Anda di halaman 1dari 6

Finite Elements in Analysis and Design 50 (2012) 201206

Contents lists available at SciVerse ScienceDirect

Finite Elements in Analysis and Design


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

A nite element/nite difference scheme for the non-classical heat conduction and associated thermal stresses
B.L. Wang n, J.C. Han, Y.G. Sun
Graduate School at Shenzhen, Harbin Institute of Technology, Harbin 150001, PR China

a r t i c l e i n f o
Article history: Received 15 December 2010 Received in revised form 25 May 2011 Accepted 23 September 2011 Available online 19 October 2011 Keywords: Non-classical heat conduction Finite element method Finite difference method Micro/nanoscale

abstract
This paper develops a nite element code based on the hyperbolic heat conduction equation including the non-Fourier effect in heat conduction. The nite element space discretization is used to obtain a system of differential equations for time. The time-related responses are obtained by solving the system of differential equations via nite difference technique. A relationship for the time step length and the element size was obtained to ensure that numerical oscillation in temperature be suppressed. Temperature-dependent material properties are taken into account in the proposed analysis model. In addition to the temperature eld, the thermal stresses are also obtained from the developed method. The thermal stresses associated with the non-classical heat conduction are found to be considerably difference from those associated with classical heat conduction. & 2011 Elsevier B.V. All rights reserved.

1. Introduction Material and structural miniaturizations of modern devices requires a better understanding of heat transfer phenomena at small dimensions. Devices having dimensions of order of microns are being developed for use of cooling of integrated circuits, and in biochemicals and biomedical applications, and cryogenics. Microelectromechanical systems and nanoelectromechanical systems have essential impact in modern technologies and industries. The activation of micro/nanoscale effects are described as the physical scale reduces and/or transient time shortens. Heat transfer at micro/nanoscale may have very different physical bases from that in macroscopic scale. The classical Fourier heat conduction law gives an innite speed of heat propagation. This implies that the body will be affected by the boundary condition or initial condition at the instant of heating. However, the speed of heat propagation at a micro/nanoscale body cannot be innite. For example, for metal at room temperature, the thermal diffusivity is k=rc $ 104 m2 =s, the thermal wave speed is C $ 105 m=s, and the physical space and time scales to correspond dimensionless x 1 and t 1 are in the order of 1 nm and 0.01 ps, respectively [1]. Therefore, it is very important to consider the non-Fourier effect when the length scale reduces to micro/nanoscale. The non-classical Fourier heat conduction was rst formulated by Cattaneo [2] and Vernottee [3], through a hyperbolic heat conduction equation coupling with the

Corresponding author. E-mail address: wangbl2001@hotmail.com (B.L. Wang).

local energy balance. Since then, considerable effort has been devoted to the non-classical Fourier heat conduction problems. Analytical solutions for the hyperbolic heat conduction equation have been obtained for relatively simple problems such as semiinnite media [4,5]. One-dimensional problems involving either a semi-innite medium or a plane, for the case of constant thermal properties were analyzed in [6]. Numerical methods for complicated problems were proposed for a multilayered medium [79], an innitely wide slab with constant properties [10]. In addition, McCormacks predictorcorrector scheme was employed to solve one-dimensional problems [11], and hyperbolic heat conduction with temperature-dependent thermal conductivity, specic heat, and thermal diffusivity [12]. In [13], the hyperbolic heat conduction equation was solved for a semi-innite medium and considered a sudden change of temperature, a sudden change of heat ux, and a pulsed heat source at the boundary surface. Recently, modication of the Fourier conduction law and its implementation that admits jumps in the primary variables across element boundaries with arbitrary orientation in space and time was reviewed [14]. A hybrid Greens function method to investigate the hyperbolic heat conduction problems was proposed [15]. Direct solutions of second-order hyperbolic heat conduction equations with nite-difference schemes have been done in [1619], and the central and backward difference schemes were applied to solve the difference equation numerically in [20]. Solutions of the second-order hyperbolic heat conduction equation are helpful for understanding the fundamental physics behind the equations. It has been shown that the primary problems encountered in solving hyperbolic heat conduction equations numerically are due to the presence of ctitious

0168-874X/$ - see front matter & 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.nel.2011.09.010

