LDC-2014-001

12/23/2014

NUMERICAL SIMULATION OF THE LAMINAR DIFFUSION

FLAME IN A SIMPLIFIED BURNER
Lawrence D. Cloutman
adastera47@yahoo.com

Abstract
The laminar ethylene-air diffusion flame in a simple laboratory burner was simulated with the COYOTE reactive flow program. This program predicts the flow field,
transport, and chemistry for the purposes of code validation and providing physical
understanding of the processes occurring in the flame. We show the results of numerical
experiments to test the importance of several physical phenomena, including gravity,
radiation, and differential diffusion. The computational results compare favorably with
the experimental measurements, and all three phenomena are important for accurate
simulations.

c
2014
by Lawrence D. Cloutman. All rights reserved.
1

Introduction

Flower and Bowman [1, 2, 3] measured the temperature and velocity fields in a simple
laboratory-scale burner. Although their primary goal was to study the sooting properties of
this burner, these measurements are suitable for validation of computational fluid dynamics
programs. The geometry is simple, the flow is laminar, and the combustion occurs in an
ethylene-air diffusion flame. The burner is a simple Wolfhard-Parker burner enclosed in a
rectangular chamber as shown in figure 1. Fuel is fed through a rectangular tube 0.8 cm wide
by 8.0 cm long, and coflowing air surrounds the tube. Flame stability requires the initial air
velocity to exceed the initial fuel velocity, and these are 22.0 and 7.0 cm/s, respectively. A
steel screen was placed 1.5-2.5 cm outside the flame to reduce flickering. The results reported
here are for a pressure of one atmosphere.
The experimental measurements of temperature and vertical velocity are the subject
of this study. A small silica-coated thermocouple was use to measure gas temperatures. The
measurements have been corrected for radiative losses and have estimated uncertainties of
4%. Velocity measurements were made with standard LDV techniques.
This report is an updated version of reference [4]. Simulations of the experiment were
performed with the COYOTE computational fluid dynamics program [5], which uses a partially implicit algorithm to solve the full transient Navier-Stokes equations on a nonuniform
grid. The model includes a real-gas caloric equation of state, arbitrary chemical kinetics,
transport coefficients from a Lennard-Jones model, a simple radiative heat loss model, and
mass diffusion based on either Ficks law or the full Stefan-Maxwell equations. The ethylene combustion is modeled by a one-step global Arrhenius kinetics rate. Also included are
three kinetic reactions for thermal NOx production and six molecular dissociation reactions
required to close the thermal NO reaction set and produce accurate temperatures.
Steady-state solutions, when they exist, are found by assuming an arbitrary initial
flow and allowing the transient to decay. While this approach is less efficient computationally
than a direct steady state approach, it has several advantages. First, it does not make an ad
hoc assumption that the final flow field is truly steady. That is, it allows the final solution
to contain quasi-periodic features such as flame flickering. If the final flow is truly steady,
then that type of solution will evolve naturally. Second, it automatically performs a flame
stability study. If, for example, the flame cannot be sustained, it should be extinguished
in the calculation (at least to the extent that the gas physics included in the calculation is
an accurate representation of the physical system). Third, it allows easy incorporation of
complex, realistic gas physics.
We show the results of numerical experiments that test the importance of several

physical phenomena, including gravity, radiative heat loss, and differential diffusion based
on the Stefan-Maxwell equations. The base case includes all three of these phenomena.
Several additional runs were made. One used zero gravity. The second neglected radiative
losses. The third used Ficks law and a unit Lewis number for all species instead of the
Stefan-Maxwell equations. The computational results are in excellent agreement with the
experimental measurements of velocity and temperature provided all three phenomena are
included. We find that the solutions are quite sensitive to gravity, NOx predictions will
require the use of the radiation model to get adequate accuracy in the temperature field,
and the solutions are sensitive to the mass transport model.
Section 2 presents the governing equations and describes the gas physics. Section
3 describes the geometry of the experimental apparatus and the problem setup. Section 4
describes several solutions for this burner. Conclusions are presented in Section 5.

