Combustion and Heat Transfer in Model Two-Dimensional

Porous Burners
C. L. HACKERT,† J. L. ELLZEY, AND O. A. EZEKOYE*
Department of Mechanical Engineering and Applied Research Laboratories, University of Texas at Austin,
Austin, TX 78712, USA

A two-dimensional model of two simple porous burner geometries is developed to analyze the influence of
multidimensionality on flames within pore scale structures. The first geometry simulates a honeycomb burner,
in which a ceramic is penetrated by many small, straight, nonconnecting passages. The second geometry consists
of many small parallel plates aligned with the flow direction. The Monte Carlo method is employed to calculate
the viewfactors for radiation heat exchange in the second geometry. This model compares well with experiments
on burning rates, operating ranges, and radiation output. Heat losses from the burner are found to reduce the
burning rate. The flame is shown to be highly two-dimensional, and limitations of one-dimensional models are
discussed. The effects of the material properties on the peak burning rate in these model porous media are
examined. Variations in the flame on length scales smaller than the pore size are also present and are discussed
and quantified. © 1998 by The Combustion Institute

NOMENCLATURE t time, sec

T temperature, K
A channel cross sectional area, m2
u axial gas velocity, m/sec
a constant in heat capacity expression
U burning rate or volume flux, m/sec
B constant in heat capacity expression
vW gas velocity, m/sec
c solid heat capacity, J/kg/K
x axial distance
cp gas heat capacity at constant pressure,
y transverse distance
J/kg/K
yf fuel mass fraction
d porous medium square unit cell side
yN nitrogen mass fraction
length, m
yo oxygen mass fraction
D mass diffusivity, m2/sec
yp combustion products mass fraction
E chemical reaction activation energy, J/
mol
Greek
F surface to burner end viewfactor
h gas specific enthalpy, J/kg a thermal diffusivity, m2/sec
hc heat of combustion, J/kg fuel e surface emissivity
J radiosity, W/m2 f equivalence ratio
k thermal conductivity, W/m/K m absolute viscosity, kg/m/sec
K surface to surface viewfactor r density, kg/m3
ṁ mass flux, kg/m2/sec v reaction rate, kg fuel/m3/sec
P pressure, Pa
q0 conduction/convection heat flux to Subscripts
surface, W/m2
avg mass flux average
q- volumetric heat flux to solid, averaged
i, j vector or matrix component
over pore, W/m3
` ambient conditions
R ideal gas constant, 8.314 J/mol/K
s stoichiometric fuel/oxygen ratio
INTRODUCTION
*Corresponding author. Address: O. A. Ezekoye, Assistant
Professor, Department of Mechanical Engineering, Univer-
sity of Texas at Austin, ETC 7.130, Austin, TX 78712. Early theoretical work on premixed flames
†
Current address: Southwest Research Institute, P. O. showed that the lean flammability limit could be
Drawer 28510, San Antonio, TX 78228-0510. extended and the burning rates increased if heat
COMBUSTION AND FLAME 116:177–191 (1999)
© 1998 by The Combustion Institute 0010-2180/99/$19.00
Published by Elsevier Science Inc. PII S0010-2180(98)00052-2
178 C. L. HACKERT ET AL.

could be recirculated from the hot products to study on an uninsulated honeycomb burner, in
the incoming reactants [1]. Takeno et al. [2] and which the propane flame is confined within a
Takeno and Sato [3] showed theoretically that ceramic penetrated by many small, straight,
conduction through a porous solid inserted into nonconnecting passages. They investigated the
a premixed flame provided this means of heat limits of stability of these flames in terms of flow
recirculation. Echigo et al. [4] showed experi- rates and equivalence ratios, and noted the
mentally and analytically that radiation from existence of two fundamentally different flame
screens in a heated flow recirculated heat from shapes. They label these flame shapes as Type I
the downstream to the upstream regions. and Type II. The Type I flame is largely flat
There has been substantial research in the across the burner, i.e., the flame is at roughly
area of porous media combustion, and a good the same distance from the inlet in every flow
review of the subject is given by Howell et al. element. The Type II flame occurs for lower
[5]. One of the first experimental studies of a flowrates and leaner flames than the Type I
laminar porous burner was conducted by Kotani flame. It is more curved in shape, with the flame
and Takeno [6], who used a ceramic tube bun- at the downstream end near the burner edge
dle as the porous material and routed the and fairly flat in the upstream half near the
exhaust gases back around the burner circum- burner center.
ference to provide preheating of the inlet mix- In addition to the practical applications, these
ture and to minimize heat losses from the burners provide an interesting area of study by
reaction zone. They were able to extend the combining complex multimode heat transfer
lean limit of their methane flame to 0.32 and and chemistry with a laminar flow field. In
noted reduced CO and NOx compared to a free principle, then, these burners can be modeled
flame. Burning rates significantly higher than without recourse to ad hoc turbulence models.
the free flame were also found. Turbulent re- Many investigators have in fact been very suc-
fractory tube burners using a similar, but larger, cessful in predicting burning rates, extended
geometry were investigated over several years flammability limits, and superadiabatic opera-
by Churchill and coworkers [7–11]. tion with one-dimensional laminar flow models
Further experiments on a methane flame [2, 3, 16 –18]. One-dimensional representations
within a sponge-like reticulated ceramic were of porous media flows, however, require models
conducted by Sathe et al. [12] and Hsu et al. [13] for the gas to solid convective heat transfer and
without the use of external exhaust gas recircu- the solid phase radiative heat transfer. While
lation, although the burners were generally well measurements of some of these properties have
insulated. Heat recirculation in these burners been made [19 –21], the values recorded are
was provided by conductive or radiative ex- generally either fairly uncertain or taken at
change between the postflame and preflame Reynolds numbers or temperatures far from the
regions. The results of these experiments are range encountered in burner applications. In
generally similar to those of Kotani and Takeno addition to the uncertain volume averaged
with respect to reduced lean limit, increased properties used for the radiation, one-dimen-
burning rate, and lowered pollutant emissions, sional radiation models must break down at
although the magnitude of the difference is length scales smaller than the pore size because
somewhat reduced. By removing the exhaust they treat the porous matrix as a continuously
gas recirculation, however, radiation from the participating medium rather than as consisting
burner surface may be captured for industrial of individual solid surfaces. Also important on
use. Experiments have shown that this type of pore scales is a related difficulty associated with
porous burner can convert a significant percent- flame curvature. It is difficult to imagine that
age of the chemical heat release into down- the flame is flat and uniform on a pore scale.
stream thermal radiation [14]. A remedy to these problems, of course, is to
A simple geometry in which the coupled use multidimensional modeling. This is the
effects of conduction and radiation modify the method employed by Sahraoui and Kaviany
flame burning properties was examined by Min [22], who compared two-dimensional simula-
and Shin [15]. They performed an experimental tions of combustion in porous media with vari-
MODEL TWO-DIMENSIONAL POROUS BURNERS 179