202

B.L. Wang et al. / Finite Elements in Analysis and Design 50 (2012) 201206

numerical oscillations near the sharp discontinuities of the temperature eld, which are physically unrealistic [10,11,15]. This paper develops a numerical technique to solve the nonclassical heat conduction equations for the time-dependent temperature eld. The method uses the nite element (FE) space discretization to obtain a system of differential equations. Finite difference (FD) scheme is used to resolve the time-dependent temperature response. Stability analysis shows that the backward nite difference scheme is an ideal method to obtain the transient temperature response. Numerical results show that the ctitious numerical oscillations can be suppressed by a proper choice of the length of the time step in FD and the space increment in the FE. The paper also establishes a criterion between the length of the time step and the size of the space increment to ensure that numerical oscillation vanishes. Temperature-dependent material properties can be considered in the numerical model. Thermal stresses associated with non-classical heat conduction, which has never been studied previously, has been obtained for a 1-D plate. Signicant differences between those obtained from the nonclassical and classical heat conductions are observed.

Eq. (4) is the so-called thermal wave equation [5]. It has been used to the problems involving transient heat conduction heated by an internal pulsed heating source, laser-induced bubble nucleation [21], thermal ink-jet printers [22,23], laser-aided wafer surface cleanup, laser diagnostics, and laser processing [5,24]. The thermal conduction Eq. (4) must be solved for prescribed boundary and initial conditions. The initial conditions specify the temperature and thermal ux distributions at time zero. These are: Txj ,0 T 0 xj ,0 and qi xj ,0 q0i xj ,0, in which the quantities with an over bar mean that the quantities are prescribed. From Eqs. (2) and (3), the initial condition for the thermal ux can also be written as

rxcx

@Txj ,0 q0i,i xj ,0 Q xj ,0: @t

In most cases, there is no internal heat generation initially and the initial temperature and thermal ux are zero. As a result, the initial conditions become Txj ,0 0, @Txj ,0 0 @t 6

2. Basic concept of non-classical heat conduction Suppose a body occupying space O and surrounded by surface S is in a coordinate system xi, where (i1, 2, 3). The temperature inside the body T(x, t) is a function of point x and time t. The heat ow is controlled by the following conduction equation @T qi,i Q rxcx @t 1

The frequently encountered heat conduction boundary conditions are specied surface temperature and specied surface heat ow: qi ni h, on boundary Sq 7a 7b

T T, on boundary ST

where qi is the component of the heat ux vector q, r(x) is the mass density, c(x) the specic heat, and Q is the internal heat generation rate per unit volume. Hereafter, the summations over the indices i and j will be assumed when appearing twice in an equation. The most popular approach to the micro/nanoscale heat transfer is the wave theory of heat conduction, which uses the relaxation behavior to describe heat transport at short time intervals [5]. The governing heat transfer equation for this situation is qi t tq kx @T , @xi 2

where Sq ST S, and the over bar represents the known value, ni are components of the unit vector n normal to the exterior of S. Eq. (7a) indicates that on the boundary Sq the thermal ux is prescribed (h is positive if it is directed towards the exterior of the body). Eq. (7b) describes the temperature boundary condition on ST.

3. Finite element/nite difference scheme For a nonhomogeneous medium, the material properties k, r and c are complex functions of spatial position x. Let the medium be divided into a nite number of elements interconnected only at nodal points. For each element occupying space Oe, the temperature at any point can be expressed in terms of their values at their nodal points by: Tx1 ,x2 ,x3 ,t N y fTg 8

in which k is the thermal conductivity, qT/qxj is the temperature gradients, and tq is the thermal relaxation time, which is related to the collision frequency of the molecules within the energy carrier. tq can be calculated using the thermal wave speed, C, and the thermal diffusivity, k=rc, as tq pk=rc1=C 2 . For metal at the room temperature, k=rc $ 104 m2 =s, tq $ 1012 s and 5 C $ 10 m=s, and the physical space and time scales to correspond dimensionless x 1 and t1 are in the order of 1 nm and 0.01 ps, respectively [1]. Therefore, non-classical heat conduction is signicant at micro/nanoscale or at extremely rapid heating. Physically, the thermal wave model, Eq. (2), means that there is a time lag between heat ux and temperature gradient. If tq is much smaller than the time interval for a particular transient process, Eq. (2) can be simplied through the rst-order Taylor series expansion. To give qi kx @T @q tq x i @xi @t 3

