Journal of Nuclear Science and Technology

ISSN: 0022-3131 (Print) 1881-1248 (Online) Journal homepage: https://www.tandfonline.com/loi/tnst20

Three-Dimensional Transient Heat Conduction

Analysis with Non-Linear Boundary Conditions by
Boundary Element Method

Makoto KOIZUMI, Motoaki UTAMURA & Koichi KOTANI

To cite this article: Makoto KOIZUMI, Motoaki UTAMURA & Koichi KOTANI (1985) Three-
Dimensional Transient Heat Conduction Analysis with Non-Linear Boundary Conditions by
Boundary Element Method, Journal of Nuclear Science and Technology, 22:12, 972-982, DOI:
10.1080/18811248.1985.9735753

To link to this article: https://doi.org/10.1080/18811248.1985.9735753

Published online: 15 Mar 2012.

Submit your article to this journal

Article views: 1115

View related articles

Citing articles: 1 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tnst20
journal of NUCLEAR
SCIENCE
and TECHNOLOGY,
221123, pp. 972-982 (December 1985).

Three-Dimensional Transient Heat Conduction

Analysis with Non-Linear Boundary Conditions
by Boundary Element Method
Makoto KOIZUMI, Motoaki UTAMURA and Koichi KOTANI
Energy Research Lab., Hitachi Ltd.*

Received May 15, 1985

A computational method based on boundary element method (BEM) has been developed
to analyze transient temperature distribution in a three-dimensional solid having non-linear
boundary condition (for example boiling or thermal radiation). To obtain high numerical
accuracy, the heat transfer coefficient which depends on the surface temperature was approxi-
mated a s a linear combination of interpolation functions with respect to time and analytical
time integration of the non-linear term included in boundary integral equation was made.
To investigate feasibility of the present method, it was applied to calculations of the
temperature distribution in an infinite flat plate cooled on one side. T h e heat transfer co-
efficient of cooling surface was in proportion to n-th power of the surface temperature.
Results are (1) the method has given a stable solution for non-linear boundary value prob-
lems such a s conventional BEM could not solve, (2) it has given an accurate solution for
large time increment except for first time step, and (3) an optimal relaxation factor decreases
with strength of non-linearity of the heat transfer coefficient.
Applying the method to thermal analysis of limiter and divertor plate installed in a
fusion test facility and fusion reactor respectively, availability of the method has been
confirmed.
KEY WORDS: thermal conduction, unsteady s tate , non-linear boundary condi-
tion, boundary element method, accuracy, thermal conductioity, computer cal-
culation

I. INTRODUCTION
Fusion and fission reactors are operated under high temperature and high heat load
conditions. Heat transfer on the structure surface of these equipments is dominated by
boiling, thermal radiation, or forced convection. Their heat transfer coefficients depend
on the surface temperature of the structure. Hence, the boundary conditions become non-
linear.
Formulations of steady and unsteady heat conduction equation by BEM (Boundary Ele-
ment Method) have been given by Brebbia & Walker") for linear problems. Bialecki et al.'"
investigated material non-linearity in steady heat conduction problems using Kirchhoff
transformation which proved to be effective in this type of non-linearity. However, this
transformation could not cover boundary value problems including non-linear boundary con-
ditions. Formulation of unsteady heat conduction problems with non-linear boundary con-
ditions by BEM has not yet been done. Conventionally, to analyze above problems, the
following method may be considered, that is, heat transfer coefficient is assumed to be
constant within a time increment. However, its defect is that large time increment might
result in ill numerical accuracy or numerical instability.
~ ~

* Moriyama-cho, Hitachi-shi 316.

-'24 -
Vol. 22, No. 12 (Dec. 1985) 973

On the other hand, if time increment is smaller than optimal time increment which is
determined from an element size, the calculation results by BEM might become unstable‘”
(in the case of finite element method or finite difference method, numerical accuracy might become
ill). In order to approximate the temperature profile in an element a higher order inter-
polation function is required when the time increment becomes small (generally, the order
of interpolation function in an element may be selected to be two a t the highest). TO obtain a
stable solution by conventional method, element size and time increment must be fine.
However it may require much CPU time. Hence, in this paper, to give high numerical
accuracy and a stable solution for large time increment, approximation of non-linear heat
transfer coefficient and time integration scheme are shown for 3-dimensional unsteady
heat conduction problems.
AND BEM FORMULATION
II. BASICEQUATION
1. Derivation of Boundary Integral Equation
The computational domain is shown in Fig. 1. The material properties of the solid
are assumed to be independent of the tem-
perature and uniform in the domain Q.
The unsteady heat conduction equation
follows :
dT
=uAT+ Q in 8 , (1)
at PC

