Computer Physics Communications 258 (2021) 107608

Contents lists available at ScienceDirect

Computer Physics Communications

journal homepage: www.elsevier.com/locate/cpc

mcrtFOAM: A mesh-agglomeration Monte Carlo ray-tracing solver for

radiative transfer in gray semitransparent solids✩
∗
Chao Fan, Xin-Lin Xia , Jian Qiu, Chuang Sun
Key Laboratory of Aerospace Thermophysics of MIIT, School of Energy Science and Engineering, Harbin Institute of Technology, No. 92, West Dazhi
Street, Harbin, Heilongjiang 150001, China

article info a b s t r a c t

Article history: Monte Carlo ray-tracing method has become a promising technique for solving radiative transfer.
Received 15 October 2019 However, large memory footprint or long CPU time hampers its further utilization. In this work, a
Received in revised form 20 August 2020 Monte Carlo ray-tracing method radiation solver called mcrtFOAM is developed in the framework of
Accepted 1 September 2020
open source CFD toolbox OpenFOAM. The radiative source term is calculated based on the radiation
Available online 11 September 2020
distribution factor to avoid repeated ray tracing. In addition, a mesh-agglomeration technique is used to
Keywords: overcome the problem of memory catastrophe for storing a huge radiation distribution factor matrix
Mesh agglomeration caused by a large number of mesh elements. The key principle is to agglomerate the refined mesh
Monte Carlo method elements to the coarse ones, and the calculation of radiation distribution factor as well as radiative
Radiation distribution factor source term is carried out in the coarse mesh, while the procedure of ray tracing as well as solving
Radiative heat transfer the heat conduction is performed in the refined mesh. After validation of the algorithm code within a
OpenFOAM cubic enclosure, a coupled conductive–radiative heat transfer problem within a cylindrical enclosure
filled with gray semitransparent solids is also tested. To test the present solver in realistic industrial
systems, a coupled conductive–radiative heat transfer within a tomographic porous structure with
gray surfaces is further investigated. All results are compared with previous studies and the solutions
by the Finite Volume Discrete Ordinate Method (FVDOM) in OpenFOAM, and they all show great
satisfaction. When the mcrtFOAM solver is used for solving the radiative transfer equation, it is found
that the CPU time and the memory footprint consumed by the present mcrtFOAM solver with mesh-
agglomeration technique are far less than that without this technique. Convincing first implementation
exercises indicate that both significant memory and CPU benefits can be expected for configurations
of industrial interest These findings will contribute greatly to achieve the calculation of coupled heat
transfer involving radiation within large number of mesh elements.
© 2020 Elsevier B.V. All rights reserved.

1. Introduction simulate radiative transfer, such as the Monte Carlo ray-tracing

method (MCRTM) [12–14]. Not only the MCRTM is unaffected by
In high temperature applications, thermal radiation has been ray effects since there is no solid angle discretization [15], but also
widely considered in heat transfer analyses [1], such as chem- is easy to be implemented in emitting, absorbing and scattering
ical reactors [2], porous burners [3], and volumetric solar re- media [16–18]. Thus, the solutions of MCRTM are usually used as
ceivers [4]. In the past few decades, lots of effective methods benchmarks [19].
have been proposed and developed to solve the Radiative Transfer Depending on whether the ray-tracing procedure needs to be
Equation (RTE). Methods for solving the RTE can be classified repeatedly performed, MCRTM can be divided into energy-based
into two categories: the deterministic methods and the stochas- MCRTM and the radiative energy absorption distribution (READ)-
tic methods [5]. The deterministic methods use ordinate-based based MCRTM [20]. For energy-based MCRTM, it is considered
techniques to deal with the RTE, such as the Finite Volume that the ray carries certain energy, and the radiative source term
Method (FVM) [6,7], the Discrete Ordinates Method (DOM) [8,9], could be directly calculated after a large number of rays are
and the spherical harmonic method (or called P-N method) [10, traced. Every time the temperature field is updated, the ray-
11]. Whereas, the stochastic methods use probability models to tracing procedure needs to be executed. The READ-based MCRTM,
however, conducts the ray tracing procedure only once, which is
✩ The review of this paper was arranged by Prof. David W. Walker. realized by decoupling the ray-tracing procedure and the solution
∗ Corresponding author. of the iterative temperature field, and the results are expressed
E-mail address: [email protected] (X.-L. Xia). in the format of radiation distribution factor [21,22]. This kind of

