You are on page 1of 7

Validation of Radiation Computations using Viewfactors and

COMSOLs Hemicube Approaches


A. F. Emery1 , R. J. Cochran2 , H. Dillon1 and A. Mescher1
1
University of Washington, Seattle, WA 98195,
2
Applied CHT, Seattle 98040

Corresponding Author: emery@u.washington.edu


Abstract: The radiation viewfactor computation is frequently a major part of the computational expense when solving thermal problems. In some situations, the user may have
access either to analytic expressions or to specialized viewfactor codes that are specifically
tailored to the problem at hand. A method for
using user supplied radiation viewfactors with
COMSOL is described. The method provides
a simple and accurate way of treating radiation problems. A short discussion of the grid
convergence is provided.
Keywords: Viewfactors, Radiation Heat
Transfer, Hemicube
1 Introduction Although there are several
different approaches to computing the radiation heat transfer between opaque diffuse surfaces, one of the most common (and the one
employed by COMSOL) is based upon radiosity which is defined as the total radiative flux
leaving a surface and is denoted by J. The
method is applicable only to isothermal surfaces with constant radiative properties of emissivity, , absorbtivity, , and reflectivity, , for
which = . The condition that =  will be
satisfied if a) either i) the irradiation of the
surface is diffuse (i.e., independent of angle)
or ii) the surface is diffuse and b) either iii)
the irradiation corresponds to emission from
a black body at the surface temperature or
iv) the surface is gray (properties are independent of wavelength). Under these conditions,
the radiation from surface i with area Ai to
surface j with area Aj is given by
Qij = Ai Fij (Ji Jj )

(1a)

and
Ai i (Ti4 Ji )
(1b)
(1 i )
where Qi is the net thermal energy leaving Ai ,
Qij is the radiation from Ai to Aj and Fij is
the viewfactor, i.e., the fraction of radiation
leaving Ai that is intercepted by Aj . Figure
1a is a schematic of the dc electrical analog
circuit for radiation between three surfaces.
Qi =

T34

J3
1 3
3 A3

1
A2 F23

J2

T24

1 2

1
A1 F13

1 2 A2
A1 F12
1 1
1 A1

J1

T14
Figure 1 Radiosity Network
The problem can be solved in terms of temperatures and either irradiation or radiosity.
Siegel and Howell [1]describe several different
approaches. In terms of radiosity, the appropriate equations for N surfaces are
N
X

[kj (1 k )Fkj ]Jj = k Tk4

j=1

N
X
j=1

1km
(2a)

(kj Fkj )Jj =

Qk
Ak

m+1 k N
(2b)

where ij is the Kronecker delta and surfaces


1 i m have prescribed temperatures and
m + 1 i N have prescribed heat input.
The problem is clearly non linear because of
the T 4 terms. In addition, the formulation
depends upon the assumption that the temperature, radiosity, and radiative properties
are constant on a surface. Even when the
temperature and the radiative properties are
constant, the radiosity is rarely constant in
real engineering systems. As a consequence
of the requirement of constancy, we often find
that the radiative surfaces are subdivided into
smaller sub surfaces, often termed facets,
for which constancy is a better assumption.
As a consequence, the number of radiating
surfaces often increases to a large number.
While the solution of Eqs. 2 is not computationally prohibitive, the computation of the
viewfactors can be extremely expensive.

radiosity varies strongly with position it does


not take too many facets to give excellent results as shown in Figure 3b.
1
J(2)
J(3)
T(2)
T(3)

1.5
2
2.5

Log (% error)

3
3.5

10

Frequently the user has available either analytic expressions for the viewfactors [2] or
a separate computer code and wishes to use
these values and avoid having COMSOL compute them by using either the hemicube or the
direct area integration methods. Some viewfactor codes are specialized and optimized for
specific structures, this is often the case when
many of the surfaces have obstructed views of
the other surfaces. The question is how to input such viewfactors into COMSOL since in
its current version (3.5a) there is no mechanism for doing so. The viewfactor matrix is
often stored using a sparse matrix format, as
well as a dense matrix format. The viewfactor
matrix must be mapped to the finite element
faces of the mesh that forms the enclosure.

4
4.5
5
5.5
6

0.5

Log

10

1.5

(Nfacets)

Figure 2b: Grid Convergence


12000
A1
A2
A3

11000
10000

Radiosity

9000
8000
7000
6000
5000
4000
3000

arc length

Figure 2a: Radiative Test Problem

Figure 3a: Radiosity


9500
9000
A1
A2
A3

8500
8000
7500

avd J

2 Radiative Test Problem As an example, let us look at the problem described in


the COMSOL Heat Transfer Module Users
Guide [3] on pages 131-141 and shown in Figure 2a. The problem was solved using the radiosity formulation and dividing the surfaces
into a number of sub-surfaces, usually termed
facets. The viewfactors were evaluated using Hottels crossed-string method [1]. Figure
2b displays the convergence of the solution as
the number of facets is increased (an equal
number was used on each surface). Convergence is measured in terms of the error in the
average value of the radiosity and temperature on surfaces 2 and 3. The convergence is
slightly less than quadratic which is not surprising because of the nonlinearity associated
with the T 4 terms. Figure 3a depicts the spatial distribution of the radiosity on each surface in terms of the arc length. Although the

