AND

MODEL FORMULATION
The numerical model solves the time-dependent equations for conservation of mass density, momentum, energy, individual species
number densities, soot number density, and
soot volume fraction:
Op
-- + V . ( pV)
at

(1)

= O,

ap V
- + V.(pW)
cTt

= -VP

+ pG-

V.r,

(2)
aE
-+ V. (EV)
Ot

= -V.PV-

V . (q~ + q~)

- V" ~,nkF;kh k + Q,

an k
at

an d

-Ot

(3)

aL
--

at

+ V. (f~V) = -V.

( ~ , f ~ ) + tof,..

Equations 1-6 are closed by the ideal gas

relations:
P = nkT,

(7)

de = p C v d T .

(8)

Equations 1-4 include terms for convection,

thermal conduction, molecular diffusion, viscosity, chemical reaction and energy release,
gravity, and radiation transport. The soot conservation equations, Eqs. 5 and 6, include terms
for convection and thermophoresis, where the
thermophoretic velocity is defined by
vt = - 0.54v

0 lnT
Or

(9)

Our solution to Eqs. 1-6 includes both radial

and axial components of the convective and

The fluid convection is solved with a high-order

implicit algorithm, Barely Implicit Correction
to Flux-Corrected Transport (BIC-FCT), that
was developed to solve the convection equations for low-velocity flows [28]. The Flux-Corrected Transport (FCT) algorithm itself [27] is
an explicit, finite-volume algorithm that is constructed to have fourth-order phase accuracy.
Through a two-step predictor-corrector algorithm, FCT ensures that all conserved quantities remain monotone and positive. However,
because FCT is an explicit algorithm, the numerical timestep required for accuracy and
stability is limited by the velocity of sound
according to the Courant-Friedrichs-Lewy condition. To filter out the sound waves from the
convection equations and therefore remove the
sound-speed (Courant) limitation on the timestep, the convection equations are usually

STRONGLY RADIATING ETHYLENE DIFFUSION FLAME

solved implicitly. Patnaik et al. [28] developed
BIC-FCT so that the timestep is limited by the
fluid velocity and not the sound speed. This
implementation has great advantages for computations of slowly evolving flows because one
BIC-FCT timestep requires the same amount
of computer time as one regular FCT explicit
timestep, but the size of the timestep might be
a factor of 50-100 times greater.
Diffusive Processes

The effects of thermal conduction, molecular

diffusion, and viscosity are evaluated using
two-dimensional finite-differencing algorithms
[7]. Temperature-dependent thermal conductivities and viscosity coefficients are calculated
from kinetic theory [29] over a 300-3000 K
temperature range, and these values are fit to
a third-order polynomial for each individual
chemical component. Mixture rules are then
applied to calculate mixture thermal conductivities [30] and mixture viscosity coefficients [31]
for each cell. In addition, temperature-dependent binary diffusion coefficients are calculated from kinetic theory [29]. Diffusion coefficients of each individual component in a
mixture are then calculated for each cell [32].
Subcycling is used in the molecular diffusion
and thermal conduction modules to ensure numerical stability [7].

pression [33],
d[C2H4]
dt

4.3 X 1012exp(-30000/RT)
x [C 2H 4]0.1[ 0 2 ]1.65
(mol/cm3-s).

(11)

The rate of depletion of ethylene is calculated

from Eq. 11. Then, based on the rate of fuel
consumption, the corresponding concentrations of oxygen, carbon dioxide, water, and
nitrogen are calculated from the appropriate
stoichiometric coefficients in Eq. 10. The heat
release rate, Q, is determined from
Q -- - A H c

d[C2H4]
dt

(12)

As discussed below in the Results section, simulations were conducted with and without radiation transport to study its effects on the flame
dynamics and structure. For simulations conducted without radiation, it was necessary to
include a calibration factor of 0.9 in the Arrhenius type reaction rate, Eq. 11, to prevent it
from increasing without limits as the temperature increased. For the simulations conducted
with radiation, this calibration factor was not
necessary--that is, the radiative losses were
large enough to prevent the reaction rate from
increasing without bonds.

Chemical Reaction and Energy Release

Soot Formation

The production and loss of species is represented by the source term in Eq. 4. Due to the
large number of computational cells required
to resolve the complex flow structure of jet
diffusion flames, it would have been prohibitive to include the full set of elementary
reactions for ethylene oxidation in this model.
Instead, we describe the chemical reaction
and energy-release process phenomenologically based on the single step reaction,

The conservation equations for soot number

density and soot volume fraction, Eqs. 5 and 6,
include terms for convection, thermophoresis,
and source terms, o)n~ and wL" These source
terms are represented by two coupled ordinary
differential equations derived by Moss et al.
[22] based on experimental measurements of
mixture fraction, temperature, and soot volume fraction in ethylene-air diffusion flames.
This model includes terms for soot nucleation,
surface growth, and coagulation on the soot
formation rate:

C 2 H 4 -k 3 0 2 + (N 2) ~ 2CO 2

+ 2 H 2 0 + (Ne),

(lO)
using a finite-rate, quasi-global Arrhenius ex-

dnd
clt = NC~ p2Ta/2Xfuele- ~/T
- C~T1/2nd2/No,

(13)

C . R . KAPLAN ET AL.

-- = --n
a p T t/Z gfuele- Ty/ r
dt
#soot

C~ C a

+- -

p2 T1/2Xfuel e -

T./T,

(14)

#soot
where the soot particle density is assumed to
be 1.8 g / c m 3, and the coefficients and activation temperatures [22] are
C a = 1.7 108 cm3/(g2K1/2s),
Ct3 = 1 x 1015 c m 3 / ( K l / e s ) ,

soot number density, wn~, includes the effects

of nucleation, coagulation, and oxidation, while
the source term for soot volume fraction, oJf,
includes the effects of surface growth, nucleation, and oxidation. The source terms are
calculated at each timestep using the current
values for the gas temperature, density, fuel
mole fraction, and soot number density, to
calculate new values for soot number density
and volume fraction. These new values are
then used in the convection and thermophoresis terms in the conservation equations, Eqs. 5
and 6.

Cr = 4.2 x 10 -11 cm3/(Kl/Zs),

Radiation Transport

C~ = 144 x 103 g,
T,~ = 46.1 103 K,
Tr = 12.6 103 K.
We have extended this model to incorporate
the oxidation mechanism of Nagle and Strickland-Constable [34],
R ox = 12

1 +kzP %

X+kBPo2(1--X)

(g/cm3-s),

(15)

where
X=

1 + (kr/kB)Po2

(16)

and where the coefficients are defined [34] as:

k A = 20 exp( - 3 0 0 0 0 / R T ) ;
kn = 4.46 1 0 - 3 e x p ( - 1 5 2 0 0 / R T ) ;
k r = 1.51 1 0 5 e x p ( - 9 7 0 0 0 / R T ) ;
k z = 21.3 e x p ( 4 1 O O / R T )

This oxidation rate is then converted from

units of g/cm2-s into appropriate units of soot
number density (number of particles oxidized/
cm a of gas-s) and soot volume fraction (cm 3
soot oxidized/cm 3 of gas-s), assuming spherical particles with an average particle diameter
of 1 10 -6 m. These oxidation rate values are
then subtracted from the right-hand side of
Eqs. 13 and 14, so that the source term for

Because direct numerical modeling of radiation transport is very expensive, a number of

simplification such as the diffusion approximation or the transparent gas approximation have
been developed. For some nonluminous weakly
radiating flames, one can reasonably approximate radiation transport with a first-order optically thin model. However, for the strongly
radiating, luminous flames studied in this article, radiation transport cannot be adequately
described using an optically thin approximation. The Monte-Carlo [35] and Zone [36]
methods are two of the more commonly used
methods for calculating multidimensional radiative heat transfer, but have not been used
extensively in combustion applications due to
their large computational costs. In this article,
we model radiation transport using the Discrete Ordinates Method (DOM), first proposed
by Chandrasekhar [37], which we believe is as
accurate as the Monte-Carlo method, but requires much less computational time and
memory. DOM is a general algorithm that can
describe radiation transport in media which
are optically thin, thick or intermediate. It is
not necessary to make a preliminary estimate
of the optical thickness of the media being
studied since the DOM algorithm is valid for
any level of opacity. Another advantage of the
DOM algorithm is the ease with which it can
be readily incorporated into multidimensional
finite-volume codes. Most recently, D O M has
been used to study the effects of radiation
transport on the mechanisms of flame stabilization over a solid fuel plate [38, 39].