https://doi.org/10.1016/j.cpc.2020.107608
0010-4655/© 2020 Elsevier B.V. All rights reserved.
C. Fan, X.-L. Xia, J. Qiu et al. Computer Physics Communications 258 (2021) 107608

technique has been applied in solving radiative heat transfer be-

Nomenclature tween surfaces by viewFactor solver (or called surface to surface
A Area of a surface element, m2 radiation model), not in semitransparent media yet. Before solv-
i, j Index of element ing the RTE, the view factors are calculated by the pre-processing
solver ‘‘viewFactorsGen’’. To improve the calculation efficiency
I Radiation intensity, W/(m sr)
as well as saving memory for storing view factors, the solver
M Memory footprint for solving RTE, MB
agglomerates the refined surface mesh elements into coarse ones.
n Refractive index
However, this solver can only work for radiative transfer between
N Radiative–conductive coefficient surfaces. In this work, the idea of agglomerating the refined mesh
Nb Bounding box elements number elements into coarse ones is applied to the calculation of radiation
Nray Rays number distribution factor, which agglomerates the refined surface and
Ns Number of surface elements volumetric mesh elements into coarse ones.
Nt Number of total elements This paper is organized as follows. Section 2 introduces the
Nv Number of volumetric elements principle and treatment of the present mcrtFOAM solver. For
RD Radiation distribution factor validation, the heat flux distribution of a cubic enclosure with
t CPU time, s uniform temperature profile and the divergence of heat flux dis-
T Temperature, K tribution of the enclosure with non-uniform temperature profile
are calculated in Section 3.1. Furthermore, a coupled radiative–
TC , TH Cold and hot wall temperature, K
conductive heat transfer problem within a cylindrical enclosure
V Volume of an volumetric element, m3
is computed in Section 3.2. After which, the CPU time and the
Greek symbols memory footprint for solving coupled conductive–radiative heat
transfer with various bounding box mesh elements numbers con-
β Extinction coefficient, m−1
sumed by mesh-agglomeration MCRTM are compared with the
ε Emissivity non-mesh-agglomeration one. In Section 3.3, the present mcrt-
κ Absorption coefficient, m−1 FOAM solver is used to solve a conductive–radiative heat transfer
λc Thermal conductivity, W/(m K) problem within a complex porous structure from micro-computer
σ Stefan Boltzmann constant, W/(m2 K4 ) tomographic (µ-CT) technique.
Subscrits
2. Numerical treatments
b Black
m Medium 2.1. Calculation methodology of radiation distribution factor by the
w Wall MCRTM in polyhedron mesh

