International Journal of Thermal Sciences 77 (2014) 186e198

Contents lists available at ScienceDirect

International Journal of Thermal Sciences

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

The solution of direct and inverse transient heat conduction problems

with layered materials using exponential basis functions
B. Movahedian, B. Boroomand*
Department of Civil Engineering, Isfahan University of Technology, Isfahan 84156-83111, Iran

a r t i c l e i n f o a b s t r a c t

Article history: In this paper we present an extension of the formulation given in (EABE, 37 (2013) 868e883) to direct
Received 1 June 2013 and inverse heat conduction problems in layered materials. Expressing the numerical solution as a series
Received in revised form of exponential basis functions (EBFs) defined in space and time, is the main idea of the presented
30 September 2013
method. We shall show that the use of the EBFs provides a versatile tool for the solution of a variety of
Accepted 30 October 2013
problems such as the classical heat diffusion with internal heat sources and the inverse problems in
Available online 8 December 2013
layered media with perfect and imperfect contacts. To this end, we choose a proper set of EBFs satisfying
the governing differential equation and interface conditions. Then a collocation technique is employed to
Keywords:
Transient heat conduction
satisfy both initial and boundary conditions. The proposed formulation can also be easily extended to the
Layered materials solution of non-Fourier heat conduction problems or the analysis of thermal wave propagation. The
Exponential basis functions capability of the presented method is investigated in the solution of some direct and inverse heat
Inverse problem conduction problems including a boundary identification problem and backward heat conduction
Heat diffusion problems. Issues pertaining to short time solution and penetration time, in inverse heat conduction
problems, are addressed in the problems solved.

2013 Elsevier Masson SAS. All rights reserved.

1. Introduction materials, usually involve the solution of non-standard eigenvalue

problems requiring effective solvers for non-linear algebraic or
The analysis of direct heat conduction problems (DHCP) in matrix equations (see Ref. [9] for instance). Nevertheless, such
layered materials is a major issue in various engineering applica- analytical solutions do not exist for many IHCPs. This means that
tions such as thermal protection systems [1,2], heat transport in the use of numerical techniques is inevitable at least for IHCPs,
biological tissues [3,4] and thermal shock analysis of the refractory letting alone the DHCPs. Beside the well-known numerical
materials in steel industry [5] (see also [6] and references therein methods such as the finite difference method (FDM) [5,20], or the
for other applications). Inverse heat conduction problems (IHCP) finite element method (FEM) is the method of fundamental solu-
have also a wide range of applications in the modeling of industrial tions (MFS), categorized as one of the meshless methods, which has
problems such as boundary identification in iron making [7], in- lately become attractive to many scientists. The MFS has been
dustrial furnaces [8] and atmospheric re-entry vehicles. recently used for the solution of direct transient heat conduction
Along with the analytical and semi-analytical methods, see Refs. problems [21,22], IHCP [23,24] and backward heat conduction
[2,9e14] for instance, some well-established techniques such as problems (BHCP) [25,26].
Beck’s Function Specification Method (FSM) [15] and Tikhonov In this paper we extend the meshless method introduced in
regularization are employed for accurate prediction of the tem- Refs. [27e29] to solve direct and inverse heat conduction problems
perature distribution, especially in IHCPs in homogenous [16] and with layered materials (see Refs. [30e34] for other applications).
multilayered materials [17e19]. Despite the availability of formal The governing transient differential equation is of concern in both
expressions for analytical solutions for DHCPs, especially for one- homogenous and non-homogenous forms. As the first step, we split
dimensional ones, numerical techniques are widely used to the solution into two parts, i.e. a homogeneous solution and a
obtain approximated solutions. This perhaps lies in the fact that the particular solution. The homogeneous solution is first approxi-
analytical solutions, particularly for problems with multi-layer mated by a summation of exponential basis functions (EBFs)
satisfying the differential equation and the interface conditions
including perfect or imperfect contacts. Then the constant co-
* Corresponding author. Tel.: þ98 311 391 3817; fax: þ98 311 391 2700. efficients of the solution series are obtained by satisfying the time
E-mail address: [email protected] (B. Boroomand). dependent boundary conditions as well as the initial conditions

1290-0729/$ e see front matter
2013 Elsevier Masson SAS. All rights reserved.
http://dx.doi.org/10.1016/j.ijthermalsci.2013.10.021
 B. Movahedian, B. Boroomand / International Journal of Thermal Sciences 77 (2014) 186e198 187

Nomenclature t non-dimensional time

x global spatial coordinate
cp heat capacity x non-dimensional spatial coordinate
Erfc($) complementary error function ~
x local spatial coordinate
E Young’s modulus
f prescribed initial condition Greek symbols
g prescribed boundary condition a coefficient of x in the exponent of EBF
h contact transfer coefficient aT thermal expansion coefficient
k heat conductivity b coefficient of t in the exponent of EBF
l length of layer k predefined factor relating Nb to Ncol
Ncol number of collocation points l coefficient of x in the exponent of EBF for the particular
Nl number of layers solution
NIx number of initial points along x n Poisson’s ratio
NDt number of Dirichlet boundary points along t r density
NNt number of Neumann boundary points along t s final time for numerical solution
Nb number of coefficient b Ub an area in the Gaussian plane for b
q source term
b
q approximated source term Repeated subscripts
qb imaginary part of b j EBF index integer
t time n layer index integer
T temperature

through a collocation method. The particular solution due to in-  

Tn ~xn ; t * ¼ fn ð~
xÞ; 0~
xn  l n ; n ¼ 1; .; Nl ; (3)
ternal heat sources is also determined by the use of another series
of EBFs. The method has been basically devised for problems whose
governing differential equations are of constant coefficients type, as ~*m ¼ x
and the time dependent temperature conditions at x ~*D
is the case in the current study, however the idea can be extended  
to more general problems using equilibrated bases functions EqBFs x*D ; t ¼ gD ðtÞ;
Tm ~ x*D ˛½0; lm ;
~ (4)
(see Ref. [35]).
The layout of the paper is as follows. In the next section the ~*m ¼ ~
or the time-varying heat flux at x x*N
model used for the heat conduction problems with layered mate-
rials is described. In Section 3 we present the formulation for the vTm  * 
~
xN ; t ¼ gN ðtÞ; x*N ˛½0; lm ;
~ (5)
solution of problems with and without source terms. In Section 4 vx
we present the collocation technique for the satisfaction of boun-
dary\initial conditions. In Section 5, the capability of the presented are needed. Here fn ð~ xÞ is the prescribed initial condition at a
method in the solution of some direct and inverse problem is particular time t*. Moreover gD(t) and gN(t) are the prescribed
illustrated. Finally in Section 6 we shall summarize the conclusions temperature and the heat flux at the specified points of the layers m
made throughout the paper. x*D and ~
and m, i.e. ~ x*N , respectively. Appropriate modeling of the
interface conditions is an important issue in the analysis of prob-
lems with layered materials [20]. In this regards, we consider two
2. Model problem sets of matching conditions. The first set is

Assuming that the problem is composed of Nl > 1 layers, the Tn ðln ; tÞ ¼ Tnþ1 ð0; tÞ; (6)
governing equation in the nth layer is as follows
and

 vT v2 T
rn cp n n  kn 2n ¼ qn ð~xn ; tÞ; 0~
xn  ln ; n ¼ 1; .; Nl : vTn vT
vt v~
x kn ðln ; tÞ ¼ knþ1 nþ1 ð0; tÞ; (7)
n
vx vx
(1)
In the above equation Tn represents the temperature field in the
nth layer which must be found in terms of the local spatial coor-
dinate ~xn and time t. Also rn, (cp)n and kn are the density, the heat
capacity and the heat conductivity coefficient, respectively, and qn
is a predefined source term varying with ~ xn and t. With ln indicating
the length of n th layer, the global spatial coordinate is related to ~
xn
as follows

