COMBUSTION AND FLAME 42:277-285 (1981) 277

Turbulent Gaseous Combustion Part H: Theory and Evaluation for

Local Properties

PHILIP J. SMITH* and L. DOUGLAS SMOOT**

Chemical Engineering Department, Brigham Young University, Provo, Utah84602

A computer model has been developed for the analysis of confined, axisymmetric, turbulent diffusion flames. The fluid
mechanics model includes the two-equation k-e turbulence model for closure. The combustion model is based on
reactions limited by mixing rate with local chemical equilibrium. The mixture fraction is assumed to have a top-hap
probability distribution, and transport equations are solved for the mean and variance of the mixture fraction. Local
mean gas phase properties, including density and species concentration, are predicted. The coupled elliptical, nonlinear,
partial differential equations are solved by an existing hybrid-differencing scheme. Predictions are made for reacting
flow systems with re.circulation, and are shown to be sensitive to grid size and inlet turbulence level. Results are
compared with experimental data reported in Part I and with data from other laboratories. Reasonable agreement is
observed between measurement and theory for species concentration in aft regions of the combustor, with H 2 and CO
comparisons being the least acceptable.

INTRODUCTION density systems has been experimentally dem-

This paper describes a study of confined, two-
dimensional, turbulent, diffusion flames with re-
circulation. The resulting model describes the mean
property fields, including velocity, temperature,
density, species concentration, and properties of the
turbulence field. The model is the basis for a
computer code for describing pulverized coal com-
bustors and gasifiers [1]. Such codes, for either
gaseous or solid fuel, can be utilized for sensitivity
analysis of physical parameters, scaling, design, and
analysis of experimental data.
Several investigators (e.g., [2-6]) have reviewed
current turbulence models and have made recom-
mendations concerning their use. We have, for the
present work, selected the k-e model [3]. Two
differential equations are solved, one for the turbu-
lence kinetic energy k and another for its dissi-
pation rate ~. Validity of this model for constant
* Assistant Professor.
** Professor and Dean, Engineering Sciences and Tech-
nology.
Copyright © 1981 by The Combustion Institute
~ublished by Elsevier North Holland, Inc.
52 Vanderbilt Avenue, New York, NY 10017
onstrated [6]. When applying the model to turbu-
lent combustion, though, other concerns remain.
For example, the question of flame-generated
turbulence [7] is not resolved.
Historically, two principal modeling approaches
have been used for the combustion process. At one
extreme, chemical kinetics are assumed to be rate
controlling and turbulent mixing is considered
unimportant [8, 9]. At the other, turbulent mixing
is treated in detail, but infinite rate chemistry is
assumed [10]. Only a few attempts [11, 12] have
been made to combine the treatment of finite rate
chemistry and detailed turbulent mixing.
Comparisons of predictions with experimental
measurements have been presented by others [13-
21], and in many cases these models predict
turbulent combustion processes with acceptable
engineering accuracy. For the present model,
mixing-limited reactions are assumed. Fluctuations
of species concentration, density, and temperature
are included by an assumed probability density
function (P.D.F.) coupled with local chemical
equilibrium [ 10, 17]. The contribution of the present
0010-2180/81/090277+09502.50
278
work lies in the attempt to predict complete chemi-
cal composition and to compare the predictions
with the measurements reported in Part I.
THEORETICAL M O D E L
Fluid Mechanics
A cylindrical reactor is fed by a coaxial jet with the
fuel in the central or primary tube and the oxidizer
in the annulus or secondary stream. Time steady
equations of continuity and motion for the axisym-
metric model are those reported [22], with
Boussinesq gradient diffusion used for turbulence
closure. The Prandtl-Kolmorgorov mean turbu-
lence energy closure model [3] relates the eddy
viscosity to the turbulence kinetic energy, and its
rate of dissipation:
~,, = c ,)li~ / ~, (1)
with differential equations for k and E taken from
Launder and Spalding [3]. The values of the
constants were those recommended by Khalil et al.
1"6]. Because of the assumed relation between the
Reynolds stresses and the mean strain rates, this
model implies that the turbulence is able to adjust
instantly to local changes in the mean flow field.
The performance of real systems, however, in-
dicates that history and action at a distance play
important roles. The k- E model also has difficulty
when the local Reynolds number of turbulence is
less than unity [23]. In addition, body forces, which
have been neglected, may have an influence on the
generation, damping and transport of turbulent
quantities.
Chemical Reactions
Our main concern has been to predict complete
species concentration profiles for turbulent gaseous
systems. The fuel and oxidizer are initially in
different streams, and must mix on a molecular
level before reaction can occur. It is assumed that
this process is rate limiting. This allows the chem-
istry to be computed from equilibrium consider-
ations, for all important species or for any subset of
possible species. Only one differential equation is
PHILIP J. SMITH and L. DOUGLAS SMOOT
required to describe the degree of"mixedness" at a
point.
For cases where there are two identifiable feed
streams with uniform properties, a conservative
scalar f (the mixture fraction) is defined:
f=Mpl(Mp+ Ms). (2)
A triangular or saw-tooth wave form for the
fluctuations infis assumed with a resulting uniform
or top-hat probability distribution. The Reynolds
averaged differential equation for the transport o f f
is [3, 10]
a (rp,~f)+fr (rp~y)
~x
8x \a, c~xJ-~ (rr ~rr =0. (3)
This equation, together with the fluid mechanics
model and the mean density, determines the mean
flow field. F r o m f a n d / ~ the time-mean values for
any arbitrary variable fl (i.e., density, temperature,
species mole fraction) are then
fl=apflp+asfls+ f~ fl(f)P(f) df. (4)
If fl is a function of any number of conservative
scalars, then it is a function only off. The variance of
the mixture fraction g is given by
g=(f_f)2 =T [ f ( t ) - f ] 2 dt, (5)
where T is large compared to the time scale of the
local turbulence. The modeled differential trans-
port equation for g is that of Launder and Spalding
[3] with model constants recommended by Khalil
et al. [6]. The element mass fractions bk are
conserved scalars and not dependent on the as-
sumption of chemical equilibrium:
bk----fbkp+(1 -- f)bks. (6)
In many situations the standard gas enthalpy is a
conserved scalar and can also be calculated from
TURBULENT DIFFUSION FLAMES II: THEORY 279