2.1.1. Radiation distribution factor

As mentioned above, the result of ray tracing procedure is
Monte Carlo method is also known as the radiation distribution given in the format of the radiation distribution factor. For gray
factor-based MCRTM, which is widely used in solving for coupled media, the radiation distribution factor RDij is defined as a frac-
heat transfer involving radiation [23,24]. However, the radiation tion of the emissive monochromatic power of element i in a
distribution factor-based MCRTM can be time and memory ex- heat transfer system that is absorbed by element j after one
pensive [25,26] due to a huge radiation distribution factor matrix or more reflections and/or scatterings. Generally, the radiation
to be stored, especially in terms of memory footprint which is distribution factor of gray medium and gray surface slightly varies
almost proportional to the square of mesh elements number. with the temperature, thus the radiation distribution factor is
Aiming to overcome these weaknesses of the MCRTM, some independent of temperature.
researchers have investigated the time-consuming problem. For For a closed system containing Nv volumetric elements and
example, parallel processing [27–29], octrees acceleration [30] Ns surface elements (thus the total number of element Nt = Ns
and bi-directionally weighted Monte Carlo [15] are commonly
+ Nv ), and the equation to characterize the radiative energy in
used to accelerate the ray-tracing procedure. These techniques
the volumetric/surface element i can be expressed by radiation
significantly reduce the cost of the ray-tracing procedure in terms
distribution factor as follows [32]:
of CPU time. However, another key problem of high memory foot-
print has scarcely been studied when involving large number of Nv Ns
∑ ∑
mesh elements. In fact, this problem is a crucial constraint for the 4κi Vi σ Ti4 = 4κj Vj σ Tj4 RDji + εk Ak σ Tk4 RDki (1a)
MCRTM when the radiative source term is calculated based on the j=1 k=1
radiation distribution factor. For instance, it is almost impossible Nv
∑ Ns
∑
to store a huge radiation distribution factor matrix with limited εi Ai σ Ti4 = 4κj Vj σ Tj4 RDji + εk Ak σ Tk4 RDki (1b)
Random Access Memory (RAM) [25] for a complex structure j=1 k=1
which is discretized into refined mesh. To the knowledge of the
authors, there are few literatures about how to reduce memory where κ and ε are the absorption coefficient and emissivity,
footprint when solving the RTE using the radiation distribution respectively, V and A are volume and area of the element, respec-
factor-based MCRTM. tively, and RDij is the radiation distribution factor.
For radiative transfer in semitransparent medium, official In the Monte Carlo ray-tracing procedure, it is required to
OpenFOAM solvers have incorporated radiation models based on simulate the random emitting point, direction and the possi-
P1 and Finite Volume Discrete Ordinate Method (FVDOM) [31]. ble transmission distance of rays. In addition, when the rays
However, there are few literatures about OpenFOAM solver based intersect with the boundary surface or scatter in the medium,
on MCRTM. At the same time, considering the RAM occupa- the direction of reflection or scattering should be determined
tion, mesh-agglomeration technique should be included in the and one can refer to the detailed description [1,33] of these
MCRTM solver in OpenFOAM. In fact, the mesh-agglomeration treatments. The following subsection introduces how to deter-
2
C. Fan, X.-L. Xia, J. Qiu et al. Computer Physics Communications 258 (2021) 107608

Fig. 1. Schematic diagram of the decomposition for a polygon surface mesh element.

Fig. 2. Schematic diagram illustrating of the decomposition of a polyhedron volume mesh element.

minate evenly distributed emitting position in a polygon surface to the weight of the triangle area, and then the member func-
element/polyhedron volumetric element, only. tion ‘‘randomPoint’’ in class ‘‘triangle’’ is invoked, as shown in
Algorithm 1. Similarly, for a polyhedron volumetric element, as
shown in Fig. 2(a), where O is the center of the element, it can also
2.1.2. Determination of emitting position in polygon/polyhedron el-
be decomposed into several pyramids (Fig. 2(b)), then the pyra-
ements
mids can be further decomposed into several sub-tetrahedrons
For the determination of emitting position in a polygon sur-
(Fig. 2(c)). The random emitting position in a tetrahedron can be
face element, the OpenFOAM class library provides ‘‘triangle’’ determined by invoking the member function ‘‘randomPoint’’ in
class, and the member function ‘‘randomPoint’’ can generate class ‘‘tetrahedron’’.
random points in a triangle, whose implementation can be seen
in Ref. [34]. In order to make the generating random emitting 2.2. Mesh-agglomeration approach for calculating radiative source
position algorithm applicable to an arbitrary polyhedron sur- term
face element, the polygon can be decomposed into many sub-
triangles. For a polygon surface element, as shown in Fig. 1(a), it To decouple the ray-tracing procedure and the solution of the
can be decomposed into three triangles: △ABE, △BCE and △ECD. iterative temperature field, the radiation distribution factor ma-
Since the emitting points are randomly and evenly distributed in trix needs to be stored when solving the combined conductive–
each volumetric element/surface element (i.e., uniform emission radiative heat transfer. From Eq. (1), the memory for storing the
distribution), a random number is first generated to determine radiation distribution factor is proportional to the square of the
which sub-triangle the random point will be located according mesh elements number. Theoretically, the memory for storing
3
C. Fan, X.-L. Xia, J. Qiu et al. Computer Physics Communications 258 (2021) 107608

the radiation distribution factor will occupy at least 1.1 TB if the Assuming that the coarse volumetric mesh element I contains
computation case involves 400 000 mesh elements, which is un- ni refined mesh elements, the corresponding temperature is Ti ,
realistic for most computers. In addition, all radiation distribution and the coarse surface mesh element II contains nj refined mesh
factors must be traversed when calculating the radiative source elements, the corresponding temperature is Tj . Then, the net
term. The traversal of a huge amount of radiative distribution radiative heat flux Qr between the two coarse mesh elements can
factors will inevitably lead to a sharp increase in CPU time. One be expressed as:
of the most effective ways to reduce memory footprint and CPU nj
ni
time is to reduce the mesh elements number. ∑ ∑
Based on the above reasons, mesh-agglomeration strategy is Qr,I,II = (RDI,II 4κi Vi σ Ti4 − RDII,I εj Aj σ Tj4 ) (3)
employed in the present mcrtFOAM solver. By agglomerating i j