X
n1
xn ¼ ~
xn þ lj ; 0~
x n  ln : (2)
j¼1

Schematic representation of a multi-layer material is shown in

Fig. 1. To solve the problem, a set of specified conditions including
the temperature distributions at time t* Fig. 1. Schematic presentation of the multi-layer material.
188 B. Movahedian, B. Boroomand / International Journal of Thermal Sciences 77 (2014) 186e198

which convey the continuity of the temperature and the heat flux. From the above equation the parameter an may be found in
The second set represents the roughness of the contact between terms of bn as
the layers and the contact resistance at the interfaces [20]. These
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi


jump conditions are defined as
r cp n r cp n
a1n ¼ þ n bn ; a2n ¼  n bn : (13)
vTn kn kn
kn ðln ; tÞ ¼ hn  ðTnþ1 ð0; tÞ  Tn ðln ; tÞÞ; (8)
vx
It must be noted that bn in the above relation may take on
and complex values. Therefore, the solution to (1) may be written as
Z    1   2 
vTnþ1
knþ1 ð0; tÞ ¼ hn  ðTnþ1 ð0; tÞ  Tn ðln ; tÞÞ; (9) TnH ð~
xn ; tÞ ¼ A1n a1n ; bn ean ~xn þbn t þ A2n a2n ; bn ean ~xn þbn t dUb :
vx
Ub
where h is the contact transfer coefficient of the nth layer. With (14)
each set of conditions, one can define direct and inverse heat
conduction problems (DHCPs and IHCPs). In well-posed DHCPs, the In the above relation Ub is an appropriate area or locus in the
initial condition, i.e. Tn ð~
xn ; tÞ is given at t* ¼ 0, and time dependent Gaussian plane. Also A1n ða1n ; bn Þ and A2n ða2n ; bn Þ are two sets of un-
PNl
boundary conditions at x ¼ 0 and x ¼ n ¼ 1 ln are specified. In
known coefficients which must be found so that boundary condi-
IHCPs one of these conditions is not given. There are two types of tions of the nth layer are satisfied. Instead of using the integral form
IHCPs; in the first one, the initial condition at t* ¼ 0 is unknown and given in (14) one may use a discrete form, for instance when the
thus the backward heat conduction problem (BHCP) determines integral is to be calculated numerically, and simply write
the temperature distribution from the information at the end of
time interval. The second type of IHCPs is concerning with the case Nn 
X
b

b H ð~ 1 ~ 2 ~
c1n;j ean;j xn þbn;j t þ c2n;j ean;j xn þbn;j t :
n xn ; tÞ ¼
in which the discrete measured data is known only on a part of the T (15)
boundary and therefore the solution is recovered from the auxiliary j¼1
data and the initial condition at t* ¼ 0. In practice, the measured b H is an approximation to T H , c1 and c2
In the above equation T n n n n
initial or boundary data may have a certain level of noise. This
represent a set of coefficients to be found from the boundary con-
causes adverse effects on the accuracy of the numerical solution.
Therefore, a numerical method should demonstrate enough sta- ditions, a1n and a2n are evaluated when a number of points, i.e. Nnb ,
bility, or reasonably low sensitivity to the noise, in recovering the are selected in Gaussian plane for bn,j so that
temperature distribution. bn;j ˛bn ¼ fbn;1 ; bn;2 ; ..bn;Nb g. Here we use identical sets for b as
n

b ¼ fb1 ; b2 ; ..bNb g in all layers, i.e. b1 ¼ b2 ¼ . ¼ bNl ¼ b

n

3. Exponential basis functions and N1b ¼ N2b ¼ . ¼ NN

b
l
¼ Nb . The reader may note that for each
b a total number of 2Nl coefficients as c1n and c2n are to be found from
The solution to the problem defined in (1) may be split into all interlayer conditions and the boundary conditions. Here we
homogeneous and particular parts as write 2(Nl  1) coefficients in terms of the last two coefficients by
satisfying the conditions in (6) and (7) or (8) and (9) for each b. This
Tn ¼ TnH þ TnP ; (10) is performed by a recursive formula with a generic relation as

where TnH is the solution to (1) with qn ¼ 0 and TnP is the solution
when qn s 0 disregarding the boundary conditions. The subscript E1n;j Cn;j ¼ E2nþ1;j Cnþ1;j ; n ¼ 1; .; Nl  1; (16)
“n” is used to denote the number of the layer, i.e. the nth layer.
Generally this summation may contain smooth and non-smooth with
parts. Here we focus on the smooth solutions so that they can be " #
1 2
expressed as a summation of some exponential functions. ean;j ln 1 ean;j ln 2
In this section we first consider problems with no source term, E1n;j¼ ;
kn a1n;j ean;j ln kn a2n;j ean;j ln
i.e. qn ¼ 0, and try to satisfy the initial and the boundary conditions  
to obtain a solution for the homogeneous differential equation. 1 1
E2nþ1;j ¼ k 1 2 ; (17)
Having introduced the scheme, we explain the way that one can nþ1 anþ1;j knþ1 anþ1;j
deal with problems with source terms.
for continuity conditions of the temperature and flux in (6) and (7)
and
3.1. Problems with no source term
" 1 1 2 2
#
kn a1n;j ean;j ln þ hn ean;j ln kn a2n;j ean;j ln þ hn ean;j ln
For the homogeneous part of the solution in each layer the E1n;j ¼ 1 2
;
following exponential form is considered (see also [36]), hn ean;j ln hn ean;j ln
 
hn hn
xn ; t; an ; bn Þ ¼ An ðan ; bn Þean ~xn þbn t :
TnH ð~ (11) E2nþ1;j ¼ hn  knþ1 a1nþ1;j hn  knþ1 a2nþ1;j ; (18)

Similar to our previous studies in Refs. [27e29], the aforemen-

tioned assumption enables us to convert the governing differential for jump conditions in (8) and (9) and
Equation (1) into an algebraic equation by inserting (11) in (1), with " # " #
qn ¼ 0, as follows c1n;j c1nþ1;j
Cn;j ¼ ; Cnþ1;j ¼ : (19)

 c2n;j c2nþ1;j
rn cp n bn  kn ðan Þ2 ¼ 0: (12)
From (16) one can find Cnþ1,j in terms of Cn,j as
 B. Movahedian, B. Boroomand / International Journal of Thermal Sciences 77 (2014) 186e198 189

Cnþ1;j ¼ ðE2nþ1;j Þ1 E1n;j Cn;j : (20) Dn;j ¼

1
Gn;j : (30)

2

kn lj  rn cp n bj
This leads to
In the next step, we use the relations in (10), (15) and (25) to
Cnþ1;j ¼ ðE2nþ1;j Þ1 E1n;j ðE2n;j Þ1 E1n1;j .ðE22;j Þ1 E11;j C1;j : (21) satisfy the matching conditions in (6) and (7) or (8) and (9) for
each bj as discussed in Section 3.1 (see Appendix A for further
With the above relation in hand one can rewrite (15) as details).
Nb h
X i  Remark 1. As mentioned earlier, we use just one set of bj for both
b H ð~ ~ 2 ~
n xn ; tÞ ¼ ean;j xn þbj t ean;j xn þbj t Hn;j C1;j ;
1
T (22) homogenous and particular solutions. After choosing appropriate bj
j¼1 (see Remark 2), for the homogenous solution, a1n;j and a2n;j are
calculated from relation (13). The parameter lj, for the particular
where solution must be selected so that the inequality (27) holds. In this
study, without loss of generality, we use lj ¼ bj for simplicity and
Hnþ1;j ¼ ðE2nþ1;j Þ1 E1n;j Hn;j ; n ¼ 1; .; Nl  1; (23) convenience.