the mixture fraction. Even when heat losses are TABLE 1

significant, this approach can be used with minor Combustor Parameters and Test Conditions
adjustments [24].
The equilibrium properties, such as species con- No. 1 No. 2
centration Yi, are now a function of f alone: Combustor [Part 1 ] [ 18 ]

Primary gas
Yi = yi(bk, h) = yi[bk(f), h(f)] = Yi(f). (7) Temperature (K) 286 344
Mass flow rate (kg/s) 0.0031 0.0032
Mean species mole fractions Yi and other properties Velocity (m/s) 21.7 21.6
such as temperature and density are obtained by Composition (molar %):
H2 - 55
Eq. (4). In this way, the unmixedness is taken into
CH 4 85.8 27
account. Even with equilibrium chemistry, the CO 2 - 8
time-averaged concentrations of fuel and air may N2 2.3 4
both be finite at the same point in space. CO - 4
Experimental evidence [25, 26, Part I] indicates C2H6 6.1 2
Ar 5.5 -
considerable overlap of the oxidant and fuel
Secondary gas
profiles. Temperature (K) 589 301
Mass flow rate (kg/s) 0.0362 0.0513
Solution Technique Velocity (m/s) 34.7 13.5
Composition (molar %):
The solution of the flow equations incorporated in N2 78.3 78
TEACH [27] has been used in this model. It is an 02 20.8 22
iterative, steady-state, finite-difference approach. Ar 0.9 -
Central differencing is used when the absolute value Parallel injection Yes Yes
Pressure (N/m2) 9.34 X i04 1.01 X 10s
of the cell Reynolds number is less than 2. Primary diameter (m) 0.016 0.020
Otherwise, upwind differencing or donor-cell dif- Secondary diameter (m) 0.029 0.078
ferencing is used for the convection terms. Mean Chamber diameter (m) 0.203 0.210
field chemical properties determined through P0O

1.0

.70 ~ ~- - ~ ~ ~ , / - - A ' '

~ .50
u.

