Sie sind auf Seite 1von 22

Journal of Quantitative Spectroscopy &

Radiative Transfer 74 (2002) 307328


www.elsevier.com/locate/jqsrt
Numerical simulation of radiative heat transfer from non-gray
gases in three-dimensional enclosures
P.J. Coelho

Instituto Superior T ecnico, Mechanical Engineering Department, Av. Rovisco Pais, 1049-001 Lisboa, Portugal
Received 21 May 2001; accepted 23 November 2001
Abstract
The discrete ordinates and the discrete transfer methods are applied to the numerical simulation of radiative
heat transfer from non-gray gases in three-dimensional enclosures. Several gas radiative property models are
used, namely the correlated k-distribution (CK), the spectral line-based weighted-sum-of-gray-gases (SLW) and
the weighted-sum-of-gray-gases (WSGG) methods. The results are compared with recently published accurate
calculations based on the statistical narrow band model. The WSGG model is computationally ecient, but
often yields relatively large errors. It should be used only if moderate accuracy is sucient. The SLW model
is the best alternative regarding the compromise between accuracy and numerical eciency. However, an
optimization of the coecients of the model is essential to reduce the computational requirements, especially
in the case of gas mixtures. The CK model is the most accurate of the methods evaluated here, but too time
consuming for engineering applications, although recent developments may partly overcome this shortcoming.
? 2002 Elsevier Science Ltd. All rights reserved.
Keywords: Non-gray gas models; Discrete ordinates method; Discrete transfer method
1. Introduction
Radiative heat transfer plays an important role in many engineering problems, including atmo-
spheric processes and combustion applications, namely, in furnaces, boilers, engines, rocket nozzles
and res. The zonal and the Monte Carlo methods are often employed in radiation problems, but
since they are very time consuming, other methods are often preferred in combustion applications.

Tel.: +351-218-4181-94; fax: +351-218-4755-45.


E-mail address: coelho@navier.ist.utl.pt (P.J. Coelho).
0022-4073/02/$ - see front matter ? 2002 Elsevier Science Ltd. All rights reserved.
PII: S0022- 4073(01)00249- 7
308 P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328
In this case, ux methods were often used during the 1970s, but they are not so popular any more
due to their poor accuracy. Nowadays, the discrete transfer method (DTM) [1], the discrete ordinates
method (DOM) [2,3] and the nite volume method (FVM) [4] are probably the most widely used
models. The FVM is very similar to the DOM in the case of rectangular enclosures. The DOM and
the DTM are used in the present work.
It is well known that the radiative properties of combustion products, in particular H
2
O and CO
2
,
vary so strongly and rapidly across the spectrum that the assumption of a gray gas is almost never
a good one. Recent surveys of models for the calculation of gas radiative properties are available
[5,6]. The line-by-line approach is the most accurate method to calculate the radiative properties of
the gases, but demands too large computing times. Narrow band models (NBM) [7] are generally
recognized as the most accurate ones suitable for practical applications, while wide band models
(WBM) [8] are also reasonably accurate and more economical.
NBM and WBM are dicult to apply to multi-dimensional problems. The NBM gives the spec-
tral transmissivity averaged over a narrow band, and the WBM yields the wide band absorptance.
Hence, these methods cannot easily be coupled to dierential solution methods of the radiative
transfer equation (RTE), such as the DOM, where knowledge of the spectral absorption coecient
or its average over an interval is required. The exact formulation of the RTE using band models
needs to account for the correlation between the spectral absorption coecient and the radiation
intensity when the RTE is averaged over a wavenumber interval. The exact (correlated) formula-
tion was applied to the DOM in [9] for both NBM and WBM, in the case of black or nearly
reecting walls. However, applications were limited to one-dimensional problems, since the exten-
sion to multi-dimensional problems is not feasible. Such an extension is only possible using either
a non-correlated formulation or a solver based on the integral representation of the RTE, e.g., a ray
tracing method. The non-correlated formulation, although satisfactory for some situations, does not
maintain good accuracy in general applications [10,11], while the ray tracing method is prohibitively
expensive for practical applications. Gray boundaries further complicate the problem, which becomes
quite involved and time consuming [12]. Band models are also dicult to apply to scattering me-
dia, and require an additional approximation, e.g., the Curtis-Godson approximation, in the case of
non-isothermal and}or non-homogeneous media.
The correlated k-distribution method (CK) [13,14], formerly developed for atmospheric radiation,
has recently received the attention of the heat transfer community [1517]. This method recognizes
that the variation of the blackbody radiation intensity over a band can be neglected, and replaces
the spectral integration over wavenumber within a band by a quadrature over the absorption coef-
cient. In this way, the application of the CK method to the solution of the RTE is much easier
than for the NBM and WBM. Unfortunately, the CK method is very computationally demanding,
particularly if applied to narrow bands and gas mixtures. Recent developments have tried to re-
duce the computational requirements by treating together groups of narrow bands using a band
lumping strategy [18], using wide bands rather than narrow bands [19], devising more ecient
quadrature schemes [20] and applying approximate treatments to overlapping bands in gas mixtures
[21].
In contrast to the band models mentioned above, global models are limited to problems with
gray walls and}or particles. The weighted-sum-of-gray-gases (WSGG) model [22] has been the most
widely used global model for the calculation of gas radiative properties in combustion applications.
In many cases, the total gas emissivity is computed and gray gas radiation is assumed. The simplicity
P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328 309
inherent to the gray gas assumption has motivated this approach, despite the well-known evidence
of its limitation. However, the WSGG may be implemented as a non-gray gas model, solving the
RTE for every gray gas, as demonstrated in [23].
In the last few years, the spectral line-based weighted-sum-of-gray-gases (SLW) model [2426]
has emerged as a more accurate alternative to the WSGG. In the SLW method, like in the CK
method, the spectral integration over wavenumber is replaced by a quadrature over absorption cross
section. However, the SLW uses an absorption-line blackbody distribution function that permits
ecient and accurate total radiative transfer calculations. A similar model, referred to as absorption
distribution function model, is described in [6].
The comparison of dierent gas property models has been generally performed in one-dimensional
geometries (see, e.g., [9,27,28]). Only a few works have been concerned with radiative trans-
fer from non-gray gases in multi-dimensional enclosures. Among these, Fiveland and Jamalud-
din [29] and Park et al. [30] do not directly use a model to compute the gas properties, but
prescribe the absorption coecient of the medium over the spectrum. The WSGG was coupled
to the DTM in [31] and to the DOM in [32]. In both cases, the evaluation of the method re-
lied on the comparison between measured and predicted results obtained from a coupled uid
ow}heat transfer code, making dicult the analysis of the role played by the gas radiation model.
The Elsasser NBM and the Edwards WBM were used together with the ray emission model in
[33] to compute radiative transfer in a three-dimensional boiler furnace with prescribed gas tem-
perature and composition. These calculations would become extremely expensive if the reactive
ow were also simulated. The statistical NBM was used together with a ray tracing method in
[10,34,35] to calculate the radiative transfer in axisymmetric cylindrical geometries.
The WSGG and the Edwards WBM were also used in [34,35], respectively. These calculations
using the ray tracing method become prohibitively expensive for routine engineering calculations,
owing to the diculties mentioned above regarding the coupling of the RTE with NBM and WBM.
The WBM and the DTM were used in [36] to model a jet ame, but the non-correlated for-
mulation was employed. The SLW model and the DOM were used in [37] to compute radiative
transfer in axisymmetric enclosures. The most extensive comparison of gas radiation models in
multi-dimensional enclosures was recently presented in [38]. Five two-dimensional enclosures were
studied. The statistical NBM model of Malkmus was used as a reference for the evaluation of
the WBM, CK, SLW and WSGG models. The RTE was solved using the DOM or a ray tracing
method.
Recently, numerical solutions of three-dimensional non-gray radiative transfer using the statistical
NBM of Malkmus have been presented [39]. The gas composition and the temperature elds of the
enclosure are prescribed, and the walls of the enclosure are black. This enables the NBM to be used
along with a ray tracing method to provide a benchmark suitable for evaluation of other models. As
pointed out in [39], the method used to obtain the benchmark results is very time consuming and not
recommended for other applications. The purpose of the present work is to couple several non-gray
gas radiation property models with the DTM and the DOM, and to evaluate their performance
regarding accuracy and CPU time requirements in three-dimensional calculations using the results
of Liu [39] as benchmark. Only gas property models that provide the absorption coecient of the
medium as the basic radiative property are used here, namely the WSGG, the SLW and the CK
methods.
310 P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328
2. Radiation and gas radiative property models
2.1. The radiative transfer equation
The RTE in an emittingabsorbing non-scattering medium may be written as
dI
v
ds
=
v
I
v
+
v
I
bv
, (1)
where I
v
is the spectral radiation intensity,
v
is the spectral absorption coecient of the medium,
s is the direction of propagation of radiation, and the subscripts v and b denote wavenumber and
blackbody, respectively.
The boundary condition for a gray surface that emits and reects diusely is given by
I
wv
(s) = c
w
I
bv
+
1 c
w

