Вы находитесь на странице: 1из 11

International Journal of Heat and Mass Transfer 60 (2013) 653663

Contents lists available at SciVerse ScienceDirect

International Journal of Heat and Mass Transfer


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

A fast non-intrusive method for estimating spatial thermal contact conductance


by means of the reciprocity functional approach and the method
of fundamental solutions
Marcelo J. Colao a,, Carlos J.S. Alves b
a
b

Federal University of Rio de Janeiro, Department of Mechanical Engineering, PEM-COPPE-UFRJ, Cx. Postal 68503, Rio de Janeiro, RJ 21941-972, Brazil
Technical University of Lisbon, Instituto Superior Tcnico Department of Mathematics Av. Rovisco Pais, 1, 1049-001 Lisbon, Portugal

a r t i c l e

i n f o

Article history:
Received 30 April 2012
Received in revised form 15 January 2013
Accepted 15 January 2013
Available online 13 February 2013
Keywords:
Inverse problems
Reciprocity function
Contact resistance

a b s t r a c t
Thermal contact conductance is very important in many heat transfer applications, such as electronic
packaging, nuclear reactors, aerospace and biomedicine, among others. The determination of the thermal
contact resistance/conductance is a very difcult task. The objective of this paper is to present a methodology to estimate the spatial variation of this parameter without intrusive measurements. The methodology presented is formulated in terms of a reciprocity functional approach together with the method of
fundamental solutions. The solution is composed of two steps. In the rst step, two auxiliary problems,
which do not depend on the thermal conductance variation, are solved. With the results of this pre-processing, different thermal conductances can be recovered by simply performing an integral. Thus, the
methodology is extremely fast and can be used to detect aws in different materials in a short time.
2013 Elsevier Ltd. All rights reserved.

1. Introduction
Thermal contact conductance is very important in many heat
transfer applications, such as electronic packaging [20], nuclear
reactors [14], aerospace and biomedicine [17], among others. It
has been recognized for several years [13] that the increasing
power density of some electronic equipment requires cooling devices able to remove great amounts of heat. In fact, there is an
interest in producing microchannel heat sinks with heat removal
capacities of more than 1 kW/cm2 [11]. An important factor in
obtaining such heat removal is to have a low thermal contact resistance between the electronics and the cooling devices. In nuclear
reactors, resistance, which occurs in the gap between the nuclear
fuel and the metallic canning, has become a limiting factor in
exploiting reactor efcacy [14].
When two materials are in contact, only fractions of them are
really touching each other. Thus, there is a discontinuity in the
temperature across the contact interface. Thermal contact resistance
is dened as the ratio of the temperature drop to the heat ow
across the interface

Rc DT=q

Thus, lower values of Rc indicate that the difference in the temperature across the interface is low, which demonstrates a good
Corresponding author.
E-mail addresses: colaco@ufrj.br (M.J. Colao), carlos.alves@math.ist.utl.pt (C.J.S.
Alves).
0017-9310/$ - see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijheatmasstransfer.2013.01.026

contact. Thermal contact conductance, in the context of this paper,


is dened as the inverse of the resistance (h = 1/Rc). Notice that both
Rc and h can vary spatially along the interface.
Some studies [15,20,10,19] used the denition given by Eq.
(1) to calculate the global thermal contact resistance of several
materials. These studies, in general, used an experimental apparatus to measure the discontinuity in temperature and the heat
ux applied. Because the discontinuity at the exact location of
the interface is difcult to measure, they took temperature measurements at several locations and extrapolated the value
of the temperature at the interface. Wolff and Schneider [6] used
the guarded hot plate method to determine temperature discontinuities across interfaces. The disadvantage of such methods is
that they only predict global values of the thermal contact
resistance/conductance. In addition, they require complicated
experimental apparatus and/or some intrusive temperature
measurements.
Milosevic et al. [14] used a non-intrusive method, the laser ash
method, together with the Gauss method to estimate a constant
value of the thermal contact resistance between two solids. In their
paper, they were able to estimate this parameter when the sample
materials were good heat conductors or when the thickness of the
layer was relatively small. In addition, the accuracy of the estimate
increased with higher values of the contact resistance. Thus, voids
with very small values of the thermal contact resistance could not
be very well captured. Milosevic also presented other results [22]
using the laser ash method.