refined mesh elements into coarse ones, two types of mesh ele- According to the reciprocity of radiation distribution factor:
ments with different number for the same computational domain nj
ni
can be obtained. Then, the calculation of radiation distribution ∑ ∑
factor and the radiative source term is carried out in the coarse RDI,II 4κi Vi = RDII,I εj Aj (4)
mesh elements, while the procedure of ray tracing and heat i=1 j=1

conduction is performed in the refined ones. From Eqs. (3) and (4), the divergence of the radiative source
A schematic map showing the agglomeration of the refined term of the coarse mesh elements can be expressed as:
mesh elements into coarse ones is shown in Fig. 3, where Fig. 3(a)
ni ni nj
is the refined mesh, whose cell center points are shown in ∑ ∑ ∑
Fig. 3(b). Fig. 3(c) is the bounding box for agglomerating the −∇ • qr,I,II = RDI,II [ 4κi Vi σ Ti − (
4
4κi Vi / εj Aj )
refined mesh elements, which is defined as bounding box mesh. i i=1 j=1
First, member function, findCell, corresponding to class fvMesh, nj
∑ ∑ni
is invoked to search which bounding box mesh element the × εj Aj σ Tj4 ]/ Vi (5)
center points of the refined mesh elements are located, as shown j i
in Algorithm 2. The refined mesh elements whose center points
located in the same bounding box mesh element are agglom- 2.3. Mesh-agglomerating approach for solving combined radiative–
erated to a coarse mesh element (Fig. 3(e) and (f)). Herein, conductive heat transfer
the agglomerated coarse mesh element is defined as the parent
element and the refine mesh element whose mesh center point The governing equation of coupled radiative-conductive heat
in this coarse mesh element is defined as its child mesh element. transfer can be expressed as:
In the ray-tracing procedure, assuming that the coarse mesh
element I contains ni refined child mesh elements (the access ∇ • (λc ∇ T ) − ∇ • qr = 0 (6)
to child mesh elements is realized by ‘childElementId’ mem- where λc is the thermal conductivity, and qr is the radiative heat
ber function, as shown in Algorithm 3), and the total bundles flux.
Ni (N1 , N2 , N3 . . . ) are emitted from each refined child mesh el-
As mentioned above, the calculation of heat conduction is
ement. The coarse mesh element II contains mi refined child
performed in the refined mesh elements, and the calculation
mesh elements, with Mi bundles (M1 , M2 , M3 . . . ) absorbed by
of the radiative source term is performed in the coarse mesh
the refined child mesh elements, then the radiation distribution
elements, whose values are then assigned to the correspond-
factor between the two coarse mesh elements can be calculated
ing refined child mesh elements. The detailed implementation
with:
is shown in Algorithm 4. A flowchart of solving the combined
RDI,II = (M1 + M2 + · · · Mn )/(N1 + N2 + · · · Nn ) (2) conductive–radiative heat transfer is shown in Fig. 4.
4
C. Fan, X.-L. Xia, J. Qiu et al. Computer Physics Communications 258 (2021) 107608

Fig. 3. Schematic map illustrating the agglomeration of the refined mesh elements into coarse ones.

3. Results and discussion and the radiative source term is obtained based on the radiation
distribution factor. All the computations were carried out on an
To check the validity of the present mcrtFOAM solver, a cu- Intel Xeon E5-2660 20 processor Windows 7 server and 16 GB
bic enclosure filled with absorbing–emitting medium is inves- RAM. In all cases, the cutoff value (i.e transmissivity [35]) is set
tigated, and the results achieved by the mcrtFOAM solver are up to 10−7 , which is carefully examined to guarantee that it can
compared with the analytical solution. Furthermore, a coupled meet the demand of calculation errors of the present MCRTM. In
radiative–conductive problem in a cylindrical enclosure filled addition, the independence of the rays number Nray (per m3 /m2 )
with absorbing–emitting medium is also investigated by the total refined elements number Nt and the bounding box mesh
present mcrtFOAM solver. The heat conduction is based on FVM, elements number Nb are checked in each case.
5
C. Fan, X.-L. Xia, J. Qiu et al. Computer Physics Communications 258 (2021) 107608

Fig. 4. Schematic diagram of solving combined conductive–radiative heat transfer.