where T is the temperature, t the time, u

the thermal diffusivity, Q the heat genera-
tion density, p the solid density and c the x
specific heat. The following boundaryton- Fig. 1 Computational domain
ditions are supposed
T=Tr, on rl 1
dT
-2-- =h(T-Tc) on 16
dn
where q is the surface heat flux, A the thermal conductivity, h the heat transfer coefficient
and Tc the coolant temperature. In Eq. ( 2 ) rl, and rz r, indicate first, second and third
kind of boundary condition respectively. Generally, h is dependent on solid surface tem-
perature. Hence r3
forms non-linear boundary condition.
The weighted residual form equivalent to Eq. ( 1 ) is

where dV indicates domain integral and W is the weighting function which is selected in
BEM for a 3-dimensional case as follows:

HI=-- 1
{4zu(T-t)} 3’2
{A
exp -17,--Y
4U(T--t)
12
,
-} (4)
where Y, is the location vector of set point i, Y the arbitrary location vector in Q and T
the arbitrary time. After twice integrations by parts in consideration of the boundary
- 25 -
974 J. Nucl. Sci. Technol.,

conditions in Eq. ( 2 ), Eq. ( 3 ) is transformed into boundary integral form.

where dS denotes surface integral. In Eq. ( 5 ) , the first term is boundary integral for
unknowns of temperature and heat flux on the boundary, the second term domain integral
for given heat generation, the third for initial condition, the fourth to sixth are boundary
integrals for boundary conditions respectively. Coefficient ci becomes unity when point i
is located in the domain, 1/2 for on the smooth boundary and zero for out of the domain.
2. Time Integration
We select the linear shape functions for the time increment
T = t / r T'+(l-t/r)To
q=t/r qr+(l-t/r)qO
Q=t/r Qr+(l--t/r)Qo,
where r is the time increment and superscript r and 0 indicate the value at the times r
and 0 within one time increment respectively.
Equation ( 5 ) includes integral of heat transfer coefficient. The heat transfer co-
efficient is empirically expressed by a function of the surface temperature that is non-
linear and its function form differs corresponding to heat transfer type. Therefore, it is
difficult to obtain a general integration form which satisfies any heat transfer type. T o
perform this integration, we approximate the heat transfer coefficient as a linear conbina-
tion interpolation functions with respect to time
h=xhi#:. (71
The order of the interpolation function is selected to be three. Because it is based on
the fact that heat transfer coefficient ever known is third powered polynomial of tempera-
ture at the highest and that temperature is assumed to vary linear within a time incre-
ment.
Interpolation functions q5: are deter-
mined so as to become unity, as shown in
Fig. 2, at each selected point 0, r/3, 2 r / 3
t
k
+i
;
,
\\/

\
$2
~ +;K<-.T I \\I -t
,
0 -
4'.-/+ T /-'O I T
and r within one time increment. Expres- 0 ,>3 2/3 113 2/3
sions of interpolation functions for heat Fig. 2 InterPoration function of heat transfer
transfer coefficient are as follows : coefficient with respect to time
&=( 1/2) (3t/r- 1)(3t+2)(t/r- 1)
&=(9/2) (3t/r-2)(t/r-l)(t/r)
1 (8)
$;=(9/2) (3t/r-l)(t/r-l)(t/r)
&=(l/Z) (3t/r-l)(3t/r-2)(t/r) I
Calculation procedure for the coefficient hi of Eq. ( 7 ) is illustrated in Fig. 3.
- 26 -
Vol. 22, No. 12 (Dec. 1985) 975

First, temperature on the non-linear boundary at

Given for Each
the next time step is assumed, and then coefficients Heat Transfer
are calculated using given empirical correlation of
heat transfer coefficient function on the basis of
selection point temperature. Coefficients are ex-
::\
h?

pressed as follows: hi
/ / '
hl=h(To,TcO) 0' 1/3 213 1
T

-
h 2= h { (2T0+T')/3, (2Tc0+Tc')/3) One Time Step

+
h =h {(TO+2T ') /3, (Tco 2Tc')/3) Fig. 3 Determination method of
coefficient in Eq. ( 7 )
h4=h(T', Tc')
With above preparations, Eq. ( 5 ) may be integrated in time analytically, which results in