and
 
1 0 4. Imposition of initial/boundary conditions
H1;j ¼ ; j ¼ 1; .; Nb : (24)
0 1
The objective here is to find the coefficients C1,j in (22) so that
the conditions (3)e(5) are satisfied. This can be performed by a
3.2. Problems with source terms collocation technique introduced in our recent researches [27e
29], for all boundary and initial conditions at Ncol points. For
As mentioned earlier, when the right hand side of (1) is non- this purpose, we consider a finite interval of time as t ˛ [0,s]
zero we split the solution as (10) where TnP is the solution when instead of t ˛ [0,N). The largeness of s may be determined by
q s 0 disregarding the boundary conditions. The idea has much in inspection, i.e. in a successive solutions one can enlarge s until
common with the dual reciprocity method (DRM) introduced in the final approximation to T(x,t) converges to a solution for
Ref. [37]. However, here we shall use EBFs for this purpose in a smaller time interval t ˛ [0,t1], t1 < s (in this paper s ¼ 1.2 t1). The
manner similar to the way we used them in Ref. [27].We construct detailed derivation of the relations and the definition of the pa-
b P by writing
T rameters used in this technique are provided in Appendix B.
n
Notes on choosing bj and s are here provided in the following two
Nb 
X  remarks.
b P ð~ d1n;j eþlj ~xn þbj t þ d2n;j elj ~xn þbj t ;
T n xn ; tÞ ¼ (25)
j¼1 Remark 2. The selection of bj is a key point to form appropriate
bases in relation (22). In Ref. [27] two strategies have been sug-
or gested one based on a projection concept and another heuristi-
cally based on numerical experiences. In this paper we use the
X
Nb
 first strategy where only the imaginary part is considered for the
b P ð~ eþlj ~xn þbj t elj ~xn þbj t Dn;j :
T n xn ; tÞ ¼ (26) bj as
j¼1

In the above relation, Nb is again the number of points in (31)

Gaussian plane for b and the pair (lj,bj) represents a set of param-
eters which are to be chosen such that they do not satisfy the where qb1 z0 and qbmax is selected based on the fluctuation of the
characteristic Equation (11) for the nth layer, i.e. boundary/initial conditions. More details about appropriately

 choosing qbmax are provided in Section 5 of [28]. In (31) we choose a
rn cp n bi  kn ðli Þ2 s0: (27) set of qbj , with equal spacing between the elements, so that the
number of the generated EBFs becomes, roughly, proportional to
It should be note that we use one set of (lj,bj) for all layers and of
the total number of the initial and boundary conditions in time and
course bj is similar to that used for the homogenous solution. The
space as
coefficients Dn,j in (26) are found by sampling the source term qn on
a grid of points in the domain [0,ln]  [0,s]. The procedure is Nb ¼ k  Ncol ; (32)
analogous to the one introduced in Ref. [27] and thus we avoid
explaining it here for the sake of brevity (see Ref. [27] for the way with k being a factor predefined by the user, and Ncol denoting the
we choose the pairs). Next we express the source term of each layer number of collocation points used in time and space. Based on
in terms of the EBFs as numerical experiments, we use 0.5  k  1.5. The increment of qbj in
(31) can be defined as below
X
Nb
1 þlj ~ 2 lj ~
xn ; tÞx b
qn ð~ q n ð~
xn ; tÞ ¼ gn;j e xn þbj t þ gn;j e xn þbj t ; (28)
2qbmax
j¼1 Dqb ¼ : (33)
Nb
or We shall present some results for the effect of changing k on the
solution in one of the numerical examples.
X
Nb

xn ; tÞx b
qn ð~ q n ð~
xn ; tÞ ¼ eþlj ~xn þbj t elj ~xn þbj t Gn;j ; (29) Remark 3. Having found a solution for a time interval as t ˛ [0,s],
i¼1 one can readily use it in a time marching algorithm, as described in
and finally, Dn,j is found by substituting (29) and (26) in (1) as Ref. [28], to solve the problem on a longer time interval.
190 B. Movahedian, B. Boroomand / International Journal of Thermal Sciences 77 (2014) 186e198

5. Numerical results the SturmeLiouville theory (see Refs. [20,41] for more details).
The exact solution is obtained using 50 terms of the series in (40).
In this section we investigate the capability of the method in the We solve the problem with 25 boundary nodes along x and the
solution of five direct and inverse heat conduction problems. The time step as Dt ¼ 0.04. 30 EBFs (Nb ¼ 30) are used by considering
verification of the numerical results is made by exact solutions (see qbj ˛f  0:1; 1; 2; .; 13; 14g in (31). This means that the final
Ref. [38] for the standards and [39,40] for the verification of the temperature solution derived from the presented method is as
exact solutions). The deviation of the numerical solution from the
exact solution is measured by defining the root mean square error
X
Nb h i
(RMSE) as b ð~
b
eqj it ea1;j ~x1
2 ~
1 x1 ; tÞ ¼ ea1;j x1 H1;j C1;j ;
1
T (41)
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi j¼1
u TP  2
u1 X
RMSE ¼ t Tk  T b
k ; (34) for the first layer, and the following one for the second layer
TP
k¼1

where TP is the number of test points for the calculation of the X

Nb h i
b ð~
b
b are the exact and approximate solutions at the
errors. Also Tk and T T 2 x2 ; tÞ ¼ eqj it ea2;j ~x2
1 2 ~
ea2;j x2 H2;j C1;j ; (42)
k
kth test point, respectively. In all examples solved, the test points j¼1
are uniformly distributed in space and time.
In order to investigate the sensitivity of the solution with where C1,j are calculated by employing the collocation method
respect to different noise levels in the measured data at the internal presented in Section 4 (see also Appendix B).
points or on the boundary, we add some randomly generated Fig. 2 depicts the variation of the difference between the
noises to them. To this end, we employ a routine for generating a set approximate and the analytical solution in (40). The results of EBFs
of random real-valued data which are applied to the exact ones as solution are also compared with those obtained by the finite dif-
ference method in Ref. [20]. As is seen, the proposed method is
~ ¼ ð1 þ e$randðqÞÞ  u;
u (35) capable of finding the solution. The small discrepancies between
the results of the EBFs and those of analytical solution at t ¼ 0.05 s
where u~ and u are the noisy and original data, respectively, e is the arise from the singularity at (x1, t) ¼ (0, 0). This effect which is seen
noise level and rand(q) is the random real number between [1, 1]. just in short time (see other results for longer times) can be
remedied by enriching the EBFs. This is clarified in the second
example.
5.1. A direct heat conduction problem with a two-layer material
In this example we further investigate the influence of grid
with and without contact resistance: sample problem 1
refinement on the solution. In this regard, the variation of RMSE
with respect to the number of collocation points in time and space
As the first sample problem, we consider a two-layer material
(see (41)) is shown Fig. 3 for the first set of interface condition and
(Nl ¼ 2) with l1 ¼ l2 ¼ 1/2 (m), (cp)1 ¼ (cp)2 ¼ 1 (J/kg  C), r1 ¼ r2 ¼ 1
different sets of EBFs. As can be seen, no significant change is
(kg/m3) and the thermal conductivities in W/m  C as k1 ¼ 1 and
k2 ¼ 0.1. The problem is solved until t ¼ 2 s. The side boundary
conditions and the initial condition of each layer are considered as
(see also [20] for a similar problem)

Tn ð~
xn ; 0Þ ¼ fn ð~
xÞ ¼ 0; n ¼ 1; 2 (36)

T1 ð0; tÞ ¼ 1; T2 ð1=2; tÞ ¼ 0: (37)

Two sets of interface conditions are assumed as (first set)

vT1 vT
T1 ð1=2; tÞ ¼ T2 ð0; tÞ; k1 ð1=2; tÞ ¼ k2 2 ð0; tÞ; (38)
vx vx

which leads to continuity of the temperature and flux, and the

following jump conditions (second set)

vT1
k1 ð1=2; tÞ ¼ 0:5  ðT2 ð0; tÞ  T1 ð1=2; tÞÞ;
vx
vT
k2 2 ð0; tÞ ¼ 0:5  ðT2 ð0; tÞ  T1 ð1=2; tÞÞ; (39)
vx

for considering the thermal contact resistance between the layers.