STRONGLY RADIATING ETHYLENE DIFFUSION FLAME

The effect of radiation transport appears in
the energy conservation equation, Eq. 3, as the
divergence of the radiative heat flux, ( 7 . qr)For a gray medium, (V qr) is evaluated from
--V'qr

a[f4I(r, lq)dll-47rlbl,

(17)

where a is the absorption coefficient, I(r, fl) is

the directional intensity, 12 is the direction of
radiation, and I b is the blackbody intensity.
For an optically thin medium, the integral term
in Eq. 17 is negligible, and the divergence of
the radiative heat flux may be calculated without having to solve the integrodifferential radiative transfer equation (RTE). However, for
strongly radiating flames that are not necessarily optically thin, one cannot neglect the integral term. To find I(r, f~)~ we solve the RTE
for gray gases,
( f l . V)I(r, 1))

sidered). In general, for two-dimensional symmetric geometry, the total number of ordinate
directions, M, is related to the order of the
approximation, N, through the relationship
M = N ( N + 2)/2 [40]. For the work presented
in this article, we use the S 4 method; hence 12
directions are evaluated. Previous tests [38, 39]
of the number of discrete ordinates required
show that the $4 approximation is accurate
while the S 2 approximation is not sufficiently
accurate. For the cases studied in Ref. 39, the
S6 and Ss approximations do not produce a
significant gain in accuracy compared with the
large increase in computational time required.
For axisymmetric cylindrical geometry, the
RTE is written for each individual ordinate
direction, m, as

txm O(ri,~)
r

Or

Oi,,
+~,,-Oz

1 0(~7,~i,~)
r

O0

= -[3i m + ai b + - ~ ~Wm'~m'mim',

= - [a + s]I(r, 12) + alb(r)

m'

(19)

- fa, _ J
+ -47r

(r, f~')~(l-l'

~ f~)dfl'.

(18)
where s is the scattering coefficient and ~(lq'
~ ) is the scattering phase function. This
says that the rate of change of intensity in the
direction of propagation, f~, is attenuated by
absorption and out-scattering (from direction
ft), and is enhanced by emission and inscattering (from direction fV to II). Since the
scattering coefficient, s, of soot particles is
negligible in comparison to the absorption coefficient, we neglect the effect of scattering in
this calculation, that is, s = 0.
We solve Eq. 18 for the intensity, I(r, ft),
using the discrete-ordinate approximation to
the RTE. This is obtained by discretizing the
entire solid angle (4~- steradians) using a finite
number of ordinate directions and corresponding weight factors. The RTE is written for each
ordinate and the integral terms are replaced by
a quadrature summed over each ordinate. This
method is sometimes referred to as the SN
approximation [40], where N represents the
order of the approximation (the number of
discrete values of direction cosines to be con-

where the values m and m' denote outgoing

and incoming directions, respectively, ~&m,~m,
and % are the direction cosines of a discrete
direction, and 0 is the angle of revolution
around the z-axis.
The $4 quadrature integrates the zero-, first-,
and second-order moments of the intensity distribution. The ordinate values for the S 4 approximation [40] are shown in Table 1. The
weight factor w = 0.5236 is used for all ordinate directions. A detailed description of the
implementation of the DOM algorithm using
the $4 approximation is presented in the Appendix.
This model includes radiative effects from
soot, CO 2, and H20. Based on their experimental data for ethylene-air diffusion flames,
Kent and Honnery [19] developed an expression for the absorption coefficient for soot as a
function of measured soot volume fraction and
temperature,
asoot = 2.66 const f,,T (cm- 1),

(20)

which we have used in our calculations (with

const = 7). Similarly, we have taken the ab-

C . R . KAPLAN ET AL.
TABLE 1
Gaussian Quadrature for the S 4 Approximation for
Axisymmetric Geometry

Direction

Radial
Component

Axial
Component

Azimuthal
Component

(m)

/~,.

~:,.

~,.

1
2
3
4
5
6
7
8
9
10
11
12

- 0.2959
0.2959
- 0.9082
- 0.2959
0.2959
0.9082
- 0.9082
- 0.2959
0.2959
0.9082
- 0.2959
0.2959

0.9082
0.9082
0.2959
0.2959
0.2959
0.2959
0.2959
0.2959
0.2959
0.2959
0.9082
0.9082

0.2959
0.2959
0.2959
0.9082
0.9082
0.2959
0.2959
0.9082
0.9082
0.2959
0.2959
0.2959

sorption coefficient for a mixture of C O 2 and

H 2 0 from the experimental and theoretical
work of Magnussen and Hjertager [41]
acoa+r~ao = 0.001(Xco= + Xr~o)

ter. The conditions of the computations discussed below are very similar.
The computational domain and initial conditions are shown in Fig. 1. The entire computational domain covers a region of 10 cm 15
cm for a grid containing 64 88 cells and
timesteps are on the order of 10 10 - 6 S.
Undiluted ethylene fuel flows at 5 c m / s
through a 1.4-cm-diameter burner, resulting in
a cold flow Reynolds number of 45. A coflowing stream of air flows at the same velocity
through a 10-cm-diameter outer ring. The
left-hand boundary is an axis of symmetry,
while the right-hand boundary is a free-slip
wall. The bottom boundary represents an inflow boundary condition, where the density,
velocity, and concentration of the incoming
species are specified. The inflow boundary corresponds to the region immediately above the

Outflow
illll11111111|no| |u | l l | | o | | a i m m
..~...moo|
Jmnlm|l.mmm.wl
.~....mHan fmHJguNm,la
.. m
m ~manJnl JwnJm| | ol ||u| lnHmamn i~ m
, . .mu m
,, ,. .. .uum. mu m
, l lammmu |nHnHmnwl au m a m l
,. ,, m
, , .mj .l.mmwl , oj m0 nnoHammwmmwmwui |wi mmH~

15 c m
(cm-l),

(2])

.~mmmmauonomm~iwn~H
,...m~m00nmnmmnnm~,n
,,.......mnn~nw~wn
..rmmmnn~Nwanmw~wn
...ummHw~oao~m~mWm~ma
',;',ll',',','g g p l ' l I g l l ~ ~ I I I I g l

where X corresponds to mole fraction. Based

on the work of Magnussen and Hjertager [41],
who summed the absorption coefficients for
soot and a mixture of CO2 and HaO to obtain
an overall absorption coefficient, we calculate
the overall absorption coefficient for each cell
as a sum of the individual absorption coefficients for soot and the mixture of CO~ and
H20.