Governing Equations

The program is based on the Navier-Stokes equations for a mixture of compressible gases.
We use the single (mass weighted) velocity representation and Eulerian coordinates.
Mass conservation is expressed by the continuity equation for each species :

+ ( u) = J + R ,
t

(1)

where is the density of species , u is the fluid velocity, J is the diffusional mass flux of
species , and R is the rate of change of species by chemical reactions. The diffusional
flux is a complex function of the flow that often is approximated by Ficks law,
J = D( /),

(2)

in some of the solutions. D is the species diffusivity (assumed here independent of ), and
is the total density. We also use the formalism of Ramshaw [6, 7], which is an approximate
treatment of the full Stefan-Maxwell equations. The R are assumed to be known functions
of the composition and thermodynamical variables. The global rate for ethylene oxidation,
for example, is
RC2 H4 = 6.4 1012 WC2 H4 [O2 ]1.65 [C2 H4 ]0.1 exp(15000/T ) g/cm3 s,

(3)

where WC2 H4 is the molecular weight of ethylene, and square brackets denote molar concentration (mol/cm3 ). This rate predicts the correct laminar flame speed under stoichiometric
conditions at one atmosphere.
The momentum equation is
X
(u )
+ (uu ) =
F P S,
t

(4)

where P is the pressure, and F is the body force per unit mass acting on species , which
in the present application is the gravitational acceleration g. The viscous stress tensor is
S = [u + (u)T ] 1 ( u) U,

(5)

where is the coefficient of viscosity, 1 = 2/3 is the second coefficient of viscosity, and
U is the unit tensor.
We use the thermal internal energy equation to express energy conservation:
X
X
(I)
+ (Iu) = P u S : u q +
H R +
F J Lrad ,
t

(6)

where I is the specific thermal internal energy, and H is the heat of formation of species .
Note that for F = g, the next-to-the-last term vanishes. The heat flux q is approximated
by the sum of Fouriers law and enthalpy diffusion:
q = T +

h J ,

(7)

where h is the specific enthalpy of species .

The radiative heat loss term Lrad is described in [8]. A complete treatment of the
radiative transfer would be extremely complex and computationally challenging, so we consider only a highly simplified model that is valid in the optically thin limit, namely a local
radiative heat sink. We use a slight generalization of the approximation used by Chao, Law,
and Tien [9], namely
Lrad = 4 KP (T 4 Tw4 ),

(8)

where is the Stefan-Boltzmann constant. The wall temperature Tw is assumed to be the

same for all walls. The function KP is related to the Planck mean opacity, P , by
KP (T, { })

= P =

T 4

Z
0

k (T, { })B (T ) d,

(9)

where { } is the set of all species densities, k is the monochromatic absorption coefficient,
and B is the Planck function. Planck mean absorption coefficients for three important
radiating species (CO, CO2 , and H2 O) were taken from [10].
The equation of state is assumed to be given as the sum of the partial pressures of
an ideal gas for each species. Transport coefficients are computed from the Lennard-Jones
model [11]. The JANAF tables [12, 13, 14] provide a homogeneous set of thermochemical
data for a large collection of materials, and these tables are used to supply the specific
enthalpy and heat of formation for each species of interest.
Chemical reactions are divided into two groups as shown in table 1. The first group is
treated kinetically, with the rates assumed to be of generalized Arrhenius form. The second
group, reactions 1e through 6e, is assumed to be in chemical equilibrium.

Table 1. One-Step Fuel Oxidation and Thermal NOx Mechanisms

1.
2a.
2b.
3a.
3b.
4.
1e.
2e.
3e.
4e.
5e.
6e.

Reaction
C2 H4 + 3 O2 2 CO2 + 2 H2 O
O2 + 2 N2 2 N + 2 NO
O2 + 2 N2 2 N + 2 NO
2 O2 + N2 2 O + 2 NO
2 O2 + N2 2 O + 2 NO
N2 + 2 OH 2 H + 2 NO
H2 *
) 2H
N2 *
) 2N
O2 *
) 2O
O2 + H2 *
) 2OH
O2 + 2H2 O *
) 4OH
*
O2 + 2CO ) 2CO2

A
6.4 1012
1.5587 1014
7.5 1012
2.6484 1010
1.6 109
2.123 1014

0.0
0.0
0.0
1.0
1.0
0.0

E/R (K)
15000 [C2 H4 ]0.1 [O2 ]1.65
67627
[O2 ]0.5 [N2 ]1.0
0
[N] [NO]
59418
[O2 ] [N2 ]0.5
19678
[O] [NO]
57020
[N2 ]0.5 [OH]