To validate the results we use the exact solution presented in Refs.
[20,41] as

X
N 2
Ti ðx; tÞ ¼ ui ðxÞ þ Cm elm t Xi;m ðxÞ; (40)
m¼1

in which ui(x) is the steady state solution and (lm, Xi,m(x)) are the
m th eigenvalues and eigen-functions of the eigen-problem Fig. 2. Comparison of the EBFs solution with the exact solution in Ref. [41] and the of
defined in Ref. [20]. The coefficients Cm can be calculated using finite difference method in Ref. [20] (a) with and (b) without contact resistance.
 B. Movahedian, B. Boroomand / International Journal of Thermal Sciences 77 (2014) 186e198 191

Table 1
Normalized CPU time for different numbers of collocation points with k ¼ 1.

Ncol Normalized CPU Normalized CPU time RMSE

time using B$B using exact solution

54 0.21 0.24 9.8  103

144 0.40 0.46 8.8  103
244 1.03 1.19 7.8  103
344 1.22 1.41 7.6  103
444 3.22 3.72 6.6  103

Table 2
The thickness and material properties of the three layers in sample problem 2 taken
from Ref. [5].

Fig. 3. Variation of RMSE, calculated by 240 test points, with respect to the number of
n ln (m) rn (kg/m3) (cp)n (J/kg K) kn (W/m K) aT,n (1/K)
collocation points (Ncol). 1 0.0125 4900 700 20 7  106
2 0.0125 3990 900 30 8.5  106
3 0.025 3680 700 2 6  106
observed in the error norms by altering k (note that by altering k the
number of EBFs changes according to (32)).
To give some insight to the computational cost, the time also the first example for a similar effect). A similar problem has
consumed in the proposed method including the collocation been studied in Refs. [5,22].
technique, is calculated. It is clear that the CPU time consumed in a Here again the solution starts with choosing 20 initial nodes for
computer code directly depends on the characteristics of the ma- the first and second layers and 40 initial nodes for the third layer.
chine used and the way that the algorithm is implemented (i.e. Time step, Dt is chosen as 0.25 s. We use 42 EBFs for the approxi-
optimized or non-optimized programming). Therefore it is logical mation of the solution within each layer. These bases are generated
to think of a normalized CPU time. In this paper we first normalize by considering qbj ˛f  0:001; 0:005; :01; .; 0:095; 0:1g in
all CPU times by considering a base CPU time for the product of two (31). The method is employed to find the solution until t ¼ 30 s. We
matrices as B $ B with B being a 2000  2000 square matrix with use the results given in Ref. [5] to compare the results obtained by
random elements (a similar discussion has been presented in Ref. EBFs with those obtained by FDM. In Fig. 4a, the variation of tem-
[29]). The results are shown in Table 1 for different sets of collo- perature along x at two time levels is illustrated. The results re-
cation points with k ¼ 1. It is worthwhile to mention that the ported for the finite difference method have been extracted from
analytical solution given in (40), and those generally presented for Ref. [5]. As can be seen, the solution obtained by the EBFs follows
DHCPs in Ref. [9] for instance, involves the solution of a non- that obtained by the FDM in Ref. [5]. The solution given in Ref. [42]
standard eigenvalue problem with a highly nonlinear character- using Laplace transformation for short time is used for longer time
istic equation. In order to solve such a characteristic equation for (t ¼ 30 s) and depicted in the same figure (see also the discussion in
each eigenvalue, an iterative algorithm, as the well-known Ref. [5] for the results of FEM which follow those of FDM).
NewtoneRaphson technique, is needed (letting alone the algo-
rithm needed to ensure that all the required eigenvalues, within the
desired range, are found). Obviously such a process is time
consuming and the effect become worse when the number of the
layers increases. To give more insight to the computational cost, we
have additionally normalized the CPU time of the numerical solu-
tions with the time consumed to obtain the exact solution. The
results are also included in Table 1. As is seen the CPU time of the
numerical solutions, in some cases, is less than that of the exact
solution. This means that, depending on the accuracy required, the
method presented in this paper may serve as a relatively fast al-
gorithm for obtaining a reasonable solution for DHCPs whose exact
solutions, in a formal form, are available.

5.2. Thermal shock analysis of a three-layer material: sample

problem 2

This example is chosen from Ref. [5] to validate the numerical

results of the presented method in a thermal shock analysis of a
three-layer material with perfect contacts. The thickness and ma-
terial properties of each layer are considered similar to those in
Ref. [5]. We repeat this data in Table 2.
The initial condition of all layers is zero and the side boundary
conditions are considered as below

T1 ð0; tÞ ¼ 900  C; T3 ð0:025; tÞ ¼ 0  C: (43)

In this case there is a singular point at ~
x1 ¼ 0 and therefore the Fig. 4. Comparison of EBFs solution with the solution in Ref. [42] and the FDM results
solution will be affected by singularity effect near this point (see of Ref [5] (a) temperature distribution and (b) stress field at t ¼ 30 s.
192 B. Movahedian, B. Boroomand / International Journal of Thermal Sciences 77 (2014) 186e198

The variation of the axial stress in each layer can also be b n ð~ b H ð~ bE ~

determined due to the temperature variation inside the domain. To
T xn ; tÞ ¼ T n xn ; tÞ þ T n ðxn ; tÞ; (47)
this end, we consider the following relations (from Ref. [5]) to
where TnE ð~
xn ; tÞ is defined as
compute the axial stress, sn(x,t), as

a E X
M h i 2pii
E E
sn ðx; tÞ ¼  T;n Tn ðx; tÞ þ TnE ð~ ebi t  ean;j ~xn ean;j ~xn $An;i ; bi ¼ 
1 2
Px þ Q; (44) xn ; tÞ ¼ : (48)
1n 1  n2 1  n2 i ¼ M
P

in which E and n are respectively the Young’s modulus and the In the above relation P is the duration of the Heaviside function
Poisson’s ratio of the material, aT,n is the thermal expansion coef- in time, here we consider P ¼ s, M represents the number of terms
ficient of the n th layer and the constant coefficients P and Q can be used in the Fourier transformation of the Heaviside function. The
calculated easily using the equilibrium conditions of the resulting coefficients of a1n;i and a2n;i are calculated from the characteristic
force and bending moment as follows, (see Ref. [5] for more details) equation in (12). The coefficients of An,i are defined as

0 11 0 0 0:0125
Z Z
0:025 Z
0:05
1 1
Z
0:05
B B aT;n E @ T3 ðx; tÞdxA C
B xdx 0:05 C C B 1n T1 ðx; tÞdx þ T2 ðx; tÞdx þ C
B C
Q 1  n2 B
B 0
C
C B
B 0 0:0125 0:025 C
C:
¼ B 0:05 C $ B 0 1C (45)
P E B Z Z
0:05 C B Z
0:0125 Z
0:025 Z
0:05 C
B C B C
@ x2 dx xdx A @ T;n @
a E
T1 ðx; tÞxdx þ T2 ðx; tÞxdx þ T3 ðx; tÞxdxAA
1n
0 0 0 0:0125 0:025