654

M.J. Colao, C.J.S. Alves / International Journal of Heat and Mass Transfer 60 (2013) 653663

Fieberg and Kneer [2] solved an inverse heat conduction problem to estimate the heat ux at the interface between two solids
and used temperature measurements at the interface to estimate
the thermal contact resistance. In their work, the measurements
were taken by an infrared camera pointed at the location of the
interface. Thus, they needed access to the location of the interface.
In addition, a time dependent global contact resistance with constant spatial distribution was estimated because no interior evaluation of the interface was performed.
Yang [21] also used an inverse heat conduction problem to estimate a time dependent contact resistance in single-coated optical
bers. Although good results were obtained, intrusive measurements were required. No spatial variation was considered.
Gill et al. [9] solved an inverse heat conduction problem to estimate the spatial distribution of the thermal contact resistance. The
authors mentioned that several models (as cited above) consider
the resistance constant, although it actually varies spatially. The results obtained by the authors were very sensitive to measurement
errors and required the use of a regularization technique. In addition, the temperatures were measured very close to the interface,
making the method very intrusive. However, the main contribution
was to estimate the spatial variation of the thermal contact resistance instead of using a constant value.
According to the discussion above, the determination of the
thermal contact resistance/conductance is a very difcult task.
The objective of this paper is to present a methodology to estimate
the spatial variation of this parameter without intrusive measurements. The methodology presented is formulated in terms of a reciprocity functional approach [16] together with the method of
fundamental solutions [18] to solve two auxiliary problems. The
solution is composed of two steps. In the rst step, auxiliary problems, which do not depend on the thermal conductance variation,
are solved. With the results of pre-processing, different thermal
conductances can be recovered by simply performing an integral.
Thus, the methodology is extremely fast and can be used to detect
aws in different species using a short computational time.
2. Mathematical formulation
Let us consider a generic domain X, divided in three parts
X = X1 [ C [ X2, where X1 is the rst domain, with a thermal conductivity K1, X2 is the second domain, with a thermal conductivity
K2, and C is the contact surface between them. The boundary of X1
is @ X1 = C0 [ C1 [ C, where the surface C0 is subjected to a prescribed heat ux and its temperature is measured. C1 is the lateral
surface of X1, and C is the contact surface between X1 and X2. On
the other hand, the boundary of X2 is @ X2 = C00 [ C2 [ C, where
C00 is the lower surface, C2 is the lateral surface of X2 and C is
the contact surface. Fig. 1 shows the geometry for a two-dimensional case.
The lateral surfaces C1 [ C2 are assumed to be thermally insulated while the lower surface C00 is subjected to a prescribed

temperature. The measurement surface C0 is assumed to have a prescribed heat ux q imposed on it. The contact surface C is assumed to
have a Robin boundary condition, i.e., K1@T1/@n = h(T1  T2), where
n is the normal derivative outward the boundary, K1 is the thermal
conductivity of region 1, T1 and T2 are the temperatures at the
interface of domains one and two, respectively, and h is the thermal
contact conductance, which varies from zero (for an insulated
boundary) to innity (for perfect contact). Typical values of thermal
contact conductances are given in Table 1.
The statement of the interface heat transfer problem in the steady-state case for constant conductivities K1 and K2 can be formulated as the following direct problem:

r2 T 1 0
@T 1
 K1
q
@n
@T 1
0
@n
@T 1
hT 1  T 2
 K1
@n
r2 T 2 0
@T 2
0
@n
T2 0
@T 2
@T 1
K2
K 1
@n
@n

in X1

2:a

at C0

2:b

at C1

2:c

at C

2:d

in X2

2:e

at C2

2:f

at C00

2:g

at C

2:h

The inverse problem consists of estimating the function h at the inaccessible contact surface C by using some extra temperature measurements Y at the boundary C0. To estimate it, we will dene
two auxiliary problems: the rst one to determine the temperature
jump T1  T2 at the interface C and the second one to determine the
heat ux K1@T1/@n at the same interface. According to Eq. (2.d),
the thermal contact conductance will be given as the ratio of these
two quantities. Note that if (T1  T2) is equal to zero, then we have a
perfect thermal contact between both surfaces and the denition of
thermal contact resistance does not make sense (h tends to innity).
2.1. Obtaining T1  T2 at C
Consider the rst auxiliary problem for some harmonic test functions F1 2 C2(X1) and F2 2 C2(X2):

r2 F 1 0
@F 1
K1
/
@n
@F 1
0
@n
F1 F2

in X1

3:a

at C

3:b

at C1

3:c

at C

3:d

r F2 0

in X2
@F 2
0
at C2
@n
F2 0
at C00
@F 2
@F 1
K2
K 1
/ at C
@n
@n

3:e
3:f
3:g
3:h

Table 1
Typical values of thermal contact conductances [5].

Fig. 1. Geometry for a two-dimensional case.

