Escolar Documentos
Profissional Documentos
Cultura Documentos
a r t i c l e
i n f o
Article history:
Received 9 March 2011
Received in revised form 12 August 2011
Accepted 14 November 2011
Available online 25 November 2011
Keywords:
Thermoelectrics
Dynamics
Non-linear nite element
Second sound
HHT algorithm
Newmark-b algorithm
a b s t r a c t
In the present work, a three-dimensional, dynamic and non-linear nite element to simulate thermoelectric behavior under a hyperbolic heat conduction model is presented. The transport equations, which couple electric and thermal energies by the Seebeck, Peltier and Thomson effects, are analytically obtained
through extended non-equilibrium thermodynamics, since the local equilibrium hypothesis is not valid
under the hyperbolic model. In addition, unidimensional analytical solutions are obtained to validate
the nite element formulation. Numerically, isoparametric eight-node elements with two degrees of freedom (voltage and temperature) per node are used. Non-linearities due to the temperature-dependence
on the transport properties and the Joule effects are addressed with the NewtonRaphson algorithm.
For the dynamic problem, HHT and Newmark-b algorithms are compared to obtain accurate results, since
numerical oscillations (Gibbs phenomena) are present when the initial boundary conditions are discontinuous. The last algorithm, which is regularized by relating time steps and element sizes, provides the
best results. Finally, the nite element implementation is validated, comparing the analytical and the
numerical solutions, and a three-dimensional example is presented.
2011 Elsevier B.V. All rights reserved.
1. Introduction
Thermoelectric materials couple electric and thermal energies
by means of three separated transport effects: Seebeck, Peltier
and Thomson. In addition, the Ohm and Fourier laws that are inherent to electric and thermal uxes are also present. Thermoelectric
materials are used as heat pumps (heating and cooling) and generators, see [1], and in the last decade miniaturized thermoelectric
devices and high-frequency processes are increasingly applied to
cool micro-electronic devices, see [2].
The classical Fourier law leads to a parabolic heat propagation
problem incorrect from a physical point of view: especially applied
to micro-devices and under rapid transient effects such as micropulses. Cattaneo [3] proposed a modication to the Fourier law that
leads to a hyperbolic heat propagation problem, avoiding the classical paradox related with the innite temperature velocity of
propagation. The modication, named sometimes second sound,
is made through the introduction of a relaxation time s, an empirical parameter dened as the time-interval between two successive collisions at the microscopic level of either holes or electrons.
From a theoretical point of view, the classical non-equilibrium
thermodynamics [4] cannot lead to hyperbolic propagation since
the local equilibrium hypothesis is not valid. For this reason, ex Corresponding author.
E-mail address: ropalgue@upv.es (R. Palma).
0045-7825/$ - see front matter 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.cma.2011.11.011
94
R. Palma et al. / Comput. Methods Appl. Mech. Engrg. 213216 (2012) 93103
Neglecting qX, assumption valid for most thermoelectric applications [19], the charge and energy balances are
The extended non-equilibrium thermodynamics theory assumes the existence of a non-equilibrium entropy s that depends
on state variables and on uxes. Since the electric energy propagates much faster than the thermal one, the hyperbolic model is
applied only for the thermal eld, and the entropy balance is given
by Jou and Lebon [4] adding the Joule irreversibility
qs_ r
q_ X r j;
qu_ r 1 j E;
2. Thermodynamic formulation
The aim of this section is develop balance and transport equations for the hyperbolic thermoelectric problem. As mentioned
above, the hyperbolic problem cannot be obtained from classical
non-equilibrium thermodynamics. This is due to the lack of validity of the local equilibrium hypothesis that states the equality of
the local relations between thermodynamic quantities in a system
in and out of equilibrium [5].