_
ns

0
|n s

|I
v
(s

) dO

, (2)
where c
w
is the emissivity of the surface, n is the unit normal vector, s is the direction of the
outgoing radiation intensity, and s

is the incoming direction associated with the elementary solid


angle dO

.
2.2. Gas radiative property models
2.2.1. CK model
The CK method divides the spectrum into bands suciently narrow to assume the Planck func-
tion as constant in each band. A function A dependent on
v
such as, for example, the spectral
transmissivity t
v
, is averaged as follows [15]:
A(
v
) =
1
v
_
v
A(
v
) dv =
_

0
A(k)[(k) dk =
_
1
0
A(k(q)) dq, (3)
where [(k) is the distribution function of
v
and q(k) is the cumulative k-distribution function given
by
q(k) =
_
k
0
[(k

) dk

. (4)
In non-uniform media, Eq. (3) relies on the scaling approximation, i.e., the spectral and spatial
dependence of the absorption coecient are separable. The reordered absorption coecient k(q)
within a band is a smooth monotonically increasing function of q, which can be easily calculated
from high-resolution spectra. Therefore, the average value of function A may be accurately computed
using a Gaussian type quadrature:
A(
v
) =
_
1
0
A(k(q)) dq
N

)=1
c
)
A(q
)
), (5)
where q
)
and w
)
are the quadrature points and weights, respectively, and N is the quadrature order.
For example, the transmissivity of a gas layer of length L averaged over a band of width v is
P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328 311
calculated as
t
v
=
N

)=1
c
)
exp(k
)
L). (6)
The absorption coecients k
)
are given by [15,40]
k
)
= X
s

T
k

)
}(1Q(1)), (7)
where X
s
is the molar fraction of the absorbing species, Q(1) is the partition function of an isolated
molecule and k

)
are parameters of the model. In a mixture of two absorbing gases, the transmissivity
is taken as the product of the transmissivities of the individual species, and a double summation
appears in Eq. (6). The data of Souani and Taine [40] was used here. It consists of a 7-point
quadrature, 43 spectral bands of variable width for H
2
O and 17 spectral bands for CO
2
. The width
of the bands has been optimized for computation eciency.
2.2.2. WSGG model
The WSGG model expresses the total gas emissivity as a weighted sum of gray gas emissivities:
c =
N
g

)=0
a
)
(1)[1 exp(
)
L)], (8)
where a
)
and
)
are the emission weighting factor and the absorption coecient for the )th gray
gas component, respectively, and L is the path length. The coecients a
)
and
)
are obtained from
a t to total emissivity with the constraint that the sum of coecients a
)
is equal to unity. The
transparent regions of the spectrum are accounted for by the term ) = 0. In the present work, the
coecients determined in [41] were used and N
g
= 3.
The WSGG model has often been used in combustion problems assuming that the medium is gray.
In such a case, the absorption coecient of the medium is calculated from the total gas emissivity,
c, as
=ln[1 c(L)]}L. (9)
To compute the local absorption coecient at a control volume of a computational grid, some
authors take L as the mean eective path length of that control volume, e.g. [42], and others take L
as the mean beam length for the whole enclosure, e.g. [32]. However, this procedure has no sound
theoretical foundations, regardless of the choice of L [5]. This arbitrariness does not exist if the
WSGG is applied as a non-gray gas model, solving the RTE for the N
g
gray gases plus one clear
gas. In fact, in this case neither Eq. (8) nor Eq. (9) are employed, and only the values of a
)
and

)
given by the model are required.
2.2.3. SLW model
The SLW model may be regarded as an improved WSGG model obtained from the line-by-line
spectra of the absorbing species. The total emissivity may also be expressed by Eq. (8). The ab-
sorbing cross-section domain is discretized into N
g
intervals, each one being represented by a single
312 P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328
value denoted by C
abs, )
. The )th gray gas absorption coecient is dened as
)
= NC
abs, )
, where N
is the molar density of the gas. The limits of the )th interval are the supplemental absorption cross
sections

C
abs, )
and

C
abs, )+1
. The coecients a
)
represent the fraction of blackbody energy in the
portion of the spectrum where the absorption cross section is between

C
abs, )
and

C
abs, )+1
, i.e., where
the eective absorption cross section is C
abs, )
. In the case of a single absorbing gas, the blackbody
weight a
)
is given by [25]
a
)
= F
s
(

C
abs, )+1
, 1
b
, 1
g
,
T
, Y
s
) F
s
(

C
abs, )
, 1
b
, 1
g
,
T
, Y
s
), (10)
where F
s
is the absorption-line blackbody distribution function, i.e., the fraction of blackbody energy
in the portion of the spectrum where the high-resolution spectral absorption cross section of the gas,
C
abs, v
, is less than a prescribed value C
abs
. In this equation 1
b
is the blackbody (source) temperature,
1
g
the gas temperature,
T
the total pressure, and Y
s
the mole fraction of the broadening species.
The correlations presented in [43,44] for the absorption-line blackbody distribution functions of H
2
O
and CO
2
were used here.
In the case of non-isothermal non-homogeneous media, a scaling approximation is employed. The
local absorption coecient is related to a local absorption cross section that depends on the local
temperature and gas composition. This local absorption cross section is calculated from an implicit
equation, as described in [25], which relates the absorption-line blackbody function at a reference
state with that at local conditions. The reference state is characterized by an average temperature
and composition over the domain.
In mixtures of CO
2
and H
2
O the blackbody weight may be computed to a good approximation
by the product of the weights of the individual species, the so-called double integration approach.
Eq. (8) involves a double summation. This implies that the computational eort for the solution of
the RTE increases quadratically compared to that of a single absorbing gas calculation. However, if
the ratio of the mole fractions of the species is spatially constant, it is possible to maintain roughly
the computational eort of a single gas calculation using a convolution approach. Both methods are
explained in detail in [26].
The calculations presented in this work were carried out using N
g
= 20 and 3. In the rst case,
the values of