liBFIB[ [ 1111101 Hi l i e [ [ O l l l l l
Ill
I~llllllll[l[lllllllllllllll]ll
IIIIlllllllllllllllllllllllllll
IIIllllllll[lll[lllllllllllllll[
I~OIIlllllllllllllllllllllllll[
IIHIIIIIIIIIIIIIIIIIIIIIIIIIIII
Ilil I I [ 1 0 0 l i l t i i Dei 11 [ [ [ l i i l l
In
IIIllllllllllllllllllllllllllll
,llllllllllllllllllllllllllllll
HIIIIIIIIIIIIIIIII[IIIIIIIIIII
411ll[llllll[llllllll[llllll/l[
I lil[ I Dl0glllDl[ [ I 0 i ] [ O I i i l [ I
11

II

...i,.mnnmmuolmHmmWl
...,....mml~am..~nH~wn
,..m..nul0lNmuWl||~na
i,.,...mlHll|lnmnuna
l.,ImmmHlln0H~UlW|.l
Ir',',l',lllllgllg:l:[| I I | ; ~ | - ' | | I
IIIIIIIIl'~Ig~|'lII:'Igl',I','.I~

.m

li Ji}Jl I I IIIIUi e l i i I I I l o l l i n g
i l[
FDIIIIII I I I I I I I I I I I I I I I I l l l l
I[I
II,ll,l[llll[lllllll[[[lllllllll
Iql~lllllOllilllll0llilllllllll[
IIIIIIIlllllllllllllll[llllllll[
,lllllllllllplllllllllllllllll[I
IIII IlU I I I l l l l l I l l [ I I I I I I I I I 1 1 1
II[ll]lllilllll0illliIOlOll[[lll
II]llllll[lllllllllllllilllllll[
IIIIIIIIIIIlllllllllllllllllllll
IHIIII[IIIIIIIIIII[II[IIIIII]I[
I?1~1111111111I l l I I I I I I I I I I I I I I I

y.

~,',llII',IIIl~'III~ I | " I I I | " - g "-I gl

I IIi1[11F I I I l l l I I [ I [ l l [ l l l l l l l l l
iiiiIIlllllllllllllllll[llllllll
I]lllllllllll[ll[lllllllllllllll
14lrlJ l 0 0 0 1 1 1 i i g l l l [ [ 0 l l l l l l l l l [
Illlll[1111111I11 I [ [ I I I I I [ l [ I I I I

BENCHMARK CALCULATIONS

',',',',gl;[g g g I g., I l l .".' .' I | -"," I I 1

FIHI n i l I [ [ l l l l l l l I I [ [ [ l [ l l l
I[[
i,iid FIll [ I [ I I I I I I I I I [ [ I I l l I I I 1 [

To provide a benchmark for the new algorithms for chemical reaction and energy release, soot formation and radiation transport
used in this model, we simulated a laminar
cofiowing ethylene-air diffusion flame and
compared the results with experimental data of
Gore and Faeth [16]. The experimental apparatus of Gore and Faeth [16] consisted of ethylene fuel flowing upward through a central
tube of 1.43 cm diameter with Reynolds numbers in the range of 45-63, while air flowed
from a concentric outer tube of 10.2 cm diame-

' , ' , ' , l l l l g l l g l : ' ; : : | | l p | | | "g~-'l

,,,,,.,,,,,,,.~ Ho ~| . . . ~ ~ n m n ,

tUttlailaluleei|iiieome~lo~im~l

',,,',,,ll~gl'.'.Ig|l:l| g l | l t

| | -'-" -'I

31cm

T
5 cm/s
C2H4

T
5 cm/s
Air

Fig. 1. Computational domain and initial conditions for

benchmark simulation of low-velocity laminar ethylene
diffusion flame. Note that this figure only shows a region
of high resolution. The full computational domain covers a
radial distance of 10 cm.

STRONGLY RADIATING ETHYLENE DIFFUSION FLAME

the mass fraction of fuel elements in the mixture at any point in the flow. Figure 2 shows
the formation of the buoyancy-driven lowfrequency structures that are convected along
the outer region of the flame. A time sequence
of oxygen mole fraction contours [42] shows
that the flicker frequency for this flame is
approximately 16 Hz. The maximum flame
temperature, 2000 K, is located in the region
where the CO 2 and H 2 0 mole fractions are
maximum. The maximum soot volume fraction
is 8 10 - 6 , and is within the high temperature region in an area where the mixture is
slightly rich of stoichiometric.
Radiation quantities at this same timestep
are shown in Fig. 3. The maximum absorption
coefficient is approximately 0.4 cm -1. These
results show that the absorption is emanating
primarily from the sooting region, so that the
soot, and not the C O 2 o r H20, is the dominant absorbing-emitting medium. The radiative
heat flux vectors show that the radiation transport is directed outward from the sooting region, and follows the curvature of the sooting
region. The length of the vector is directly
proportional to the magnitude of the radiative
heat flux; hence, the strongest radiative flux is
emanating from the sooting region.

nozzle exit; that is, the computational domain

does not include the nozzle itself. The top
boundary is a zero-gradient outflow boundary
condition. Boundary conditions for the radiation transport model include diffusely emitting-reflecting surface conditions at the inflow,
outflow, and right-hand boundaries,

(1
i,, = ewibw

Ew )

77"

G,, <0

[~m,lWm,im, ,

~m > 0.

(22)

At the axis of symmetry, the specularly reflecting boundary condition is used such that i m =
ira,, /,Zm, = --/Jt~m, ~m, = ~m, rim, ~- nm.

Figure 2 shows instantaneous contours of

mixture fraction, oxygen mole fraction, temperature, CO 2 mole fraction, HzO mole fraction, and soot volume fraction at timestep
40000. These results are not in steady state,
even for such a low-velocity flame; however,
this timestep is well past any critical stages of
transient development. The dashed line with
solid circles that are superimposed on the contours shows the location of the stoichiometric
flame surface. Mixture fraction is defined as
Mixture

Oz Mole

Temperature

Fraction

Fraction

(K)

14 cm

C O 2 Mole
Fraction

H20 Mole
Fraction

Soot Volume
Fraction X 10 -7

I
,+
i

L _ _ t _ _

3 cm

Fig. 2. Instantaneous contours at timestep 40000 for benchmark simulation. The location of the stoichiometric flame
surface is represented by the dashed line with solid circles.

10

C . R . KAPLAN ET AL.

Absorption
Coefficient
(cm-l)
14 cm

Magnitude
of Radiative
Heat Flux
(kW/m2)

Radiative
Intensity
(kW/m2)
I

Radiative Heat
Flux Vectors
I

~,##itl#

t i l t l l l l l l
, / / l l t l l l / I f
I t l l l l l l t / t l l

I l l / / / l l l l l l t

~tlllllllll/,
Illlllllill/.
///i/il/.~

:IT

7t t

i
r

I
i

0
3 cm
Fig. 3. Instantaneous contours of radiation quantities at timestep 40000 for benchmark simulation. The location of
the stoichiometricflame surface is represented by the dashed line with solid circles.