Contacting faces

Conductance [W/(m2C)]

Iron/ Aluminum
Cooper/ Cooper
Aluminum/ Aluminum
Stainless steel/ Stainless steel
Stainless steel/ Stainless steel (evacuated gaps)
Ceramic/ Ceramic

45,000
10,00025,000
2,20012,000
2,0003,700
2001,100
5003,000

655

M.J. Colao, C.J.S. Alves / International Journal of Heat and Mass Transfer 60 (2013) 653663

There is no boundary condition for F1 at C0. However, there are two


boundary conditions at C. For F2, there are boundary conditions for
all boundaries of X2. Thus, the problem for F1 is a Cauchy problem.
Note that / appearing in Eq. (3.b) is a generic basis function, which
will be dened later.
Let us write the following identity for the domain X1:

F 1 r2 T 1  T 1 r2 F 1 dX1

RF 1 K 1
4

X1

Using Eqs. (2.a) and (3.a), both Laplacians are zero such that Eq. (4)
vanishes. By using Greens second identity, we can, however, obtain


Z 
@T 1
@F 1
F 1 r2 T 1  T 1 r2 F 1 dX1
F1
 T1
d@ X1
@n
@n
X1
@ X1


Z
@T 1
@F 1
d@ X1
F1
0
 T1
@n
@n
C0 [C1 [C
Z



@T 1
@F 1
F1
 T1
d@ X1
@n
@n
C0 [C

Using Eq. (2.b) and the fact the some measurements Y are available
at the boundary C0, such that T1 = Y at C0, we obtain

  


Z 
q
@F 1
@T 1
@F 1
Y
F1
F1
dC0
 T1
dC
K1
@n
@n
@n
C0
C

Let us now consider another identity, for the domain X2:

F 2 r2 T 2  T 2 r2 F 2 dX2

15

Now take F1,j such that K1@F1,j/@n = /j at C [Eq. (3.b)], where (/j) is a
L2(C) orthonormal basis system. Then, taking the projection of Eq.
(15) over /j, the discontinuity T1  T2 can be written as

X
X
hT 1  T 2 ; /j iL2 C /j
RF 1;j K 1 /j
j

and using the boundary conditions on C1, Eqs. (2.c) and (3.c),



@F 1
T 1  T2; K1
@n L2 C

T 1  T 2 C

For the calculation of the reciprocity functional, no information


regarding the boundary C is needed. In addition, once the function
F1 is specied, only the conductivity K1, the imposed heat ux q and
the measured temperatures Y are needed for the calculation of
RF 1 . Using Eqs. (13) and (14), we obtain

16

To obtain the functions F1,j, we must solve a Cauchy problem with


double conditions K1@F1,j/@n = /j and F1,j = F2,j at the boundary C.
The problem has to be solved thus for several functions /j. The solution of the auxiliary problem is independent of the direct problem,
except by the geometry and the thermal conductivity K1. Thus, once
the auxiliary problem is solved and the functions F1,j are obtained,
different discontinuity congurations T1  T2 can be obtained by
simply evaluating a different integral in Eq. (14). Eq. (16) also allows
to identify situations of perfect thermal contact (T1 = T2) where we
can avoid the division in the denition of h.
Remark A. We can identify the situation where T1 = T2 everywhere
in C, by considering the analytical solution found in Section 2.4.
This case occurs when the measured temperatures (Y) in C0 are
equal to q(a/K2 + b/K1), where a is the height of the domain X2 and
b is the height of the domain X1

X2

where the Laplacians are taken from Eqs. (2.e) and (3.e) such that
Eq. (8) vanishes. By using Greens second identity, as well as Eqs.
(2.f), (2.g), (3.f) and (3.g), we can obtain

Z 
C


@T 2
@F 2
F2
 T2
dC
@n
@n

As K1 and K2 are constants, summing Eqs. (7) and (9), we obtain

  



Z
q
@F 1
@T 1
@F 1
Y
dC0 K 1 F 1
 T1
dC
K1 F1
K1
@n
@n
@n
C0
C


Z
@T 2
@F 2
dC 0
 K2 F2
 T2
@n
@n
C

10

Consider now the second auxiliary problem, for some harmonic


test functions G1 2 C2(X1):

r2 G1 0

in X1

17:a

G1 /
@G1
0
@n
@G1
0
@n

at C

17:b

at C1

17:c

at C

17:d

Following the same procedure to obtain Eq. (11), we have

  


Z 
q
@G1
@T 1
Y
K 1 G1
K 1 G1
dC0
dC
K1
@n
@n
C0
C

Z 
@G1

K1T1
dC
@n
C

or

  


Z 
q
@F 1
@T 2
@T 1
dC0
dC
Y
K1 F1
K 2 F 2
 K1F1
K1
@n
@n
@n
C0
C

Z 
@F 2
@F 1

K2T2
K1T 1
dC
@n
@n
C

Using Eqs. (2.h) and (3.d), the rst term on the right hand side is
equal to zero, so we have

  


Z 
q
@F 1
@F 2
@F 1
Y
K1 F1
K2T2
dC0
K1T 1
dC
K1
@n
@n
@n
C0
C

12

  

Z
q
@F 1
@F 1
Y
K1 F1
K1
dC0
T 1  T 2 dC
K1
@n
@n
C0
C

13

  

q
@F 1
Y
F1
dC0
K1
@n
C0

14

19

Now we can dene RG1 as the reciprocity functional in terms of


the test functions G1 as

Z
C0

Now we can dene RF 1 as the reciprocity functional, a notion


used by Andrieux and Abda [16], in terms of the test functions F1 as

  