This validity is closely related with the Deborah number De =
s/sM < 1, where sM is the macroscopic experimental time (effect
duration) and s the time interval between two successive collisions
between particles. Large Deborah numbers occur if: s is large, for
materials such as polymers, and/or sM is short, for very fast phenomena such as ultrasound propagation. The extended non-equilibrium thermodynamics [5] must be used for transport equations
when De P 1.
qu_ r 1 j E:
r j 0;
1
T
j rV
1
s
2 1_ ;
1 r
T
T
jT
js
1
T
rs
j rV
1
s
1 r
2 1_ :
T
T
jT
j crV acrT;
1 s1_ jrT aTj;
in which dimensions are: a [V/K], c [A/mV], j [W/mK] but T is introduced in Celsius (C) degrees, see [21].
rV aTrT
Fig. 1. Energy balance in a domain X: the three effects, reversible couplings (arrows
without border) and irreversible thermal and electric uxes (arrows with border).
cT
where the rst term in the right-hand represents the thermal energy converted into electric by the Seebeck effect. Due to restrictions
R. Palma et al. / Comput. Methods Appl. Mech. Engrg. 213216 (2012) 93103
95
RX
RX
C
X
dT sqcTd
_ X
dT qcTd
13
dTqc dC 0:
3.2. Discretization
10
The continuum domain X is discretized into elements Xe, see
nel
Fig. 2. Therefore, X [ei
Xe where nel is the total number of elements, each one delimited by npe = 8 nodes. In turn, each node relates to two degrees of freedom (voltage, temperature) per node.
For an element e, degrees of freedom and spatial coordinates x
are approximated using 3D isoparametric shape functions Ne
n
o
n
o
where Ve V 1e ; . . . ; V ne pe , Te T 1e ; . . . ; T ne pe . Be is the discretized
gradient matrix of the element composed of npe submatrices with
the form:
3. FE formulation
BtA
11
V V h Ne V e ;
T T h Ne Te ;
rV rV Be Ve ;
T_ T_ h Ne T_ e ;
rT rT h Be Te ;
e ;
T T h Ne T
14
x xh Ne xe ;
@N A
@x1
@N A
@x2
@N A
@x3
i
;
15
dV dV h Ne dVe ; dT dT h Ne dTe ;
rdV rdV h Be dVe ; rdT rdT h Be dTe :
16
Introducing these discretizations into the transport equations (6)top and (11)-bottom
je cBe Ve acBe Te ;
qe jBe Te acNe Te Be Ve a2 cNe Be T2e :
17
VV
on CV ;
j n jc
on Cj ;
TT
on CT ;
q n qc
on Cq ;
R
R
dV r jdX 0;
R
R
_
X dT r qdX X dTj rVdX X dT qc TdX
R
R
X dT s@ t fj rVg dX X dT sqcTdX 0:
X
12
The s and q-terms are new with respect to the formulation developed in [7]. The divergence theorem is now applied to the rst term
of both equations, and Dirichlet boundary conditions are introduced
Fig. 2. Domain discretization in nite elements Xe. Isoparametric three-dimensional and interface ux elements. Two degrees of freedom per node.
96
R. Palma et al. / Comput. Methods Appl. Mech. Engrg. 213216 (2012) 93103
R
R
Xe
Bte je dXe
Xe
Bte qe dXe
R
R
Nte
Ce
Nte jc dCe 0;
V
K VV
AB @ V B fRA g
Xe
Ce
Nte qc dCe 0:
18
k
@RA
RkA
dgkA
g
19
where A, B are the global numbering of two nodes, k the NR iteration counter and gA derivatives of the degrees of freedom at node A:
_ A fV_ A ; T_ A gt , and second
value as UA = {VA, TA}t, rst derivative as U
t
k
@RA
c1 KAB c2 CAB c3 MAB :
@gB
20
c
for
In Table 1 the parameters c1, c2 and c3 are given in terms of b,
for the HHT algorithm.