We consider E ¼ 400  109 Pa and n ¼ 0.2 for all layers. The Anþ1;i ¼ ðE2nþ1;j Þ1 E1n;j An;i ; A1;i ¼ 0 a1;i
T
; n ¼ 1; .; Nl  1;
obtained stress field at t ¼ 30 s, the results of the solution in Ref.
(49)
[42] and those of the FDM are shown in Fig. 4b.
As mentioned earlier, the presence of singularity at ~
x1 ¼ 0 may
where a1,i is the ith coefficient of the Fourier transformation of the
cause some discrepancies from the analytical solution in a short
Heaviside function. By adding such enrichment EBFs, the rest of the
time. In the last part of this example, we compare the results of
procedure is to modify the initial and boundary conditions (see
EBFs solution with two analytical solutions. The first one is the
Appendix B for the modification of the boundary conditions due to
analytical asymptotic solution for short times based on applying
the presence of source terms). The results of such enrichment are
inverse Laplace transform in Ref. [42] used earlier for longer time.
shown in Fig. 6 for two sets of solutions with M ¼ 50, 150. As is seen,
The second one is a simple analytical solution for times less than
enrichment of the solution with extra EBFs can properly take into
the so-called deviation time td defined in Refs. [43,44]. Such a so-
account the singularity at (x1,t) ¼ (0,0). Noteworthy is that adding
lution stems from a simplifying assumption conveying that, at short
such an enrichment solution has no effect on the results of t ¼ 30 s
times, the disturbance of the other layers on the first layer can be
shown in Fig. 4 (not shown). It may also suffice to mention that we
neglected. In that case, the temperature solution is in fact the well-
have used such a strategy in Example 5.1, see Fig. 2b, and obtained
known transient solution of a 1D semi-infinite body with the
new results matching with those of the analytical solution for small
constant prescribed temperature condition at ~ x1 ¼ 0 given in Ref.
times.
[44] as

! 5.3. An inverse heat conduction problem with a three-layer

x 0:1 ð2l1  xÞ2 material: sample problem 3
Tðx; tÞzErfc pffiffiffiffiffiffiffiffiffiffiffi ; for t  td ¼ ; (46)
2 mt n m
As the third sample problem, we investigate the capability of the
proposed method in the solution of an inverse heat conduction
where n indicates the errors less than 10n , (n ¼ 2; 3; ::: ; 10) and
problem defined by a three-layer material with l1 ¼ 0.43 mm,
m ¼ k/(r  cp) is the thermal diffusivity of the first layer. Fig. 5 de-
picts the variation of T1(0.006,t) for times less than td. It can be seen
that the results of the analytical solution in Ref. [42] and those
obtained from relation (46) coincide while the results obtained
from EBFs show some discrepancies from the other two (a similar
effect is seen in the previous example in short time). As already
mentioned, the main reason of the difference between the EBFs
solution and the analytical solution lies in the presence of a sin-
gularity at (x1,t) ¼ (0,0). In other words, there are two different
values at this points from the initial conditions, T1(0, 0) ¼ 0  C, and
the boundary condition, T1(0,0) ¼ 900  C.
As a remedy for such an effect, one may use extra EBFs to enrich
the solution. The role of enriched part of the solution is to produce a
step function in time. Such an enrichment solution appears as a
particular solution, pertaining to the boundaries, in the formula- Fig. 5. Comparison of EBFs solution with the analytical solution in Equation (46)
tion. We therefore add a function as TnE ð~xn ; tÞ to the solution (n ¼ 3) and the analytical solution of Ref [42] in short time (t  td ¼ 2.28667 s).
 B. Movahedian, B. Boroomand / International Journal of Thermal Sciences 77 (2014) 186e198 193

Fig. 6. The effect of enrichment function with EBFs in short time (t  td ¼ 2.28667 s).

l2 ¼ 9.14 mm and l3 ¼ 25.4 mm. The thermal conductivity and the

volumetric heat capacity, D ¼ r  cp, of the layers are taken as those
used in an experiment in Ref. [45] as (k1,k2,k3) ¼ (0.142,3.40,0.088)
(W/m  C), and (D1,D2,D3) ¼ (2.03,1.42,0.419)  106 (J/m3  C). The
aim here is to estimate the heat flux values at the ~ x1 ¼ 0 from the
zero flux condition at ~ x3 ¼ 25:4 and the auxiliary temperature
measurement at ~ x1 ¼ 0:43 as T1(0.43,t) ¼ g(t), where g(t) has been
extracted from the data given in Ref. [45] and illustrated in Fig. 7. In
this figure, g1(t), g2(t) and g3(t) are the fitted functions as

g1 ðtÞ ¼ 31:0356189  0:103695  t þ 0:758476

 t 2  1:048064  t 3 þ 0:524369  t 4  0:101327
 t 5 þ 0:006757  t 6 ; t  5:3155;
(50)
Fig. 7. The measured temperature variation from Ref. [45].

2
g2 ðtÞ ¼ 20:686943 þ 5:684967  t  0:902997  t
þ 0:091964  t 3  0:005350  t 4 þ 0:000163
 t 5  0:000002  t 6 ; 5:3155  t  22:5478;
(51)

g3 ðtÞ ¼ 29387:556224  6670:004477  t þ 630:618418

 t 2  31:718962  t 3 þ 0:894761  t 4  0:013418
 t 5 þ 0:000084  t 6 ; 22:5478  t  30:
(52)
Here, we solve this problem using the non-dimensional form of
Equation (1) by defining x ¼ x=L and t ¼ D1  t=L2 , where L is the Fig. 8. Estimated heat flux values from the EBFs solution, the Tikhonov regularization
total length of the material. The corresponding set of b for non- [45], the function speciation method [45], and the measured data in Ref. [45].
dimensional time variable, was also chosen as
qbj ˛f  0:2; 50; 100; .; 700g. Again we use 200 initial nodes
and Dt ¼ 0.025 to arrange the nodal points. The solution is per-
formed until t ¼ 30 s. In Fig. 8, the estimated heat flux by the
Tikhonov regularization method and the function specification
method (FSM) together with the measured data in Ref. [45], are
compared with the results of EBFs solution. As can be seen, the
presented method yields some results in the range of those pro-
duced by the other two methods.
In order to investigate the sensitivity of the solution with
respect to a perturbation of the measured data, we add some levels
of noise to the input data at ~x1 ¼ 0:43. As can be seen in Fig. 9, the
accuracy of the recovered and heat flux remains intact up to
e ¼ 0.05.
In the last part of this example, we test the capability of the
presented method in computing the so called “penetration time” Fig. 9. Estimated heat flux values from the EBFs solution with noisy data.
194 B. Movahedian, B. Boroomand / International Journal of Thermal Sciences 77 (2014) 186e198

Table 3
The penetration time computed from EBFs solution and relation (53).

x (mm) n ¼ 2 n ¼ 3

tpen EBFs solution tpen in (53)a tpen EBFs solution tpen in (53)a

x ¼ 5.00 2.73685 3.04846 2.24862 2.69897

x ¼ 9.57 4.02535 4.83700 3.15624 3.89134
x ¼ 22.27 24.3244 24.8737 16.8326 17.2491
a
Two seconds were added to the values obtained from (53) because the heat process
starts at t ¼ 2 s.

introduced in Ref. [43]. The penetration time, tpen, is defined as the

time taken, in the heating process exited from the boundary sur-
face, for the temperature or heat flux at an interior point, i.e. x, to
exceed a certain level defined as 10n  vT1 ð0Þ=vx. This parameter
was introduced in Ref. [46] for a multi-layered material as
2 32


0:1 6 6 nX
1
lj  1  dnj x  xn1 7
ffi þ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 7
tpen ðxÞz 
n 4 j¼1 


 ffi5 ;
kj = rj  cp j kn = rn  cp n
Fig. 11. Variation of moving boundary with different noise levels.
for x  xn1 ; n ¼ 1;.;Nl ; (53)

where dnj is the Kronecker symbol and x is the position of the point
of interest. In Table 3, we present the results for the penetration
Table 4
time for when the heat flux exceeds the level of
The material properties of the five-layer media in sample problem 5.
10n  7800ðW=m2 Þ at three different points (note that two sec-
onds were added to the times obtained by (57) since the heat n rn (kg/m3) (cp)n (J/kg K) kn (W/m K)
process starts at t ¼ 2 s). The table shows that the penetration 1 1 1 1
times calculated by EBFs and those form (53) are in reasonable 2 1/2 1/10 1/2
agreement, although (53) serves as a formula for approximation. 3 1 1 1
4 1/2 1/10 1/2
5 1 1 1
5.4. Moving boundary identification in a three-layer material:
sample problem 4