Z 
q
@G1
@T 1
dC0
dC
Y
K 1 G1
K 1 G1
K1
@n
@n
C0
C

RG1

Using Eq. (3.h), we nally obtain

18

Using Eq. (17.d), we obtain

11

RF 1

2.2. Obtaining K1@T1/@ n at C

G1

 

q
@G1
Y
dC0
K1
@n

20

Using Eqs. (19) and (20), we obtain



@T 1
RG1 K 1  G1 ; K 1
@n L2 C

21

As before, taking G1,j such that G1,j = /j at C [Eq. (17.b)], where (/j) is
a L2(C) orthonormal basis system, the term K1@T1/@n can be written as

656

M.J. Colao, C.J.S. Alves / International Journal of Heat and Mass Transfer 60 (2013) 653663




X @T 1
X
@T 1
 K1

K1
/j
RG1;j K 1 /j
; /j
@n C
@n
L2 C
j
j

22

To nd the functions G1,j, we must solve a Cauchy problem with double


conditions G1,j = /j and @G1,j/@ n = 0 at the boundary C. Once again, the
solution of the auxiliary problem is independent of the direct problem,
except in terms of the geometry, which is the same. Thus, once the auxiliary problem is solved and the functions G1,j are obtained, different
heat uxes K1@T1/@n can be obtained by simply evaluating a different
integral in Eq. (20), which is the main advantage of this method because
the pre-processing involved in solving the auxiliary problems is performed only once. Then, different thermal contact conductances can
be recovered by simply solving different integrals.
Remark B. Note that this auxiliary problem is independently
solvable. The extension of the solution to C0 by some harmonic
functions (as for example when one uses the method of fundamental solutions that will be described in Subsection 2.5) insures
the compatibility of this solution in the whole domain.

2.3. Obtaining h at C
From the previous results, the value of h can be obtained from
Eqs. (2.d), (16) and (22) as

X
RG1;j /j
j

hX
RF 1;j /j

23

whenever the denominator is non-zero. Since we rst nd the value


of RF 1 by using Eq. (16), if it happens that (T1  T2) is equal to
zero, then we have a perfect contact between both surfaces and
Eq. (23) is not necessary (h tends to innity).

the second auxiliary problem. In both problems, the solution for


the domain X2 is straightforward and can be obtained by any traditional method, such as nite difference, integral transform, etc.
However, the solution for the domain X1 is more involved because
there is no boundary condition at C0. Instead, there are two boundary conditions at C, and a special technique is required to solve this
Cauchy problem. One possible strategy is to use the method of fundamental solutions [4,12].
The method of fundamental solutions (MFS) is a meshless technique, which is integration free, for the numerical solution of certain elliptic boundary value problems. It was rst proposed by
Krupadze and Aleksidze [18], and it is applicable if the fundamental solution of the governing equation is known. The MFS has been
used to solve different kinds of problems such as biharmonic problems, radiation-type boundary conditions, diffusive-convective
problems, free boundary problems, potential problems and elastostatic, acoustic problems [7], and steady-state heat conduction in
composite materials [8,1]. The MFS is also known as the desingularized method, the charge simulation method or the superposition
method in the mathematical and engineering literature [3].
The main idea of the MFS consists of approximating the solution
of the problem by means of a linear combination of fundamental
solutions with respect to some source points that are placed on a
ctitious boundary outside or inside the domain. Good results
are usually obtained with equally distributed collocation points
on the boundary and a similar distribution of source points on
the articial boundary.
The solutions of the auxiliary problems are then approximated
by

F 1 x

N
X
bi M i x  xi for x 2 X1

29:a

i1

G1 x

N
X

ci Mi x  xi for x 2 X1

29:b

i1

2.4. Analytical 1D solution


If we consider a constant value of thermal contact conductance
h along the horizontal direction, an analytical solution can be
found. Considering the boundary C1 thermally insulated, the temperatures in the domains X1 and X2 can be written as

T 1 y C 1 y C 2

26:a

T 2 y C 3 y C 4

26:b

where the Cs are constants. If measured temperatures Y and an imposed heat ux q are available at boundary C0, these constants can
be obtained and the value of h can be simply obtained by the following equation

q
Y  qa=K 2 b=K 1

1
ln jx  xi j for 2D
2p
1
for 3D
Mi x  xi
4pjx  xi j

Mi x  xi 

30:a
30:b

Eqs. (29) and (30) can be then used to solve the auxiliary problems
for the domain X1 with a proper choice of source points located
outside of the domain.

27
3. Results

where a is the height of the domain X2 and b is the height of the


domain X1. Note that T1 = T2 all over C if

  
q
K1
a
Y
b
K1
K2

where bi and ci are the unknown coefcients to be determined,


Mi(x) is the fundamental solution of the elliptic partial differential
equation considered and N is the number of source points. For the
Laplace equations, such as the case considered in this paper, the
fundamental solutions in two and three dimensions are given as

28

This condition allows us to identify a perfect transmission contact.