ous one-dimensional formulations. Because of

computational limitations and the large number
of simulations performed, they did not include
either multistep chemistry or radiation heat
transport. Research by Hsu and Matthews [23]
has shown that including detailed chemistry can
be important for near stoichiometric and rich
flames, but is less important in modeling very
lean flames. Radiation heat transport, however, Fig. 1. Burner geometry for honeycomb burner. Dotted line
is known to be critical in establishing the char- shows single-channel computational domain.
acter of porous burner flames. Although similar
effects can be achieved using a high solid con-
ductivity, in some geometries it is primarily the is shown in Fig. 2. The plate width is specified to
radiation heat transport which provides most of be 10% of the pore size d, so that the porosity
the preheating of the unburned mixture. The of the medium is 95%. The porous burner is
work presented in this paper extends that of assumed to be of finite length in the axial ( x)
Sahraoui and Kaviany by including surface-to- direction, but is assumed to be infinite in the
surface radiation. transverse ( y) direction. The base case for this
In this study, we simulate two different po- study is a 1.28 mm pore size, or 20 ppi (pores
rous geometries. The first is a honeycomb po- per inch) medium about 25 mm long, with solid
rous media consisting of parallel nonconnecting emissivity of 0.4 and conductivity of 2.2 W/m/K.
passages similar to the experimental work of Since the flow is steady and at low Reynolds
Min and Shin [15]. For this geometry, solid
number, we assume that there is no pore-to-
conduction and radiation from the tube walls
pore variation in the transverse direction. This
heat the solid upstream of the flame. In the
allows the use of symmetry planes (marked in
second geometry, the walls are parallel but
Fig. 2) to cut the computational domain to
broken such that there is no longer a continuous
solid path. The importance of solid conduction one-half pore in the cross direction. The com-
is minimized and surface-to-surface radiation is putational domain extends 5 mm upstream and
the only means of transferring heat upstream. 5 mm downstream of the porous medium.
These results are compared to experimental
measurements from burners with reticulated
porous ceramics.

NUMERICAL TECHNIQUE

Geometry and Computational Domain

The geometry for the honeycomb burner is

shown in Fig. 1. The dimensions and properties
are those from the burner of Min and Shin [15].
The channel width is 1.1 mm and the wall
thickness is 0.17 mm resulting in a porosity of
75%. The total length is 20 mm. The solid
emissivity is 0.4 and the solid conductivity is 2.2
W/m/K. The computational domain is shown
with dotted lines. It extends 5 mm upstream and
5 mm downstream of the porous solid. Fig. 2. Burner geometry for separated plates burner with
The second porous geometry consisting of planes of symmetry, computational domain, and method of
very thin plates infinitely long in the z direction radiation transport.
180 C. L. HACKERT ET AL.

Governing Equations v 5 3.90 3 10 10~ y fr ! 1.3~ y or ! 0.9e 2E/RT (9)

3
In this problem, we solve conservation equa- for propane combustion. Here v is kg/m of fuel
tions for mass, momentum, gas phase energy, consumed per second, E is the activation energy
solid phase energy, and species. We neglect for both reactions (125.6 kJ/mol), R is the ideal
body forces, the Soret and Dufour effects, and gas constant, and T is the local gas temperature
in Kelvin. The mechanisms were validated
gas radiation. This last assumption is justified by
against experimental data of Yamaoka and
a mean beam length analysis [24] which shows
Tsuji [26] and Yu et al. [27] for equivalence
that for radiant length scales on the order of a
ratios between .55 and 1.0. The measured flame
millimeter the gas radiation is under 1% of the
speeds of Yu et al. are approximately 10%
radiation from adjacent solid surfaces. The
higher than those of Yamaoka and Tsuji for
equations are then
methane. Our model predicts flame speeds
Continuity:
which are between the two measured values
­r except at an equivalence ratio of 0.9 where it
W z ~ r vW ! 5 0,
1¹ (1) predicts a flame speed approximately 2% lower
­t
than that measured by Yamaoka and Tsuji. For
Momentum: propane, our model predicts flame speeds which
are within 2% of the values of Yamaoka and
­ ­P
W z ~ r vW v i! 5 ¹
~ r v i! 1 ¹ W z ~m¹
W v i! 2 (2) Tsuji except at equivalence ratios near 1.0
­t ­ xi where the variation is 7%.
Gas phase energy: The density is obtained directly from the ideal
gas law using the pressure, the temperature, and
­
­t
W z ~ r vW h! 5 ¹
~ r h! 1 ¹ W z
k
S D
cp
W h 1 h cv
¹ (3)
the mixture composition. Other fluid properties
are modeled on the known temperature-depen-
dent values for air [28]. Good agreement with
Solid phase energy:
the viscosity of air is found by setting m 5
­ 3.371 3 10 27 T 0.7 , where T is in Kelvin and m
W z ~k¹
~ r cT! 5 ¹ W T! (4) is in kg/m/sec. Similarly, the thermal diffusion is
­t
assumed to be given by ra 5 k/c p 5 4.816 3
Fuel diffusion: 10 27 T 0.7 , where units are as above. This
corresponds to a fixed Prandtl number of 0.7.
­
W z ~ r vW y f! 5 ¹
~ r y f! 1 ¹ W z ~ r D¹
W y f! 2 v (5) The Lewis numbers for all species are assumed
­t to be unity at all temperatures, and so r D
Oxygen diffusion: follows the same temperature dependence as
ra. The heat capacity of the mixture is modeled
­ by c p 5 a exp(BT). In this expression, a 5
W z ~ r vW y o! 5 ¹
~ r y o! 1 ¹ W z ~ r D¹
W y o! 2 s v (6)
­t 947 1 0.12 f J/kg/K for methane and a 5
947 1 0.095 f for propane fuels, and B 5
Product diffusion:
1.83 3 10 24 K21. In the limit of f 5 0 this
­ expression does very well in matching the
W z ~ r vW y p! 5 ¹
~ r y p! 1 ¹ W z ~ r D¹
W y p! 1 ~s 1 1! v known values of c p for air. For other values of f,
­t
(7) this expression follows the adiabatic flame tem-
peratures for the two fuels.
A single-step global reaction mechanism for
the complete combustion of fuel is used. Fol- Boundary and Interface Conditions
lowing the form outlined by Westbrook and
Dryer [25], the reaction rate is given by At the inlet to the computational domain which
is 5 mm upstream of the burner inlet, the
v 5 8.45 3 10 11~ y fr ! 1.8~ y or !e 2E/RT (8)
velocity, temperature (300K), and mass fraction
for methane combustion; and (based on equivalence ratio) of the incoming
MODEL TWO-DIMENSIONAL POROUS BURNERS 181