the N-b algorithm and a
The consistent tangent, capacity and mass matrices, are derived
for each iteration
KAB
@RA
;
@UB
CAB
@RA
;
@ U_ B
MAB
@RA
:
B
@U
21
RVA
RTA
BtA jA dX
C N A jc dC;
R
t
B
q
d
X
X A A
X N A j A BA V A dX
R
R
X NA s@ t fjA BA V A gdX X NA sqcT A dX
R
R
NA qcT_ A dX NA q dC:
V
K VT
AB @ T B fRA g
K TV
AB
BtA @ V B fjgdX;
BtA @ T B fjgdX;
23
@ V B fjg cBB ;
@ T B fjg acBB @ T fagNB cBB T B
@ T fcgBB V B aBB T B NB ;
24
n o
V
C VV
0;
AB @ V_ B RA
C TT
AB
n o
V
C VT
0;
AB @ T_ B RA
R
NA @ tV_ B fjgBB V B dX s X NA jB BB dX;
R
R
@ T_ B fRTA g s X NA @ tT_ B fjgBB V B dX X NA qcN B dX;
T
C TV
AB @ V_ B fRA g s
25
where the derivatives are again obtained applying the chain rule to
the discretized transport Eq. (17)
26
V
M VT
AB @ T B fRA g 0;
T
MTV
fRA g 0;
AB @ V
B
R
T
MTT
AB @ T B fRA g X N A sqcN B dX:
27
"
T
V
22
X BtA @ V B fqgdX
R
X NA @ V B fjgBB V B dX X N A jA BB dX
R
R
X NA s@ V B fjgBB V_ B dX X NA s@ t fjgBB dX;
R t
T
K TT
AB @ T B fRA g X BA @ T B fqgdX
R
R
X NA @ T B fjgBB V B dX X NA s@ T B fjgBB V_ B dX:
@ V B fRTA g
k1
c1 KTV c2 CTV
c1 KVT
c1 KVV
k
k
T
dT
:
V
dV
dT
dV
RT
RV
)k
;
28
The present formulation is implemented as a user element in the
non-linear nite element code FEAP, see [16]. Each element uses
a standard, in the nite element method (FEM) sense, eight-node
isoparametric element with two degrees of freedom per node A:
voltage VA and temperature TA.
4. Validations
Table 1
Tangent matrix parameters for Newmark and HHT algorithms.
Method
Parameters
c1
c2
c3
N-b
HHT
a
c
Dt
b
a c
Dt
b
1
Dt 2
b
1
Dt 2
b
R. Palma et al. / Comput. Methods Appl. Mech. Engrg. 213216 (2012) 93103
Table 2
Unidimensional validation cases, simplications and considered effects: F - Fourier, S Seebeck, J - Joule, Th - Thomson, ss - second sound.
Case
Simplications
Effects
I
II
III
IV
s = 0; j = 0; a, c, j = cte
s = 0; j 0; a, c, j = cte
s = 0; j 0; a(T); c, j = cte
s 0; j 0; a(T); c, j = cte
F,
F,
F,
F,
S
S, J
S, J, Th
S, J, Th, ss
I:
II :
29
The fourth and fth terms of the right side of case IV (second line)
represent the second sound effects: irreversibility, (5)-right and
hyperbolicity, respectively.
4.1. Cases IIII
For these cases, s = 0 and the thermal strong forms (29) are diffusive: second order parabolic partial differential equations. For I,
the situation is linear and homogeneous; for II linear and nonhomogeneous and for III non-linear and non-homogeneous.
Analytical solutions for the thermal eld in cases I and II are given in [24], while the electrical elds can be calculated from (9),
see the Appendix. For III, the thermal and electrical solutions are
given in [25].
Numerical solutions are obtained using a structured and coarse
mesh of 11 elements in the x3 direction. Only one element is used
97
For a linear problem and according to [13], no numerical oscillations will appear if the dynamic algorithm is regularized by the
relationship Dt 6 Ch/v, where C is the Courant number [26] and
98
R. Palma et al. / Comput. Methods Appl. Mech. Engrg. 213216 (2012) 93103
Fig. 4. Voltage (left) and temperature (right) distributions along thermoelement length for cases I (top row) and coincident II, III (bottom row) at three times in [s], see Table
2. Analytical results with lines, nite element with circles.
Fig. 5. Thermal uxes for cases I (left) and coincident II, III (right) at three times in [s], see Table 2. Analytical results with lines, nite element with circles.
v j=qcs,
R. Palma et al. / Comput. Methods Appl. Mech. Engrg. 213216 (2012) 93103
99
Fig. 6. Analytical temperature solution for t = 0.06, s = 0.02 [s], C = 1/6 (top).