Note that if K1 = K2, the material is the same, and the condition Y =
(q/K1)(a + b) is just the compatible condition between the Dirichlet
and Neumann data. In that case, we have perfect transmission and
the solution is the same T1(y) = T2(y) = (q y)/K1.
2.5. Solution methodology
The previous sections showed how to obtain the discontinuity
T1  T2 by the rst auxiliary problem and the term K1@T1/@n by

In this paper, we considered a steady-state heat conduction


problem in a medium composed of two layers, as shown in
Fig. 1. Boundaries C1 and C2 are thermally insulated while Co is
subjected to a heat ux q. The boundary C00 is kept at a constant
temperature equal to 0oC. The domains X1 and X2 may have different thermal conductivities. The contact between the two domains
is made by the boundary C, which is assumed to have a Robin
boundary condition, i.e., K1@T1/@n = h(T1  T2), where n is the
normal derivative outward of the boundary, K1 is the thermal conductivity of region 1, T1 and T2 are the temperatures at the interface of domains one and two, respectively, and h is thermal
contact conductance, which varies from zero (for an insulated
boundary) to innity (for a perfect contact). The inverse problem
thus is to determine h by means of the reciprocity functional

657

M.J. Colao, C.J.S. Alves / International Journal of Heat and Mass Transfer 60 (2013) 653663

Besides, we used a Tikhonov regularization method to solve the


system Ax = b:

Table 2
Heat transfer coefcient proles.
Prole

h[W/(m2C)]

1
2

hmax for x < L/4 and x > 3L/4; 0 for L/4 < x< L/2
hmax for (x < L/4) and (L/2 < x< 3L/4); 0 for (L/4 < x < L/2) and
(x > 3L/4)
sin (px)
sin (2px)
hmax
hmax for (x < L/4) and (L/2 < x < 3L/4); hmax/2 for (L/4 < x < L/2); 0 for
(x > 3L/4)

3
4
5
6

kI AT Ax AT b

31

where the Tikhonov parameter k was initially chose small (k = 108)


and it was continually updated to higher value (k = 2k) if the Discrepancy principle fails (i.e. if the mean squared difference between
the calculated and the measured temperature is less than the variance of the measured data), or if the inverse of the condition number of ATA is zero in oating point representation.
In the above procedure we used the IMSL subroutines DLFIRG
and DLFCRG for solving and estimating the condition number of
the matrices, respectively. This approach was able to produce stable solutions, as we will show later.
Six different thermal contact conductance proles were used in
this paper. Table 2 shows the denitions of the proles, where L is
the length of the material and hmax is the maximum value of h. Note
that test case 5 corresponds to the one used in the analytical solution presented in Section 2.4.
The reciprocity functional approach requires the solution of the
auxiliary problems only once. Once the solution is obtained, the same
solution can be used to capture different heat transfer coefcient proles, even with different imposed heat uxes and different measured
temperatures.
The measured temperatures were obtained by solving the direct
problem with a known value of h. The solution was obtained by the
nite-difference approach. Because different methodologies were
used for the solution of the direct and inverse problems, the socalled inverse crime was avoided. A grid convergence analysis
was performed for a case where the heat transfer coefcient was
given by prole 1 in Table 2, where hmax was set to 10,000 W/
(m2 C) in a geometry with a length equal to 0.04 m and a total
height equal to 0.02 m (0.01 m for domain 1 and 0.01 m for domain 2). It was assumed that the thermal conductivity was
54 W/(m C) for both domains, and a constant heat ux of
10 W/m2 was imposed. After some tests, a grid with 120  60
points was considered converged for this problem.
To verify the stability of the solution, experimental errors were
introduced in the measured temperatures according to Eq. (32)

approach. To circumvent the lack of information on the boundary


C, temperature measurements, Y, at the boundary C0 are also
available.
As discussed before, the reciprocity functional approach requires the solution of two auxiliary problems: one for the determination of (T1  T2) and the other for the determination of K1@T1/
@n, both at the interface C. Thus, the heat transfer coefcient can
be obtained simply by dividing one quantity by the other. For the
solution of the auxiliary problem related to the boundary X2,
which is a boundary value problem, we used a nite-difference approach, where a GaussSeidel approach with SOR was used to
solve the linear system. The problem for the boundary X1 is more
complex because it is a Cauchy problem, and we used the method
of fundamental solutions (MFS), described in the previous section.
In this work, in order to alleviate the ill-conditioned character of
the inverse problem, we rst ltered the measured temperature,
by using the method of fundamental solutions. In other words,
we used the measured temperature and the imposed heat ux to
solve a Cauchy problem for the domain X1 (the boundary condition
at C is still unknown, but we have two boundary conditions at C0).
Then we used the obtained solution of this procedure to calculate
(using again the MFS) new temperatures at the boundary C0 that
are compatible with the imposed heat ux. It is important to note
that the MFS will in this way regularize the measured temperatures in order to guarantee this compatibility condition. Those
new temperatures were then used in place of the measured ones.