Here we choose an example from Ref. [24] to solve the inverse

heat conduction problem for finding the position of the moving
boundary at the right side of a three-layer material. The following
initial conditions is considered for each layer

Fig. 10. Variation of moving boundary estimated using EBFs solution and the MFS Fig. 12. Variation of (a) temperature Tb ðx; tÞ and (b) the numerical error Tb ðx; tÞ  Tðx; tÞ
solution with the Tikhonov regularization from Ref. [24]. (sample problem 5).
 B. Movahedian, B. Boroomand / International Journal of Thermal Sciences 77 (2014) 186e198 195

1 23 We solve this problem to evaluate the initial temperature

x1 ; 0Þ ¼  ~
T1 ð~ x þ ; 0~
x1  0:2; (54) distribution using 50 internal nodes and Dt ¼ 0.025 s. 125 EBFs
3 1 30 b ðx; tÞ
are selected to form the solution series. The variation of T
and its difference with the exact solution are shown in Fig. 12
1 7
x2 ; 0Þ ¼  ~
T2 ð~ x þ ; 0~
x2  0:2; (55) using four values for time. It can be observed that the numeri-
2 2 10 cal error of the recovered initial temperature is smaller than
0.0015.
3
T3 ð~
x3 ; 0Þ ¼ ~
x3 þ ; 0~
x3  sð0Þ; (56)
5
6. Conclusions
where s(t) indicates the position of the moving boundary which
should be found in such a way that T3(s(t),t) ¼ 0. Two boundary We have presented a method based on using exponential basis
conditions of Dirichlet and Neumann types exist at ~
x1 ¼ 0 as functions (EBFs) for the solution of inverse and direct transient
heat conduction problems in layered materials. The EBFs are
vT1 1 restricted to satisfy the time dependent differential equation and
ð0; tÞ ¼  ; 0  t  1; (57) the interface conditions between the adjacent layers. The initial
vx 3
and boundary conditions are satisfied using a collocation
while approach. The capability of the presented method is shown in the

8
< 0:764977 þ 0:042094  t  0:074438  t 2  10:147233  t 3 þ 13:901787  t 4 ; 0 < t  1=3;
T1 ð0; tÞ ¼ 2:812241  15:100077  t þ 36:536971  t 2  39:774638  t 3 þ 16:384512  t 4 ; 1=3 < t  2=3; (58)
:
3:894573 þ 18:700002  t  29:104415  t 2 þ 19:404098  t 3  4:817190  t 4 ; 2=3 < t  1:

The density and the heat capacity of all layers are equal to 1, solution of five sample problems. In order to show the capabilities
while (k1,k2,k3) ¼ (3,1,1) (W/m  C). With such characteristics and of the method, the issues pertaining to short time solution and
features, the exact solution of s(t) is as penetration time, in inverse heat conduction problems, have been
8 addressed in the problems solved. Promising results have been
< 3=5  t; 0 < t  1=3; obtained in all cases.
sðtÞ ¼ 2=3; 1=3 < t  2=3; (59)
:
1  0:5t; 2=3 < t  1:
Appendix A
Here, we solve this problem with qbj ˛f  0:01; 2; 4; ::::; 12g
which gives 42 EBFs. We also use 20 initial nodes and Dt ¼ 0.05 to The details of satisfying matching conditions for problems with
arrange the nodal points. The solution is performed until t ¼ 1 s. source terms, see Section 3.2, are given in this appendix. Upon
The results of EBFs solution and the MFS solution with the Tikhonov finding the coefficients Dn,j in (30), we use the relations in (10), (15)
regularization from Ref. [24] are compared in Fig. 10. As can be seen, and (25) to satisfy the matching conditions in (6) and (7) for each bj
the approximate solution and the exact solution for the moving as
boundary position are in excellent agreement.
The stability of the presented method is also investigated when " #" #  " #
some noisy data are used as the initial and boundary conditions. Fig.11
1
ean;j ln 1
2
ean;j ln 2 c1n;j eþlj ln elj ln d1n;j
þ
depicts the variation of moving boundary with different noise levels. It kn a1n;j ean;j ln kn a2n;j ean;j ln c2n;j þkn lj eþlj ln kn lj elj ln d2n;j
can be observed that the results are not sensitive to the noise level.
 " 1 #  
1 1 cnþ1;j 1 1
5.5. A backward heat conduction problem with a five-layer ¼ k 1 2 þ
nþ1 anþ1;j knþ1 anþ1;j c2 þknþ1 lj knþ1 lj
material: sample problem 5 nþ1;j

" #
In this example a BHCP with a five-layer material is considered d1nþ1;j
 ; (A-1)
as a new benchmark problem. The characteristics of each layer are d2nþ1;j
specified in Table 4. The prescribed boundary conditions at
x ¼ 0,5 (m) and the internal conditions at t* ¼ 1 are calculated from
the following exact solution

8
> e0:707107~x1 þ0:5t 0~
x1 1
>
>
>
< 7:42752  e 0:223607~
x 2 þ0:5t  5:39941  e0:223607~ x2 þ0:5t 0~
x2 1
~
Tn ðxn ; tÞ ¼ 3:56126  e0:0:707107 ~
x 3 þ0:5t þ 1:40992  e 0:707107 ~
x3 þ0:5t 0~
x3 1 (60)
>
>
>
> 24:60051  e 0:223607~x þ0:5t  16:6827  e0:223607 ~
x 4 þ0:5t 0~
x4 1
:
4

~ þ0:5t 0:707107 ~ 5 þ0:5t

12:19926  e 0:707107 x 5 þ 5:22567  e x 0~
x5 1
196 B. Movahedian, B. Boroomand / International Journal of Thermal Sciences 77 (2014) 186e198

or jump conditions in (8) and (9) as

" 1 1 2 2
#" # " #" #
kn a1n;j ean;j ln þ hn ean;j ln kn a2n;j ean;j ln þ hn ean;j ln c1n;j kn lj elj ln þ hn elj ln kn lj elj ln þ hn elj ln d1n;j
þ
hn ean;j ln
1 2
hn ean;j ln c2n;j hn elj ln hn elj ln d2n;j

 " #  " #
hn hn c1nþ1;j hn hn d1nþ1;j
¼ hn  knþ1 a1nþ1;j hn  knþ1 a2nþ1;j þ : (A-2)
c2nþ1;j hn  knþ1 lj hn þ knþ1 lj d2nþ1;j

The above relations can be expressed in a compact form as ~*D defined in the mth layer and
for the temperature condition at x

Nb h
X i 
~ 1 D ¼ E2
E1n;j Cn;j þ E ~2 * *
nþ1;j Cnþ1;j þ Enþ1;j Dnþ1;j ; n ¼ 1; .; Nl  1: a1m;j eam;j ~xN þbj t a2 eam;j ~xN þbj t Hm;j C1;j ¼ gN ðtÞ; ~x*N ˛½0; lm ;
1 2
n;j n;j
m;j
(A-3) j¼1

As before, we can rewrite Cn,j in terms of only two sets of un-

(B-3)
known coefficients, for instance, C1,j, as follows
for the heat flux condition at ~x*N defined in the m th layer. Here we
x internal points (at t*) in each layer
sample both sides of (B-1) at NI;n
Cn;j ¼ Hn;j C1;j þ Pn;j (A-4)
and both sides of (B-2) and (B-3) at ND t and N t boundary points,
N
where respectively, along t to write