gas flow are specified. At the downstream end traveling straight until it strikes a surface with
of the computational domain the gradients of identical properties several pores to one side.
velocity, temperature, and mass fraction are Every ray emitted under this method is tracked
zero. At the transverse or radial boundary, there until impacts a surface (considering the burner
is no heat loss except when otherwise specified. exit and inlet as surfaces). Since all rays are
Several interface conditions are applied at the accounted for, and the viewfactor is the fraction
gas–solid interface within the computational of rays impacting on a given surface, the sum of
domain. The first is a no-slip condition, such all viewfactors for a given surface is one. The
that gas velocities are zero at the solid surface. radiation boundary condition enforced in Eq. 11
Secondly, the solid is treated as being imperme- is that the burner surfaces exchange radiation
able to species and noncatalytic. This implies with a large, cold (300 K) environment at both
that the normal gradient of species concentra- the upstream and downstream ends.
tion at the solid surface is zero. Finally, the
energy interface condition includes the effects Solution Technique
of gas-to-solid heat conduction and the effects
of solid-to-solid radiation. This is expressed as The base code solves Eqs. 1–7 using an alter-
nating direction implicit (ADI) finite volume
k solid
­T
­n
U solid
5 k gas
­T
­n
U gas
formulation, and the pseudotransient method
for steady problems [29]. The pressure field is
solved using the SIMPLE method [30].
e The surface radiation equations (11) are
1 ~ J 2 s T 4solid! (10)
12e solved on the same grid as the other equations.
Once the viewfactors are calculated, the radia-
where n is a direction normal to and out of the
tion equation is with a discretization equal to
surface. The-two dimensional model formula-
the grid size. This yields 300 to 600 radiating
tion allows the elimination of the arbitrary
surfaces in the burner model, depending on
gas-to-solid heat transfer coefficient that must
geometry and pore size.
be used in one-dimensional models. The solid
Typical grid spacing is 30 – 40 mm in the
surfaces are considered to be gray and diffuse
burner region, with slightly larger finite volumes
and the surface radiosities J are solved as a
upstream and downstream of the burner. A
matrix equation,
typical computational domain would then be
J i 5 es T 4i 1 ~1 2 e !@K ij J j 1 F is T 4`#. (11) approximately 800 by 30 grid points. This grid-
ding system was proven to be sufficient by
In the honeycomb geometry, analytic expres- testing several grid sizes. The reduction in error
sions for the viewfactors K and F are available with smaller grids is shown to be of order Dx 1.9 ,
[24]. We solve for the radiation in the separated consistent with the order Dx 2 method used. A
plates geometry using the symmetry planes change in grid size from 20 mm to 40 mm
shown in Fig. 2. Since the geometry is fairly resulted in less than a 0.01% change in burning
complicated, the viewfactors for the thermal rate.
radiation exchange are calculated using a
Monte Carlo method [24]. In this method, a
large number of rays are emitted from each of HONEYCOMB BURNER
the discretized solid surfaces in random direc-
tions and are tracked until they reach another Comparison to Experiments
surface. When a statistically significant number
of rays have been emitted, the impacts are While the experiments of Min and Shin [15]
counted and the viewfactors relating the sur- were performed using square channels, this
faces calculated. As shown in Fig. 2, a ray obviously cannot be done for two-dimensional
emitted from a surface is reflected across the computations. Instead, two types of geometry
symmetry planes until it strikes a solid object or will be used: the cylindrical tube and the parallel
leaves the burner. This is equivalent to the ray planes. The tube is a good choice for modeling
182 C. L. HACKERT ET AL.

quite different from predictions from one-di-

mensional models which volume average the
solid and gas in any given cell. It possible
however to compare our calculations with both
one-dimensional simulations and experiments
which are based upon average conditions by
defining a mass averaged temperature.
The computational gas temperatures (Fig. 3)
may be presented as a mass averaged gas tem-
perature at a particular x location