Problem Description

The geometry of the experiment is shown in figure 1. The burner is nothing more than a
rectangular metal tube with a 0.8 by 8.0 cm cross section through which ethylene flows. The
tube is surrounded by the axial air flow, which is confined in a rectangular chamber whose
walls are several centimeters from the tube. The high aspect ratio of the open end of the
tube allows us to assume the flame is uniform along most of the length of the opening. The
two dimensional simulations were made in a plane perpendicular to the long direction of the
slot. We also assume bilateral symmetry about a line bisecting the slot the long way, so
we place the lower left hand corner of the grid at the center of the slot, but 0.5 cm below
the opening. The wall of the slot is assumed to be a sheet of metal 0.1 cm thick and is
represented in the two-dimensional plots as a series of xs. The base case uses a uniform grid
of 1 mm square zones and has 30 by 60 zones. One run was made with 0.5 mm square zones.
The fuel, ethylene, is allowed to flow into the bottom of the mesh through horizontal zones
numbers 2 through 5 in the coarse grid (number 1 is the fictitious zone at the left side of the
mesh), and air flows in through zones 6 through 31. An obstacle representing the edge of
the burner occupies the first 6 zones vertically (including the bottom fictitious zone) of the
6th column.
Our inflow boundary condition is the type (ii) of Rudy and Strikwerda [15] with
specified density. We assume that the inflowing gases have a temperature of 300 K and a
pressure of 1.013 106 dynes/cm2 . The inflow velocities are 7 cm/s for the ethylene and 22
cm/s for the air velocity. The inflow density is 1.131 103 g/cm3 for the ethylene. The
air was assumed to be a mixture of five species with densities of 2.688 104 g/cm3 O2 ,
8.766 104 N2 , 5.292 107 CO2 , 7.217 108 H2 O, and 1.489 105 argon. Unless
otherwise noted, the gravitational acceleration is assumed to be -980 cm/s2 .

Numerical Solutions

A series of six cases were run out to steady state using a variety of numerical parameters
and physical submodels. The base case has a resolution of 1 mm and Ficks law is used for
the species diffusion. The same diffusivity is used for all species and is calculated from the
mixture viscosity and a Schmidt number of 0.7. The radiation model was included. The
same problem was run with 0.5 mm zones. There was no significant change in the solution,
suggesting grid independence of the solutions. It is somewhat surprising that the solution
with 1 mm resolution is so well converged. Another variation of the base case was to use a
fuel oxidation rate that is half that of equation (3). This had no effect on the solution. This
is not surprising since it takes approximately 0.01 s for the fluid to cross a computational
zone, but the chemical time scale for oxidizing the fuel is approximately three orders of
magnitude smaller. A similar insensitivity to the reaction rates, however, is not expected for
the much slower thermal NO reactions.
Another case was the same as the base case except Ficks law was replaced by the
detailed mass transport model. A comparison between these two cases is given in the next
five figures. Figure 2 shows the isotherms for both cases. The multicomponent diffusion
model produces a higher peak temperature (2422 K, as compared to 2317 K with Ficks
law). The height of the flame was not measured in the experiment, but it was observed to
be well beyond the height of this computational grid (6.0 cm), as predicted here. The axial
temperature gradient in the core of the flame is significantly smaller in the multicomponent
case.
Figure 3 shows the calculated and experimental horizontal temperature profiles. The
left edge of each plot is at the center line of the flame, and the peak temperature occurs
approximately above the edge of the slot. The Ficks law calculation does very well except
in the center of the flame. The multicomponent calculation does better in the center, but
is systematically a little hotter than the experimental values. The experimental flame was
intended to produce sooting conditions, and the calculations do not yet have a soot production or soot radiation model. The fact that the multicomponent calculation is systematically
slightly too hot is consistent with the radiative losses expected from the soot.
Figure 4 shows the horizontal profile of axial velocity for the experiment and both
calculations. Both calculations agree with the experiment out to 1.0 cm. It is not known
with certainty why the outer flow of air is systematically higher than the 22 cm/s flow speed
reported by Flower and Bowman, but it seems likely that the experimental air inflow speed
was closer to 30 cm/s in the cases at 1 atm pressure.
Figure 5 shows mass fractions of H2 for both calculations. Since differential diffusion