P .30
x
i .2o
®
¢

®
. .10
®
o .07

.05 I I I I I I I I I I
2 3 5 7 10 20 30 50 70 100 too

Axial Location Ix/rpl

Fig. 1. Effect of grid size on centerline mixture fraction decay in laboratory combustor
No. 1 (Table 1). Grid size: A = 31 X 31, B = 25 X 25, C = 20 X 20, D = 15 X 15, E =
20× 31, F = 2 0 X 2 5 .
280 PHILIP J. SMITH and L. DOUGLAS SMOOT

1.0
.7O
.50

.30
measurements E ~ O ~~
i .2o
==

•~ .lO
ID
o .07
.05 I i i I ! I I I I I
2 3 5 7 10 20 30 50 70 100 200

Axial Location [X/rpl

Fig. 2. Effect o f inlet turbulent intensities on centerline mixture fraction decay in
laboratory combustor No. 1 (Table 1). Turbulent intensities: (A) Ip = 2%, I s = 2%;
(B) I p = 2%, I s = 6%; (C) Ip = 6%, I s = 6%; (D) l p = 10%, Is = 6%; ( E ) I p = 10%,I s =
10%. Data from Part 1.

are calculated by Gaussian quadrature from the
instantaneous equilibrium properties. The equilib-
rium subroutine uses a minimization of the local
Gibbs free energy 1-28].
C O D E PARAMETERS
Grid Size
These predictions were performed for the labo-
ratory combustor described in Part I. In the
particular tests of interest the combustor was fired
with natural gas from the center jet stream and
preheated air from the surrounding annulus. The
parameters and test conditions for this reactor
(combustor No. 1) are summarized in Table 1.
Fig. 1 shows centerline mixture fraction decay for
several different symmetric and asymmetric grid
sizes. The size of the grid used is identified as I x J,
where I and J are the number of main nodes in the

~ ~
1.0

.70 Ref.14" ' "

._g
,~ .50
M "~ 0~ ~,~-Re,. 18
lx

~ 3O ThisstudyJ ~ = ~ ~

i .2o

o
@
.10

.07
O.t. ,., %0000
.05 i I I I I I I i I i

2 3 5 7 10 20 30 50 70 100 200

Axial Location [X/rp]

Fig. 3. Comparison o f experimental data o f the centerline mixture fraction with
three combustion models. (Table 1, eombustor No. 2).
TURBULENT DIFFUSION FLAMES II: THEORY 281

Mixture Fraction 02 tO
-.2

~ ~ . o 0-
2.5-
y~Xy~c~'C~3CrCt
0-
-.2 5-
¢N
o 0-
10"
-,2
5-
tO

15- 0[]0
10-
'q' -0
5"
20"

15-

10-
,, .4
5-
~' .2
tO ~ o 0. m
~ o
-- : E 2 0 .
rnD
15-

10-

5- ~

~
%8 o
20 I"7
- . 6

15
• -.4
10
Lf)
-.2
5
~ 0

Radial Radial
Location Icml Location Icm]

Fig. 4. Comparison o f (left) mixture fraction and (right) oxygen mole fractions from
theory and measurements for natural gas combustion. Data points are measurements
from combustor No. 1 o f Table 1.

axial and radial directions, respectively. Grid size
has a significant impact on the results. By contrast,
calculations [24] performed for a nonreacting flow
device showed virtually no deviation for grid sizes
of 20 x 20 and 31 × 31. In this second example,
numerical error due to grid resolution was elim-
inated with a grid size of 20 x 20. The main
difference between the two results seems to be in the
size of the combustor relative to the primary or
secondary inlet tubes. For smaller ratios, a coarser
grid is adequate.
Turbulence Parameters
Turbulent intensities of the inlet streams are not
always known. Fig. 2 illustrates the effect of inlet
turbulent intensity on the centerline mixture frac-
tion. All computations were performed with a
282 PHILIP Jo SMITH and L. DOUGLAS SMOOT

P.N.
CO2

~&&&&&&&&
5"
10- t~

1y0 "7~ ~ ' ' ~ ~ 5 " ~'

o
10' b- o Q

e= o
Q.
o o ,,mR&&
o

-40

-20

-0
-80
.60

40

-20 10- f
o 0 O-
lO ~ ~ ~ I I
Radial Radial
Location Icrnl Location Icm)
Fig. 5. Comparison of (left) methane arid (right) carbon monoxide mole fractions
from theory and measurements for natural gas combustion. Data points are measure-
ments from combustor No. 1 of Table 1.

constant grid size of 20 x 30. The strong effect of
initial turbulent intensity on the apparent core
length and the decay slope is obvious, but for these
experiments curves B, C, and D represent the most
reasonable values for the inlet intensities.
COMPARISON WITH M E A S U R E M E N T S
Predictions were made for the two laboratory
combustors of Table 1. Data from combustor 2 has
also been used for comparison with predictions by
Elghobashi [18] and Gosman et al. [14]. Fig. 3
shows the predictions of the centerline mixture
fraction decay and the experimental measurements
[14] for combustor 2. Each of the models uses the
k - ~ theory for turbulence closure but a different
basis for describing the chemical reactions. The
present model predicts the mixing phenomena for
reacting gases in the same range as do the predic-
tions of [14] and [18] and in the same range as is
TURBULENT DIFFUSION FLAMES II: THEORY 283