C
abs, )
were dened from a division of the absorption cross-section domain between
3 10
5
and 120 m
2
mol
1
into 20 logarithmically spaced intervals, and the values of C
abs, )
were
calculated from:
C
abs, )
= exp[0.5 (ln

C
abs, )
+ ln

C
abs, )+1
)]. (11)
In the second case, Eq. (11) is not used and the eective and supplemental absorption cross sections
are obtained from an optimized t to total emissivity for the reference state and range of path lengths
of the problem under consideration.
2.3. Radiation models
The solution of the RTE using the DOM and the DTM is described in the literature. Therefore,
only the features related to the incorporation of the gas radiative property models in the solution
algorithms are addressed below.
P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328 313
2.3.1. Discrete ordinates method
When the WSGG or the SLW methods are used, the RTE for the )th gray gas component may
be expressed as [23,37]:
dI
)
ds
=
)
I
)
+
)
a
)
I
b
. (12)
In the WSGG a
)
is a function of temperature and in the SLW a
)
is calculated from Eq. (10) or its
generalization for gas mixtures.
In the present work, the angular discretization was carried out using the TN quadrature [45] to
allow a direct comparison with the benchmark data, which was also obtained using that quadrature.
The equation for the mth direction is written as

m
dI
m
)
dx
+ p
m
dI
m
)
d,
+
m
dI
m
)
d:
=
)
I
m
)
+
)
a
)
I
b
, (13)
where
m
, p
m
and
m
are the direction cosines. The total radiation intensity is calculated from
I
m
=
N
g

)=0
I
m
)
. (14)
Discretization of Eq. (13) over a control volume was carried out using the CLAM discretization
scheme [46,47]. This is a bounded second-order scheme, signicantly more accurate than the STEP
scheme, and free from spurious oscillations and unphysical negative intensities. The discretized
equation may be written as follows, using the deferred correction procedure [48] to implement the
CLAM scheme:
I
m
P, )
=
a
)

)
I
b
J +|
m
|A
x
I
m
W, )
+|p
m
|A
,
I
m
S, )
+|
m
|A
:
I
m
B, )
+ S
m
dc

)
J +|
m
|A
x
+|p
m
|A
,
+|
m
|A
:
, (15)
where the subscripts P, W, S and B refer to the points at the center of the control volume and at
the upstream cell faces along the x, , and : directions, respectively. The areas of these cell faces
were denoted by A
x
, A
,
and A
:
, and the volume of the cell by J. The deferred correction term S
m
dc
is dened as
S
m
dc
=|
m
|A
x
(I
m
P
I
m
E, )
) +|p
m
|A
,
(I
m
P
I
m
N, )
) +|
m
|A
:
(I
m
P
I
m
T, )
), (16)
where the subscripts E, N and T stand for the downstream cell faces along x, , and : directions,
respectively. The radiation intensity at the downstream cell faces is calculated according to the
CLAM scheme using the radiation intensities computed at the previous iteration for downstream
grid nodes, and at the current iteration for upstream grid nodes. The STEP scheme is recovered
if the deferred correction term is set to zero. The solution algorithm follows standard practices, as
described in [3].
The heat ux incident on the boundary surface x = constant may be written as
q
w, x
=
N
g

)=0
M

m=1
(s
m
n0)
w
m
|
m
|I
m
)
, (17)
314 P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328
where M is the quadrature order and w
m
is the quadrature weight. The summation extends over
all the directions pointing towards the boundary. Similar equations may be written for the other
boundaries.
The divergence of the radiative heat ux is given by
q =
N
g

)=0

)
_
4a
)
I
b

m=1
w
m
I
m
)
_
. (18)
If the CK model is used, the RTE for quadrature point q
)
in a band of width v
i
is given by
dI
i, )
ds
=k
i, )
I
i, )
+ k
i, )

I
b, v
i
, (19)
where

I
b, v
i
is the spectral blackbody intensity averaged over the width of the band. The total
radiation intensity is given by
I =
N
b

i=1
v
i

I
v
i
=
N
b

i=1
v
i
N

)=1
c
)
I
i, )
, (20)
where N
b
is the total number of bands over which the spectrum is divided, each one with width
v
i
. The counterpart of Eqs. (15) and (16) may easily be obtained, while the heat ux at surface
x = constant and the radiative heat source are determined from
q
w, x
=
N
b

i=1
v
i
_

_
N

)=1
c
)
_
_
_
_
_
_
_
M

m=1
(s
m
n0)
w
m
|
m
|I
m
i, )
_
_
_
_
_
_
_
_

_
, (21)
q =
N
b

i=1
v
i
_
_
N

)=1
c
)
k
i, )
_
4

I
b, v
i

M

m=1
w
m
I
m
i, )
_
_
_
. (22)
2.3.2. Discrete transfer method
The RTE for the )th gray gas component in the WSGG or SLW methods (Eq. (12)) may be
discretized along a line of sight yielding
I
+
)
= I

)
e

)
s
+ a
)
I
b
(1 e

)
s
), (23)
where the superscripts + and denote the intensity leaving and entering a control volume, respec-
tively. The total gas intensity is given by
I =
N
g

)=0
I
)
. (24)
P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328 315
It is worth to point out that, contrary to the gray gas formulation, the spectral formulation does not
require the prescription of a path length for the determination of the absorption coecients, since
these are directly provided by the WSGG or the SLW models.
The heat ux incident on a boundary surface is calculated as
q
w
=
_

0
_
ns

0
I

v
(s

)|n s

| dO

dv
N
g

)=0

k
I
k
)
cos 0
k
O
k
, (25)
where 0
k
is the angle between the incident radiation along the kth direction and the normal to the
surface. The index k runs over all the directions resultant from the angular discretization of the
hemisphere around the incident point on the wall.
The divergence of the radiative heat ux constitutes a sink term of the energy conservation equation
and is approximated by
_
J
q dJ =
_