The simulations were also analyzed to determine if a state relationship exists between the
major gas species and fuel-equivalence ratio
and these were compared with the published
experimental data of Gore and Faeth [16].
Scatter plots were prepared in which the mass
fraction of the major gas species were plotted
as a function of local fuel-equivalence ratio at
three axial locations in the flame: 5, 8, and 12
cm. Figure 4 shows the state relationships that
were obtained from the simulations and from
the experiments of Gore and Faeth [16]. In
comparing the simulation results with the experimental measurements, it should be noted
that the numerical model neglects the formation and depletion of carbon monoxide. In the
stoichiometric region, ~b ~ 1, both the fuel and
oxygen are nearly depleted, as shown by the
simulation data and the experimental [16] data.
In the lean region, (th < 1), the simulation

results for the ethylene mass fraction closely

match those of the experiments [16]; in the
fuel rich regions, mainly for ~b > 2, the simulations slightly underpredict the ethylene mass
fraction. The agreement between simulations
and experiment [16] is quite good for the oxygen mass fraction as well. Although there is a
higher degree of scatter in the CO 2 and H 2 0
data, there is still reasonable agreement between the simulation and experimental measurements [16]. The maximum mass fractions
of CO2 and H 2 0 are located in the region of
stoichiometry, at ~b ~ 1, and decrease in the
lean and rich regions. In the rich region where
~b > 2, the H 2 0 mass fraction is slightly underpredicted compared to the experiments.
Figure 5, which shows the state relationship
data for soot volume fraction, shows more scatter for both the experimental [16] data and
simulation results. The higher degree of scatter

STRONGLY R A D I A T I N G E T H Y L E N E DIFFUSION F L A M E

11

!.

~Q
O
o

&

100

3"10"'

,2qe~'L.e,m-~

....

100

3"I0"

lO t

l~

m.
O

= Gore a n d F a e t h d a t a
A = Simulation

eq

-.

Izl

o.

3"10"

I I II

I II

10

10I

glq

e$
O
I

3 . 1 0 "t

S i I I

to*

III

to'

Fuel E q u i v a l e n c e Ratio
Fig. 4. Mass fraction of C2H 4, 02, CO 2, and H 2 0 versus local fuel equivalence ratio shows state relationships for
major gas species for benchmark simulation. Gore and Faeth data are extracted from Ref. 16.

can be attributed to effects of finite-rate chemistry and hydrodynamics [16]. The scatter in
the simulation data points is not systematic,
and could be due to variations in temperature
and fuel mole fraction for the three radial
traverses. Figure 5 shows that most of the soot
is formed in the region slightly rich of stoichiometric, and that the maximum value of soot
volume fraction is around ( 8 - 1 0 ) 10 -6,
which is in reasonable agreement with the experimental data [16].
UNSTEADY HIGH-VELOCITY
ETHYLENE FLAME
We now proceed to use the numerical model
to study unsteady higher-velocity flames. In
particular, we consider an axisymmetric flame
formed between a high-velocity (5 m / s ) fuel
jet flowing into a 30-cm/s air stream. This jet
velocity was chosen for this study as it repre-

sents the beginning of our evaluation of flame

liftoff phenomena, where the jet velocities
range from 5 to 50 m / s . Simulations were
conducted for two fuel jet mixtures: undiluted
ethylene and a nitrogen-diluted ethylene mixture (C2Ha:N2/3:l). The Reynolds number
(based on cold flow conditions at the fuel jet)
for this system is approximately 5000.
Computational Grid and Boundary Conditions
A considerably more resolved computational
grid was required for this high-velocity unsteady jet diffusion flame in order to resolve
the instabilities in the shear layer of the jet.
The region near the jet and the initial conditions are shown in Fig. 6. The full domain is
167 172 cm and is modeled on a 128 224
variably spaced computational grid. Cells of
0.02 cm 0.02 cm are concentrated around
the jet exit. The grid is then stretched [7, 8]

12

C . R . KAPLAN ET AL.
= Gore a n d Faeth data
= Simulation

Outflow
34.1 cm

!!ll~llr$1fllllllllllll
!!r:[l$1111111fll l il
iUiJJIJiillflililllll

:l;rll[lllrllllllllllllll
iiillllllllllllll[lllllll
:ilHIIIIIIIIIIIIIIIl|ll

~llllllllfflllllll|l|l
~l,iilJtlflllllllllifll
hrilllililllfliiillllil
I;IrrlrlllPHflfflllllfll
Ifll~lJJJlllllllllJUll|l
!fIilrlllllllll[llllllll

~"

a 8
~o

IIIIllll[llllllllllllll
i hllPlllllllllllllllll
~J~lllJllllllllllllllll
iIIiIllllrllllllllllllll
iJiiJiiiiiiiiiiiiiiin
:111111111111111111111
IlllllJIIIIIIllllllll
PIIHrlrlllllllllllllll
i i,llllllllllllllllllll
ilqlleltlllUlllllllll
ililllllillPlili~lllUl

i,;iiiiiifffi!iii$$$$i

.............
,,,,,,,,
.......... ,,,,,,,,,,,,
........ ,,,,,,,,,,,,,,,

o
0

'!::ll:llll~llllllff~l
l:ll~:l:~lllllllllll

?!?i:,!?il}i]]]ii}]]~]]

:::::::::::::::::::::
:=::::::::::::::::
A

0.5
*

10-~

1#

10'
Fuel Equivalence

I III

Ratio

Fig. 5. State relationship for soot volume fraction for

benchmark simulation. Gore and Faeth data and the solid
line (fit to the data in Ref. 16) are extracted from Ref. 16.

both radially and axially. The fuel flows through

a jet of radius 0.5 cm at 5 m / s , while air flows
through the outer annular region at 30 cm/s.
The boundary conditions for these simulations are the same as those described for the
benchmark case, except for the addition of a
pressure-control feedback process at the outflow boundary. The basic idea is that the axial
velocity at the outflow boundary is adjusted to
allow the flow to relax the ambient atmosphere, if the pressure within the flame becomes larger than atmospheric. Although the
height of this flame is approximately 1 m, the
grid is stretched to 1.7 m in the axial direction
to prevent disturbances created at the zero
gradient outflow boundary from affecting the
flame upstream. Although more elegant outflow boundary conditions [43] have been developed for high-velocity jets using an explicit
FCT algorithm [27], we have not yet developed
such boundary conditions for the implicit
(BIC-FCT) [28] algorithm.
F l a m e Structure and Radiative Properties

Figure 7 shows the instantaneous contours of

fuel mole fraction, temperature, soot volume

0.0
0.0

4.5

cm

Air, 30 cm/s

Fuel Mixture

5 m/s
Fig. 6. Computationaldomain and initial conditionsfor a
5 m/s C2H4-N2 jet flowinginto a 30 cm/s air stream.
Note that the figure only shows the part of the computational domainwith high resolution.The full computational
domain covers a region of 167 172 cm.

fraction, divergence of the radiative heat flux,

radiative intensity, and magnitude of the radiative heat flux and radiative flux vectors at
timestep 40,000, is well past any transient stages
of the flame development. The dashed line
with solid circles superimposed on the contours shows the location of the stoichiometric
flame surface. The maximum temperature of
the flame, 2050 K, is located at an axial distance of 4.5 cm from the base of the flame.
Figure 7 also shows the buoyancy-driven lowfrequency structures that are convected along
the outer region of this transitional flame.
The sooting region is located within the
high-temperature region, in an area which is
slightly rich of stoichiometric. Soot volume
fraction increases with axial position up to a
maximum value of 9 10 -6 at a height of 11
cm, and then gradually decreases at higher
axial positions. The region in the flame where
V . q r attains a maximum positive value lies in
the same region as the sooting zone, indicating
again, that soot, and not the CO 2 or H 2 0 , is
the dominant absorbing-emitting medium. As