Fig. 3. Calculated two-dimensional isotherms for f 5 0.55,

flowrate of twice the laminar flame speed.
T avg 5
1
ṁA E r uT dA (12)

where A is the cross sectional area and u is the

normal velocity. This is, of course, less than the
a single square element. A square channel of peak gas temperature for any axial location.
side length d has the same hydraulic diameter as Although Min and Shin made experimental gas
a tube of diameter d. Also, considering a single temperature measurements using thermocou-
flow channel only, the porosity of a square cell ples which in theory should represent a point
with a given wall thickness is the same as the measurement more than the mass averaged
porosity of a cylindrical channel with the same value, there are several points which make an
wall thickness. In the parallel plates case, the exact comparison difficult. The location of the
hydraulic diameter is twice the plate separation bead is not exactly stated, and the thermocouple
distance, and the porosity for a given wall support and bead sizes may be intrusive. Ther-
thickness and hydraulic diameter is within a few mocouple bead sizes were 25 mm for the gas
percent of the square channel case. Reynolds temperature measurements and 0.1 mm for the
numbers are uniformly low. The highest Reyn- solid temperature measurements [15]. In addi-
olds number (based on hydraulic diameter) is tion, each thermocouple was supported by a
113 for any result presented below, and a typical 0.5-mm ceramic tube. For comparison, recall
Reynolds number is 47. Modeling a tube ele- that the channel size is 1.1 mm. Because of
ment as a parallel plate channel of the same these complications the simplest assumption is
hydraulic diameter introduces little additional to compare our average temperatures with their
error. The burning rates for the two cases lie experimentally measured values.
very close to each other for all flame locations. In Fig. 4 we compare the computed average
We present our results in terms of the burning gas and solid temperatures to those measured
rate (cm/sec), which is the volume flux of mix- by Min and Shin. In the experiments, a series of
ture (m3/sec/m2 5 m/sec) supplied to the burn- walls were removed from the square channels to
er; and the radiant fraction, which is the fraction allow for easier placement of thermocouples.
of energy released by the combustion reaction The resulting passage was a parallel planes
that is radiated out of the burner through either channel surrounded by square channels. As
the upstream or downstream end. We refer to previously noted, it is impossible to exactly
the radiant fraction at the downstream end as simulate the actual three-dimensional experi-
the radiant output and the radiant fraction at ment with a two-dimensional model. As an
the upstream end as the radiant loss. approximation, the computational model used
The calculated temperature distribution is is that of a single cylindrical channel with radi-
two-dimensional, and a condition where the ally insulated walls.
flow rate is set at twice the laminar flame speed The agreement between prediction and ex-
and f 5 0.55 is shown in Fig. 3. Note that in periment is good for temperatures downstream
spite of the small length scale, the temperature of the flame, but there is a discrepancy in the
field is significantly curved. This prediction is peak gas temperature. There are several possi-
MODEL TWO-DIMENSIONAL POROUS BURNERS 183

Fig. 5. Range of stable operation of honeycomb burner as a

function of equivalence ratio.
Fig. 4. Comparison of spatially averaged calculated temper-
atures for radially insulated element to experiment, f 5
0.55, flowrate is about twice the laminar flame speed. burner inlet, with predicted temperatures ex-
ceeding 1800 K for some cases.
In Fig. 5, we compare the range of flowrates
ble causes for this, including the modified ex- that produce stable flames in radially insulated
perimental geometry and the lack of detailed elements to the experimentally measured values
chemistry in the present calculations. Further- for the Type I flame of Min and Shin [15]. The
more, recall that the computed gas temperature lower computed limit for the radially insulated
is an average of the true, two-dimensional tem- case compares well with the Type I experiments,
perature distribution previously shown in Fig. 3. supporting Min and Shin’s contention that the
The peak computed gas temperature in Fig. 3 is Type II shaped flame occupies a regime domi-
higher than that shown in Fig. 4. Another nated by radial heat losses. The computed up-
contributing factor to the discrepancy is that in per limit of operation is significantly above the
the experiment some of the upstream radiation experimental value. This increase is attributed
from the burner (more than 40 kW/m2) is to the complete absence of radial heat losses in
captured by the inlet plenum and preheats the these computations, which provides a more
inflowing gases. If half of this heat flux is adiabatic burner than is physically realizable.
transferred to the inflowing gases, these gases Min and Shin found the best agreement be-
would have a temperature of about 375 K tween their one-dimensional simulations and
instead of 300 K. Raising the inlet temperature experiments when they included an arbitrary
does decrease the discrepancy but it also causes heat loss factor to model radial losses. The
an increase in the burning rate to a level far computed lower edge of the stability diagram is
above that observed in the experiment. This is somewhat poorly defined because of the high
likely due to the use of single step chemistry. gradient of burning rate as a function of flame
The model does, however, predict a burning location near the inlet region. The upper limit
rate within 5% of the experimental value for the of the burning rate can be found much more
same flame location. readily, because it does not change much with
The peak computed gas temperature for this flame location. Nevertheless, there is good
case is about 1580 K, less than the adiabatic agreement for the lower limit from f 5 0.52 to
flame temperature. The experimental results f 5 0.7. At the lower equivalence ratios, radial
show superadiabatic operation; that is, the peak heat losses again become significant and the
gas temperature is above the nominal adiabatic computed values fall below the experiment. In
flame temperature (;1610 K) at this equiva- the experiments of Min and Shin, the burner
lence ratio. Superadiabatic operation is pre- lean limit was found to be f 5 0.49. In the
dicted by the numerical model, however, for present computations, the lean limit for gener-
flames located further downstream from the ating stable flames was found to be f 5 0.425.
184 C. L. HACKERT ET AL.