in which [Ny] is known as the shape function matrix and is a function of spatial positions, {T} is a vector which contains the temperature values at the nodal points of the element. It follows that the temperature gradients @T/@xj at any point in region Oe can be written as f@T g f@T=@x1 ,@T=@x2 ,@T=@x3 gT By fTg 9

where By LN y and [L] denotes a differential operator matrix. Substituting Eq. (9) into Eq. (1), the heat uxes qi in the element can be obtained as follows: fqg fq1 ,q2 ,q3 gT kBy fTgtq f@q=@tg 10

Substitution of Eq. (3), and its derivative with respect to xi into Eq. (1) yields the governing equation in terms of temperature:

rxcxtq x

@2 T @T @Q 2 kxr T tq x Q : rxcx @t @t @t 2

Suppose that the medium undergoes a virtual temperature change dT (dT is zero on the boundary ST). By multiplying Eq. (4) by dT and integrating in the entire space domain O, and with the substitution of Eqs. (8) and (10), the nite element approximation of the heat equation can be obtained as (after assemblage) _ DfT g CfT g KfTg fpg 11

B.L. Wang et al. / Finite Elements in Analysis and Design 50 (2012) 201206

203

where the dot represents differentiation with respect to time. The element matrices and external heat load vector are given by Z D rxcxtq xNy T N y dO, 12a
O

_ the approximation of fTt m g and fT t m g, and that we want to compute {T}m 1. Replacing the derivative by difference quotients _ fT t m 1 g % fTgm 1 fTgm , Dt fT t m 1 g % _ _ fT gm 1 fT gm , Dt 13

Z C
O

rxcxNy T N y dO,
By T kxBy dO,

12b

Z K
O

the differential Eq. (11) at the point ttm 1 would then become D D _ fTgm 1 fTgm fT gm C K T m 1 pt m 1 fTgm 1 fTgm Dt Dt Dt2 14 For m0, Eq. (14) can be used directly to give T 1 since the initial _ _ conditions T 0 and T 0 are already known. Once T 1 is known, T 1 can be obtained from the rst of Eq. (13) and T 2 can be calculated from Eq. (14). Thus, Eq. (14) represents a general family of recurrence relations and is the so-called backward difference scheme. According to the theory of differential equation [27,28], Eq. (14) for the time portion is unconditionally stable and converges with the truncation error of order Dt. The number of time steps in Eq. (14) can be chosen such that the required precise can be achieved. If Eq. (11) is satised at ttm, the system will be the forward difference. The average of the forward difference and the backward difference would be the so-called central difference. The central difference and the forward difference usually produce more signicant oscillation than backward difference. It should be mentioned that the matrix differential Eq. (11) can also be solved via other techniques such as mode superposition and exact time integral. However, these techniques require considerable computation effect and produce obvious numerical oscillations, and thus are not effective in non-classical Fourier heat conduction analysis. A nite element code is made in the programming environment MATLAB, to nd the transient solution of Eq. (11). Eight types of elements are constructed as shown in Table 1, in which NTROD2, NTROD2N, NTROD3, and NTROD3N are one-dimensional elements, and NTQUAD4, NTQUAT4N, NTQUAD9, and NTQUAD9N

12c

R R fpg S Ny T hds O Q Ny T dO Z Z @h @Q ds tq x N T dO tq xNy T @t @t y S O

12d

Note that the classical Fourier solution is recovered if the matrix D is replaced with zero. In Eqs. (12), the material properties r, c and [k] can be functions of spatial coordinates xi, and {T} and {p} can be functions of time t. The numerical integration scheme, such as Gauss-Legendre integration, can be used to evaluate the integrals involved in Eqs. (12). The computational accuracy can be improved by increasing the number of elements. The element matrices [C], and [K] are same as these given in [25,26] for the classical Fourier heat conduction. If the material is considered as homogeneous in an element, the element matrix [D] for the non-classical heat conduction is related to the element matrix [C] through D tq C. The problem now is to solve the matrix differential Eq. (11). To this end, the direct numerical difference will be used. Since we cannot determine the nodal temperature {T} from Eq. (11) for all values of time t in an interval [0, t0], we will have to be satised with computing approximations {T}m of {T(tm)} for some points t m M 0 m in the interval. We assume that the points are equidistant, i.e., that tm m(Dt), m0,y,M, where the time step length Dt is dened as Dtt0/M for an integer M. To make the derivation somewhat more _ general, we assume that we have known fTgm and fT gm , which are
Table 1 Element types. Element name Element shape Description