0

(I
+
v
I

v
) cos 0


N
g

)=0

(I
+
)
I

)
) cos 0

. (26)
The summation runs over all the radiation beams crossing the control volume under consideration
and each term represents the variation of radiative energy of a beam within that control volume [1].
In the case of the CK method, the discretization of Eq. (19) is straightforward. The incident heat
ux on a boundary surface and the radiative heat source are calculated from
q
w
=
N
b

i=1
v
i
N

)=1
c
)
_

k
I
k
i, )
cos 0
k
O
k
_
, (27)
_
J
q dJ =
N
b

i=1
v
i
_
_
_
N

)=1
c
)
_

(I
+
i, )
I

i, )
) cos 0

_
_
_
_
. (28)
The solution algorithm is not described here, since it can be easily adapted from the gray gas
formulation of the RTE.
3. Results and discussion
The models described above were applied to three radiative transfer problems in three-dimensional
enclosures. Accurate solutions for these problems were generated by Liu [39] using a ray tracing
method for the solution of the RTE and a statistical narrow band model (SNB) for the calculation of
the gas radiative properties, and will be used as a reference. All the test cases consist of a rectangular
enclosure of 2 m 2 m 4 m surrounded by black walls at 300 K and containing a gas mixture at
atmospheric pressure. In all cases the grid size is the same used in the benchmark. In the DOM the
calculations were performed using the T4 quadrature, like in [39], yielding M = 128 directions. In
the DTM an angular discretization with 4 angles per octant in the polar and azimuthal directions was
used yielding also a total of 128 directions per unit sphere. The calculations were performed using
an Alpha EV5 21164 processor at 600 MHz. The results of the DOM calculations are presented and
316 P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328
-100
-80
-60
-40
-20
0
0 0.5 1 1.5 2
x [m]
0 0.5 1 1.5 2
x [m]
0 0.5 1 1.5 2
z [m]
0 0.5 1 1.5 2
z [m]
20
25
30
35
H
[
k
W
/
m
2
]
H
[
k
W
/
m
2
]
-120
-100
-80
-60
-40
-20
0
20
25
30
35
SNB
CK
SLW-20
SLW-3
WSGG WSGGgg
WSGGgl
3
_
.
q
[
k
W
/
m
]
3

.
q
[
k
W
/
m
]
_
(a)
(c) (d)
(b)
Fig. 1. Predicted proles of radiative heat source and incident heat ux calculated by the DOM compared with the ray
tracing SNB benchmark data [39] for test case 1: (a) Radiative heat source along the centerline (1 m, 1 m, :); (b) incident
heat ux along (2 m, 1 m, :); (c) radiative heat source along (x, 1 m, 0.375 m); (d) incident heat ux along (x, 1 m, 4 m).
discussed rst. Then, the DTM results are presented and the dierences from the DOM results are
pointed out.
3.1. Discrete ordinates method
3.1.1. Test case 1
In this case the gas is pure water vapor maintained at a uniform temperature of 1000 K. An
11 11 16 uniform grid was used. The radiative heat source along the centerline (1 m, 1 m, :) and
along (x, 1 m, 0.375 m), and the incident heat ux along (2 m, 1 m, :) and (x, 1 m, 4 m) are shown in
Fig. 1. The average and maximum relative errors along these lines are summarized in Table 1.
The gray gas implementation of the WSGG generally yields very poor predictions. If the local
absorption coecient is estimated from Eq. (9) taking L as the mean eective path length of the
control volume under consideration (WSGGgl), the average relative error for both the radiative source
P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328 317
Table 1
Average and maximum relative errors along lines (1 m, 1 m, :), (2 m, 1 m, :), (x, 1 m, 0.375 m) and (x, 1 m, 4 m), which
are denoted by subscripts 14, respectively, and CPU time for test case 1
DOM DTM
WSGG SLW-20 SLW-3 CK WSGG SLW-20 SLW-3 CK

E
rel, 1
(%) 28.30 3.82 10.10 1.11 28.85 4.92 9.40 2.90
E
max, 1
(%) 40.06 6.37 14.45 4.18 35.80 9.05 14.28 8.80

E
rel, 2
(%) 11.47 2.20 1.29 0.61 11.11 1.77 0.81 0.19
E
max, 2
(%) 11.89 5.04 5.46 3.42 11.99 2.11 2.01 0.35

E
rel, 3
(%) 33.94 6.33 4.05 4.43 29.32 3.80 5.21 2.48
E
max, 3
(%) 42.04 9.20 8.18 7.89 37.72 7.67 8.81 5.21

E
rel, 4
(%) 10.93 2.33 2.26 0.64 10.37 1.79 1.73 0.23
E
max, 4
(%) 11.95 4.64 5.66 2.94 12.24 2.38 2.52 0.51
t
CPU
(s) 5.6 23.5 5.0 476.0 3.8 12.6 2.4 357.4
and the heat ux exceed 50%, and the maximum errors are often larger than 100%. The predicted
incident heat uxes are not plotted in Fig. 1 because they are beyond the limits of the graphic. If L
is taken as the mean beam length for the whole enclosure (WSGGgg), the mean relative error of the
radiative source is also very large, and the predicted proles are much atter than in the benchmark,
failing to reproduce the eect of the side walls. Nevertheless, the predicted incident heat uxes are
fairly accurate, except in the vicinity of the side walls. In fact, the average relative error is 4.6%
in Fig. 1(b) and 6.4% in Fig. 1(d), and the maximum error does not exceed 20%. This level of
accuracy is probably achieved because the medium is isothermal and homogeneous. Average relative
errors above 10% were found in test case 2, where the medium is non-homogeneous, and above
40% in test case 3, where the medium is non-isothermal. Therefore, although computationally fast,
this approximation should not be used in practical applications owing to its poor accuracy, specially
in the calculation of the radiative source, and it will not be further investigated here.
The non-gray implementation of the WSGG, simply referred to as WSGG in the gures, is
by far the least accurate of the three non-gray gas models under investigation. The incident heat
uxes in Figs. 1(b) and (d) are underpredicted by more than 10%, while the radiative sources in
Figs. 1(a) and (c) are underestimated by about 30%, as reported in Table 1. This represents a
signicant improvement over the gray gas approximation regarding the radiative source prediction.
The proles are qualitatively correct, but the errors are still high. The SLW model yields much
better results than the WSGG model. If no optimization of the absorption cross sections is attempted
and N
g
= 20 (SLW-20), the average relative error of the incident heat uxes is lower than 2.5%,
while the error of the radiative heat source is larger, but does not exceed 6.5%. Maximum errors
are lower than 10%. If the number of gray gases is reduced to 3 using optimized absorption cross
sections (SLW-3), the errors are comparable, except for the radiative source along the centerline,
whose average and maximum errors are 10.1% and 14.5%, respectively. The CK method yields the
most accurate predictions, with average relative errors lower than 1% for the radiative heat uxes,
and below 4.5% for the radiative source.
The computational requirements of the dierent methods are also summarized in Table 1. The
CPU time for the WSGG and SLW-3 methods is similar because in both cases the RTE is solved
318 P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328
-140
-120
-100
-80
-60
-40
-20
0
0 0.5 1 1.5 2
z [ m]
(CLAM, 11x11x16, T4)
(CLAM, 22x22x32, T6)
(STEP, 11x11x16, T4)
(STEP, 22x22x32, T6)
20
25
30
35
0 0.5 1 1.5 2
z [ m]
H
[
k
W
/
m
2
] 3
_