 
~1 D  E ~2 X
Nb
Pnþ1;j ¼ ðE2nþ1;j Þ1 E1n;j Pn;j þ En;j n;j nþ1;j Dnþ1;j ; eIn;j Hn;j C1;j ¼ f n ; (B-4)
j¼1
n ¼ 1; .; Nl  1; (A-5)

while as the collocated values at t ¼ t* and

P1 ¼ 0: (A-6) X
Nb
eD
m;j Hm;j C1;j ¼ gD ; (B-5)
In view of (22), the final solution of each layer is then found as j¼1

Nb h
X i  as the collocated values at ~ x*D and
xm ¼ ~
b n ð~ ~ 2 ~
xn ; tÞ ¼ ean;j xn þbj t ean;j xn þbj t Hn;j C1;j
1
T
j¼1
X
Nb
Nb h
X i eN
m;j Hm;j C1;j ¼ gN ; (B-6)
a1n;j ~xn þbj t 2 ~
þ e ean;j xn þbj t Pn;j (A-7) j¼1
j¼1

X
Nb
 as the collocated values at ~
xm ¼ ~x*N . In the above relations fn, gD and
þ eþlj ~xn þbj t elj ~xn þbj t Dn;j : gN are three vectors containing the values of fn ð~xn Þ, gD(t) and gN(t) at
j¼1 x , N t and N t sampling points, respectively .The definitions of eI ,
NI;n D N n;j
eD N
m;j and em;j are as follows
Appendix B
2 3T
~ * a1n;j ð~xn ÞNx þbj t *
ean;j ðxn Þ1 þbj t
1
Here, we present the details of the collocation technique used in . e
¼ 4 5 ;
I;n
eIn;j (B-7)
this study. In this regard, firstly problems with no source term, i.e. 2 ~ * a2n;j ð~xn ÞNx þbj t *
ean;j ðxn Þ1 þbj t . e I;n
TnP ¼ 0, are considered. As the first step of the collocation tech-
nique, we substitute (22) in (3)e(5) which leads to the following 2 3T
*
sets of equations ~
*
a1m;j ~xD þbj tNt
eam;j xD þbj t1
1
. e
¼ 4 5 ;
p
eD
m;j * (B-8)
2 ~
*
a2m;j ~xD þbj tNt
N h
X b
i  eam;j xD þbj t1 . e p

~ ~ ¼ fn ð~ 0~
* *
ean;j xn þbj t ean;j xn þbj t xn Þ; xn  ln ;
1 2
Hn;j C1;j
j¼1 2 * * 3T
1 am;j xN þbj t1
1 ~ a1m;j ~xN þbj tNt
(B-1) 6 am;j  e . a1m;j  e p
7
eN
m;j ¼ 4 * * 5 : (B-9)
a2m;j ~xN þbj t1 a2m;j ~xN þbj tNt
a2m;j  e . a2m;j  e p
for the temperature distributions at time t ¼ t* and
A new arrangement of Equations (B-4)e(B-6) may be rewritten as

Xh
Nb i 
~
*
2 ~
*
x*D ˛½0; lm ;
~ X
Nb
eam;j xD þbj t eam;j xD þbj t Hm;j C1;j ¼ gD ðtÞ;
1
Vj C1;j ¼ U; (B-10)
j¼1
j¼1
(B-2)
 B. Movahedian, B. Boroomand / International Journal of Thermal Sciences 77 (2014) 186e198 197

where

  T T

T
T
T
Vj ¼ H1;j ðeI1;j ÞT . HNl ;j ðeINl ;j ÞT Hm;j ðeD
m;j Þ
T
Hm;j ðeN
m;j Þ
T
; (B-11)

References
and
[1] S.Y. Zhao, B.M. Zhang, S.Y. Du, Probabilistic modeling of transient heat transfer
and assessment of thermal reliability of fibrous insulation under aerodynamic
U ¼ ½f T1 .f TNl jgTD jgTN T : (B-12) heating conditions, Int. J. Therm. Sci. 48 (2009) 1302e1310.
[2] M. Ferraiuolo, O. Manca, Heat transfer in a multi-layered thermal protection
The coefficients C1,j may now be found as (see also [47e49]) system under aerodynamic heating, Int. J. Therm. Sci. 53 (2012) 56e70.
[3] K.C. Liu, Y.N. Wang, Y.S. Chen, Investigation on the bio-heat transfer with the
dual-phase-lag effect, Int. J. Therm. Sci. 58 (2012) 29e35.
C1;j ¼ VTj RU: (B-13) [4] K.C. Liu, Thermal propagation analysis for living tissue with surface heating,
Int. J. Therm. Sci. 47 (2008) 507e513.
P l
where R is a (Ncol)2 projection matrix, with Ncol ¼ ð N x
n ¼ 1 NI;n Þþ
[5] J. Hein, J. Storm, M. Kuna, Numerical thermal shock analysis of functionally
t t graded and layered materials, Int. J. Therm. Sci. 60 (2012) 41e51.
ND þ NN , which can be evaluated (see Refs. [27e29] for more de-
[6] C. An, J. Su, Improved lumped models for transient combined convective and
tails) by inserting relation (B-13) in (B-10) as radiative cooling of multi-layer composite slabs, Appl. Therm. Eng. 31 (2011)
2508e2517.
2 3þ [7] T.P. Fredman, A boundary identification method for an inverse heat conduc-
X
Nb
tion problem with an application in ironmaking, Heat Mass Transf. 41 (2004)
R ¼ 4 Vj VTj 5 : (B-14) 95e103.
j¼1 [8] M.A. Marois, M. Désilets, M. Lacroix, What is the most suitable fixed grid
solidification method for handling time-varying inverse Stefan problems in
In the above equation [$]þ denotes the pseudo-inverse of the high temperature industrial furnaces? Int. J. Heat Mass Transf. 55 (2012)
matrix (see Refs. [27e29]). With the coefficients C1j in hand, one 5471e5478.
[9] M.N. Özişik, Heat Conduction, second ed., John Wiley & Sons, Inc., NY, 1993.
may find the final homogeneous solution as [10] F. de Monte, An analytic approach to the unsteady heat conduction processes
0 1 in one-dimensional composite media, Int. J. Heat Mass Transf. 45 (6) (2002)
Nb h
X i 1333e1343.
b H ð~ @ a1n;j ~xn þbj t
ean;j xn þbj t Hn;j Vj ARU:
2 ~ T
n xn ; tÞ ¼
T (B-15) [11] F. de Monte, Multi-layer transient heat conduction using transition time
e scales, Int. J. Therm. Sci. 45 (2006) 882e892.
j¼1
[12] G. Gonzalez de la Cruz, Yu.G. Gurevich, Heat transfer in two-layered systems
excited by a pulsed laser, Int. J. Therm. Sci. 51 (2012) 1e6.
The proposed collocation technique can be easily employed for [13] M. Tian, S. Zhu, Q. Chen, N. Pan, Effects of layer stacking sequence on tem-
the problems with source term. Since the particular solution has a perature response of multi-layer composite materials under dynamic condi-
projection on the boundaries, the boundary and initial conditions of tions, Appl. Therm. Eng. 33-34 (2012) 219e226.
[14] S.M. Becker, H. Herwig, One dimensional transient heat conduction in
the problem are satisfied by defining U instead of U in (B-12) as segmented fin-like geometries with distinct discrete peripheral convection,
Int. J. Therm. Sci. 71 (2013) 148e162.
h T T
iT [15] J.V. Beck, B. Blackwell, S.R. St. Clair Jr., Inverse Heat Conduction; Ill-posed
U ¼ f 1 .f Nl gTD gTN (B-16) Problems, John Wiley and Sons, New York, 1985.
[16] Z.C. Feng, J.K. Chen, Y. Zhang, Real-time solution of heat conduction in a finite
slab for inverse analysis, Int. J. Therm. Sci. 49 (2010) 762e768.
where f n , gD and gN are there vectors containing the values of [17] J.C. Liu, T. Wei, The method of lines to reconstruct a moving boundary for a
f n ð~ x , N t and N t sampling points, respec-
xn Þ, g D ðtÞ and g N ðtÞ at NI;n D N one-dimensional heat equation in a multilayer domain, J. Eng. Math. 71
tively. These modified functions are defined as follows (2011) 157e170.
[18] J. Pailhes, C. Pradere, J.L. Battaglia, J. Toutain, A. Kusiak, A.W. Aregba,
J.C. Batsale, Thermal quadrupole method with internal heat sources, Int. J.
Nb h
X i Therm. Sci. 53 (2012) 49e55.
f n ð~
xn Þ ¼ fn ð~ ~ ~
* *
xn Þ  ean;j xn þbj t ean;j xn þbj t
1 2
Pn;j [19] C.S. Liu, Solving two typical inverse Stefan problems by using the Lie-group
j¼1 shooting method, Int. J. Heat Mass Transf. 54 (2011) 1941e1949.
(B-17) [20] R.I. Hickson, S.I. Barry, G.N. Mercer, H.S. Sidhu, Finite difference schemes for
N b h
X i  multilayer diffusion, Math. Comput. Model. 54 (2011) 210e220.
eþlj ~xn þbj t elj ~xn þbj t
* *
 Dn;j ; [21] B.T. Johansson, D. Lesnic, A method of fundamental solutions for transient
j¼1 heat conduction, Eng. Anal. Bound. Elem. 32 (2008) 697e703.
[22] B.T. Johansson, D. Lesnic, A method of fundamental solutions for transient
heat conduction in layered materials, Eng. Anal. Bound. Elem. 33 (2009)
Nb h
X i 1362e1367.
* 2 ~*
g D ðtÞ ¼ gD ðtÞ  ~
eam;j xD þbj t eam;j xD þbj t Pm;j
1 [23] S. Chantasiriwan, B.T. Johansson, D. Lesnic, The method of fundamental so-
lutions for free surface Stefan problems, Eng. Anal. Bound. Elem. 33 (2009)
j¼1
(B-18) 529e538.
N b h
X i  [24] T. Wei, Y.S. Li, An inverse boundary problem for one-dimensional heat
* *
 eþlj ~xD þbj t elj ~xD þbj t Dm;j ; equation with a multilayer domain, Eng. Anal. Bound. Elem. 33 (2009) 225e
232.
j¼1
[25] N.S. Mera, The method of fundamental solutions for the backward heat con-
duction problem, Inverse Prob. Sci. Eng. 13 (2005) 65e78.
and [26] B.T. Johansson, D. Lesnic, T. Reeve, A comparative study on applying the
method of fundamental solutions to the backward heat conduction problem,
Nb h Math. Comput. Model. 54 (2011) 403e416.
X * * i
~ ~ [27] B. Boroomand, S. Soghrati, B. Movahedian, Exponential basis functions in
a1m;j eam;j xN þbj t a2m;j eam;j xN þbj t Pm;j
1 2
g N ðtÞ ¼ gN ðtÞ  solution of static and time harmonic elastic problems in a meshless style, Int. J.
j¼1 Numer. Methods Eng. 81 (2010) 971e1018.
(B-19) [28] B. Movahedian, B. Boroomand, S. Soghrati, A Trefftz method in space and time
Nb h
X i 
* * using exponential basis functions: application to direct and inverse heat
 lj eþl j~
xN þ j t b lj el j~
xN þ j tb Dm;j : conduction problems, Eng. Anal. Bound. Elem. 37 (2013) 868e883.
j¼1
198 B. Movahedian, B. Boroomand / International Journal of Thermal Sciences 77 (2014) 186e198