Newmark-b and different HHT parameters to study Gibbs phenomena (middle),
see Table 3. Adjusting by trial and error the Courant number in Test b to avoid
oscillations (bottom).
Table 3
Newmark-b and HHT parameters used for numerical testing.
Test
Algorithm
q
b
c
a
N-b
N-b
N-b
N-b
1
3/2
1/2
0
1/4
9/14
4/9
1
1/2
16/49
5/6
3/2
HHT
HHT
3/2
1/2
9/14
4/9
16/49
5/6
6/7
2/3
except for a small region around the wave front at t = 0.06 [s]; this
is due to the intrinsic oscillations from the discrete time integration scheme and is very difcult to avoid.
Fig. 8 compares the semi-analytical and FEM thermal uxes for
t = 0.06 [s]. Two peaks due to the discontinuous boundary conditions at the wave front are evident, representing the propagation
Fig. 7. Voltage (left) and temperature (right) vs. thermoelement length at three times in [s]. For voltage, only nite element results; for temperature, analytical results with
lines and nite element with circles.
100
R. Palma et al. / Comput. Methods Appl. Mech. Engrg. 213216 (2012) 93103
Fig. 9. Front wave thermal ux vs. thermoelement length at several times for s = 0.02 (top) and relaxation time inuence in the wave shape (bottom) at t = 0.06, both in [s].
Only nite element results represented.
101
R. Palma et al. / Comput. Methods Appl. Mech. Engrg. 213216 (2012) 93103
PARABOLIC MODEL
Time = 7.74E-03
Time = 3.44E-02
Time = 6.51E-02
Time = 1.29E-01
3.47E+00
HYPERBOLIC MODEL
-8.43E-09
2.30E+01
2.30E+01
2.90E+01
2.90E+01
3.50E+01
3.50E+01
4.10E+01
4.10E+01
4.70E+01
4.70E+01
5.00E+01
5.33E+01
2.78E+01
5.93E-01
2.30E+01
2.30E+01
2.90E+01
2.90E+01
3.50E+01
3.50E+01
4.10E+01
4.10E+01
4.70E+01
4.70E+01
5.00E+01
5.28E+01
3.00E+01
3.00E+01
2.30E+01
2.30E+01
2.90E+01
2.90E+01
3.50E+01
3.50E+01
4.10E+01
4.10E+01
4.70E+01
4.70E+01
5.00E+01
5.34E+01
3.00E+01
3.00E+01
2.30E+01
2.30E+01
2.90E+01
2.90E+01
3.50E+01
3.50E+01
4.10E+01
4.10E+01
4.70E+01
4.70E+01
5.00E+01
5.00E+01
Fig. 11. Temperature distribution inside the truncated pyramid thermoelement for the parabolic (left) and hyperbolic (right) models at several instants. For parabolic model
s = 0, for hyperbolic s = 0.02 [s] assumed. Boundary conditions and geometry shown in Fig. 10.
Fig. 12. Analytical temperature distribution at t = 0.06 [s] for cases IIV vs. thermoelement length. Boundary conditions and prescribed values displayed in Fig. 3.
micro-devices under fast operation modes. For instance, an electronphonon thermal mismatch due to different thermal and electrical carriers was reported in [31] for cooling microthermoelectrics. This non-equilibrium has until now been modeled
by molecular dynamics or Monte Carlo techniques; however, the
present combined parabolichyperbolic model could simulate
these effects through the introduction of an empirical value of s.