20
observed experimentally. However, the magnitude l I I I

H20
of the differences is of concern. 0
A comparison of the predicted centerline axial
decay of the mixture fraction and the results of Part IS 7o z~
I was shown in Fig. 2, curve C. Fig. 4 (left) shows N2 ~
predicted and measured radial profiles of the
c
0
mixture fraction at various axial locations.
I0 60 L~,
Deviations from experimental measurements are CO2
_o
greatest in regions of steep gradients where the o o o o 0
0 0 0 0 0 0
lowest accuracy existed in the experimental data.
Figs. 4 (right) and 5 show comparisons of selected
species concentration profiles at several combustor
axial stations. Agreement between theory and
._ co ..72.1.0
measurement is least accurate in the early regions of
the reactor near the centerline. The agreement in the
recirculation zone is reasonable. 0 .02 .04 ,06 .08 .10
Although all other computations were performed Radial Location [m]
with full equilibrium among all important species, Fig. 6. Radial variations of the mole fractions at an a×~l
the predictions in Figs. 4 (fight) and 5 were location of 137.5 cm. Predictions include H 2 and CO
made by eliminating any formation of CO and formation.

H2. Predictions which included H 2 and CO in the

computation [24] showed concentrations of these
species to far in excess of measured values in
certain regions of the reactor. In particular, predic-
tions suggested that the fuel was pyrolyzing in hot,
fuel-rich regions. The theory shows H 2 mole
fractions in excess of 15%, whereas the measure-
ments showed less than 0.1% [Part I]. An estimate
of the kinetic limitations was made with a detailed
laminar flame propagation model [29]. The com-
puted characteristic reaction time was ten times
larger than the estimated eddy life time. While other
major species (CO2, H20, 02, CH 4) had reached
their equilibrium values, the computed H 2 and CO
levels were well below equilibrium values. It was
concluded that initial H 2 formation is most likely
reaction rate limited and does not form in equilib-
rium amounts.
In the aft regions of the reactor, the CO2 and
CH 4 agreement is not particularly good. This is due
to the restriction on H 2 and CO formation.
Experimental measurements show about 2% H 2
and CO in these regions which account for the
discrepencies. In these later regions in the reactor,
equilibrium is more closely approached, even for
H 2 and CO. Fig. 6 shows the model predictions at
an axial position of 137.5 cm when CO and H 2 are
included in the calculation. This should be com-
pared with the plot at the same location in Fig. 5.
Agreement for Fig. 6 is much improved. Fuel and
oxidizer overlap in all the regions of the reactor (see
Fig. 5). This predicted result, which is due to
unmixedness, is also observed experimentally.
CONCLUSIONS
The combustion model, which uses a mixture
fraction P.D.F. and assumes local instantaneous
equilibrium (i.e., micromixing limited), permits pre-
diction of complete species composition in gaseous
combustion chambers. The statistical P.D.F. pre-
dicts the observed overlap of fuel and oxidiTing
species. The technique is sensitive to initial turbu-
lence levels of the primary and secondary streams.
Agreement between the measured and predicted
mixture fraction is good. Model predictions for
species composition compare reasonably well with
measurements except in areas of step gradients
where the accuracy of the probe measurements is
questionable. However, H 2 and CO predictions
were noticeable deviants. Preliminary indications
show that these species are likely kinetically limited
in forward regions of the combustor.
284 PHILIP J. SMITH and L. DOUGLAS SMOOT