Table 3
Test cases analyzed.
hmax = 1  103[W/(m2C)]

Prole

hmax = 1  104[W/(m2C)]

q = 10W/m

q = 100,000W/m

L = 0.04 m

r=0
1
2
3
4
5
6

Case
Case
Case
Case
Case
Case

1a
2a
3a
4a
5a
6a

hmax = 1  103[W/(m C)]

L = 0.004 m

r = 0.001jYmaxj

r = 0.005jYmaxj

r=0

Case
Case
Case
Case
Case
Case

Case
Case
Case
Case
Case
Case

Case
Case
Case
Case
Case
Case

1b
2b
3b
4b
5b
6b

1c
2c
3c
4c
5c
6c

1d
2d
3d
4d
5d
6d

L = 0.04 m

Case
Case
Case
Case
Case
Case

1e
2e
3e
4e
5e
6e

r = 0.001jYmaxj

r = 0.001jYmaxj

Case
Case
Case
Case
Case
Case

Case
Case
Case
Case
Case
Case

1f
2f
3f
4f
5f
6f

1g
2g
3g
4g
5g
6g

Table 4
Maximum values of the measured temperatures (C).
Prole

hmax = 1  103 [W/(m2C)]


q = 10 W/m2
L = 0.04 m

hmax = 1  103 [W/(m2C)]


q = 10 W/m2
L = 0.004 m

hmax = 1  104 [W/(m2C)]


q = 10 W/m2
L = 0.04 m

hmax = 1  103 [W/(m2C)]


q = 100,000 W/m2
L = 0.04 m

1
2
3
4
5
6

0.026
0.027
0.020
0.020
0.014
0.022

0.021
0.021
0.016
0.016
0.010
0.017

0.0072
0.0083
0.0058
0.0055
0.0047
0.0075

261
271
202
197
137
225

658

M.J. Colao, C.J.S. Alves / International Journal of Heat and Mass Transfer 60 (2013) 653663

Fig. 2. Estimated K1(T1  T2) for test cases 16a.

Fig. 3. Estimated K1@T1/@n for test cases 16a.

Y TC0 er

32

where e is a random variable with a Gaussian distribution and r is


the standard deviation of the measurements. In order to generate
the Gaussian random number, with zero mean and unit variance,
we used the Box-Muller transform

q
2 lnu cos2pv

30

where u and v are two uniformly distributed random numbers (between 0 and 1) generated by the Fortran intrinsic function
random_number ().

M.J. Colao, C.J.S. Alves / International Journal of Heat and Mass Transfer 60 (2013) 653663

659

Fig. 4. Estimated h for test cases 16a.

Fig. 5. Estimated h for test cases 16b (r = 0.1 % of jYmaxj).

As discussed before, the solution of the auxiliary problems involves their solution for a set of orthonormal functions. In this
study, we chose a combination of sine and cosine-wave functions.
As we will see, this choice does not provide the best estimate for
discontinuous functions, mainly due to the Gibbs phenomenon.

The number of test functions being used were choose such


as the reciprocity functions given by Eqs. (15) and (21) reached a
stable value. Thus, the two auxiliary problems were solved
several times for different values of /j, as showed in Eqs. (3.b)
and (17.b).

660

M.J. Colao, C.J.S. Alves / International Journal of Heat and Mass Transfer 60 (2013) 653663

Fig. 6. Estimated h for test cases 16c (r = 0.5 % of jYmaxj).

Fig. 7. Estimated h for test cases 16d (smaller domain).

Different test cases were chosen to test different values of hmax,


q (the imposed heat ux), the size of the material, and the levels of
uncertainties. Table 3 shows the test cases analyzed, and Table 4
presents the maximum values of the measured temperatures (Y)
for each test case.

Initially, we will analyze the results of (T1  T2) for cases 16a.
Fig. 2 shows these values using the heat transfer coefcients for
the six proles presented in Table 1, where hmax was set to
1,000 W/(m2 C) in a geometry with a length equal to 0.04 m and
a total height equal to 0.02 m (0.01 m for domain 1 and 0.01 m

M.J. Colao, C.J.S. Alves / International Journal of Heat and Mass Transfer 60 (2013) 653663

661

Fig. 8. Estimated h for test cases 16e (larger hmax).