Material properties within the element H N H N H

GaussLegendre points 2 3 3 3 22

NTROD2 NTROD2N NTROD3 NTROD3N NTQUAD4

One-dimensional 2-node linear element Same as NTROD2 One-dimensional 3-node quadratic element Same as NTROD3 Two-dimensional 4-node linear element

NTQUAD4N

Same as NTQUAD4

33

NTQUAD9

Two-dimensional 4 $ 9 node quadratic element (Any one or more of the nodes 5 to 9 can be vacant)

33

NTQUAD9N

Same as NTQUAD9

33

H: Homogeneous element (the properties inside the elements are constants and are their values at the center of the element). N: Non-homogeneous (the properties inside the element are calculated from their values at the nodes of the element).

204

B.L. Wang et al. / Finite Elements in Analysis and Design 50 (2012) 201206

are two-dimensional iso-parametric elements. In addition, element types NTROD2N, NTROD3N, NTQUAT4N, and NTQUAT9N having spatially varied material properties inside the elements (i.e., in evaluation the integrals in Eqs. (12), r, c and [k] are considered as functions of spatial coordinates xi). For one-dimensional thermal conductivity problems, NTQUAD4 is equivalent to NTROD2, NTQUAD4N is equivalent to NTROD2N, NTQUAD9 is equivalent to NTROD3, and NTQUAD9N is equivalent to NTROD3N. Also given in Table 1 is the number of Gauss integration points used to numerically evaluate the integrals involved in Eqs. (12). If material properties are temperature-dependent, r, c and [k] are functions of t. The matrix equations have the same forms as those given above, provided that the coefcient matrices [D], [C], and [K] are functions of temperature T and/or its gradients. The system become non-linear and one can use additional assumption that material properties (and then [D], [C], and [K]) at time interval (tm, tm 1) are functions of the temperature vector {T}m at time tm, which has already been known in each time step. 4. Numerical examples To demonstrate the applicability of the numerical method, we consider two problems: a one-dimensional temperature rise in a plate along its length direction and an axially symmetric temperature change in a cylinder along its thickness direction. Unless other specied, the backward difference scheme is used. 4.1. A plate

1.2 1 0.8 0.6 T/T0 0.4 0.2 0 -0.2


Non classical solution for
t = 0.25

Classical solution for


t = 0.25

Classical solution for t = 0.5 , and


t = 0.75

Non classical solution for Non classical solution for


t = 0 .5 t = 0.75

0.1

0.2

0.3

0.4

0.5 x

0.6

0.7

0.8

0.9

Fig. 1. Temperature prole in a one-dimensional plate for t 0:25, 0:5, 0:75 (the number of time steps is Dt=tq 1000; for curves with oscillations, the number of elements in x-direction is 100; for curves without oscillations, the number of elements in x-direction is 500).

1.4 1.2 1

t =1.5, thermally insulated at x = b

Denote b as the length of the plate (therefore, x A 0,b). In this case, Eq. (4) can be re-written in dimensionless form as 15