NOMENCLATURE 4. Mellor, G. L., and Herring, H. J.,AIAA J. 11:590-

599 (1973).
5. Spalding, D. B., in Turbulence Mixing in Nonreactive
b element mass fraction
and Reactive Flows, Ed. S. N. B. Murthy, Plenum,
C "universal" constant New York, 1975.
f mixture fraction 6. Khalil, E. E., Spalding, D. B., and Whitelaw, J. H.,
g mean square fluctuation of f Int. J. Heat Mass Trans. 18:775-791 (1975).
h enthalpy (J kg-1) 7. Bray, K. N. C., Seventeenth Symposium (Interna-
tional} on Combustion, The Combustion Institute,
I turbulent intensity
Pittsburgh, Pa., 1978, p. 223.
k kinetic energy of turbulence 8. Bowman, C. T., Progr. Energy Comb. Sci. 1:33-46
(in 2 S-2) (1975).
M mass of fluid atoms (kg) 9. Caretto, L. S., Progr. Energy Comb. Sci. 1:47-72
P probability density function (1976).
10. Bflger, R. W., Progr. Energy Comb. Sci. 1:87-109
r radial direction (m)
(1976).
t time (s) 11. Borghi,R.,Adv. Geophysics 18B:349 (1974).
u axial velocity component (m s-1) 12. Spalding,D. B.,J. Energy 2:16-23 (1978).
v radial velocity component (m s-1) 13. Spalding, D. B., in Turbulent Combustion, Ed. L.
x axial position (m) A. Kennedy, Progr. Astronautics and Aeronautics
58:105-116 (1978).
y mole fraction of species
14. Gosman, A. D., Lockwood, F. C., and Salooja, A.
~t intermittency factor P., Seventeenth Symposium (International) on Com-
fl arbitrary property (variable) bustion, The Combustion Institute, Pittsubrgh, Pa.,
E dissipation rate of turbulent kinetic 1978, p. 747.
energy (m 2 s-3) 15. Libby, P. A., Comb. Sci. Tech. 13:79-98 (1976).
# dynamic viscosity (kg m - 1 s-1) 16. Spalding,D. B., Chem. Eng. Sci. 26:95-107 (1971).
17. Becker, H. A., Fifteenth Symposium (International)
p density (kg m - 3 ) on Combustion, The Combustion Institute, Pitts-
a Schmidt or Prandtl N u m b e r burgh, Pa., 1974, p. 601.
18. Elghobashi, S., in Studies in Convection, Vol. 2,
Subscripts Ed. B. E. Launder, Academic Press, London, 1977,
p. 141.
e effective 19. Hutchinson, P., Khalil, E. E., and Whitelaw, J. H., in
f mixture fraction Turbulent Combustion Ed. L. A. Kennedy, Progr. in
i gas species A stronautics and Aeronautics 58:211-222 (1978).
20. Abou Ellail, M. M. M., Gosman, A. D., Lockwood, F.
k element
C., and Megahed, I. E. A., in Turbulent Combustion
p primary stream Ed. L. A. Kennedy, Progr. Astronautics and Aero-
s secondary stream nautics 58:163-190 (1978).
p viscosity 21. Patankar, S. V., in Studies in Convection, Vol. 1,
Ed. B. E. Launder, Academic Press, London, 1975,
p. 1.
Superscripts
22. Smoot, L. D., and Pratt, D. T.,(Eds.),Pulverized Coal
Combustion and Gasification, Plenum, New York,
-- Reynolds mean
1979.
23. Jones, W. P., and Launder, B. E., Int. J. Heat Mass
Trans. 15:301-313 (1972).
REFERENCES 24. Smith, P. J., Ph.D. Dissertation, Department of
1. Smith, P. J., Fletcher, T. H., and Smoot, L. D., Chemical Engineering, Brigham Young University,
Provo, Utah, (1979).
Eighteenth Symposium (International) on Combus-
25. Hawthorne, W. R., Weddell, D. S., and Hottel, H.
tion, The Combustion Institute, Pittsburgh, Pa.,
C., Third Symposium on Combustion, Flame and
1980.
Explosion Phenomena, Williams and Wilkins, Balti-
2. Spalding, D. B.,Combust. ScL Tech. 13:3-25 (1976).
more, 1949, p. 228.
3. Launder, B. E., and Spalding, D. B., Mathematical
Models of Turbulence, Academic Press, London, 26. Kent, J. H., and Bilger, R. W., "Turbulent Diffusion
1972. Flames," Kolling Report F-37, Department of Me-
TURBULENT D I F F U S I O N FLAMES II: THEORY
chanical Engineering, the University of Sydney
(1971).
27. Gosman, A. D., and Pun, W. M., "Lecture Notes for
Course entitled'Calculationof Reeirculating Flows'",
Imperial College, London (December 1973).
285
28. Pratt, D. T., in Studies in Convection, Vol. 2, Ed. B.
E. Launder, Academic Press, London, 1977, p. 191.
29. Smoot, L. D., Hecker, W. C., and Williams, G. A.,
Combust. Flame 26:323-342 (1976).
Received 8 October 1979; revised 10 July 1980