7000
6500
6000
5500
5000
4500

10

15

20

25

Number of Facets on each Surface

Figure 3b: Effect of the Number of Facets


on the average Radiosity

In order to compute just the radiative problem


using COMSOL we reduced the block height
from 1 to 0.1 and used an orthotropic conductivity of 1 in the direction parallel to the surface and 10000 perpendicular to the surface.
Two meshes were used, a fine one (hauto=2,
2912 quadratic Lagrangian triangles, 6898 dof)
and a coarse one (hauto=5, 182 triangles, 637
dof). Figure 4 compares the solutions. The
temperatures, radiative flux, and irradiation
matched, as shown on Figure 4a. However,
there was a disagreement for the radiosity for
surface 1, Figure 4b. The values plotted were
extracted from COMSOL running under a Matlab m file and it was characteristic of the COMSOL solution that there was an anomolous behavior at the ends of the surfaces (this will be
discussed further in section 5). Table 1 compares the results.
3000
A1
A2
A3

2000

Radiative Flux

1000

Table 1 COMSOL results


Coarse
Fine
Exact
Mesh
Mesh
Values
637
6898
dof
dof
R
ndflux
-10931
-10966
-11000
1
R
rflux
-10932
-10968
-11000
1
R
G1
29167
29255
29337
T1
300
300
300
J
4559
4572
4584
1
R
6000
6000
6000
R ndflux2
rflux
6000
6000
6000
2
R
G2
21047
21081
21159
T2
653
653
654
J
9016
9027
9053
R2
5008
4948
5000
R ndflux3
rflux
5000
5000
5000
3
R
G3
31679
31767
31876
T3
603
603
604
J3
7336
7353
7375
R
R
where ndflux and rflux are the
integrated normal conductive and radiative
and T is the average temperature.

1000

3 COMSOL Sample Problem

2000

The radiative test problem is not characteristic of general heat transfer problems in which
there is substantial conduction in the parts.
To examine the effect of conduction we solved
the sample problem according to the directions given in the manual, i.e., thick layers,
isotropic conductivity of 400 using the fine
mesh specified as specified, Figure 5a and a
coarse mesh, Figure 5b (this is the default
mesh available in the GUI with hauto=5)..

3000

4000

0.5

1.5

2.5

3.5

4.5

arc length

Figure 4a: Radiative Flux, Exact (heavy lines)


COMSOL results (light lines)
12000
A1
A2
A3

11000
10000

Radiosity

9000
8000
7000
6000
5000
4000
3000

0.5

1.5

2.5

3.5

4.5

arc length

Figure 4b: Radiosity, Exact (heavy lines)


COMSOL results (light lines)
Figure 4a Fine Mesh for
the Sample Problem

8000

7500

Irradiation

7000

6500

6000

5500

5000

4500

Figure 4b Coarse Mesh for the


the Sample Problem

Table 2 Fine and Coarse Mesh Results


Coarse Mesh Fine Mesh
270
3013
triangles
triangles
663
6730
dof
dof
R
ndflux
-10896
-10895
1
R
-10894
-10894
R rflux1
G1
29419
29244
T1
307
307
J
4588
4588
1
R
6000
6000
R ndflux2
rflux
6000
6000
2
R
G2
19390
19303
T2
645
644
J
8434
8434
2
R
4993
4993
R ndflux3
rflux
5000
5000
3
R
G3
30709
30523
T3
601
600
J3
7105
7104

Arc Length

Figure 5a Comparison of Irradiation,


fine mesh (heavy lines) and coarse mesh
(light lines)
3000

2000

Radiative Flux

Using the approach described in the COMSOL


manual and computing the radiation viewfactors by the hemicube method, we obtain the
results given in Table 2 for the fine mesh recommended in the manual, Figure 4a, and the
coarse mesh, Figure 4b. The results are listed
in Table 2 and show very good agreement between the two meshes. The only substantive
difference between the results were for the irradiation and radiative flux as shown in Figure
5.

1000

1000

2000

3000

Arc Length

Figure 5b Comparison of Radiative Flux,


fine mesh (heavy lines) and coarse mesh
(light lines)

4 Using User Defined Viewfactors The


user supplied viewfactors can be used in two
different ways, both starting with the same
procedure.
1 Identify all of the radiating surfaces, A(i), 1
im
2 Define the average temperature and area of
each surface, T R(i), AR(i)
3 Define the boundary condition for each
surface as one of a prescribed heat flux, qR(i).