STRONGLY RADIATING ETHYLENE DIFFUSION FLAME

in the intermediate regime between optically

thin and thick. The medium outside of the
sooting region is optically thin.
Figure 7 also shows the radiative flux vectors
and contours of the magnitude of the
2
radiative heat flux, defined as (qr. rad~a~ +
qr, ax~at2",1/2
S . The radiative heat flux vectors point
predominantly in the radial direction, normal
to the surface of the sooting region. The length
of each of the radiative flux vectors is proportional to its magnitude; the largest flux vectors
are located near the region of maximum soot
concentration. Although the radiative flux vectors point predominantly in the radial direction, they do have a significant axial component near the bottom tip of the sooting region.
Figure 8 shows the decrease in radiative flux
with radial distance from the flame at an axial
location of 10 cm. Within the heavily sooting
region, the simulations show radiatve heat
fluxes on the order of 90 k W / m 2. However, as
shown in Fig. 8, the heat flux decreases significantly with radial distance from the sooting
region. At 15 cm radial distance, the radiative
heat flux has already decreased by an order of
magnitude, and by 70 cm radial distance, it has
decreased by two orders of magnitude to a

shown in the energy conservation equation,

Eq. 3, a positive value of V'qr results in a
decrease in total energy density. Hence, the
region of maximum V.qr corresponds to the
region where the energy losses due to radiation are greatest.
Figure 7 shows that the radiative intensity is
greatest in the sooting region where it reaches
a maximum value of 2 10 z k W / m 2 and then
decreases with radial distance. These contours
show the strong attenuation of radiative intensity in the heavily sooting region. The opacity
of a gas is a measure of the ability of a given
path length of gas to attenuate radiation of a
given wavelength and is defined [35] as

iA(S) =

ia(O)exp[- Ka(S)],

(23)

where Ka(S) is the opacity of a layer of thickness S. Simulation results show that the maximum thickness of the sooting layer is approximately 0.5 cm. After passing through this
sooting layer, the radiative intensity is attenuated to approximately 50% of its original value,
corresponding to an opacity of approximately
0.7. Hence, in the heavily sooting region, the
medium is neither optically thin nor thick, but

Call4 Mole
Fraction

Temperature
(K)

Soot Volume
Fraction x 10"

13

V.q,)

x 102
(kW/rn 3)

Radiative
Intensity
(kW/m ~)

Magnitude of
Radiative Heat
Flux x 10~
(kW/m 2)

Radiative Heat
Flux Vectors

19 cm

- - 1
i

~'

L'* 'I~I~,\\',,\N\ \ \ x x x \ ' ~ x '

l
:
Lr,

,__

:7;:7'277
I
3

0
3 cm
Fig. 7. Instantaneous contours of flame properties for 5 m / s fuel jet coflowing into air. The location of the stoichiometric
flame surface is represented by the dashed line with solid circles.

14

C. R. K A P L A N E T AL.
T e m p e r a t u r e (K)

Soot Volume
Fraction x ltY~
r

o
~
,J

0.0

15.0
30.0
45.0
60.0
75.0
R a d i a l D i s t a n c e (era)
Fig. 8. Radiative heat flux (kW/m 2) versus radial distance
from sooting layer.
value of near 1 k W / m 2. Experimental measurements in e t h y l e n e - a i r diffusion flames also
show total radiative heat flux values ranging
from approximately 3 to 1 k W / m 2 at radial
distances ranging from 10 to 60 cm [16].
The radiative heat loss from the flame is
calculated by summing the product of the
radiative flux with the corresponding cross sectional area along the outer boundary (righthand side and outflow boundaries) of the computational domain. The radiative loss from the
flame is 6.5 kW for the undiluted fuel jet, and
5 kW for the nitrogen-diluted
case
(C2Hn:N2/3:l). These radiative heat loss values represent approximately 3 5 % - 4 0 % of the
chemical heat released. As expected, the radiative heat loss is greatest for the undiluted fuel
jet case as more soot is generated in that case.

19 cm

'-o

fi

Effect o f R a d i a t i o n on F l a m e Properties

Figure 9 shows contours of t e m p e r a t u r e and

soot volume fraction for a case where radiation
was excluded from (Case A) and included in
(Case B) the simulation. W h e n radiation transport is not included, the maximum flame temperature is 2200 K, and soot volume fractions
reach values of 25 x 10 -6 (when radiation is
not included in the calculation, the flame temperatures actually increase without bound due
to the Arrhenius type reaction rate; however,
by scaling the reaction rate by a factor of 0.9,

2 cm
Fig. 9. Instantaneous contours of temperature and soot
volume fraction at timestep 40000 for simulations conducted with and without radiation. The contour interval is
deliberately maintained at the same value for each contour
type (200 K for temperature contours, 20 10 7 for soot
volume fraction contours) to show the effect of radiation
on flame sheet and sooting layer thickness.

STRONGLY RADIATING ETHYLENE DIFFUSION FLAME

the temperature leveled off at a value of 2200
K). However, when radiation is included, the
maximum flame temperature decreases to 2050
K and the maximum soot volume fraction attained is 9 10 -6, a decrease by a factor of
three.
Figure 9 quantifies the fact that one of the
most significant effects of radiation transport is
to shrink the flame. When radiation is included, the radiative heat losses reduce the
flame temperature, which reduces the chemical
heat release rate, which, in turn, reduces the
volumetric expansion, thus causing the flame
to shrink. As the flame shrinks, the overall
temperature distribution in the flame changes,
which, in turn, changes the distribution of
species concentrations and soot volume fraction.
Figure 10 shows radial profiles of temperature and soot volume fraction at an axial location of 10 cm. The high-temperature region of
the flame is significantly narrower when radiation is included. Likewise, the sooting region is
narrower, and the quantity of soot is considerably reduced due to the lower temperatures.
When radiation is included, the sooting region
is located closer to the flame centerline, as the
overall flame width is narrowed by the radiative losses.

15

Radiative Versus Conductive Heat Fluxes

Figure 11 shows radial profiles of the magnitudes of radiative heat flux (qr, as defined in
the radiation transport section) and conductive
heat flux (qc -- - k c AT) for cases of an undiluted and nitrogen-diluted (C2H4:N2/3:l) fuel
jet, at axial locations of 4 cm (below the sooting region) and 10 cm (within the heavy sooting region). The magnitude of qc is maximum
inside of the flame sheet where the radial
temperature profile is sharply increasing. Then
qc decreases approximately two orders of magnitude within the flame sheet itself, where the
maximum temperature is maintained and
therefore the thermal gradient is reduced, and
then increases immediately outside of the flame
sheet where the radial temperature profile
sharply declines. Further outside the flame
sheet (toward the coflow region), there is a
very small thermal gradient and q~ is again
very small.
The behavior of the radiative heat flux, qr, is
very different. The value of qr is low at the
flame centerline, then sharply increases in the
sooting region, and then gradually decreases
with distance from the sooting region toward
the coflow region. For the undiluted fuel case,
more soot is generated and the radiatve flux is

o = without radiation
A = with radiation
i
tl,
0

0.2

0.6
1.0
1.4
l.B
2~.
Radial Distance (cm)

0.2

0.6
1.0
1.4
1.8
2.2
Radial Distance (cm)

Fig. 10. Radial profile of temperature and soot volume fraction at 10 cm axial distance for cases with and
without radiation.

16

C . R . KAPLAN ET AL.