.
q
[
k
W
/
m
]
(a) (b)
Fig. 2. Comparison of DOM predictions for test case 1 using two dierent spatial}angular discretizations: (a) Radiative
heat source along the centerline (1 m, 1 m, :) using the CK method; (b) incident heat ux along (2 m, 1 m, :) using the
SLW-20 method.
four times, namely for a transparent medium and for each one of the N
g
=3 gray gases. The SLW-20
method is much slower, since the RTE is solved N
g
+ 1 = 21 times. The CK method is the slowest
one, because the RTE is solved for the N
b
bands, and for N quadrature points within each band. The
results given in Table 1, as well as those shown in Fig. 1, refer to calculations carried out using the
CLAM scheme. However, if the STEP scheme is used, the CPU time is signicantly lower (1.0, 2.9
and 99.2 s for WSGG, SLW-20 and CK methods, respectively). In fact, the CLAM scheme involves
the correction term given by Eq. (16), which is treated implicitly, implying an iterative scheme even
if the walls are black, as in the present work. On the contrary, the STEP scheme does not require
an iterative scheme for black walls and prescribed temperature of the medium.
The sensitivity of the predicted results to spatial and angular renement is illustrated in Fig. 2. It
shows the radiative source along the centerline (1 m, 1 m, :) calculated by using the CK method and
the incident heat ux along (2 m, 1 m, :) computed with the SLW-20 method. The previous results
are compared with those computed by doubling the number of control volumes in all directions and
increasing the quadrature from T4 (16 directions per octant) to T6 (36 directions per octant). Both
the STEP and CLAM discretization schemes were used. It can be seen that the results for the coarse
and ne grids are in excellent agreement if the CLAM scheme is used. However, the predictions
obtained using the STEP scheme show an inuence of the grid renement, denoting an evolution
towards the CLAM solution as the grid is rened. The average relative error for the coarse grid in
Fig. 2(a) is 19.1% for the STEP scheme and 1.1% for the CLAM scheme. The corresponding errors
in Fig. 2(b) are 3.4% and 2.2% for the STEP and CLAM schemes, respectively. Similar trends have
been found for the remaining proles displayed in Fig. 1, and for the dierent radiation gas models
employed. It is concluded that the CLAM scheme is much more accurate than the STEP scheme, as
expected, particularly in the calculation of the radiative source, and the results displayed in Fig. 1
may be considered grid independent. The CLAM scheme was used in all the DOM calculations
reported below for the following test cases.
3.1.2. Test case 2
In the second test case the gas temperature remains uniform at 1000 K, but the medium is a
non-uniform mixture of H
2
O and N
2
with X
H
2
O
changing along the : direction according to :(1:}4).
P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328 319
-40
-35
-30
-25
-20
-15
0 0.5 1 1.5 2
z [m]
0 0.5 1 1.5 2
z [m]
10
15
20
25
30
35
-80
-70
-60
-50
-40
-30
-20
15
17
19
21
23
25
0 0.5 1 1.5 2
x [m]
0 0.5 1 1.5 2
x [m]
H
[
k
W
/
m
2
]
H
[
k
W
/
m
2
]
3
_


.
q
[
k
W
/
m
]
3

_

.
q
[
k
W
/
m
]
SNB
CK
SLW-20
SLW-3
WSGG
(a)
(c)
(d)
(b)
Fig. 3. Predicted proles of radiative heat source and incident heat ux calculated by the DOM compared with the ray
tracing SNB benchmark data [39] for test case 2: (a) Radiative heat source along the centerline (1 m, 1 m, :); (b) incident
heat ux along (2 m, 1 m, :); (c) radiative heat source along (x, 1 m, 0.24 m); (d) incident heat ux along (x, 1 m, 4 m).
The results obtained using an 111125 uniform grid and the T4 quadrature are displayed in Fig. 3,
and the average and maximum relative errors are given in Table 2.
The WSGG method gives relatively high errors (30.0% on average, and 50.4% at maximum) for
the radiative source along the centerline, and does not predict the signicant decrease of the radiative
source towards the wall at : = 0 m. Better predictions are obtained along (x, 1 m, 0.24 m), although
large errors (about 20%) are again observed in the vicinity of the wall at x = 0 m and x = 2 m.
The errors in the predicted heat uxes are smaller, particularly along (x, 1 m, 4 m). The SLW-20
method yields accurate results for both the radiative source and the incident heat ux, with average
relative errors always below 9% and often much lower, and maximum relative errors marginally
exceeding 10%. The SLW-3 also gives good results, except along the centerline where a maximum
error of 23% was found and the prole exhibits a qualitatively wrong shape. The CK method yields
the lowest relative errors for the radiative heat source, and errors comparable to those of the SLW
method for the incident heat ux. The dierences between the computational requirements of the gas
radiation models are similar to those reported for test case 1. Fig. 4 shows that doubling the number
320 P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328
Table 2
Average and maximum relative errors along lines (1 m, 1 m, :), (2 m, 1 m, :), (x, 1 m, 0.24 m) and (x, 1 m, 4 m), which are
denoted by subscripts 14, respectively, and CPU time for test case 2
DOM DTM
WSGG SLW-20 SLW-3 CK WSGG SLW-20 SLW-3 CK

E
rel, 1
(%) 30.04 3.35 8.09 1.16 30.38 4.29 8.57 2.28
E
max, 1
(%) 50.40 9.68 23.09 4.90 44.46 14.67 29.60 5.76

E
rel, 2
(%) 6.11 2.57 3.17 4.86 6.66 2.84 2.94 5.50
E
max, 2
(%) 12.52 2.84 5.26 6.60 16.88 4.67 4.37 8.91

E
rel, 3
(%) 8.45 8.86 6.26 5.27 12.30 6.50 4.68 3.30
E
max, 3
(%) 19.53 10.65 8.75 7.15 21.37 9.26 8.87 5.64

E
rel, 4
(%) 1.81 1.23 2.21 1.91 1.63 1.62 2.81 1.43
E
max, 4
(%) 4.26 2.03 3.00 3.52 4.49 3.11 3.94 3.04
t
CPU
(s) 21.2 100.2 20.6 1642.8 6.4 24.2 4.5 592.5
-40
-35
-30
-25
-20
-15
-10
CK (11x11x25, T4)
CK (22x22x50, T6)
SLW-20 (11x11x25 , T4)
SLW-20 (22x22x50, T6)
10
15
20
25
30
35
0 0.5 1 1.5 2
z [m]
0 0.5 1 1.5 2
z [m]
H
[
k
W
/
m
2
]
3
_

.
q
[
k
W
/
m
]
(a)
(b)
Fig. 4. Comparison of DOM predictions for test case 2 using two dierent spatial}angular discretizations: (a) Radiative
heat source along the centerline (1 m, 1 m, :); (b) incident heat ux along (2 m, 1 m, :).
of control volumes in each coordinate direction and switching from the T4 to the T6 quadrature has
a negligible inuence on the predictions.
3.1.3. Test case 3
In the last test case the temperature and gas composition are typical of those in a gas-red furnace
with one burner placed at the center of one of the walls (x = 1 m, , = 1 m). The gas composition
is a uniform mixture of 10% CO
2
, 20% H
2
O and 70% N
2
on a molar basis. The gas temperature is
dened as 1 =(1
c
(:) 400)

[(r}R) +800, where 1


c
(:) increases linearly from 400 K at : =0 m to
1800 K at : = 0.375 m, and decreases linearly farther downstream from 1800 to 800 K at the exit
section (: = 4 m). In this equation r is the distance from a point to the burner axis, R = 1 m, and
[(r}R) = 1 3(r}R)
2
+ 2(r}R)
3
if r}R61, and [(r}R) = 0 otherwise. The results computed using a
non-uniform 171724 grid are displayed in Fig. 5, and the average and maximum relative errors
are shown in Table 3.
P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328 321
-600
-500
-400
-300
-200
-100
0
100
SNB
CK
WSGG
-600
-500
-400
-300
-200
-100
0
100
5
10
15
20
25
0 2 1 3 4
z [m]
0 2 1 3 4
z [m]
0 2 1 3 4
z [m]
0 2 1 3 4
z [m]
5
10
15
20
25
SNB
SLW-di-20
SLW-cv-20
SLW-di-3
SLW-cv-3
SNB
SLW-di-20
SLW-cv-20
SLW-di-3
SLW-cv-3
SNB
CK
WSGG
H
[
k
W
/
m
2
]
H
[
k
W
/
m
2
]
3
_