Fig. 7. Burning rate and radiant output fraction as a

function of wall emissivity for radially insulated tube ele-
ment with f 5 0.6. Flame location is x/L 5 1/4.
Fig. 6. Burning rate, radiant output fraction, and front end
radiant loss fraction as a function of flame location for
radially insulated tube element with f 5 0.6. dependence on flame location is primarily a
function of radiant losses. Upstream radiant
heat losses can be quite high depending on the
Burning Rate and Radiant Fractions flame location. The walls upstream of the flame
are cooled convectively by the unburned mix-
In Fig. 6, we present the burning rate, radiant ture, and so most upstream radiation exiting the
loss fraction, and radiant output fraction for a f burner is direct or reflected from the postflame
5 0.6 flame as a function of flame location. The region. Flames very near the inlet have up-
corresponding firing rates at this equivalence stream radiant loss fractions of almost 0.20, but
ratio are 428 kW/m2 for the lower burning rate this rapidly drops to less than 0.005 as the flame
of 21 cm/sec, and 1350 kW/m2 for the maximum is moved downstream.
burning rate of 66.1 cm/sec. The burning rate is Radiant output decreases slightly as the flame
defined as the average velocity of the cold moves downstream, and then increases again as
unburned gases just ahead of the porous me- the burning rate decreases. The actual radiant
dium inlet. Thus, the average velocity of un- power output increases slowly but continuously
burned gases in the porous medium is higher by as the flame moves downstream, but in calcu-
a factor of one over the porosity. The flame lating the fraction this increase is offset by the
location is defined as the leading edge of the increase in burning rate. Radiant output contin-
1000 K isotherm. The radiant output is the ues to increase in this region until it becomes so
fraction of chemical energy released in the large that a flame cannot be supported.
burner which is emitted as radiation from the Radiant output for flames in the stable region
downstream end of the burner. Similarly, the lie in the narrow range from 0.15 to 0.17,
radiant loss is the fraction of chemical energy compared with measured output of 0.15 to 0.25
radiated from the upstream end of the burner. reported for reticulated ceramics [17], and
For this flame, the burning rates for flames about 0.20 reported for sooting diffusion flames
stabilized near the inlet of the burner are sig- [31]. The one-dimensional model of Sathe et al.
nificantly lower than those for flames stabilized [32] also reports peak radiant efficiencies of
further inside. Flames in regions where the about 15%.
gradient of burning rate with respect to location
is negative are not stable. If a flame in this Wall Thermal Properties
region is perturbed downstream, its burning
rate decreases and the flame then continues to For a fixed flame location and channel size, the
propagate further downstream. Conversely, if burning rate and radiant output increase with
the flame is perturbed upstream in this region wall emissivity (Fig. 7). The increase is rapid
then its burning rate increases and it continues near e 5 0, but flattens out at higher emissivi-
to propagate in that direction. The burning rate ties. This is expected because the case of no
MODEL TWO-DIMENSIONAL POROUS BURNERS 185

radiation is very different from that of slight

radiation. Since the channel has a very high
aspect ratio, radiation in the middle of the
channel will likely be reflected several times
before exiting. This provides greater opportu-
nity for reabsorption, and thus the channel has
an effective emissivity much greater than the
actual wall emissivity. This effect would be most
pronounced at low e, and so provides something
of a minimum radiative power until e becomes
very small. For higher emissivities, the wall
Fig. 8. Burning rate and radiant output fraction as a
radiosity rapidly approaches the black body function of solid conductivity for radially insulated tube
value because it is difficult for radiation to element with f 5 0.6. Flame location is x/L 5 1/4.
escape the middle of the channel without being
reabsorbed. Further increases in emissivity have
little effect on the radiation field except near the maximum: continued increases in solid conduc-
channel ends. This is indicated by the much tivity always lead to increases in burning rate.
slower increases in burning rate and radiant This will be contrasted later with the separated
output. plates case, in which solid thermal conduction is
Although not shown, the radiant losses from no longer a long length scale phenomenon. The
the upstream end of the burner have the oppo- radiant fraction, however, decreases with in-
site trend. The losses are highest at low emis- creased solid conductivity. This is mostly due to
sivity (almost 3% for e 5 0.01) and rapidly the increased burning rate. While the down-
decrease with increasing wall emissivity. This is stream radiant flux (W/m2) remains relatively
due to the fact that surfaces of low emissivity constant for a given equivalence ratio, the
must be highly reflecting. For reasons explained chemical energy release is directly proportional
above, the radiosity in the middle of the burner to the burning rate. Thus, the radiant fraction,
will be much higher than es T 4 because of which is the ratio of these two numbers, will
internal reflection and reabsorption. When ra- decrease.
diation from this region of high radiosity travels
upstream of the flame, the fraction reflected Multiple Channel Geometry
further upstream increases as emissivity de-
creases. At very low emissivities, significant In the previous results every channel in the
amounts of radiation will be reflected out of the burner is assumed to operate identically, and so
burner. As the emissivity increases, more and there is no heat transfer through the channel
more of this radiation will be absorbed into the walls. In real operation, the burner is of finite
wall. It should be noted that the radiation model size and will lose heat radially through the sides
here considers diffusely reflecting surfaces. The as well as upstream and downstream. This de-
presence of specular reflection would greatly pends on the burner radius but most laboratory
enhance this effect, although the effects of scale burners are relatively small. Furthermore,
radiation polarization should also then be in- Min and Shin observed macroscopic curvature
cluded. of the flame in the Type II region of operation,
Increased wall conductivity has a similar ef- and attribute this to heat losses through the
fect as increased emissivity in terms of the walls. To investigate this phenomenon, a mul-
upstream transport of heat. The effect of axial tichannel burner was simulated numerically.
conductive transport is shown in Fig. 8 for a Although a two-dimensional simulation of this
fixed flame location and emissivity of 0.4. As conjugate heat transfer problem is computa-
expected, the burning rate increases with in- tionally intensive, we simulate this case for two
creased solid conductivity as more heat is trans- different conditions in order to clarify the con-
ported upstream of the flame. Unlike the case trolling heat transfer mechanisms. This model
of increased emissivity, there is no theoretical burner consists of 13 parallel plate channels,
186 C. L. HACKERT ET AL.

Fig. 9. Isotherms for a flame in a 13-channel burner with

heat losses from the edges. The furthest upstream isotherm
shown is 500 K. The far downstream isotherm is 1500 K, and Fig. 10. Isotherms for a flame in a 13-channel burner with
the maximum isotherm is 1600 K. Burning rate is 41.1 heat losses from the edges. The furthest upstream isotherm
cm/sec. shown is 400 K. The far downstream isotherm is 1450 K, and
the maximum isotherm is 1800 K. Burning rate is 62.8
cm/sec.