10 cm

I ~

10 cm

gr,

,.,I

-1
0.0

'~

0.6

1.2

1.8

2.4

3.0

4cm [

ro,

,
0.0

0.6

12

1.8

2.4

3.0

Radial Distance (em)

"~ 0.0

0.6

1.2

1.8

2.4

3.0

4cm

8to
~

:
0.0

0.6

1.2

1.8

2.4

3.0

Radial Distance (era)

Fig. 11. Magnitudeof radiative and conductiveheat fluxesat heightsof 10 cm (withinthe sootinglayer)and
4 cm (outsideof sootinglayer).
greater than for the case where the fuel mixture is diluted. Figure 11 shows that in the
heavily sooting region (10 cm axial height), the
maximum value of qr is slightly higher than
the maximum value of qc for the undiluted
fuel case, where large quantities of soot are
generated. For the nitrogen-diluted case, where
less soot is generated, q > qr. Hence, the importance of radiation transport increases (in
comparison to conductive heat transport) as
the amount of diluent decreases. The magnitude of q~ within the sooting region (10 cm
height) is approximately an order of magnitude
greater than that within a nonsooting region of
the flame, as shown at a height of 4 cm.
SUMMARY AND DISCUSSION
A solution of the time-dependent NavierStokes equations, coupled with submodels for

ethylene combustion, soot formation, and radiation transport, has been performed to evaluate the importance of radiation transport on
the dynamics of strongly radiating luminous
flames. The unique feature of this model is the
coupling of multidimensional radiation transport, using the DOM algorithm, to one for
multidimensional fluid dynamics in axisymmetric geometry. One of the most significant advantages of the DOM model is that it can be
used to solve general radiation transport problems for any level of opacity ranging from
optically thin to thick.
In this model, we assume that the flame can
be represented by axisymmetric geometry. The
major assumptions used in the submodels are
as follows:
1. Chemical reaction model is represented by
a single-step rate Arrhenius type algorithm,

STRONGLY RADIATING ETHYLENE DIFFUSION FLAME

and the major species tracked include only
C2H4, 02, H20, CO2, and N 2.
2. The soot formation algorithm is represented
by two coupled ordinary differential equations, which include empirically derived coefficients in the representation of surface
growth, nucleation and coagulation. The
phenomenological soot oxidation model is
based only on temperature and oxygen partial pressure.
3. The radiation transport model assumes:
a. Radiation transport is independent of
wavelength (gray-gas approximation to
the RTE).
b. Scattering is negligible compared to absorption.
c. We consider the radiative effects of soot,
CO 2, and H 2 0 only, and neglect the
radiative effects of the C2H 4 fuel.
d. The soot absorption coefficient can be
represented by a Planck mean. Although
this is the appropriate mean for determining the outgoing radiation in the divergence calculation, it can differ from
the incident mean absorption coefficient
[44, 45], which is required for the incoming radiant intensity. We make the assumption that the two mean values are
equal.
e. The gas absorption coefficient for the
combination of CO 2 and H 2 0 is also a
Planck mean.
Results from a simulation of a coflowing
undiluted laminar 5 c m / s ethylene-air diffusion flame were compared with experimental
data of Gore and Faeth [16] to provide a
benchmark for the chemical reaction, energy
release, and soot formation algorithms. These
simulation results showed that a universal state
relationship exists between each of the major
gas species and local fuel-equivalence ratio,
indicating that the chemical reaction and energy release algorithm was adequately describing the species conversion processes. The state
relationship for soot volume fraction shows
more scatter than that for the major gas
species. Soot was predominately formed in the
region slightly rich of stoichiometric, and the
quantity of soot generated was approximately
the same as that observed in experiments. Although the simulation does demonstrate the

17

correct universal state relationships for species

concentrations and soot volume fraction, this
does not provide experimental validation of
the spatial and temporal behavior of the simulation.
Simulations were conducted for a highervelocity (5 m / s ) undiluted ethylene jet diffusion flame with and without radiation
transport. The results showed that radiation
transport reduces the maximum flame temperature and maximum soot volume fraction in
the flame. But, more importantly, the decrease
in temperature (due to radiative heat loss)
causes a decrease in the chemical heat release
rate, which, in turn reduces the volumetric
expansion, causing the flame to shrink. Hence,
the overall temperature, species concentration
and soot volume fraction distributions in the
flame changed due to radiation transfer. Radiative losses were approximately 35%-40% of
the chemical heat released.
The sooting region within the flame is quite
narrow and is located in the high-temperature
region near the fuel-rich side. The radiative
flux vectors were directed at an angle normal
to the surface of the sooting region. The flux
vectors were directed primarily in the radial
direction; however, in regions where the sooting region curved, the radiative flux vectors
had a significant axial component.
Radiative intensity was greatest within the
heavily sooting region, and was attenuated to
approximately 50% of its original value after
passing through the 0.5-cm-thick sooting layer.
Therefore, the opacity in the sooting region is
around 0.7, which corresponds to the intermediate regime between optically thin and thick.
For an undiluted fuel jet in the regions
where large quantities of soot are generated,
the effects of radiative heat transfer dominate
those of conductive heat transfer. For a nitrogen-diluted fuel jet mixture, conduction becomes more important than radiation, as
smaller quantities of soot are generated. Another measure of the importance of radiation
is a radiation-convection parameter defined by
Bhattacharjee and Grosshandler [46], in which
the authors state that radiation must be accounted for if the value of the parameter is
significantly larger than unity. For the undiluted ethylene-air flame considered here, their
parameter attains values of approximately 40.

18
Hence, by this criterion, radiation is a dominant mechanism for heat transfer compared to
conduction or convection for the undiluted jet
flame presented in this paper. Other investigations of radiation transport [16, 24-26] have
also demonstrated the importance of including
accurate radiation transfer models in strongly
radiating luminous flames.
Future work includes evaluation of some of
the major assumptions made in the formulation of this model to determine their overall
effect on the simulation results. Although it is
impossible to include the full set of elementary
reactions in the chemical reaction and energy
release model, we intend to consider a reduced
mechanism including approximately 10-15 reactions. Based on the results of the reduced
mechanism, we can then determine under what
conditions the single-step chemistry model is
sufficient for estimating the radiative loss fraction. We may also consider replacing the
chemical reaction model by convecting a conserved scalar, such as mixture fraction, and
using flamelet models to calculate resulting
flame temperature and species.
Future work will also investigate alternative
radiation transport models. First, we plan to
examine the maximum level of DOM algorithm necessary for this flame--that is, we will
try lower-order (S:) and higher-order ($6, S8)
DOM approximations to determine their effect
on the results. We also will compare DOM
simulation results with those obtained by using
an optically thin radiation model. Another
low-order model that we plan to evaluate is the
variable Eddington approximation, in which the
radiation field is treated in terms of a local
radiation energy density. In this case, the radiation energy density itself satisfies a scalar
transport equation, which includes a quantity,
fE, the dimensionless variable Eddington factor
[27]. The variable Eddington factor ranges between values of one for optically thin media
and one third for optically thick media, and
gives the model its ability to approximate the
local radiation flux in regions of varying opacity. Generally, one does not know the value of
the variable Eddington factor for the particular
application, although it can be calculated from
a detailed solution of the equations of radia-

C. R. KAPLAN ET AL.
tive transfer. Using the results of the DOM
model (S 4 approximation), preliminary calculations of the variable Eddington factor show
that it is near unity (optically thin) in a large
portion of the flame and quickly decreases to
values as low as 0.4 (intermediate regime between optically thick and thin) near the sooting
regions of the flame.
Computation times for simulations without
radiation on the highly-resolved grid were 35
)< 10 -6 CPU-seconds per timestep per cell,
requiring approximately 7 h of CPU-time per
simulation on a Cray Y-MP. Including the
DOM radiation model (which was not specifically optimized for Cray Y-MP architecture)
increased the computational time by a factor
of two. A large part of the extra cost is from
calculating radiation transport over the very
fine computational grid required to resolve the
flame structure. The DOM algorithm itself is a
very efficient method and does not require
such high resolution to obtain accurate radiation transport predictions. One possible way to
reduce the cost would be to use two grids: a
fine one to resolve the flame structure and a
coarse one for the DOM model, and then
interpolate between the two grids. In this case,
we trade off computer time for programming
complexity, but this approach would be especially useful when higher-order approximations
(S 6, S 8) to radiation transport are needed.