When the heat transfer coefficient is independent of the temperature in Eq. (lo), H J
and H I become equal to hJ, and h I l , respectively.
3. Spatial Integration
To calculate the boundary integral equation (lo),solid surface is divided into boundary
elements and the internal region in the solid is also divided into volumetric elements.
An eight-node isoparametric element is used for the boundary element and a twenty-
node hexahedral element is used for the volumetric element. Final form of Eq. (10) is

- 27 -
 916 J . Nucl. Sci. Technol..

where y j and gj are shape functions for boundary and volumetric element respectively, m
is the number of element and j the number of node in an element.
Gauss-Legendre numerical integral formula is adopted for the element ir.tegration.
The integrands of Eq. (13) have l / r singularity decreasing accuracy of the numerical inte-
gration, so that, the numerical integration method used by Higashimachi ~t al.") is adopted
when the set point i belongs to the element in integration. The isoparametric element is
integrated in the polar coordinate system in which the center is placed at the set point i.
After numerical integration, the following linear equation system is obtained :

where A is a coefficient matrix of boundary tem-

perature, B the coefficient matrix of boundary
heat flux, Ac the heat transfer matrix, D the co-
efficient matrix of initial temperature in the region
and E the coefficient matrix of heat generation
in the region. Solid temperature, coolant tempera-
ture, heat flux and heat generation vectors are
denoted by T , Tc, q and Q respectively. A cal-
culation scheme of the present program is described
in F'ig. 4. Matrices A, B, D and E are evaluated
before entering time increment cycle. After heat
transfer coefficient matrix Ac is evaluated, grobal
matrices are constructed and temperature distribu-
tion is calculated. To obtain coefficients hi of lL+
Eq. ( 9 ), the temperature distribution on the sur-
Print
face at the time equal to t+r is needed because
coefficients hr depend on the surface temperature. NO

Therefore the reconstruction cycle of both heat

transfer coefficient matrix and grobal matrices are Fig. 4 Calculation scheme

-28-
Vol. 22, No. 12 (Dec. 1985) 977

repeated updating temperature distribution until solution is converged. Then calculation

proceeds until time reaches its maximum.
We adopt a kind of relaxation method within one time increment cycle. That is, the
temperature k-th iteration cycle is determined using k-1-th solution of surface temperature,
T k - - ’global
, matrix 2 and known value vector b”.
k = o(i-lb”-Tk-l), (15)
where o is relaxation factor. In the present method it is assumed to be constant. This
is because, if the relaxation factor is given by
o=z(a(~T-b)/aT}-’, (16)
Eq. (15) becomes Newton-Raphson method. It takes much CPU time to calculate inverse
Jacobian @ ( i T - b ) / a T ]--’ a t every iteration cycle.

m. EFFECT OF TIME INCREMENT

T o investigate feasibility of the present method, it was applied to one-dimensional
heat conduction problem in an infinite flat plate of which one side was cooled and the
other side insulated. The heat generation density was assumed to be uniform in the flat
plate. The heat transfer coefficient h was assumed to have following form:
h=A(T-Tc)”. (17)
Part of the infinite plate modeled a s shown in
Fig. 5. All material properties were assumed to
be unity. The exponent n of Eq. (17) was assumed
to be 2 to represent boiling heat transfer co-
efficient.
Conventionally, heat transfer coefficient for
non-linear boundary has been assumed to be con-
stant within one time increment. If non-linearity
is weak, stable numerical solution may be obtained
by conventional method. However non-linearity
becomes strong, the numerical solution by conven-
tional method becomes unstable ; thus, to avoid the
numerical instability, the time increment should be
less than a value given by stability condition. It
is difficult to obtain this numerical stability condi-
tion for conventional BEM. Therefore, this condi-
tion was obtained analogously from implicit finite
difference method which has non-linear boundary
condition. The strong and weak non-linearity is 0 Nodal point, @ Double nodal point
defined as follows: Fig. 5 Modeling of infinite flat plate

For strong non-linearity h >R/&n -1)

For weak non-linearity h<R/d(n-l),
where 6 is a half element size for second order element.
Numerical stability condition of time increment for strong non-linearity is estimate<
by the following equation :
- 29 -
978 J. Nucl. Sci. Technol.,

Two cases were selected. For one case, the numerical solution by conventional method
may become stable and for the other, it may become unstable. They are corresponding
to weak and strong non-linearity respectively. The first case was that coefficient A was
assumed to be 10 W/cmP.K3 and heat generation density Q was set 2 W/cmS and time
increment r was 0.1 s. T h e other case was that A=100 W/cm‘.K3, Q=lO W/cm2 and
r=0.05s. Calculation results by the present method were compared with that by con-
ventional method. Figure 6 depicts time histories of temperature on the cooled surface
of the plate. Solid line indicates the result by the present method and broken line that
by conventional method. T h e present method gave a smooth curve. On the other hand
conventional method resulted in more or less oscillated temperature in early stage of cal-
culation process. However, conventional method has not given large error for weak non-
linearity case.