with plate separations of 0.55 mm. This yields

the same hydraulic diameter per channel as the SEPARATED PLATES BURNER
experiments. The wall thickness is 0.17 mm, and
all properties are as previously stated. The outer In the next geometry, we consider a more
wall is assumed to radiate to a large ambient complex medium consisting of separated paral-
environment of 300 K, and is also assumed to be lel plates, as shown in Fig. 2. The base case for
subject to a convection coefficient of 20 this study is a 1.28 mm pore size, or 20 ppi
W/m2/K. This model burner has a total trans- (pores per inch) medium about 25 mm long,
with solid emissivity of 0.4 and conductivity of
verse width of less than 10 mm, so it cannot be
2.2 W/m/K. The equivalence ratio of the incom-
said that this is a realistic model for a normal
ing methane mixture is 0.6 unless specified
sized burner. Nevertheless, the effects of trans-
otherwise. The peak Reynolds number for a
verse heat loss through multiple channels is a
flow is about 70, well within the steady laminar
valuable observation and may perhaps be ex-
regime reported for flow through packed beds
trapolated to larger sizes.
in Kaviany [33]. The peak firing rate at f 5 0.6
The effect of flame position in the multichan-
is 869 kW/m2, corresponding to a burning rate
nel burner is demonstrated in Figs. 9 and 10. In
of 45 cm/sec.
Fig. 9, the flame is positioned at about 0.4 mm The two-dimensional character of the flame is
at the leading edge. The flame is fairly flat at most evident in the temperature field and ve-
this location, and the overall temperatures are locity vectors, which are shown for the base case
reduced from the single insulated tube case. (Fig. 11). The burning rate for this case is 40
The burning rate for this case is 41.1 cm/sec, cm/sec. Significant curvature of the isotherms
again slightly below the value for a flame in an within the pore can be observed, and in fact the
insulated channel with a comparable leading flame leading edge (roughly corresponding to
edge (44 cm/sec). Figure 10 displays a flame at a the 1000 K isotherm) appears to occupy most of
deeper location, with a flame leading edge at
about 7 mm into the burner. The burning rate
for this case is 62.8 cm/sec, as compared to 66
cm/sec for the insulated channel. As expected,
the overall flame curvature is also greatly en-
larged. This supports the hypothesis that trans-
verse heat losses are a limiting criterion for the
maximum burning rate attainable in a porous
burner, and the reason why the experimentally
measured peak burning rate is so much lower Fig. 11. Isotherms and velocity vectors for a flame in
than that predicted for the adiabatic channel. separated plates porous geometry.
MODEL TWO-DIMENSIONAL POROUS BURNERS 187

Fig. 13. Burning rate, downstream radiant output, and

upstream radiant loss as a function of flame location in the
Fig. 12. Average gas and solid temperatures for flame in
base case burner.
Fig. 11.

a pore. The flame ignites from the high temper- a function of flame location are shown in Fig.
ature solid surface, and the rapid expansion of 13. The radiant output fraction is approximately
the hot gases helps to move the flame away from constant at 0.22. The radiant loss fraction de-
the solid and out to the channel centerline. creases approximately exponentially with the
The transversely averaged (one-dimensional) depth of the flame into the burner. For this
temperature profiles corresponding to the flame reason, the burning rate is significantly lowered
of Fig. 11 are shown in Fig. 12. Because the solid near the upstream end. At about two pore
is broken, solid phase temperatures do not exist lengths (;2 mm) from the upstream end, the
for some axial locations. The gas temperature radiant loss is 45%, so it is not surprising that no
plotted is a mass average gas temperature, steady flame could be sustained numerically
defined at each axial location as in Eq. 12, so further upstream. We have assumed that each
that areas of high mass flux are weighted more end of the burner radiates to a cold (300 K)
strongly. The effect of preheating is clear in this black environment. For the upstream end, this
figure. Significant heat release occurs at a down- may be an unrealistic boundary condition in an
stream distance of approximately 7 mm which experiment. The high upstream radiant losses
increases the gas temperature to its maximum tend to rapidly heat the upstream plenum unless
value. For locations upstream of the heat re- some cooling method is employed. Higher tem-
lease zone the solid temperature is greater than peratures just upstream of the burner will re-
the gas temperature. The solid plates in this duce the radiant losses and increase the burning
region have been heated due to radiative ex- rate at the upstream end closer to the peak
change with the solid downstream of the flame. value.
Heat transfer from the solid to the gas increases Once the radiant losses approach zero, the
the gas temperature from its inlet value of 298 K burning rate does not change. The fact that
to ;1200 K at 0.007 m. The adiabatic flame radiant losses approach zero implies that the
temperature for complete combustion for this upstream radiation from the postflame solid is
equivalence ratio is 1684 K. The mass average almost completely absorbed by the preflame
gas temperature peaks at 1752 K, and the actual solid, which then gives up this heat to the
peak gas temperature (Fig. 11) is higher still unburned gases. This obviously provides the
(1770 K). Downstream of the reaction zone, the maximum possible preheating, and so should
temperatures of the solid and gas decrease. yield the maximum possible burning rate. The
peak burning rate in this model burner is 44
Burning Rates and Radiant Fractions cm/sec. For comparison, the adiabatic flame
speed at this equivalence ratio is 11.1 cm/sec.
The burning rate and radiant fractions from the The positive effects of decreasing the radiant
upstream and downstream ends of the burner as losses and increasing the burning rate continues
188 C. L. HACKERT ET AL.