Method 1
Use a routine that calculates the radiosity and
heat flux for each radiating surface using Equations 2. The computed heat fluxes are qR(i).
This is most easily done if COMSOL is run under Matlab so that the solution to Equations
2 can be programmed as a Matlab m file.
Method 2
This method is probably the easiest to implement, certainly with geometries input by hand
or through an m file. It depends upon the use
of exchange factors, usually termed script F,
F . In terms of F the radiative exchange between any two surfaces is given by
Qij = Ai Fij (Ti4 Tj4 )

Figure 6 Exchange Factor Network


The results of using one to three facets on each
surface are compared to the COMSOL results
for the coarse mesh in Figure 7. For one facet,
the results are constant on each surface and as
the number of facets increases the distribution
begins to approach that of COMSOL using the
hemicube method.

(3a)

and the flux boundary condition, qR(i) is given


by

3500
User Viewfactors, 1 Facet
User Viewfactors, 2 Facets
User Viewfactors, 3 Facets
COMSOL Hemicube

3000
2500

Aj Fji (Tj4 Ti4 )/AR(i)

(3b)

Flux

N
X

qR(i) =

j=1

2000
1500
1000

Script F is computed by (note that Ai Fij =


Aj Fji )

500
0

F = D1 M + e

(4a)

-500

where with = 1 

D=

(2)
(2) F12

..
.

..
.

(1)
(1) Fm1

(2)
(2) Fm2

...
..
.
...

Figure 7a Radiative Flux on A3

(m)
(m) F1m
..
.

1
(m)

1 F12 . . . F1m
..
..
..
..

M=
.
.
.
.
Fm1 Fm2 . . . 1

(1) 0 0
0

.. . .
..
e = ...
.
.
.
0
0 0 (m)
PN
While j=1 Fij = 1 the exchange factors satPN
isfy j=1 Fij = i

Arc Length

1
(1)

608
User Viewfactors, 1 Facet
User Viewfactors, 2 Facets
User Viewfactors, 3 Facets
COMSOL Hemicube

606

Temperature

604
602
600
598
596
594
592

Arc Length
Figure 7b Temperature on A3

Comsol Coarse Mesh

Comsol Coarse Mesh

3000

8000
A6
A9
A3

2000

7500

Irradiation

Radiative Flux

7000
1000

6500
A6
A9
A3

6000

1000
5500
2000

3000

5000

4500

arc length

arc length

a) Radiative Flux

b) Irradiation
Figure 8 Coarse Mesh

5 Identifying Radiating Surfaces We are


relatively new users of COMSOL and although
we have spent significant time in plumbing the
depths of the code, we have not been able to
find an easy automatic way to identify the radiating surfaces. Probably the best approach
is to input the geometry using a CAD program
in which the radiating surfaces have a unique
identifier. We hope that some interested users
can develop the logical steps within the COMSOL model to implement steps 1-3 in either
a Matlab m file or a COMSOL script file.
The problem is to relate the boundary number with the facet number used in the viewfactor program. Depending upon the geometry, COMSOL assigns a boundary number.
Small changes in the geometry usually result
in a change in the boundary number. Thus
the user cannot be assured that the boundary number is unique. See our paper Effect
of a Correlated Surface Roughness on Conduction through a Slab for an example of the
steps necessary for a user to be able to define a
boundary with a unique and unchanging number.
6 Concerns about COMSOLs approach
An examination of Figure 5 shows that the
COMSOL results display an anomalous behavior at the ends of the surfaces. Unfortunately we do not have access to the proprietary code and were unable to determine the
cause of this behavior. The coarse mesh results shown on Figure 8 show this effect much
clearer.

Although the COMSOL calculations give excellent results for the average and total quantities, this behavior at the ends of the surfaces
may lead to difficulties if the system being analyzed depends upon the computed fluxes as
input to other features.
To gain a better understanding of this effect
we examined the simpler system shown in Figure 9 which because of the boundary conditions should have zero radiative heat flux on
all surfaces.

Figure 9 Schematic of Isothermal Cavity

Figure 10 displays the radiative flux on the


floor of the cavity. As for the sample problem, there is a discontinuity at the ends of the
surface. The temperature and the radiosity
are correctly computed and the integrated

radiative flux is also correct, but the flux and


the irradiation at the corners is incorrect. Without access to the code it is not possible for us
to suggest any modifications that would eliminate this problem. Note that the value plotted
on these figures are the nodal values extracted
from COMSOL and are not what you would
see from the Domain plot, line extrusion postprocessing

Radiative Flux

3.0
2.5
2.0
1.5
1.0
0.5
0.0
-0.5
-0.1

0.0

0.1

0.2

0.3

0.4

0.5

0.6

Arc Length
Figure 10 Radiative Flux on the Cavity Floor

Acknowledgment These results were obtained


as part of the research supported by the
National Science Foundation through Grant
0626533.
References
1 Siegel, R. and Howell, J. R., Thermal Radiation Heat Transfer, Taylor and Francis,
1992
2 Howell, J. R., A Catlog of Radiation Heat
Transfer Configuration Factors,
http://www.me.utexas.edu/ howell/
3 COMSOL Version 3.5a, Heat Transfer
Module Users Guide,COMSOL Multiphysics,
2008

You might also like