0.8 T/T0 0.6 0.4 0.2 0

t =1, thermally insulated at x = b

@2 T

p where t t=tq , x x=l0 , and l0 ktq =rc characterize length parameter. 4.1.1. Temperature eld Either element type NTQUAD4 and element type NTROD2 are used to model the one-dimensional heat conduction in the plate. The nite element method uses 500 or 100 linear elements in the x direction. For the two-dimensional element type (NTQUAD4), the number of elements in the direction perpendicular to the x direction can be arbitrary (since the temperature eld is one-dimensional and varies only with x). The time step Dt tq =1000 is used for the nite difference calculation. Variation of temperature with time are displayed in Fig. 1 for t Zb and Fig. 2 for t o b. The normalized length of the plate is b 1. It is found that the value of the space increment strongly inuences the accuracy of the numerical results. As pointed previously, the numerical oscillations in the vicinity of sharp discontinuities is physically unrealistic thus should be depressed. A rigorous proof on the stability of the problem is beyond the scope of this paper. Through comparisons of different calculations, it is found that for each specied length of time step (nondimensional), Dt=tq , when the length scale of the elements (nondimensional), Dx=l0 , is such that Dx=l0 o 2Dt=tq , the ctitious oscillation has been depressed. Once the condition Dx=l0 o2Dt=tq is satised, further decrease in Dx results in little change in the temperature prole. The numerical results approach the analytical solution accurately when the time step Dt and the space length Dx decrease simultaneously and satisfy the condition Dx=l0 o2Dt=tq . Finally, the results of Fig. 2 suggest that in some situations, the temperature inside the medium can be higher than the applied thermal shock temperature. Such a phenomenon cannot be produced by the classical Fourier heat conduction.

@T @2 T 2: 2 @t @x @t

t =1.5, zero temperature at x = b

0.1

0.2

0.3

0.4

0.5 x

0.6

0.7

0.8

0.9

Fig. 2. Temperature prole in a one-dimensional plate.

In order to evaluate the quality of the backward scheme, results from the central difference scheme and the backward difference scheme are compared and are plotted in Fig. 3. Clearly, the central difference scheme gives strong oscillation for the temperature eld. Result from the backward difference scheme is a smooth curve of temperature-time relationship (note that there is sharp drop of the curve at temperature discontinuous point). The backward nite difference scheme in the time step is found to be an efcient method to overcome the numerical oscillation in hyperbolic heat conduction equation (at least more efcient than the central difference scheme). In fact, it is well known that the application of backward nite difference scheme can effectively suppress the numerical oscillation, not only in the problem addressed in this paper, but also in other pones of other disciplines.

4.1.2. Thermal stress To our best knowledge, there is no research on the thermal stresses associated with non-classical heat conduction. Here, we calculate the thermal stresses in the plate using the transient

B.L. Wang et al. / Finite Elements in Analysis and Design 50 (2012) 201206

205

1 0.8 0.6 0.4 T/T0 0.2 0 -0.2 -0.4


Central difference,
t = 0.25

0.5
t = 0.75

t = 0.25

Dimensionless stress

Backward difference,

Central difference, Backward difference,


t = 0.75

0 t=0.9 -0.5 t=0.6 t=0.3

-1

0.5

1 1.5 Dimensionless x

0.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x
Fig. 3. Comparison of the backward difference and central difference for the calculation of temperature prole in a one-dimensional plate for t 0:25 and 0:75 (the number of time steps is Dt=tq 1000; the number of elements in x-direction is 500).

Dimensionless stress

t=1.5 0.1 0 -0.1 -0.2 t=6 t=3

temperature eld from the non-Fourier heat conduction. Since the thickness b of the plate is small as compared with the length of the plate, the normal strain in the length direction of the plate can be assumed as ex A BxaTx, where A and B are unknown constants and a is thermal expansion coefcient. Accordingly, the thermal stress eld in the length direction of the plate is sx EA BxaTx, in which E is the Youngs modulus. The Rb equilibrium of the plate requires that 0 sxdx 0 and Rb sxxdx 0. This determines the unknown constants A and B. 0 As a result, the stresses in the plate are obtained. Thermal stress distribution in the plate at different time are shown in Fig. 4. Thermal stress is normalized by s=EaT 0 and position is normalized by x=b (or x=b). For the short thermal shock time, the stress near the thermal shock surface (x 0) is compressive. The compressive stress decreases with time rstly, and then becomes tensile. This is apparently different from the stress associated with classical heat conduction. In that situation, the stress on the thermal shock surface is always compressive and decreases monotonously with increasing time. Similar to the case of classical heat conduction, the stress in the entire plate becomes zero as time is sufciently large. 4.2. A homogeneous cylinder Denote a and b as the inner and outer radius of the cylinder, r as the radical coordinate (therefore, r A a,b). In this case, the nonFourier heat conduction equation is   @2 T @T 1@ @T r , 16 tq 2 rc k @t r @r @r @t whose non-dimensional form is   @T 1 @ @T : k r 2 @r r @r @t @t

0.5

1 1.5 Dimensionless x

Fig. 4. Thermal stress distributions in a plate subjected to a thermal shock on its surface x 0 (the temperature on the surface x b of the plate is 0, thickness of the plate is b 2; dimensionless x is x=b, dimensionless stress is sx=EaT 0 ; note that time t in the gure has been normalized with t q ).