Fig. 15. Pore level variations in burning rate and radiation

for separated plates porous geometry. Shaded area shows
Fig. 14. Variation of peak burning rate and radiant output location of solid in burner pore.
with equivalence ratio, with experimental data and model
results of Hsu et al. (1993).
ments of Hsu et al. [13]. Comparison is also
made with the one-dimensional model of Hsu et
until the flame is located at about the halfway al. [13], and some improvement may be seen.
point in the burner; at this point the burning The one-dimensional model shows a concave-
rate begins to drop from its peak value. Beyond downward behavior with equivalence ratio,
a given depth into the medium, there is an while the experiments and present model show
insufficient length of the postflame zone to a concave upward behavior. The radiant output
allow the maximum possible convective heat fraction gradually but continuously decreases
transfer rates from the solid to the gas. with increasing burning rate, as shown above
In all of the results, the radiant output frac- and in the experiments of Khanna et al. [34].
tion is relatively constant near a value of 0.22, Experiments at equivalence ratios above 0.7
regardless of flame position, pore size, or were not performed because the high resulting
burner material properties. The computed val- temperatures strongly degrade the ceramic
ues are consistent with the values measured in a used. Computations for equivalence ratios
longer 10 ppi burner by Khanna et al. [34] in above 0.7 are not shown because the high flow
both magnitude and trend. The slight decrease rates and high temperatures degrade the accu-
in radiant output fraction with increasing burn- racy of the model. Detailed chemistry and tur-
ing rate is because the outlet temperature, bulence models (or at least transient calcula-
raised to the fourth power, does not increase as tions for the higher Re flows) will be needed to
much as the heat release. Even though the accurately model flames at higher equivalence
radiant output power (in W/m2) has increased, ratios.
the radiant output fraction has decreased
slightly. The physical pore geometry of the Pore Length Scale Variations
computational porous media is very different
than the physical geometry of the experimental The lack of smoothness in the curves plotted in
medium. That the results are so similar seem to Fig. 13 is due to local pore scale variations in the
imply that the exact pore geometry may not play steady burning rate and radiation field. These
a large role in the radiant output. Nevertheless, variations are plotted in Fig. 15 for a pore near
it seems likely that improving the models de- the center of the burner (where the overall
scription of the multidimensional geometry burning rate does not change much with flame
would result in a better match with experiment. position). The region occupied by the solid is
The peak burning rate as a function of equiv- indicated on the graph. The flame location
alence ratio is shown in Fig. 14. This gives an indicated in the figure is here defined as the
idea of the operating range of the burner. The leading edge of the 1000 K isotherm, which
computed values compare well with the experi- roughly corresponds to the beginning of the
MODEL TWO-DIMENSIONAL POROUS BURNERS 189

reaction zone. The nominally stable flame loca-

tions are those where the burning rate slope
dU/dx is positive. This is consistent with the
computational results reported by Sahraoui and
Kaviany [22], in that a flame alongside of, or just
in front of, a solid region is unstable. The
natural flame position in porous media with
separated solid surfaces is just downstream of a
solid region. Note also that examination of the
pore level variations in radiant fraction shows a
difference of up to 0.01 in the radiant output
fraction depending on the local flame location
in the pore. Fig. 16. Computed local Nusselt numbers for flame near the
center of 1.28 mm pore size (20 ppi) porous burner.

Gas–Solid Heat Transfer

viscosity. This is, however, a very complicated
One of the most severe limitations of the one- situation with both the gas conductivity and
dimensional approach is the necessity for a viscosity changing with temperature in two di-
model for the gas-to-solid heat transfer. Multi- mensions, and strong radiative coupling to the
dimensional simulations not only do away with solid.
the need for such a model, but also allow the The exception to the uniformity of the Nu
calculation of what the heat transfer coefficient pattern occurs near the flame front, where
or pore scale Nusselt number should be. Actu- calculated Nu are seen to become negative. The
ally, there are two Nusselt numbers which may reason for this odd behavior can be seen by
be calculated. We can define the conventional examining Fig. 11 and recalling the definition in
Nusselt number use. Because of viscous effects the maximum
velocities occur along the centerline of each
Nu 5 q0d/DT/k, (13) pore. The mass average gas temperature used in
the calculation of Nu is thus heavily weighted
where q0 is the local heat flux to the solid plate,
toward the gas temperature at the pore center-
d is the pore size, DT is the difference between
line. Since the flame front is convex, the center-
the local mass average gas temperature (Eq. 12)
line temperature at the leading flame front is
and the local solid temperature, and k is the gas
significantly cooler than that of the combusting
conductivity at the mass average gas tempera-
mixture near the wall. The hot burned gases
ture. This is plotted as a function of x for the 20
drive heat into the wall, raising the wall temper-
ppi case for flames near the middle of the
ature significantly. For a small range of spatial
burner (Fig. 16). Of course, q0 is zero for those
locations then, there is heat transfer into the
x locations for which no solid exists. Nu is
wall but the wall temperature is higher than the
technically undefined there, since there is no mass average gas temperature for these axial
solid at these locations; the Nu is plotted as locations. This results in the “negative” Nusselt
zero, however, since there is no heat transfer. number. Of course, a negative Nusselt number
For each plate the heat transfer is higher near is neither useful nor physically meaningful, but
the leading and trailing edges because of the it does demonstrate the failure near the flame
developing boundary layer. Somewhat contrary front of the single gas temperature model with
to expectation, the calculated Nu does not constant Nusselt number.
change much either with pore size (calculations A volumetric Nusselt number and heat trans-
were made for 5 pore sizes) or from pore to fer coefficient can also be determined by aver-
pore downstream. By implication, the calcu- aging over pores,
lated Nu value does not change much with
Reynolds number, as Re is a function of both q-d 2 2q# 0~d/ 2! d 2 q# 0d
Nu v 5 5 5 , (14)
the pore size and temperature-dependent gas DTk d2 DTk DTk
190 C. L. HACKERT ET AL.