3.1. Cubic enclosure

The first case, as shown in Fig. 5, is a radiative transfer prob-

lem within a cubic enclosure full of absorbing–emitting medium.
The medium within the cubic enclosure is hot as Tm and three
different absorption coefficients (κ = 0.1, 1.0 and 10.0 m−1 ,
respectively) are investigated, and all the walls are black and cold
(Tw = 0 and ε = 1.0). The wall heat flux along the centerline of
a wall in this cubic enclosure case is investigated to check the
accuracy of the present mcrtFOAM solver. The exact intensity at
an arbitrary location is given by the following formula when the
medium temperature is known and uniform [36]:
I(s) = Ibw e−κ s + Ib (1 − e−κ s ) (7)
where I(s) is the radiation intensity of any location, Ibw is the radi-
ation intensity of the black walls and Ib is the radiation intensity
of the medium. It is found that the results are independent of the
mesh elements number and the rays number when Nt = 39 950
and Nray = 6 × 107 m−3 .
Fig. 5. The cubic enclosure (unit: m). To study the influence of bounding box mesh on the accuracy,
the computational domain is respectively discretized into uni-
form mesh and non-uniform mesh using GAMBIT software, and
6
C. Fan, X.-L. Xia, J. Qiu et al. Computer Physics Communications 258 (2021) 107608

the mesh elements number Nb = 15 × 15 × 15. The coarse mesh than those of the uniform one. This indicates that reducing the
elements agglomerated by uniform and non-uniform bounding bounding mesh size adjacent to the walls can effectively improve
box mesh are respectively shown in Figs. 6(c) and 7(c). the calculation accuracy.
Fig. 8 shows the results of the cubic enclosure case and the To validate the present mcrtFOAM solver for non-uniform
exact values along the centerline for three different absorption temperature profile, the temperature profile of the enclosure is
coefficients κ = 0.1, 1.0 and 10.0 m−1 . In the figure, the uniform set according to Ref. [35], which is given as:
mesh and non-uniform mesh represent uniform bounding box
mesh and non-uniform bounding box mesh, respectively. The Tm = (sin π x • sin π y • sin π z • π/σ )0.25 (8)
results in this case are satisfactory with the average relative errors
less than 1.0%. The CPU time for the present MCRTM with /with- Two different absorption coefficients are analyzed (κ = 0.5
out mesh-agglomeration is 177 s and 503 s, respectively. The and 5.0 m−1 ). The refined mesh elements number and rays num-
reason for the decrease in CPU time is attributed to the shrinking ber are Nt = 323 and Nray = 6.55 × 107 m−3 , respectively. For
of hard disk I/O writing as a result of decrease in radiation the bounding box mesh elements number Nb , it is found that
distribution factor. In addition, in terms of the error distribution, the results are independent of it when Nb = 153 . Fig. 9 shows
the relative errors near the walls are larger than those in the the divergence of radiative flux distribution of this present work,
middle, which is consistent with the conclusions presented in analytic solution [38] and Monte Carlo ray tracing method with
Ref. [37]. By comparing results of the uniform and non-uniform GPU acceleration [35]. As can be seen from this figure, the results
bounding box meshes, it can be seen that the relative errors of the obtained for this simulation are in excellent agreement with the
non-uniform bounding box mesh near the wall are much lower relative errors less than 0.8%. This implies the present mcrtFOAM
7
C. Fan, X.-L. Xia, J. Qiu et al. Computer Physics Communications 258 (2021) 107608