.
q

[
k
W
/
m
]


3
_

.
q

[
k
W
/
m
]


(a)
(c) (d)
(b)
Fig. 5. Predicted proles of radiative heat source and incident heat ux calculated by the DOM compared with the ray
tracing SNB benchmark data [39] for test case 3: (a) Radiative heat source along the centerline (1 m, 1 m, :) using WSGG
and CK methods; (b) radiative heat source along the centerline (1 m, 1 m, :) using SLW method; (c) incident heat ux
along (2 m, 1 m, :) using WSGG and CK methods; (d) incident heat ux along (2 m, 1 m, :) using SLW method.
Table 3
Average and maximum relative errors along lines (1 m, 1 m, :) and (2 m, 1 m, :), which are denoted by subscripts 1 and
2, respectively, and CPU time for test case 3

E
rel, 1
(%) E
max, 1
(%)

E
rel, 2
(%) E
max, 2
(%) t
CPU
(s)
DOM WSGG 21.60 30.04 6.02 9.59 45.6
SLW-di-20 3.16 14.54 1.73 4.20 4885.3
SLW-di-3 8.23 29.03 9.83 12.85 212.4
SLW-cv-20 10.25 17.55 5.33 8.31 200.0
SLW-cv-3 6.35 33.92 5.28 14.34 30.9
CK 2.53 12.30 3.27 4.55 25808.3
DTM WSGG 17.82 36.93 3.18 6.09 14.4
SLW-di-20 9.46 38.90 1.39 3.05 1287.6
SLW-di-3 10.08 33.62 7.04 8.95 44.0
SLW-cv-20 10.97 33.11 2.85 4.90 89.8
SLW-cv-3 12.09 37.71 7.94 12.30 11.7
CK 8.42 37.88 0.70 2.18 9420.2
322 P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328
-600
-500
-400
-300
-200
-100
0
100
0 1 2 3 4
z [m]
(17x17x24, T4)
(34x34x48, T6)
3
_

.
q

[
k
W
/
m
]


Fig. 6. Comparison of radiative heat source along the centerline (1 m, 1 m, :) for test case 3 using the DOM, SLW-di-20
and two dierent spatial}angular discretizations.
The CK predictions closely reproduce the SNB benchmark data with an average relative error
below 3% for the radiative source and 3.5% for the incident heat ux. The maximum relative
error is larger than in the other test cases, since the source in the vicinity of the front and back
walls (: = 0 and 4 m, respectively) is small. This originates high local relative errors even though
the absolute errors may be small. The SLW calculations were performed using both the double
integration (SLW-di) and the convolution (SLW-cv) approaches, with (SLW-3) or without (SLW-20)
optimization of the absorption cross sections. The optimization was always based on total emissivities
computed from the double integration approximation. In all cases the results are in good agreement
with the SNB data, with average relative errors below 10%, except in one case where this value is
marginally exceeded. The convolution approach is somewhat less accurate than the double integration
for N
g
=20, even though the medium is homogeneous. If N
g
=3 there is a decrease of accuracy for
the double integration approximation, as expected. However, the convolution approach for N
g
= 3
gives more accurate results than for N
g
= 20. The WSGG predictions are the least accurate ones,
like in the previous problems, particularly in the calculation of the radiative source. Fig. 6 shows
that the results may be regarded as grid independent.
The CK and the SLW-di methods are both signicantly more computationally demanding in the
case of mixtures of two absorbing gases. The CK method requires N
2
quadrature points within a
band, while the SLW-di method requires (N
g
+ 1)
2
gray gases. Therefore, as a rst approximation,
the CPU time in the case of a binary mixture increases by a factor of N for the CK method, and by
a factor of (N
g
+1) for the SLW-di method, compared to the case of a single absorbing gas. These
trends are consistent with the computing times summarized in Table 3. The SLW-cv method, which
involves only (N
g
+ 1) gray gases, like in the case of a single absorbing gas, is signicantly faster
than the SLW-di method. If the optimization of absorption cross sections is used, further signicant
time savings are achieved. It is worth noting that the CPU time for the SLW-cv-3 and WSGG
methods is comparable, but the rst one is much more accurate.
3.2. Discrete transfer method
In general, the predictions obtained using the DTM are similar to those obtained using the DOM,
but there a few aspects that are worth noting. The predictions for test case 1 are not shown, since
P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328 323
-40
-35
-30
-25
-20
-15
0 0.5 1 1.5 2
z [m]
0 0.5 1 1.5 2
x [m]
0 0.5 1 1.5 2
z [m]
10
15
20
25
30
35
-80
-70
-60
-50
-40
-30
-20
15
17
19
21
23
25
0 0.5 1 1.5 2
x [m]
H
[
k
W
/
m
2
]
H
[
k
W
/
m
2
]
3
_

.
q

[
k
W
/
m
]


3
_

.
q

[
k
W
/
m
]


SNB
CK
SLW-20
SLW-3
WSGG
(a) (b)
(c) (d)
Fig. 7. Predicted proles of radiative heat source and incident heat ux calculated by the DTM compared with the ray
tracing SNB benchmark data [39] for test case 2: (a) Radiative heat source along the centerline (1 m, 1 m, :); (b) incident
heat ux along (2 m, 1 m, :); (c) radiative heat source along (x, 1 m, 0.24 m); (d) incident heat ux along (x, 1 m, 4 m).
they are hardly distinguishable from those plotted in Fig. 1. The predictions for test cases 2 and
3 are shown in Figs. 710. The average and maximum errors, and the CPU time are listed in
Tables 13.
The incident heat uxes predicted by the DTM are as accurate as, or marginally more accurate
than, the heat uxes calculated using the DOM. In fact, once the computational domain is spatially
discretized, the DTM integrates the RTE analytically along a line of sight. On the contrary, the
DOM does not perform an exact integration of the RTE, and requires a discretization scheme to
compute the radiation intensity at the faces of every control volume.
The predicted radiative heat source is generally less accurate if the DTM is used, and the proles
often show unphysical oscillations, as shown in Figs. 7 and 9. These oscillations are related to the
way the DTM computes the radiative heat source. The calculation is based on the variation of the
radiation intensity of the radiation beams crossing a control volume, according to Eqs. (26) or (28).
The number of radiation beams, i.e., the number of terms in the inner summation in Eqs. (26) and
324 P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328
-40
-35
-30
-25
-20
-15
-10
0 0.5 1 1.5 2
z [m]
10
15
20
25
30
35
0 0.5 1 1.5 2
z [m]
CK (11x11x25, T4)
CK (22x22x50, T6)
SLW-20 (11x11x25 ,T4)
SLW-20 (22x22x50, T6)
H
[
k
W
/
m
2
]
3
_

.
q

[
k
W
/
m
]