Fig. 5. Finite element mesh for a cylinder.

@2 T

17 cylinder is considered. The nite element mesh is shown in Fig. 5, which has 5 elements in the circumferential direction and 100 elements in the t direction (denoted as 5 100 elements). The system has 500 elements and 2211 notes. The time step Dt tq =200 is used for the nite difference calculation. Fig. 6 plots the radial temperature distributions for t 1,2. It seems that there is a sharp drop of temperature at br t, where br is the distance from the heating surface to the temperature drop

p where t t=tq , t r=l0 and l0 ktq =rc characterize length parameter. The values of a and b are a 2 and b 5, respectively. Initially, the cylinder is at zero temperature. At time 0, there is a sudden temperature change on its outer surface and the inner surface is maintained at zero temperature. In order to model the curve boundary accurately, two dimensional iso-parametric elements NTQUAD9 are used. Because of symmetry, only 1/4

206

B.L. Wang et al. / Finite Elements in Analysis and Design 50 (2012) 201206

Acknowledgments We are grateful to the National Science Foundation of China for the support of this work through projects (#10972067, #11172081). References
[1] D.Y. Tzou, The generalized lagging response in small-scale and high-rate heating, Int. J. Heat Mass Transfer 38 (1995) 32313240. [2] C. Cattaneo, Sur une forme de lequation de la chaleur eliminant le paradoxe dine propagation instantanee, C. R. Acad. Sci. 247 (1958) 431433. [3] P. Vernotte, Les paradoxes de la theorie continue de lequation de la chaleur, C. R. Acad. Sci. 246 (1958) 31543155. [4] M.N. Ozisik, D.Y. Tzou, On the wave theory in heat conduction, ASME J. Heat Transfer 116 (1994) 526535. [5] D.Y. Tzou, Macro-to Microscale Heat Transfer: The Lagging Behavior, Taylor & Francis, Washington, DC, 1997. [6] H.T. Chen, J.Y. Lin, Numerical analysis for hyperbolic heat conduction, Int. J. Heat Mass Transfer 36 (1993) 28912898. [7] J. Li, P. Cheng, G.P. Peterson, J.Z. Xu, Rapid transient heat conduction in multilayer materials with pulsed heating boundary, Numerical Heat Transfer, Part A 47 (2005) 633652. [8] W.B. Lor, H.S. Chu, Effect of interface thermal resistance on heat transfer in a composite medium using the thermal wave model, Int. J. Heat Mass Transfer 43 (2000) 653663. [9] M.A. Al-Nimr, M. Naji, R.I. Abdallah, Thermal behavior of a multi-layered thin slab carrying periodic signals under the effect of the dual-phase-lag heat conduction model, Int. J. Thermophys. 25 (2004) 949966. [10] H.Q. Yang, Characteristics-based, high-order accurate and nonoscillatory numerical method for hyperbolic heat conduction, Numer. Heat Transfer B 18 (1990) 221241. [11] D.E. Glass, M.N. Ozisik, D.S. McRae, B. Vick, On the numerical solution of hyperbolic heat conduction, Numer. Heat Transfer 8 (1985) 497504. [12] B. Pulvirenti, A. Barletta, E. Zanchini, Finite-difference solution of hyperbolic heat conduction with temperature-dependent properties, Numer. Heat Transfer A 34 (1998) 169183. [13] D.E. Glass, M.N. Ozisik, D.S. McRae, B. Vick, Hyperbolic heat conduction with temperature-dependent thermal conductivity, J. Appl. Phys. 59 (1985) 18611865. [14] S.T. Miller, R.B. Haber, A spacetime discontinuous Galerkin method for hyperbolic heat conduction, Comput. Methods Appl. Mech. Engrg 198 (2008) 194209. [15] T.M. Chen, A hybrid Greens function method for the hyperbolic heat conduction problems, Int. J. Heat Mass Transfer 52 (2009) 42734278. [16] C. Weber, Analysis and solution of the Ill-Posed inverse heat conduction problem, Int. J. Heat Mass Transfer 24 (1981) 17831791. [17] A. Kar, C.L. Chan, J. Mazumder, Comparative studies on nonlinear hyperbolic and parabolic heat conduction for various boundary conditions: analytical and numerical solutions, ASME J. Heat Transfer 114 (1992) 1420. [18] S.H. Pulko, A.J. Wilkinson, A. Saidane, TLM representation of the hyperbolic heat conduction equations, Int. J. Numer. Modell 15 (2002) 303315. [19] A.L. Koay, S.H. Pulko, A.J. Wilkinson, Reverse time TLM modeling of thermal problems described by the hyperbolic heat conduction equation, Numer. Heat Transfer B 44 (2003) 347363. [20] G.F. Carey, M. Tsai, Hyperbolic heat transfer with reection, Numer. Heat Transfer 5 (1982) 309327. [21] J. Li, Z.F. Zhang, D.Y. Liu, Experimental and Theoretical Study on the Rapid Transient Nucleated Boiling Heat Transfer, Prog. in Natural Sci 11 (2001) 529535. [22] A. Asai, Bubble dynamics in boiling under high heat ux pulse heating, ASME J. Heat Transfer 113 (1991) 973979. [23] H.K. Park, X. Zhang, C.P. Grigoropoulos, C.C. Poon, A.C. Tam, Transient temperature during the vaporization of liquid on a pulsed laser-heated solid surface, Trans. ASME 118 (1996) 702708. [24] T.Q. Qiu, C.L. Tien, Femtosecond laser heating of multi-layer metal-I. analysis, Int. J. Heat Mass Transfer 39 (1994) 27892797. [25] B.L. Wang, Z.H. Tian, Application of nite elementnite difference method to the determination of transient temperature eld in functionally graded materials, Finite Elem. Anal. Des. 41 (2005) 335349. [26] B.L. Wang, Transient one-dimensional heat conduction problems solved by nite element, Int. J. Mech. Sci. 47 (2005) 303317. [27] Kenneth H. Huberner, Earl A. Thornton, Byrom Ted G., The Finite Element Method for Engineers, Wiley-Interscience Publication, NewYork, Wiley, 1995. [28] Sadik Kaka, Yaman Yener, Heat Conduction, Taylor & Francis, Washington, DC, 1993.