[29] S.H. Hashemi, B. Boroomand, B. Movahedian, Exponential basis functions in [39] J.V. Beck, R. McMasters, K.J. Dowding, D.E. Amos, Intrinsic verification methods
space and time: a meshless method for 2D time dependent problems, in linear heat conduction, Int. J. Heat Mass Transf. 49 (2006) 2984e2994.
J. Comput. Phys. 241 (2013) 526e545. [40] K.D. Cole, J.V. Beck, A. Haji-Sheikh, B. Litkouhi, Heat Conduction Using Green’s
[30] M. Shahbazi, B. Boroomand, S. Soghrati, A mesh-free method using expo- Functions, second ed., CRC Press, Taylor & Francis, 2011.
nential basis functions for laminates modeled by CLPT, FSDT and TSDT e part [41] R.I. Hickson, S.I. Barry, G.N. Mercer, Critical times in multilayer diffusion. Part
I: formulation, Compos. Struct. 93 (2011) 3112e3119. 1: exact solutions, Int. J. Heat Mass Transf. 52 (2009) 5776e5783.
[31] M. Shahbazi, B. Boroomand, S. Soghrati, A mesh-free method using [42] Z.H. Jin, G.H. Paulino, Transient thermal stress analysis of an edge crack in a
exponential basis functions for laminates modeled by CLPT, FSDT and functionally graded material, Int. J. Fract. 107 (2001) 73e98.
TSDT e part II: implementation and results, Compos. Struct. 94 (2011) [43] F. de Monte, J.V. Beck, D.E. Amos, Diffusion of thermal disturbances in two-
84e91. dimensional Cartesian transient heat conduction, Int. J. Heat Mass Transf. 51
[32] B. Shamsaei, B. Boroomand, Exponential basis functions in solution of lami- (2008) 5931e5941.
nated structures, Compos. Struct. 93 (2011) 2010e2019. [44] F. de Monte, J.V. Beck, D.E. Amos, Solving two-dimensional Cartesian unsteady
[33] S.M. Zandi, B. Boroomand, S. Soghrati, Exponential basis functions in solution heat conduction problems for small values of the time, Int. J. Therm. Sci. 60
of incompressible fluid problems with moving free surfaces, J. Comput. Phys. (2012) 106e113.
231 (2012) 505e527. [45] J.V. Beck, B. Blackwell, A. Haji-Sheikh, Comparison of some inverse heat conduction
[34] S.M. Zandi, B. Boroomand, S. Soghrati, Exponential basis functions in solution methods using experimental data, Int. J. Heat Mass Transf. 39 (1996) 3649e3657.
of problems with fully incompressible materials: a meshfree method, [46] G. Pontrelli, F. de Monte, A multi-layer porous wall model for coronary drug-
J. Comput. Phys. 231 (2012) 7255e7273. eluting stents, Int. J. Heat Mass Transf. 53 (2010) 3629e3637.
[35] B. Boroomand, N. Noormohammadi, Weakly equilibrated basis functions for [47] B. Boroomand, F. Mossaiby, Generalization of robustness test procedure for
elasticity problems, Eng. Anal. Bound. Elem. 37 (2013) 1712e1727. error estimators. Part I: formulation for patches near kinked boundaries, Int. J.
[36] F.B. Hildebrand, Advanced Calculus for Applications, second ed., Prentice-Hall, Numer. Methods Eng. 64 (2005) 427e460.
NJ, 1976. [48] B. Boroomand, F. Mossaiby, Generalization of robustness test procedure for
[37] P.W. Partridge, C.A. Brebbia, L.C. Wrobel, The Dual Reciprocity Boundary error estimators. Part II: test results for error estimators using SPR and REP,
Element Method Southampton: Computational Mechanics, 1992. Int. J. Numer. Methods Eng. 64 (2005) 461e502.
[38] ASME, Standard for Verification and Validation in Computational Fluid Dy- [49] B. Boroomand, F. Mossaiby, Dynamic solution of unbounded domains using
namics and Heat Transfer, ASME V&V 20e2009, The American Society of finite element method: discrete Green’s functions in frequency domain, Int. J.
Mechanical Engineers, 2009. Numer. Methods Eng. 67 (2006) 1491e1530.