where q# 0 is the average heat flux into one side of havior as a function of equivalence ratio are all
the plate. The factor of 2 is needed because the predicted without the use of any ad hoc adjust-
plate has two sides, and d/ 2 is the length of the able parameters. Transverse heat losses from
plate. Thus, the volumetric Nusselt number is the uninsulated burner are shown to reduce the
the average over the plate of the local, normal peak burning rate and cause flame curvature on
Nusselt number. By coincidence, Nu v for this a burner scale. Upstream radiant losses are
model geometry is equal to the average Nu over shown to control the burning rate in the up-
the solid regions. Since the normal Nusselt stream half of the burner. Flame curvature on a
number does not seem to vary much from pore pore scale is shown for all geometries. The pore
to pore, Nu v is calculated using an average over scale, volumetric Nusselt number is predicted to
the length of the burner. For all pore sizes and be 5.4 6 0.3 for the separated plates geometry,
flow rates examined, we found Nu v 5 5.4 6 regardless of burning rate or pore size. This
0.3. This is in the same order of magnitude as number fits in with the wide range of values
the calculated Nu v of Sahraoui and Kaviany presented in the literature. More applicable
[22], who determined Nu v for their model po- results would probably be obtained if the pore
rous media (90% porosity) to be roughly con- geometry was redesigned to match the experi-
stant at 8.6 6 0.2 over a range of Re from 5 to mental geometry more closely.
90. (Their values are adjusted to be consistent
with the current definition of Nu v .) A slightly This work was supported by the Independent
higher value is expected from that study since Research and Development Program of the Ap-
the square solid cylinders have a larger front for plied Research Laboratories, University of Texas
stagnation point heat transfer. The Nu v mea- at Austin, and by the Office of Naval Research.
sured by Younis and Viskanta [19] all were
taken at higher Re than the current values, but
extrapolation of their correlations would yield
REFERENCES
Nu v ranging from 2.3 to 8.5 depending on pore
size. Recall, however, that these measurements 1. Hardesty, D. R., and Weinberg, F. J., Combust. Sci.
Technol. 8:201 (1974).
varied by a factor of 2–3 when taken using
2. Takeno, T. Sato, K., and Hase, K, Eighteenth Sympo-
porous media by a different manufacturer. sium (International) on Combustion, The Combustion
This reinforces an important point, that the Institute, 1981, p. 465.
heat transfer coefficient is likely to be quite 3. Takeno, T., and Sato, K., Combust. Sci. Technol. 20:73
geometry-dependent. The computational po- (1979).
4. Echigo, R., Yoshizawa, Y., Hanamura, K., and To-
rous medium has an artificial pore shape and a
mimura, T., Proceedings of the 8th International Heat
higher porosity (at 95%) when compared with Transfer Conference, San Francisco, 1986, Vol. II, p.
the physical porous media used in experiments 827.
(85–90% porosity). Another issue arises with 5. Howell, J. R., Hall, M. J., and Ellzey, J. L., Prog. Energy
respect to the pore size. In physical ceramics the Combust. Sci. 22:121 (1996).
6. Kotani, Y., and Takeno, T., Nineteenth Symposium
flow blockage area may change while the poros-
(International) on Combustion, The Combustion Insti-
ity remains constant. This may account for tute, Pittsburgh, 1982, p. 1503.
variations in both the heat transfer and burning 7. Churchill, S. W., Chem. Eng. Technol. 12:249 (1989).
rate as a function of pore size. 8. Choi, B., and Churchill, S. W., Seventeenth Symposium
(International) on Combustion, The Combustion Insti-
tute, Pittsburgh, 1979, p. 917.
9. Pfefferle, L. D., and Churchill, S. W., Combust. Flame
CONCLUSIONS 56:165 (1984).
10. Kansuntisukmongkol, R., Miyachi, H., Ozoe, H., and
Results from the two-dimensional model com- Churchill, S. W., Combust. Flame 108:158 (1997).
pare well with experimentally measured porous 11. Brown, M. A., Chan, C., Targett, M. J., Holtzmuller,
M., and Churchill, S. W., Combust. Sci. Technol.
burner attributes, in spite of differences be-
115:207 (1996).
tween the computational and experimental pore 12. Sathe, S. B., Kulkarni, M. R., Peck, R. E., and Tong,
geometries. The peak burning rate for insulated T. W., presented at Western States Section of the
burners, downstream radiant output, and be- Combustion Institute, Oct. 23–24, 1989.
MODEL TWO-DIMENSIONAL POROUS BURNERS 191

13. Hsu, P.-F., Evans, W. D., and Howell, J. R., Combust. 25. Westbrook, C. K., and Dryer, F. L., Combust. Sci.
Sci. Technol. 90:149 (1993). Technol. 27:31 (1981).
14. Ellzey, J. L., and Goel, R., Combust. Sci. Technol. 26. Yamaoka, I., and Tsuji, H., (1984) Twentieth Sympo-
107:81 (1995). sium (International) on Combustion, The Combustion
15. Min, D. K., and Shin, H. D., Int. J. Heat Mass Trans. Institute, Pittsburgh, 1984, p. 1883.
34:341 (1991). 27. Yu, G., Law, C. K., and Wu, C. K., Combust. Flame
16. Yoshizawa, Y., Sasaki, K., and Echigo, R., Int. J. Heat 63:339 (1986).
Mass Trans. 31:311 (1988). 28. Incropera, F. P., and DeWitt, D. P., Fundamentals of
17. Sathe, S. B., Kulkarni, M. R., Peck, R. E., and Tong, Heat and Mass Transfer, 3rd ed., John Wiley, New
T. W., Twenty-Third Symposium (International) on York, 1990.
Combustion, The Combustion Institute, Pittsburgh, 29. Fletcher, C. A. J., Computational Techniques for Fluid
1990, pp. 1011–1018. Dynamics 1: Fundamental and General Techniques,
18. Tong, T. W., and Sathe, S. B. (1988). ASME HTD Springer-Verlag, Berlin, 1991.
04:147. 30. Patankar, S. V., Numerical Heat Transfer and Fluid
19. Younis, L. B., and Viskanta, R., Int. J. Heat Mass Flow, Hemisphere, New York, 1980.
Transfer 36:1425 (1993). 31. Markstein, G. H., and DeRis, J., Twenty-Third Sympo-
20. Hsu, P.-F., and Howell, J. R., Exp. Heat Transfer 5:293 sium (International) on Combustion, The Combustion
(1992). Institute, Pittsburgh, 1990, p. 1685.
21. Hendricks, T. J. (1993). Ph.D. dissertation, University 32. Sathe, S. B., Peck, R. E., and Tong, T. W., Int. J. Heat
of Texas at Austin. Mass Transfer 33:1331 (1990).
22. Sahraoui, M., and Kaviany, M., Int. J. Heat Mass 33. Kaviany, M., Principles of Heat Transfer in Porous
Transfer 37:2817 (1994). Media, Springer-Verlag, New York, 1991.
23. Hsu, P.-F., and Matthews, R. D., Combust. Flame 34. Khanna, V., Goel, R., and Ellzey, J. L., Combust. Sci.
93:457 (1993). Technol. 99:133 (1994).
24. Siegel, R., and Howell, J. R., Thermal Radiation Heat
Transfer, 3rd ed., Hemisphere, Washington, 1992. Received 27 March 1997; accepted 1 April 1998