-
--*--
I

Present Method

Conventional Method
Calculation results for Q-10 W/cm”, A=100W/
crn2.K3 and r=0.05s are shown in Fig. 7.

,+- Conventional Melhod(r=O 05 s)

Present Method(r=0.05 s)

.
-CT

” (r=0 1 S)

A (,=-O 2 I)

Time (s)
Time (8)
Fig. 6 Comparison of present method
with conventional method for Fig. 7 Comparison of present method with conven-
weak non-linearity tional method for strong non-linearity

The present method has given a smooth curve. However, the solution by the con-
ventional method become unstable. This case could not be calculated by conventional
method, because time increment could not satisfy two stability condition simultaneously.
One is determined by non-linearity and the other by element size(3). Time increment
should be less than former condition (about 0.01 for this case) and larger than latter condi-
tion (about 0.02 for this case).
Iteration cycles between one time increment in the present method found to be about
5 or 6. In the present method calculation for large time increment has been carried out
and stable result has been obtained. The case for r=O.1 s is shown in Fig. 7. The solu-
tion except first time step agreed with that for the case of r=0.05s. Same trend appears
in the case of larger time increment ( r = O . 2 ~ ) . This is because, change of actual surface
temperature rising ratio within first time increment has too large to approximate surface
temperature with linear interporation function for large time increment. As mentioned
above, in the present method time increment is not limited by non-linearity of the heat
transfer coefficient.

- 30 -
Vol. 22, No. 12 (Dec. 1985) 979

Iv. EVALUATION
OF RELAXATION
FACTOR
If relaxation factor w of Eq. (15) is nearest to the value calculated by Eq. (16), it might
become optimql. That is, optimal relaxation factor might depend on the exponent n of
heat transfer coefficient (Eq. (17)). We investigated the dependence of o on n. Figure 8
shows the dependence of the number of iteration cycles on relaxation factor o. The
exponent n was varied from 1 to 3 in this case. The minimum iteration cycle became
six and the value of relaxation factor o giving minimum number of iteration cycles was
shown to vary with the exponent n. This dependence of w on n is indicated in Fig. 9.
As shown in Fig. 9 optimal relaxation factor w decreases with increase of exponent n of
heat transfer coefficient.

3
L
* O r 0
I
0
U
C
.-
0
m
-
m
al
IT

0
0 0.5 1 .o 00 1 2 3
Relaxation Factor w Exponent n

Fig. 8 Number of iteration cycles Fig. 9 Optimal relaxation factor for

with relaxation factor exponent of heat transfer
( h= 100(T-Tc) ") coefficient (h=lOO(T-Tc)")

v. TO FUSION
APPLICATION REACTORCOMPONENTS
In order to demonstrate Configuration of Limiter
the applicability of the
method, two different ex- Ion Injection Vacuum
amples have been solved. In Chamber

one case, the boundary con- Plasma

dition is dominated by ther- Toroidal
mal radiation ( n = 3 and in Coil

the other case, by nucleate

boiling (n=2).
1. Analysis of Limiter
4
t , Insulated Surface
Inlet
We applied the present Thermal Radiation Heat Loss
(Non-Linear 6.C.)
method to thermal analysis qo
of a limiter which is installed .5cm

in a fusion test facility. A Region to Receive Heat Load

configuration of the limiter Time(ms)

is shown in Fig. 10. The Fig. 10 Application to thermal analysis of

limiter made of molybdenum limiter for fusion test facility

- 31 -
980 J. Nucl. Sci. Technol.,

is a rectangular flat plate which has round shape on one side. It is set up in a vacuum
chamber and plays a role of controlling plasma shape by direct contacting. Hence, contact
with plasma results in extremly high heat load on the limiter surface. In this phenomenon,
thermal radiation should be considered as boundary condition on the limiter surface be-
cause of its high temperature (T>500'C).
Transient temperature distribution after sudden contact with plasma was modeled as
the heat load condition as shown in Fig. 10. The number of boundary elements and nodes
were 76 and 220 respectively. Time increment was selected to be 0.05s.
Isotherms on the limiter
surface are displayed in Fig.
11. Broken lines indicate
boundary elements. Tem-
/
,
100°C
perature increases until the
heat load ceases and maxi-
mum temperature reaches
Time=
about 1,OOO"Cnear the center
\ 700C
of geometrical symmetry.
After that, thermal radiation
and diffusion lower the sur-
face temperature rapidly. In
this case, iteration cycles
required were three or four
between one time increment
calculation when relative
convergence error for itera- ' 10007:

tion was set to be lo-'. Fig. 11 Transient temperature profile on limiter surface
2. Analysis of Divertor Plate
A divertor is used for impuriy control and ash exhaust in a plasma. The divertor
plate receives high energy charged particle beams and transforms their kinetic energy
into heat. It is composed of many water-cooled copper tubes tightly arranged in a parallel
manner.
The divertor plate has been modeled as shown in Fig. 12. . Incident heat flux has
been assumed to be 500 W/cmZ and cooling conditions have been assumed as follows, that
is, coolant pressure is 1MPa, coolant velocity 1m/s and coolant temperature 60°C. The
heat transfer coefficient is given by forced convection or nucleate boiling whether the
surface temperature becomes lower or higher than the saturation temperature (180'C). T h e
heat transfer coefficient for forced convection is approximated by Dittus-Boelter's equation
(which is independent of the surface temperature) and for boiling by Nishikawa-Yamagata's
equation (which is in proportion to square of the difference between surface temperature and the
coolant saturation temperature). Number of elements and nodes have been 100 and 302 re-
spectively. Initial temperature has been set at 60°C and time increment has been selected
0.25 s.
Results at times 0.5 and 1.0s are shown in Fig. 13. AS maximum temperature of
cooling surface at time 0.5s is lower than saturation temperature of coolant, boiling has
not occurred yet. On the other hand, at time LOs, boiling has occurred in the nearest
heated surface (where is presented by hatched lines in Fig. 13).
- 32 -
Vol. 22, No. 12 (Dec. 1985) 981

Divertor Plate

Toroidai Field Coil

Configuration of Divertor plate

0 Part Model

Poroidal Cross Section of Fusion Reactor Incident Heat Flux

(500 W / d )

Fig. 12 Application to thermal analysis of divertor plate for fusion reactor

220%

Non-Linear

Time equal to 0.5 s (Forced Convection) Time equal to 1.O s (Boiling Heat Transfer)

Fig. 13 Transient temperature profile of divertor plate

VI. CONCLUDINGREMARKS
A computational method of BEM for the transient temperature of 3-dimensional struc-
tures has been developed. To obtain high numerical calculation accuracy, the heat transfer
coefficient was approximated as a linear combination of interpolation functions with respect
to time and each coefficient of the functions was determined from the surface tempera-
ture. Time integration was made analytically with respect to non-linear boundary condi-
tion term included boundary integral equation.
It is able to calculate temperature distribution in solids which have non-linear boundary
conditions such as the heat transfer coefficient can be approximated as a polinomial of the
surface temperature.
To investigate characteristics of the present method, it was applied to calculations of
the temperature distribution in the infinite flat plate of which one side was cooled and the
-3 i
 982 J. Nucl. Sci. Technol.,

other was insulated. T he heat transfer coefficient of cooling surface was in proportion to
n-th power of surface temperature. T he following results were obtained :
(1) The method has given stable solution for non-linear boundary value problem such
as conventional method that heat transfer coefficient was assumed to be constant
within one time increment could not solve.
(2) It has given accurate solution for large time increment except for first time step.
(3) The optimal relaxation factor decreases with strength of non-linearity of the heat
transfer coefficient.
Applying the method to analysis of limiter and divertor plate installed in a fusion test
facility and fusion reactor respectively, it has been confirmed availability of the method.

-REFERENCES-
(I) BREBBIA,C. A , , WALKER,S. : “Boundary Element Techniques i n Engineering”, Chap. 6 , (1980),
Butterworth.
(2) BIALECKI,R., et 0 1 . : Boundary value problems in heat conduction with non-linear material and
non-linear boundary conditions, Appl. Math. Modelling, 5, 417-421 (1981).
(3) KOIZUMI, M., et a l . : Boundary elements, Proc. 5 t h Int. Conf., p. 143-152 (1983), Spinger-Verlag.
(4) HIGASHIMACHI, T., et al.: Proc. Annu. Mtg. on Jpn. Soc. Mech. Engrs., (in Japanese), No. 830-1,
p. 7-9 (1983).

- 34 -