(a) (b)
Fig. 8. Comparison of DTM predictions for test case 2 using two dierent spatial}angular discretizations: (a) Radiative
heat source along the centerline (1 m, 1 m, :); (b) incident heat ux along (2 m, 1 m, :).
-600
-500
-400
-300
-200
-100
0
100
-600
-500
-400
-300
-200
-100
0
100
5
10
15
20
25
5
10
15
20
25
0 1 2 3 4
z [m]
0 1 2 3 4
z [m]
0 1 2 3 4
z [m]
0 1 2 3 4
z [m]
3

.
q
[
k
W
/
m
]
3
_

.
q
[
k
W
/
m
]
SNB
CK
WSGG
SNB
SLW-di-20
SLW-cv-20
SLW-di-3
SLW-cv-3
H
[
k
W
/
m
2
]
SNB
SLW-di-20
SLW-cv-20
SLW-di-3
SLW-cv-3
SNB
CK
WSGG
H
[
k
W
/
m
2
]
(a) (b)
(c) (d)
Fig. 9. Predicted proles of radiative heat source and incident heat ux calculated by the DTM compared with the
ray tracing SNB benchmark data [39] for test case 3: (a) Radiative heat source along the centerline (1 m, 1 m, :) us-
ing WSGG and CK methods; (b) radiative heat source along the centerline (1 m, 1 m, :) using SLW method; (c) inci-
dent heat ux along (2 m, 1 m, :) using WSGG and CK methods; (d) incident heat ux along (2 m, 1 m, :) using SLW
method.
P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328 325
-600
-500
-400
-300
-200
-100
0
100
0 2 1 3 4
z [ m]
(17x17x24, T4)
(34x34x48, T6)
3
_

.
q
[
k
W
/
m
]
Fig. 10. Comparison of radiative heat source along the centerline (1 m, 1 m, :) for test case 3 using the DTM, SLW-di-20
and two dierent spatial}angular discretizations.
(28), as well as the distance traveled by a radiation beam within a control volume, change from
control volume to control volume. Large changes between neighboring control volumes may occur if
the angular discretization is not ne enough, causing the observed wiggles. The only way to reduce
them is to further rene the angular discretization. Figs. 8 and 10 show that the oscillations are
only slightly reduced by using 6 6 instead of 4 4 solid angles per octant, revealing that a ner
discretization would be needed to further reduce them.
The CPU time for the DTM is lower than for the DOM for the present test cases, as shown in
Tables 13. However, it is important to realize that the DTM does not require an iterative procedure
for black walls, as in these cases, while the DOM needs an iterative procedure for the CLAM
scheme, but not for the STEP scheme. The calculations performed using the DOM and the STEP
scheme are faster than the DTM calculations. As an example, in test case 3 the DOM}STEP}CK
calculations required 2503 s, compared to 9420 s for DTM}CK, and 25808 s for DOM}CLAM}CK.
In the case of gray walls, all the methods are iterative, and the DOM may become faster than the
DTM, since one iteration of the DOM is faster than one iteration of the DTM.
4. Conclusions
Three non-gray gas radiation models that provide the absorption coecient of the medium as the
basic radiative property have been used to solve radiative transfer problems in three-dimensional
enclosures using the discrete ordinates and the discrete transfer methods. Results obtained using a
ray tracing method and the statistical narrow band model were taken as a reference. The following
conclusions may be drawn from the analysis carried out:
1. The gray gas implementation of the WSGG generally yields unsatisfactory results. The calculation
of the local absorption coecient based on a mean beam length for the whole enclosure may
give fair predictions of the incident heat uxes in the case of homogeneous and isothermal media,
but the errors become signicant in non-homogeneous or non-isothermal media. Moreover, large
326 P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328
errors in the radiative heat source are consistently found. Therefore, this simple and fast approach
should not be used in practical applications.
2. The WSGG model provides substantial improvements over the gray gas approach, but it is by
far the least accurate of the non-gray gas models. Average relative errors up to 30% and local
relative errors up to 50% have been found. The WSGG model should be used only if errors of
this order of magnitude are tolerable.
3. The CK model is very accurate, but too time consuming for many engineering problems, partic-
ularly in the case of mixtures with more than one absorbing gas.
4. The SLW model is quite accurate, but the computational requirements are also high in the case of
two absorbing gases if the double integration approximation is employed. Signicant savings in
CPU time are achieved using the convolution approach, but this is restricted to spatially constant
molar ratio of the two gases. The optimization of the absorption cross sections is an eective
way to signicantly reduce the computational requirements. These become comparable to those of
the WSGG in the case of a single absorbing species or two absorbing species together with the
convolution approach. The SLW method is the best alternative regarding the compromise between
accuracy and numerical eciency.
5. The evaluation of gas radiation models requires spatial and angular independent solutions. This
is dicult to achieve for the DOM with the commonly used STEP scheme. This diculty is
overcome using a high-order resolution discretization scheme, such as the CLAM scheme.
6. The DTM often yields proles of radiative heat source with unphysical oscillations. These can
only be suppressed by signicantly rening the angular discretization.
7. In the case of black walls, the DOM}STEP is faster than the DTM, and this is faster than the
DOM}CLAM. However, the last method is iterative, and one iteration of the DOM}CLAM is
faster than one iteration of the DTM.
Acknowledgements
The PRAXIS XXI Program of the Portuguese Ministry of Science and Technology supported this
work under contract PRAXIS}P}EME}12034}1998.
References
[1] Lockwood FC, Shah NG. A new radiation solution method for incorporation in general combustion prediction
procedures. Proceedings of the 18th International Symposium on Combustion, The Combustion Institute, 1981.
p. 140514.
[2] Chandrasekar S. Radiative transfer. New York: Dover Publications, 1960.
[3] Fiveland WA. Discrete-ordinates solutions of the radiative transport equation for rectangular enclosures. J Heat
Transfer 1984;106:699706.
[4] Raithby GD, Chui EH. A nite volume method for predicting a radiant heat transfer in enclosures with participating
media. J Heat Transfer 1990;112:41523.
[5] Lallemant N, Sayre A, Weber R. Evaluation of emissivity correlations for H
2
OCO
2
N
2
}air mixtures and coupling
with solution methods of the radiative transfer equation. Prog Energy Combust Sci 1996;22:54374.
[6] Taine J, Souani A. Gas IR radiative properties: from spectroscopic data to approximate models. In Hartnett JP,
Irvine Jr TF, editors, Advances in heat transfer, vol. 33. San Diego: Academic Press, 1999. p. 295414.
P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328 327
[7] Ludwig CB, Malkmus W, Reardon JE, Thompson JAL. In Hartnett JP, Irvine Jr TF, editors, Handbook of infrared
radiation from combustion gases. Technical Report NASA SP-3080, Washington, DC, 1973.
[8] Edwards DK. Molecular gas band radiation. In: Advances in heat transfer, vol. 12. New York: Academic Press,
1976. p. 11593.
[9] Kim TK, Menart JA, Lee HS. Nongray radiative gas analyses using the SN discrete ordinates method. J Heat
Transfer 1991;113:94652.
[10] Zhang L, Souani A, Taine J. Spectral correlated and non-correlated radiative transfer in a nite axisymmetric system
containing an absorbing and emitting real gas-particle mixture. Int J Heat Mass Transfer 1988;31:226172.
[11] Liu F, G ulder