effects are largest for the very light species, we expect some differences between these two
plots. Not only are the details of the internal distributions in the flame different, but the
peak values are different by a factor of nearly two.
Figure 6 shows mass fractions of NO. As in the case of H2 , there are some significant
differences in the spatial distribution. With Ficks law, the peak NO mass fraction occurs
in a sheet in the outer part of the flame, coincident with the high temperatures. In the
multicomponent model, the peak NO mass fraction occurs in an island just above the lip of
the burner, and there is nearly a factor of four difference in the peak NO mass fraction. The
NO flow rates at the top of the grid are 2.5 106 g/s-cm and 6.4 106 g/s-cm for the
Ficks law and multicomponent solutions, respectively. These results strongly suggest that
burner models that are expected to produce reliable solutions that include any significant
level of chemical detail must include a detailed mass transport model.
The base case was rerun without the radiative heat loss model. The peak temperature
is 2355 K, close to the adiabatic stoichiometric flame temperature of 2380 K. The isotherms
are qualitatively the same as in the base case. Temperatures are typically 50 to 100 K higher
at points in the flame zone as compared to the base case. The H2 mass fraction peaks at
8.6 104 , a little over twice the base case value. The peak NO mass fraction is 3.5 104
and has a different spatial distribution than in the first two cases. Far more NO production
is occurring as the hot gases rise through the grid than in the base case. The NO flow rate
is 6.8 106 g/s-cm and clearly would be higher if the grid were taller.
The importance of buoyancy forces is demonstrated by rerunning the base case with
zero gravitational acceleration. Figure 7 shows the isotherms and NO mass fractions for the
zero-gravity case. The flame is now much wider than before, and the lower temperatures
suppress NO production.

Conclusions

The COYOTE hydrodynamics program has been used to simulate reactive flows in a simplified experimental burner. One objective of this study is validation of the program and
demonstration of its capability to simulate laminar diffusion flames. The second objective is
to study the importance of several physical submodels in burner simulations.
While detailed comparison of the computational and experimental results is still in
progress, we can make the following general observations:
1) The model successfully predicts the velocity and temperature fields with surprisingly coarse zoning (1 mm resolution).
2) It is critical to include gravity in the calculations as buoyancy effects are quite
pronounced. Without buoyancy forces to help lift hot combustion products away from the
burner, the flame becomes much broader and takes on a different velocity profile.
3) A comparison of the results with Ficks law mass diffusion and the Stefan-Maxwell
equations shows that both the temperature and velocity fields are predicted more accurately
with the latter model. The difference is as much as several percent. Moreover, the spatial
distributions of several radical species change significantly between the two models. This
behavior suggests the importance of using the detailed transport model in applications where
accurate predictions of some of the minor species are important, such as in the case of
emissions of NO, formaldehyde, and PAHs. In general, we recommend the use of the more
complex model whenever possible for laminar flame simulations.
4) The radiative cooling model lowers local combustion temperatures on the order of
50-100 K for non-sooting flames. While this change has little effect on the overall dynamics
of the flow, it makes approximately a factor of two difference in the thermal NO production
rate due to the high temperature sensitivity of the thermal NO mechanism.
5) Future work will concentrate on more detailed chemical kinetics, an improved
radiative transfer model, and development of a soot model (both the chemistry and the
radiative properties).