solver in this manuscript also works well for the non-uniform refined mesh elements is constant 44 587, and the results solved
temperature field case. by the non-agglomerated MCRTM are considered as reference
(Ref. [39] proved that the results of the case with over 16 686
3.2. Cylindrical enclosure mesh are independent of mesh elements number).
The reference temperature Tre is equal to TH and the radiative–
The second case deals with a coupled radiative–conductive conductive coefficient is defined as N = λc β /(4σ n2 Tre
3
). For each
heat transfer problem, which is confined in a cylindrical enclosure case, three different radiative–conductive coefficients N = 0.1,
containing absorbing–emitting medium, as shown in Fig. 10. In 1.0 and 10.0 are investigated. Fig. 11 shows the comparison of the
this case, the temperature of the top wall and the lateral wall of results between the present mesh-agglomeration MCRTM solver
the cylinder is maintained at TC = 500 K, whereas the bottom and the non-agglomeration MCRTM solver, where the number in
wall is at TH = 1000 K. The absorption coefficient is 1.0 m−1 this figure represents the bounding box mesh elements number.
and all the walls are black. Three bounding box mesh elements As can be seen from the figure, the maximum differences for
number Nb = 1989, 3488, 5454 are tested, where the rays nondimensional temperature profile obtained from these bound-
number in all cases is 3 × 107 m−3 . In these cases, the number of ing box mesh levels are less than 0.2%. This implies that the
8
C. Fan, X.-L. Xia, J. Qiu et al. Computer Physics Communications 258 (2021) 107608

third grid level (the largest grid number of each case) presents a latter with coarse mesh elements might possibly have been
reasonable precision for the numerical calculations. In addition, buried by the former.
with the increase of the radiative–conductive coefficient, the Table 1 shows the CPU time, the memory footprint of the
gap of relative errors between the present mesh-agglomeration non-mesh-agglomeration MCRTM and the mesh-agglomeration
MCRTM and the non-mesh-agglomeration MCRTM gradually nar- MCRTM with various radiative–conductive coefficients. From this
rows. When the radiative–conductive coefficient is N = 10, table, it can be seen that the CPU time and the memory footprint
the maximum relative errors are less than 0.003%. The large of the present mesh-agglomeration MCRTM are far less than
radiative–conductive coefficient means that thermal conduction those of the non-mesh-agglomeration MCRTM. This demonstrates
overwhelms thermal radiation, and the numerical errors of the that the present mesh-agglomeration MCTM in this work can
9
C. Fan, X.-L. Xia, J. Qiu et al. Computer Physics Communications 258 (2021) 107608

Fig. 6. The refined mesh and uniform bounding box mesh in Fig. 5.

Fig. 7. The refined mesh and the non-uniform bounding box mesh in Fig. 5.

Table 1
CPU time as well as memory footprint for the cases of Fig. 10.
Bounding box mesh Radiative– M / MB t / s
elements number conductive
coefficient
Non-mesh- – 0.1 14 612 340
agglomeration – 1.0 14 612 114
MCRTM – 10.0 14 612 113
Mesh- 0.1 72 2.6
agglomeration 1989 1.0 72 1.4
MCRTM 10.0 72 1.3
Mesh- 0.1 201 3.3
agglomeration 3488 1.0 201 1.8
MCRTM 10.0 201 1.8
Mesh- 0.1 442 4.4
agglomeration 5454 1.0 442 2.6
MCRTM 10.0 442 2.5

significantly reduce the memory footprint and CPU time under walls of the computation domain are treated as symmetry. It
limited errors. is found that the results are independent of the refined mesh
elements number and total rays number when they are up to
3.3. Tomographic porous structure Nt = 39 8431 and Nray = 5.0 × 107 mm−2 , respectively. Three
different emissivities ε = 0.1, 0.5 and 0.9 are investigated. For
In order to verify the robustness of the solver for the complex the purpose of assessing the accuracy of the present mcrtFOAM
models, a coupled radiative-conductive heat transfer problem solver, the FVDOM of the official OpenFOAM solver is used to
within a tomographic porous structure (Fig. 12) is tested by the solve this problem also.
present mcrtFOAM solver. The skeleton is treated as opaque and Fig. 13 shows the results of various bounding box mesh lev-
its surface emissivity is ε . The thermal conductivity λc of the els. As can be seen from the figure, the maximum differences
skeleton is 10 W/(m K). In this case, the temperature of the left for temperature profile obtained from these grid levels are less
and right walls is 1000 K and 500 K, respectively, and the lateral than 1%, which implies that the first grid level (the largest grid
10
C. Fan, X.-L. Xia, J. Qiu et al. Computer Physics Communications 258 (2021) 107608

Fig. 8. Comparison of radiative heat flux along line A–A in Fig. 5: (a) κ = 0.1, (b) κ = 1.0, (c) κ = 10.0.

emissivity of the skeleton surface, the average temperature gradi-

ent of the skeleton increases and the boundary temperature slip
decreases. This is because, increase in surface emissivity reduces
the direct equivalent space thermal resistance between boundary
and skeleton, thus, promoting heat transfer and resulting in an
increase in temperature gradient and a decrease in the boundary
temperature slip.