This work was sponsored by the Naval Research Laboratory through the Office of Naval
Research. Computing time was provided by Numerical Aerodynamic Simulator (NAS) and the
Naval Research Laboratory. The authors would
like to thank Dr. F. Grinstein for his assistance
with the graphing routines, and Drs. F. Williams,
P. Tatem, and J. Boris for providing the resources
and environment necessary for this work to be
accomplished, and to the referees of this manuscript for their thorough review and helpful
comments.
REFERENCES
Chamberlin, D. S., and Rose, A., First Symposium on
Combustion and Flame and Explosion Phenomena, The
Combustion Institute, Pittsburgh, 1928, p. 27.

STRONGLY RADIATING ETHYLENE DIFFUSION FLAME

2. Yule, A. J., Chigier, N. A., Ralph, S., Boulderstone,
R., and Ventura, J., AL4A J. 19:752-760 (1981).
3. Chert, L. D., Seaba, J. P., Roquemore, W. M., and
Goss, L. P., Twenty-Second Symposium (International)
on Combustion, The Combustion Institute, Pittsburgh, 1988, p. 677.
4. Buckmaster, J., and Peters, N., Twenty-First Symposium (International) on Combustion, The Combustion
Institute, Pittsburgh, 1986, p. 1829.
5. Roquemore, W. M., Chen, L.-D., Seaba, J. P., Tschen,
P. S., Goss, L. P., and Trump, D. D., Phys. Fluids
30:2600-2601 (1987).
6. Laskey, K. J., Ellzey, J. L., and Oran, E. S., AIAA
Paper 89-0572, AIAA, Washington, D.C. 1989.
7. Ellzey, J. L., Laskey, K. J., and Oran, E. S., Combust.
Flame 84:249-264 (1991).
8. Ellzey, J. L. and Oran, E. S., Twenty-Third Symposium
(International) on Combustion, The Combustion Institute, Pittsburgh, 1990, p. 1635.
9. Davis, R. W., Moore, E. F., Roquemore, W. M.,
Chen, L.-D., Vilimpoc, V., and Goss, L. P., Combust.
Flame 83:263-270 (1991).
10. Faeth, G. M., Gore, J. P., Chuech, S. G, and Jeng,
S. M., in Annual Review of Numerical Fluid Mechanics
and Heat Transfer (C. L. Tien and T. C. Chawla,
Eds.), Hemisphere, New York, 1989, p. 1.
11. Sivathanu, Y. R., Gore, J. P., and Dolinar, J., Cornbust. Sci. Technol. 76:45-66 (1991).
12. Grosshandler, W. L., Int. J. Heat Mass Transl.
23:1447-1457 (1980).
13. Bilger, R. W., Combust. Flame 30:277-284 (1977).
14. Faeth, G. M., and Samuelson, G. S., Prog. Energy.
Combust. Sci. 12:305-373 (1986).
15. Kounalakis, M. E., Sivathanu, Y. R., and Faeth, G.M.,
J. Heat Transl. 113:437-445 (1991).
16. Gore, J. P., and Faeth, G. M., Twenty-First Symposium
(International) on Combustion, The Combustion Institute, Pittsburgh, 1986, p. 1521.
17. Santoro, R. J., Semerjian, H. G., and Dobbins, R. A.,
Combust. Flame 51:203-218 (1983).
18. Santoro, R. J., and Semerjian, H. G., Twentieth Symposium (International) on Combustion, The Combustion Institute, Pittsburgh, 1984, p. 997.
19. Kent, J. H., and Honnery, D. R., Combust. Flame
79:287-298 (1990).
20. Honnery, D. R., and Kent, J. H., Combust. Flame
82:426-434 (1990).
21. Kent, J. H., and Honnery, D. R., Combust. Sci. Technol. 75:167-177 (1991).
22. Moss, J. B., Stewart, C. D., and Syed, K. J., TwentySecond Symposium (International) on Combustion,
The Combustion Institute, Pittsburgh, 1988, p. 413.
23. Syed, K. J., Stewart, C. D., and Moss, J. B., TwentyThird Symposium (International) on Combustion, The
Combustion Institute, Pittsburgh, 1990, p. 1533.
24. Gosman, A. D., and Lockwood, F. C., Fourteenth
Symposium (International) on Combustion, The Combustion Institute, Pittsburgh, 1973, p. 661.
25. Tamanini, F., Seventeenth Symposium (International)
on Combustion, The Combustion Institute, Pittsburgh, 1979, p. 1075.

19

26. Gore, J. P., Ip, U.-S., and Sivathanu, Y. R., J. Heat

Transl. 114:487-493 (1992).
27. Oran, E. S., and Boris, J. P., Numerical Simulation of
Reactive Flow, Elsevier, New York, 1987, pp. 339, 477.
28. Patnaik, G., Guirguis, R. H., Boris, J. P., and Oran,
E. S., J. Comput. Phys. 71:1-20 (1987).
29. Hirschfelder, J. O., Curtiss, C. F., and Bird, R. B.,
Molecular Theory of Gases and Liquids, Wiley, New
York, 1954.
30. Kee, R. J., Dixon-Lewis, G., Warnatz, J., Coltrin,
M. E., and Miller, J., SAND86-8246, Sandia National
Laboratory, 1986.
31. Wilke, C. R., J. Chem. Phys. 18:578-579 (1950).
32. Bird, R. B., Stewart, W. E., and Lightfoot, E. N.,
Transport Phenomena, Wiley, New York, 1960.
33. Westbrook, C. K., and Dryer, F. L., Combust. Sci.
Technol. 27:31-43 (1981).
34. Nagle, J., and Strickland-Constable, R. F., Proceedings of the Fifth Carbon Conference, Pergamon, New
York, 1962, Vol. l, p. 154.
35. Siegel, R., and Howell, J. R., Thermal Radiation Heat
Transfer, Hemisphere, New York, 1981, Vol. 2, p. 426.
36. Hottel, H. C., and Sarofim, A. F., Radiative Transfer,
McGraw-Hill, New York, 1967.
37. Chandrasekhar, S., Radiative Transfer, Dover, New
York, 1960, p. 318.
38. Kim, J. S. , Baek, S. W., and Kaplan, C. R., Combust.
Sci. Technol., 88:133-150 (1992).
39. Kim, T. Y., and Baek, S. W., Int. J. Heat Mass Transl.
34:2265-2273 (1991).
40. Carlson, B. G., and Lathrop, K. D., in Computing
Methods in Reactor Physics (H. Greenspan, C. Kelber,
and D. Okrent, Eds.), Gordon and Breach, New York,
1968, p. 167.
41. Magnussen, B. F., and Hjertager, B. H., Sixteenth
Symposium (International) on Combustion, The Combustion Institute, Pittsburgh, 1977, p. 719.
42. Kaplan, C. R., Baek, S. W., and Oran, E. S., AIAA
Paper No. 93-0109, American Institute of Aeronautics and Astronautics, Washington, D.C., 1993.
43. Grinstein, F. F., submitted.
44. Cess, R. D., and Mighdoll, P., Int. J. Heat Mass
Transl. 10:1291-1292 (1967).
45. Grosshandler, W. L., and Thurlow, E. M., J. Heat
Transf. 114:243-249 (1992).
46. Bhattacharjee, S., and Grosshandler, W. L., Combust.
Flame 77:347-357 (1989).