OL, Smallwood GJ. Non-grey gas radiative transfer analyses using the statistical narrow-band model.
Int J Heat Mass Transfer 1998;41:222736.
[12] Menart JA, Lee HS, Kim TK. Discrete ordinates solutions of non-gray radiative transfer with diusely reecting
walls. J Heat Transfer 1993;115:18493.
[13] Goody RM, West R, Chen L, Chrisp D. The correlated-k method for radiation calculations in non-homogeneous
atmospheres. JQSRT 1989;42:53950.
[14] Lacis AA, Oinas V. A description of the correlated k-distribution method for modeling non-gray gaseous absorption,
thermal emission, and multiple scattering in vertically inhomogeneous atmospheres. J Geophys Res 1991;96:
902763.
[15] Rivi ere P, Scutaru D, Souani T, Taine J. A new ck data basis suitable from 300 to 2500 K for spectrally
correlated radiative transfer in CO
2
H
2
O-transparent gas mixtures. Proceedings of the 10th International Heat Transfer
Conference, vol. 2. Brighton, UK, 1994. p. 12934.
[16] Tang KC, Brewster MQ. K-distribution analysis of gas radiation with non-gray, emitting, absorbing, and anisotropic
scattering particles. J Heat Transfer 1994;116:9805.
[17] Liu F, Smallwood GJ, G ulder

OL. Radiation heat transfer calculations using the SNBCK method. AIAA Paper
99-3679, 1999.
[18] Liu F, Smallwood GJ, G ulder

OL. Band lumping strategy for radiation heat transfer using a narrowband model.
J Thermophys Heat Transfer 2000;14:27881.
[19] Marin O, Buckius RO. Wideband correlated-k method applied to absorbing, emitting, and scattering media.
J Thermophys Heat Transfer 1996;10:36471.
[20] Liu F, Smallwood GJ, G ulder

OL. Application of the statistical narrow band correlated-k method to low-resolution
spectral intensity and radiative heat transfer calculationseects of the quadrature scheme. Int J Heat Mass Transfer
2000;43:311935.
[21] Liu F, Smallwood GJ, G ulder

OL. Application of the statistical narrow band correlated-k method to non-grey gas
radiation in CO
2
H
2
O mixtures: approximate treatments of overlapping bands. JQSRT 2001;68:40117.
[22] Hottel HC, Sarom AF. Radiative transfer. New York: McGraw-Hill, 1967.
[23] Modest MF. The weighted-sum-of-gray-gases model for arbitrary solution methods in radiative transfer. J Heat
Transfer 1991;113:6506.
[24] Denison MK, Webb BW. A spectral line-based weighted-sum-of-gray-gases model for arbitrary RTE solvers. J Heat
Transfer 1993;115:100412.
[25] Denison MK, Webb BW. The spectral line-based weighted-sum-of-gray-gases model in non-isothermal
non-homogeneous media. J Heat Transfer 1995;117:35965.
[26] Denison MK, Webb BW. The spectral-line weighted-sum-of-gray-gases model for H
2
O}CO
2
mixtures. J Heat Transfer
1995;117:78892.
[27] Pierrot L, Souani A, Taine J. Accuracy of narrow-band and global models for radiative transfer in H
2
O, CO
2
and
H
2
OCO
2
mixtures at high temperature. JQSRT 1999;62:52348.
[28] Marakis JG. Application of narrow and wide band models for radiative transfer in planar media. Int J Heat Mass
Transfer 2001;44:13142.
[29] Fiveland WA, Jamaluddin AS. Three-dimensional spectral radiative heat transfer solutions by the discrete-ordinates
method. J Thermophys Heat Transfer 1991;5:3359.
[30] Park HM, Lee JH, Park JH. Analysis of spectral radiative heat transfer in furnaces using an ecient computational
technique. JQSRT 1999;62:45975.
328 P.J. Coelho / Journal of Quantitative Spectroscopy & Radiative Transfer 74 (2002) 307328
[31] Bresslo NW, Moss JB, Rubini PA. CFD prediction of coupled radiation heat transfer and soot production in
turbulent ames. Proceedings of the 26th International Symposium on Combustion, The Combustion Institute, 1996.
p. 237986.
[32] Liu F, Becker HA, Bindar Y. A comparative study of radiative heat transfer modelling in gas-red furnaces using
the simple grey gas and the weighted-sum-of-gray-gases models. Int J Heat Mass Transfer 1998;41:335771.
[33] Maruyama S, Guo Z. Radiative heat transfer in arbitrary congurations with non-gray absorbing, emitting, and
anisotropic scattering media. J Heat Transfer 1999;121:7226.
[34] Souani A, Djavan E. A comparison between weighted-sum-of-gray-gases and statistical narrow-band radiation
models for combustion applications. Combust Flame 1994;97:24050.
[35] Marakis JG, Kakaras E, Kakatsios X, Papageorgiou N. Application of the direct numerical integration method in a
cylindrical geometry containing non-gray pressurised gases. Proceedings Eurotherm #56, Heat Transfer in Radiating
and Combusting Systems-3, Delphi, Greece, 13 April 1998. p. 26374.
[36] Cumber PS, Fairweather M, Ledin HS. Application of wide band radiation models to non-homogeneous combustion
systems. Int J Heat Mass Transfer 1998;41:157384.
[37] Denison MK. A spectral line-based weighted-sum-of-gray-gases model for arbitrary RTE solvers. Ph.D. Thesis,
Mechanical Engineering Department, Brigham Young University, Provo, UT, 1994.
[38] Goutiere V, Liu F, Charette A. An assessment of real-gas modelling in 2D enclosures. JQSRT 2000;64:299326.
[39] Liu F. Numerical solutions of three-dimensional non-grey gas radiative transfer using the statistical narrow-band
model. J Heat Transfer 1999;121:2003.
[40] Souani A, Taine J. High temperature gas radiative property parameters of statistical narrow-band model for H
2
O,
CO
2
and CO, and correlated-k model for H
2
O and CO
2
. Int J Heat Mass Transfer 1997;40:98791.
[41] Smith TF, Shen ZF, Friedman JN. Evaluation of coecients for the weighted sum of gray gases model. J Heat
Transfer 1982;104:6028.
[42] Gosman AD, Lockwood FC. Incorporation of a ux model for radiation into a nite dierence procedure for furnace
calculations. Proceedings of the 14th International Symposium on Combustion, The Combustion Institute, 1973.
p. 66171.
[43] Denison MK, Webb BW. An absorption-line blackbody distribution function for ecient calculation of total gas
radiative transfer. JQSRT 1993;50:499510.
[44] Denison MK, Webb BW. Development and application of an absorption-line blackbody distribution function for
CO
2
. Int J Heat Mass Transfer 1995;38:181321.
[45] Thurgood CP, Pollard A, Becker HA. The 1
N
quadrature set for the discrete ordinates method. J Heat Transfer
1995;117:106870.
[46] Van Leer B. Towards the ultimate conservation dierence scheme. II Monoticity and conservation combined in a
second-order scheme. J Comput Phys 1974;14:36170.
[47] Jessee JP, Fiveland WA. Bounded, high-resolution dierencing schemes applied to the discrete ordinates method.
J Thermophys Heat Transfer 1997;11:5408.
[48] Khosla PK, Rubin SG. A diagonally dominant second order accurate implicit scheme. Comput Fluids 1974;2:2079.

Das könnte Ihnen auch gefallen