4. Conclusion

In this work a mesh-agglomeration Monte Carlo ray-tracing

solver for radiative transfer in gray semitransparent solids is
developed in the framework of OpenFOAM. The idea of mesh-
agglomeration technique for calculating view factors between
surfaces is introduced into the calculation for radiation distri-
bution factor. By agglomerating the refined mesh elements into
Fig. 9. Verification of the present mcrtFOAM solver code for a gray medium (at coarse ones, thereby reducing the mesh elements number, the
y = z = 0.5 [m]). CPU time and the memory footprint are significantly decreased.
In addition, in the ray-tracing procedure for arbitrary polyhedral
mesh elements, mesh decomposition approach is employed. The
number) could satisfy the precision for the numerical calcula- polygon surface mesh elements (polyhedral volumetric mesh el-
tions. ements) are decomposed into several sub-triangle surface mesh
Figs. 14 and 15, respectively, show the temperature contour elements (sub-tetrahedron volumetric mesh elements). Based on
and average cross-sectional temperature along the z-axis for the the present mcrtFOAM solver, radiative heat transfer problems
three different emissivities. In each case, the results of this study within a cubic and cylindrical enclosure are calculated. Besides,
yield good agreements with the solutions obtained by the FV- a coupled radiative–conductive heat transfer problem within a
DOM. As can be seen from Fig. 15, with the increase of the tomographic porous structure is also investigated.
11
C. Fan, X.-L. Xia, J. Qiu et al. Computer Physics Communications 258 (2021) 107608

Fig. 10. The cylindrical enclosure (unit: m).

Fig. 11. Comparison of nondimensional temperature along line B–B in Fig. 10: (a) N = 0.1, (b) N = 1.0, (c) N = 10.0.

The results are all compared with previous studies and the The numerical findings of this work show that CPU time and
solution of FVDOM, and they all show great satisfaction. Convinc- the memory footprint can be greatly reduced by the proposed
ing first implementation exercises indicate that both significant mcrtFOAM solver with mesh-agglomeration technique when
memory and CPU benefits can be expected for configurations of
industrial interest. Future work will include theoretical analysis to solving the radiative transfer equation by the Monte Carlo ray-
get systematic conclusions about the application range for which tracing method. This new approach will facilitate the calculation
such benefits will effectively observed. of radiative transfer with large number of mesh elements.
12
C. Fan, X.-L. Xia, J. Qiu et al. Computer Physics Communications 258 (2021) 107608

Fig. 12. The tomographic porous structure and its boundary conditions (unit: mm).

Fig. 13. Coarse bounding box mesh elements number independent test in Fig. 12: (a) ε = 0.1, (b) ε = 0.5, (c) ε = 0.9.

13
C. Fan, X.-L. Xia, J. Qiu et al. Computer Physics Communications 258 (2021) 107608

Fig. 14. The temperature contour of skeleton with different emissivities (unit: m).