Fig. 9. Estimated h for test cases 16f (larger hmax and r = 0.1 % of jYmaxj).

for domain 2). In addition, the thermal conductivity was assumed


to be 54 W/(m C) for both domains, and a constant heat ux equal
to 10 W/m2 was imposed. Fig. 2 shows that the general behavior
of (T1  T2) is well captured. In addition, even for functions where h
is discontinuous, the variation of (T1  T2) is continuous.
Fig. 3 shows the estimated values for K1@T1/@ n corresponding
to the test cases analyzed in Fig. 2. Here, the estimate of K1@T1/@n

is worse than the one for (T1  T2). It can be observed however that
the oscillations are reasonably well captured, although the integration process smoothes the discontinuities.
Finally, Fig. 4 shows the estimated value of the thermal contact
conductance, which is basically a computation of the previous values of K1@T1/@n and (T1  T2). The general behavior of the function
is well captured. Although the discontinuity is not predicted by this

662

M.J. Colao, C.J.S. Alves / International Journal of Heat and Mass Transfer 60 (2013) 653663

Fig. 10. Estimated h for test cases 16g (larger heat ux and r = 0.1 % of jYmaxj)

technique, the location of the maximum and minimum values of h


is predicted well in all cases. This problem is mainly due to the
choice of the orthogonal functions (Fourier series). Thus, discontinuity functions could be better estimated by a proper choice of the
orthogonal functions in Eqs. (3.b) and (17.b). For test function
number 5, the analytical solution obtained by Eq. (27) is also
showed, where it can be seen that it perfectly match the exact value of h.
Next, we will analyze the inuence of the measurement errors
on the estimated values of the thermal conductance. Figs. 5 and
6 show the results for a standard deviation r equal to 0.1%
and 0.5% of jYmaxj, respectively, corresponding to test cases 16b
and 16c. The real function is reasonably well captured for a
standard deviation of 0.1% of jYmaxj, although the estimate is not
as good for 0.5% of jYmaxj. However, even for this value of r, results
could be used to identify the regions of maximum and minimum
values of the thermal conductance.
For the next result, we used a domain with a length of 0.004 m
and a total height of 0.002 m (0.001 m for domain 1 and 0.001 m
for domain 2), corresponding to cases 16d, to verify the sensitivity
of the method with respect to the distance between the measurements and the unknown boundary condition interface. All other
physical parameters remained constant, including the number of
grid points. For this case, we considered measurements without errors, and the results are presented in Fig. 7, corresponding to test
cases 16d. These gure shows little variance with the respect to
the previous gures presented in Fig. 4 for cases 16a, demonstrating that the estimate is not dependent on the thickness of the
material.
Fig. 8 shows the inuence of hmax on the solution, corresponding
to test cases 16e. This gure shows the results for the original domain (the same used to obtain the results up to Fig. 6) but for
hmax = 10,000 W/(m2C). These gures show that the estimate becomes worse as the discontinuity increases, which was expected.
In addition, larger values of the thermal conductance indicate an

interface with lower thermal resistance. Thus, in the limiting case


of an innite h, it is expected that the method will not be able to
identify it because it indicates a perfect continuous material.
Fig. 9 show the results for test cases 16f, which are the same as
those presented in Fig. 8 but with measurement errors with a standard deviation of 0.1% of the maximum temperature. These results
demonstrate the stability of the method with respect to the errors
in the measurements.
As presented in Table 4, the maximum value of the measured
temperature in all test cases is extremely low because the applied
heat ux is not sufcient to heat the material above 1 C. The rst
set of results then showed the capability of estimating the unknown thermal conductance prole, even for extremely small variations of the temperature. However, it is not practical because no
instrument is able to measure such low values of temperatures
with small measurement errors. Thus, Fig. 10 shows the results
for test cases 16g, where the applied heat ux was increased to
100 kW/m2, resulting in a maximum value of the measured temperature of approximately 200 C. For this case, a standard deviation for the measurement errors of 0.1% of the maximum
measured temperature (0.2 C) was assumed. This value is representative of the measurement errors found in modern infrared
cameras. The analysis of Fig. 10 shows that the estimated value
of h is very good, even in the presence of experimental errors. Thus,
this method could be used in real experiments by simply increasing the value of the applied heat ux such that the measurement
errors can t within the maximum values required for the stability
of the method (0.1% of jYmaxj in the present case).
4. Conclusions
In this study, we used an approach based on the reciprocity
functional together with the method of fundamental solutions to
estimate the spatial variation of the thermal contact conductance
between two materials. The solution is composed of two steps. In