Received 16 June 1992; revised 9 July 1993

APPENDIX: IMPLEMENTATION OF DOM

For axisymmetric cylindrical geometry, the
RTE is written for each individual ordinate
direction, m, as
t*., O ( r i m )
- - + ~ , .
r

Or

Oi.,
Oz

| O(rlmi m)
r

&b

20

C . R . KAPLAN ET AL.
,.

I~>0

i,,

Fig. A1. Orientation of angular coordinates system for

cylindrical geometry for radiation transfer calculation.

s
- 3 i m + ai b + ~

EWm'f~m'mim"
rn'

~<0

~<0

~<0

ip

i,

~>0

~>0

Fig. A2. Control volume and direction cosines for DOM

calculation.

X ( ~ m + l / 2 i p , m + l / 2 -- Otm_l/2ip, m _ l / 2 )

(A1)

= --[3imp A V + aibp A V

sAV

where the values m and m' denote outgoing

and incoming directions, respectively, and qt is
the angle of revolution around the z axis.
Although the influence of scattering is neglected in our calculation, we include the scattering term in the following equations to show
how it is incorporated in the general DOM
model. As shown in Fig. A1, the direction
cosines of a discrete direction, l~m, are/zm, ~,,,
and ~/m, respectively, and satisfy the identity
/.~m2 + ~rn2 + 'r/m2 = 1. The direction f~,, can
be pictured as a point on the surface of a unit
sphere with which a surface area, Win, is associated. The wm represent angular Gaussian
quadrature weights and satisfy the requirement
that the weights sum to the surface area of the
unit sphere. A total of M directions are chosen
and the angular areas are measured in units of
47r, such that E m
M= 1Wra = 1.
Integrating Eq. A1 over an arbitrary control
volume, as shown in Fig. A2,, yields:
I.~m(hnimn - Zsims ) + ~m(Zeime - A.,im~)

1
-(A n -A,)--

Wm

+ m4,n.
, - - ~ ~Pm'mWm' i,n 'p,

(A2)

where An, A,, Ae, and A w are the corresponding areas of the control volume sides for
the north, south, east, and west faces, respectively, AV is the volume of the control element, and p is the node of interest for which
we are calculating the intensity, imp. The terms
iron, ins, ime, and imw are the intensities for the
individual mth direction at the north, south,
east, or west nodes, respectively (Fig. A2). The
term ip, m_l/2 is the intensity at the point
of interest in the angular direction ( m 1/2) where the direction m + 1 / 2 defines the
edges of the angular range denoted by the
Gaussian quadrature wm. Hence, the term
(Otm+l/2ip, rn+l/2 -- Otm_l/2ip, m _ l / 2 ) r e p r e s e n t s the flow out of and into the angular
range. The a terms appear in Eq. A2 to preserve the conservation of intensity in the curved
coordinate and are determined from radiative
equilibrium conditions [37].
To solve Eq. A2 for imp, interpolation relationships are used to express values of intensity
at unknown points to values of intensity at
known points. For example, for a direction m

STRONGLY R A D I A T I N G E T H Y L E N E DIFFUSION F L A M E
which has positive direction cosines (~.tm > 0,
~,, > 0), we can eliminate the intensities at the
unknown points iron , ime , and ip, m+l/2, by
expressing them in terms of intensities at
known points i .... i,,,w, and i t.... -1/2, using

21

rectional evaluation [37], we first solve the intensity at each grid point for each of three
directions with positive direction cosines (~-m
> O, ~m > 0), then for each of the three directions with direction cosines /z,, < 0, ~m > 0,
then for each of the three directions where
imp = ~i .... + (1 -- ~ ' ) i m s ,
P~m < 0, ~m < 0, and finally for each of the
three directions where tx,, > 0, ~m < 0, as
ira1, = ~im,. + (1 -- ~")imw ,
shown in Fig. A2. For a positive set of direction cosines (Iz > 0, ~ > 0), the calculation
ira1, = (il .... +1/2 + (1 -- ~ ) i p , m _ l / 2 ,
(A3)
starts at the left-hand bottom corner of the
computational domain and proceeds to the top
where ff is an interpolation factor. Initially,
right-hand corner; known intensities are at the
central differencing (~ = 0.5) is used in the
south and west nodes, and interpolation formuinterpolation relationships. Substituting Eq. A3
las (as in Eq. A3) are used to eliminate the
into Eq. A2 and rearranging gives an expresintensity at the north and east nodes from Eq.
sion for i,, u in terms of known variables (for
A2
to calculate the intensity at the point of
~,,, > 0, ~,, > 0),
interest, imp. For direction cosines ~m < 0,
iml ~ = ( tZmAr i .... + ~mAzimw - [ ( A n - A s ) / W m] ~,,, > 0, the calculation proceeds from the
right-hand bottom corner to the top left-hand
XA,fip, m_l/2 + ;SAV)/(lzmA
n + ~mAe corner, and known intensities are at the south
and east nodes. For direction cosines I'~m < O,
- [ ( A n - A~)/wmlo~m+~/2 + [3 a v ) ,
~m < 0, the calculation proceeds from the top
right-hand corner to the bottom left-hand cor(A4)
ner, and known intensities are at the north and
east nodes. And finally, for direction cosines
where
tzm > 0, ~m < 0, the calculation proceeds from
the top left-hand corner to the bottom rightA r =As~+
(1 - ~ ) A n ,
hand corner, and known intensities are at the
A~ = A w ~ + (1 - g')Ae,
north and west nodes. Iterations are performed
over all ordinate directions and cells
A,, = a m _ l / 2 ~ + (1 - ~')am+l/2,
until the calculated intensity at each computas
tional cell does not change within a given
S = aibp + ~ ~ f~m,mWrn,im,p.
constraint
depending on the individual direcm'
tion considered.
Once
imp is calculated from Eq. A4, the
Even if all terms in Eq. A4 are positive, the
remaining unknown surrounding intensities,
interpolation relationships, Eq. A3, used to
extrapolate across a grid point could produce
ira,, ime, and ip, m+l/2 a r e found from the
interpolation relations, Eq. A3. These intensiunphysical negative intensities. In practice, this
ties are then used as inputs for the calculation
situation usually occurs in the presence of steep
of intensity at the next cell in the correspondgradients or when the spatial resolution is ining direction, and the solution proceeds recuradequate [29, 38]. In our simulations, if a negasively through the entire grid for this direction
tive intensity is calculated, the interpolation
[37].
procedure is changed from central differencing
For the S 4 approximation, intensities of all
(~ = 0.5) toward upwind differencing (~" gradof the points in the computational domain are
ually is increased from 0.5 to a maximum value
calculated for each of the twelve directions
of 1) until a stable solution is achieved and
considered. According to the principle of dithere are no negative intensities.