10

References
[1] W. L. Flower and C. T. Bowman, Measurements of the Structure of Sooting Laminar
Diffusion Flames at Elevated Pressures, 20th Symposium (International) on Combustion, The Combustion Institute, Pittsburgh, 1984, 1035-1044.
[2] W. L. Flower and C. T. Bowman, Measurements of the effect of elevated pressure on
soot formation in laminar diffusion flames, Comb. Sci. Tech. 37, 1984, 93.
[3] W. L. Flower, The Effect of Elevated Pressure on the Rate of Soot Production in
Laminar Diffusion Flames, presented at the Spring Meeting of the Western States
Section of the Combustion Institute, 1985.
[4] L. D. Cloutman, Numerical Simulation of the Laminar Diffusion Flame in a Simplified
Burner, Lawrence Livermore National Laboratory report UCRL-JC-122619, 1995.
[5] L. D. Cloutman, COYOTE: A Computer Program for 2-D Reactive Flow Simulations,
Lawrence Livermore National Laboratory report UCRL-ID-103611, 1990.
[6] J. D. Ramshaw, Self-consistent effective binary diffusion in multicomponent gas mixtures, J. Non-Equilib. Thermodyn. 15, 295 (1990).
[7] J. D. Ramshaw, Hydrodynamic theory of multicomponent diffusion and thermal diffusion in multitemperature gas mixtures, J. Non-Equilib. Thermodyn. 18, 121 (1993).
[8] L. D. Cloutman, Numerical Simulation of Radiative Heat Loss in an Experimental
Burner, Lawrence Livermore National Laboratory report UCRL-JC-115048, presented
at the 1993 Fall Meeting of the Western States Section Meeting of the Combustion
Institute, 1993.
[9] B. H. Chao, C. K. Law, and J. S. Tien, Twenty-Third Symposium (International) on
Combustion, The Combustion Institute, 1990, 523.
[10] M. M. Abu-Romia and C. L. Tien, Appropriate mean absorption coefficients for infrared radiation of gases, J. Heat Transfer 89, 321 (1967).
[11] L. D. Cloutman, A Database of Selected Transport Coefficients for Combustion Studies, Lawrence Livermore National Laboratory report UCRL-ID-115050, 1993.
[12] D. R. Stull and H. Prophet, JANAF Thermochemical Tables, 2nd ed. (U. S. Department
of Commerce/National Bureau of Standards, NSRDS-NBS 37, June 1971).

11

[13] M. W. Chase, J. L. Curnutt, A. T. Hu, H. Prophet, A. N. Syverud, and L. C. Walker,

JANAF Thermochemical Table, 1974 Supplement, J. Phys. Chem. Ref. Data 3:311
(1974).
[14] M. W. Chase, Jr., C. A. Davies, J. R. Downey, Jr., D. J. Frurip, M. A. McDonald,
A. N. and Syverud, JANAF Thermochemical Tables, Third Edition, Parts I and II.
Supplement No. 1, J. Phys. Chem. Ref. Data 14 (1985).
[15] D. H. Rudy J. C. and Strikwerda, Computers & Fluids 9:327 (1981).
[16] L. D. Cloutman, Numerical Simulation of Turbulent Mixing and Combustion Near the
Inlet of a Burner, Lawrence Livermore National Laboratory report UCRL-JC-112943,
1993.

12

grid

1 1
t T

air

ethylene

12cm/s

7 cm/s

air

air

Fig. 1. Top and side views of the slot burner.

grid is also shown.

The location of the two-dimeneional computational

13

TEMPERATURE
Ficks Law

Multicomponent

MIN = 2,7103990+02

MIN = 2.550626D!-02

I.AAX= 2.316704D+-03

MAX =

2.4215020+03

F]g. 2. Isotherms for the bsse cuse (Ficks law maas transport) and multicomponent msss transport
solutions.

14

Ficks Law

02

0.4

0.6

0.8

Lo

t~

x (an)

:
.

0.0

0.2

0.+

0.6

O.B

LO

1.2

x (an)

3. Horizontal profiles of temperature 2.0 cm above the burner.

15

L4

I-6

Fides Law

1
Calc.

Multicomponent

0
:
z:0
:
;.
.
:
0
s

---

Calc. =

0.

s
0
0

0.0

02

0.4

0.6

0.8
* (cm)

,0

L2

4. Horizontal profiles of axial velocity 2.0 cm above the burner.

16

,.+

*.6

H2 MASS FRACTION
Multicomponent

L
MAX =

3.841191D-04

MAX =

6.438925D-04

>

Fig. 5. Hz mass ikction contours for the base case (Ficks law mass transport) and multicomponent
mass transport solutions.

17

NO MASS FRACTION
.

Multicomponent

Ficks Law
1
~

/
1 : 3
I
I 1I

MAX= 9.680672005
.

Fig. 6. NO mass fraction contours for the bsse case (Ficks law mass transport) and multicomponent
mass transport solutions.

18

MAX = 2.313397D+03
8

MAX = 9.079066D-05

Fig. 7. Isotherms and NO mass fraction contours for the zero-gravity solution.

19

...