M.J. Colao, C.J.S. Alves / International Journal of Heat and Mass Transfer 60 (2013) 653663

the rst step, the auxiliary problems, which do not depend on the
thermal conductance variation, are solved. With the results of preprocessing, different thermal conductances can be recovered by
simply performing an integral. Thus, the methodology is extremely
fast and can be used to detect aws in different samples using low
computational time. The results obtained were very good, even for
data with measurement errors. In order to improve the quality of
the results, especially for the discontinuous thermal conductances,
different orthonormal functions shall be investigated. Also, the
authors are already working on the extension of the method proposed to transient heat conduction in composite materials. Preliminary results are very good and shall be published soon.
Acknowledgements
The authors thank the Brazilian agencies, Conselho Nacional de
Desenvolvimento Cientco e Tecnolgico (CNPq), Coordenao de
vel Superior (CAPES) and Fundao
Aperfeioamento de Pessoal de N
Carlos Chagas Filho de Amparo Pesquisa do Estado do Rio de Janeiro
(FAPERJ), for fostering science and for nancial support for this
work. This work was part of an international cooperation project
between Brazil and Portugal (CAPES/FCT 305).
References
[1] A. Karageorghis, D. Lesnic, Steady-state nonlinear heat conduction in
composite materials using the method of fundamental solutions, Comput.
Methods Appl. Mech. Engrg. 197 (2008) 31223137.
[2] C. Fieberg, R. Kneer, Determination of thermal contact resistance from
transient temperature measurements, Int. J. Heat Mass Tran. 51 (2008)
10171023.
[3] C.J.S. Alves, C.S. Chen, A new method of fundamental solutions applied to
nonhomogeneous elliptic problems, Adv. Comput. Math. 23 (2005) 125142.
[4] C.J.S. Alves, M.J. Colao, V. Leito, N.F. Martins, H.R.B. Orlande, N.C. Roberty,
Recovering the source term in a linear difusion problem by the method of
fundamental solutions, Inverse Probl. Sci. Eng. 16 (2008) 10051021.
[5] Dassault Sustmes Solid Works Corporation, Solid Works User Guide, 2012.
[6] E.G. Wolff, D.A. Schneider, Prediction of thermal contact resistance between
polished surfaces, Int. J. Heat Mass Tran. 41 (1998) 34693482.

663

[7] G. Fairweather, A. Karageorghis, The method of fundamental solutions for


elliptic boundary value problems, Adv. Comput. Math. 9 (1998) 6995.
[8] J.R. Berger, A. Karageorghis, The method of fundamental solutions for heat
conduction in layered materials, Int. J. Numer. Methods Eng. 45 (1999) 1681
1694.
[9] J. Gill, E. Divo, A.J. Kassab, Estimating thermal contact resistance using
sensitivity analysis and regularization, Eng. Anal. Boundary Elem. 33 (2009)
5462.
[10] J. Liu, H. Feng, X. Luo, R. Hu, S. Liu, A simple setup to test thermal contact
resistance between interfaces of two contacted solid materiais, in: 11th
International Conference on Electronic Packaging Technology & High Density
Packaging, 2010.
[11] L. Jiang, M. Wong, Y. Zohar, Forced convection boiling in a microchannel heat
sink, J. Microelectromech. Sys. 10 (1) (2001) 8087.
[12] M.F. Valle, M.J. Colao, F. Scofano Neto, Estimation of the heat transfer
coefcient by means of the method of fundamental solutions, Inverse Probl.
Sci. Eng. 16 (2008) 777795.
[13] M.J. Edmonds, A.M. Jones, P.W. OCallaghan, S.D. Probert, The prediction and
measurement of thermal contact resistance, Wear 50 (1978) 299319.
[14] N.D. Milosevic, M. Raynaud, K.D. Maglic, Estimation of thermal contact
resistance between the materials of double-layer sample using the laser
ash method, Inverse Probl. Eng. 10 (1) (2002) 85103.
[15] P.J. Halliday, R. Parker, R.B. Piggott, A.C. Smith, D.C. Steer, Estimation of the
thermal contact resistance between potato granules and steel, J. Food Eng. 28
(1996) 261270.
[16] S. Andrieux, A.B. Abda, The reciprocity gap: a general concept for aw
identication problems, Mech. Res. Commun. 20 (5) (1993) 415420.
[17] T.H. McWaid, E. Marschall, Application of the modied greenwood and
williamson contact model for the prediction of thermal contact resistance,
Wear 152 (1992) 263277.
[18] V. Krupadze, M.A. Aleksidze, The method of functional equations for the
approximate solution of certain boundary value problems, Zh. Vychisl. Mat.
Mat. Fiz. 4 (4) (1964) 683715.
[19] X. Luo, H. Feng, J. Liu, M.L. Liu, S. Liu, An experimental investigation on thermal
contact resistance across metal contact interfaces, in: 12th International
Conference on Electronic Packaging Technology & High Density Packaging,
2011.
[20] X. Zhang, P. Cong, S. Fujiwara, M. Fujii, A new method for numerical simulation
of thermal contact resistance in cylindrical coordinates, Int. J. Heat Mass Tran.
47 (2004) 10911098.
[21] Y.-C. Yang, Estimation of thermal contact resistance and thermally induced
optical effects in single-coated optical bers, Opt. Commun. 278 (2007) 8189.
[22] N.D. Milosevic, M. Raynaud, D. Maglic, Simultaneous estimation of the thermal
contact resistance of thin solid lms and coatings using the two-dimensional
ash method, Int. J. Thermophys. 24 (3) (2003) 799819.

Вам также может понравиться