6. Conclusions
This article has presented a non-linear and transient nite element formulation to simulate thermoelectric behavior under parabolic and hyperbolic heat conduction. The formulation has been
implemented into the computational code FEAP. Numerically,
non-linearities and transient hyperbolicity have been addressed
102
R. Palma et al. / Comput. Methods Appl. Mech. Engrg. 213216 (2012) 93103
1
P
Vx3 ; t a DLT x3 DT
cn SxeKt ;
n1
1
P
DT
L
Tx3 ; t T c x3
cn SxeKt ;
n1
1
P
q3 x3 ; t j DLT
cn Cx nLp eKt ;
A:1
n1
2
cn
L
DT
x3 Sxdx3 :
Tc
L
A:2
Case II
Vx3 ; t jc3 Lx a
A2
L
DT
L
1
P
A1 x3 Lx
cn SxeKt ;
n1
1
P
x3 T C
cn SxeKt ;
n1
1
P
q3 x3 ; t j AL2 2A3 x3
cn Cx nLp eKt aTx3 ; tj3 ;
Tx3 ; t
A3 x23
n1
A:3
cn
2
L
A2
x3 T c A3 x23 Sx dx3 :
L
A:4
Case IV
The T distribution along the TE was reported in [8] using the
Laplace transform technique. However, we believe there are
several errors in this work, for which the result is developed
again. From the reference, we use the dimensionless parameters
n = x3/L and g = tj/ qcL2 but (10) is rewritten in an amenable
form to be solved by the Laplace technique
A:5
Tn; 0 0;
T0; g T h ;
T1; g T c :
Wn 0;
W1 Tsh ;
W0 Tsc ;
A:6
P11
F2
T c
ssF 3 s2 s s
F2
sF3 s2 s
ek11 ek12
q
k11
12 A1 A21 4B1 ;
k12
A1 F 1 t h L2 AajssF 4 ;
A:7
B1 t 2h L4 s F 3 s2 s:
R. Palma et al. / Comput. Methods Appl. Mech. Engrg. 213216 (2012) 93103
[8] M. Alata, M. Al-Nimr, M. Naji, Transient behaviour of a thermoelectric device
under the hyperbolic heat conduction model, Int. J. Thermophys. 24 (6) (2003)
17531768.
[9] J. Gembarovic, V. Majernik, Non-fourier propagation of heat pulses in nite
medium, Int. J. Heat Mass Transfer 31 (5) (1988) 10731080.
[10] H. Chen, J. Lin, Numerical analysis for hyperbolic heat conduction, Int. J. Heat
Mass Transfer 36 (11) (1993) 28912898.
[11] A. Barletta, E. Zanchini, Hyperbolic heat conduction and local equilibrium: a
second law analysis, Int. J. Heat Mass Transfer 40 (5) (1997) 10071016.
[12] M. Manzari, M. Manzari, On numerical solution of hyperbolic heat conduction,
Commun. Numer. Methods Engrg. 15 (1999) 853866.
[13] R. Ciegis, Numerical solution of hyperbolic heat conduction equation, Math.
Modell. Anal. 14 (1) (2009) 1124.
[14] H. Gmez, I. Colominas, F. Navarrina, M. Casteleiro, A discontinuous Galerkin
method for a hyperbolic model for convectiondiffusion problems in CFD, Int.
J. Numer. Methods Engrg. 71 (2007) 13421364.
[15] S. Bargmann, P. Steinmann, Theoretical and computational aspects of nonclassical thermoelasticity, Comput. Methods Appl. Mech. Engrg. 196 (2006)
516527.
[16] R. Taylor, FEAP, A Finite Element Analysis Program: User Manual, University of
California, Berkeley, 2010. <http://www.ce.berkeley.edu/feap>.
[17] H. Hilber, T. Hughes, R. Taylor, Improved numerical dissipation for time
integration algorithms in structural dynamics, Earthquake Engrg. Struct. Dyn.
5 (1977) 282292.
[18] J. Jackson, Classical Electrodynamics, John Wiley and Sons Inc., 1962.
[19] A. Eringen, G. Maugin, Electrodynamics of Continua I, Springer-Verlag New
York, Inc., 1990.
103