References

[1] J. Howell, R. Siegel, M. Mengüç, Thermal Radiation Heat Transfer, fifth ed.,
Hemisphere Pub. Corp., 2010.
[2] W. Fuqiang, T. Jianyu, J. Huijian, L. Yu, Energy 91 (2015) 645–654.
[3] M.M. Keshtkar, S.A. Gandjalikhan Nassab, J. Quant. Spectrosc. Radiat.
Transfer 110 (2009) 1894–1907.
[4] F. Bai, Int. J. Therm. Sci. 49 (2010) 2400–2404.
[5] H.-F. Sun, F.-X. Sun, X.-L. Xia, J. Quant. Spectrosc. Radiat. Transfer 205
(2018) 135–146.
[6] G.D. Raithby, E.H. Chui, J. Heat Transfer 112 (1990) 415.
[7] E. Chui, G. Raithby, Numer. Heat Transfer B 22 (3) (1992) 251–272.
[8] C.-H. Wang, Y.-Y. Feng, K. Yue, X.-X. Zhang, Int. Commun. Heat Mass
Transfer 108 (2019) 104287.
[9] S. Chandrasekhar, New York Dover 17 (9) (1960) 237–266.
[10] F.M. Modest, J. Heat Transfer 112 (1990) 819.
[11] J. Jeans, Mon. Not. R. Astron. Soc. 78 (1917) 28–36.
[12] Yong Huang, Guo-Dong Shi, Ke-Yong Zhu, Astrophys. J. (2016).
[13] A.V. Gusarov, J.P. Kruth, Int. J. Heat Mass Transfer 48 (2005) 3423–3434.
[14] J.R. Howell, M. Perlmutter, J. Heat Transfer 86 (1964) 116.
[15] H.F. Sun, F.X. Sun, X.L. Xia, Numer. Heat Transfer B 71 (2017) 202–215.
Fig. 15. Comparison of average cross-sectional temperature along the z-axis in [16] B.T. Wong, M.P. Mengü, J. Quant. Spectrosc. Radiat. Transfer 84 (2004)
Fig. 12. 437–450.
[17] R. Vaillon, B.T. Wong, M.P. Mengü, J. Quant. Spectrosc. Radiat. Transfer 84
(2004) 383–394.
[18] B.T. Wong, M.P. Mengü?, J. Quant. Spectrosc. Radiat. Transfer 111 (2010)
Declaration of competing interest
399–419.
[19] P.F. Hsu, J.T. Farmer, J. Heat Transfer 119 (1997).
[20] W.J. Yang, Adv. Heat Transfer 27 (1995).
The authors declare that they have no known competing finan- [21] J.R. Mahan, Radiation Heat Transfer, Wiley, New Your, 2002.
cial interests or personal relationships that could have appeared [22] L.H. Liu, X. Xu, J. Quant. Spectrosc. Radiat. Transfer 84 (2004) 349–355.
to influence the work reported in this paper. [23] X. Chen, X.L. Xia, C. Sun, Y. Li, Int. J. Heat Mass Transfer 91 (2015)
1060–1068.
[24] S. Yong, F.Q. Wang, X.L. Xia, H.P. Tan, Y.C. Liang, Int. J. Hydrogen Energy
36 (2011) 12148–12158.
Acknowledgments [25] D.N. Trivic, Int. J. Heat Mass Transfer 70 (2014) 298–312.
[26] L. Trovalet, G. Jeandel, P.J. Coelho, F. Asllanaj, J. Quant. Spectrosc. Radiat.
Transfer 112 (2011) 2661–2675.
This study was supported by the National Natural Science [27] P.D. Santos, A. Lani, Comput. Phys. Comm. 202 (2016) 233–261.
[28] C.F. Nielsen, Comput. Phys. Comm. 252 (2020) 107128.
Foundation of China (Nos. 51536001 and 51776053). In addition,
[29] S. Collange, M. Daumas, D. Defour, Comput. Phys. Comm. 178 (2008)
our sincere gratitude goes to Mr. Olayemi Dudu for his input in 135–143.
improving the English writing level of this paper. [30] W. Saftly, M. Baes, P. Camps, Astron. Astrophys. 561 (2013).

14
C. Fan, X.-L. Xia, J. Qiu et al. Computer Physics Communications 258 (2021) 107608

[31] www.openfoam.org. [36] K. Kim, E. Lee, T.H. Song, J. Quant. Spectrosc. Radiat. Transfer 109 (2008)
[32] L.M. Ruan, H.P. Tan, Y.Y. Yan, Numer. Heat Transfer 42 (2002) 253–268. 2579–2589.
[33] M.F. Modest, Radiative Heat Transfer, Academic press, 2013. [37] J.T. Farmer, J.R. Howell, Adv. Heat Transfer 31 (1998) 333–429.
[34] G. Turk, in: A.S. Glassner (Ed.), Graphics Gems, Morgan Kaufmann, San [38] A. Sakurai, T.H. Song, S. Maruyama, H.K. Kim, JSME Int. J. 48 (2005)
Diego, 1990, pp. 649–807. 259–264.
[35] S. Silvestri, R. Pecnik, A fast GPU Monte Carlo radiative heat transfer [39] C. Fan, X.-L. Li, X.-L. Xia, C. Sun, J. Quant. Spectrosc. Radiat. Transfer 225
implementation for coupling with direct numerical simulation, 2019. (2019) 110–118.

15