Fig. 6. Temperature prole in a one-dimensional cylinder for t 1,2.

surface. Here the higher order elements NTQUAD9 is used to model the cylinder and no oscillation of temperature with radial coordinate is found. This means that the high-order elements can model the non-Fourier heat conduction more accurately. In fact, temperature eld T calculated with 1111 notes and 10 100 (10 elements in the circumferential direction and 100 elements in the r direction) linear NTQUAD4 elements is also obtained and some oscillations of T with r in the vicinity of sharp temperature drop was indeed observed. In addition to the use of high-order elements, the sharp drop in the temperature prole can be modeled efciently with ner mesh (more elements, smaller space increment).

5. Concluding remarks Non-Fourier effect is important in heat conduction at micro/ nanometer scale or in the situation of rapid transient heating. A main difculty of the numerical solutions of hyperbolic heat conduction problems is the numerical oscillation in the vicinity of sharp temperature gradient or discontinuities. In this paper, a nite element/nite difference scheme is developed to study the non-classical Fourier heat conduction problems. Several 1-D and 2-D element types are developed. The main advantage of the proposed method is that it can treat temperature dependent material properties. The backward nite difference scheme for the time portion is unconditionally stable therefore is an ideal method for the transient solution of the temperature eld. Through proper selection of element size in the FE and time step in the FD, the oscillation in the vicinity of sharp temperature change can be essential suppressed. Mesh rening and/or using of high-order elements, together with the nite difference, benet to the suppression of numerical oscillation. A relationship between the length of the time step and the element size is identied ensuring that numerical oscillation in the temperature vanishes or be suppressed. Finally, the thermal stresses associated with non-classical heat conduction show dramatic difference from those associated with classical heat conduction.