See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/259894214

Numerical Inversion of Laplace Transform for Time Resolved Thermal

Characterization Experiment

Article  in  Journal of Heat Transfer · April 2011

DOI: 10.1115/1.4002777

CITATIONS READS
18 1,068

7 authors, including:

Jean-Luc Battaglia J. Toutain

University of Bordeaux Ecole Nationale Supérieure de Chimie, de Biologie et de Physique (ENSCBP)
200 PUBLICATIONS   1,875 CITATIONS    52 PUBLICATIONS   383 CITATIONS   

SEE PROFILE SEE PROFILE

Christophe Pradere Jérôme Pailhès

Institut de Mécanique et d'Ingénierie de Bordeaux Ecole Nationale Supérieure d'Arts et Métiers
137 PUBLICATIONS   1,033 CITATIONS    89 PUBLICATIONS   660 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

water content measurement at local scale View project

Integrated Product and Process Design View project

All content following this page was uploaded by J. Toutain on 14 January 2019.

The user has requested enhancement of the downloaded file.

PROOF COPY [HT-10-1093] 003102JHR
1 Numerical Inversion of Laplace assumption, the temperature at any point M, for any transient 48
excitation denoted here by Q共t兲, can be calculated using two dif- 49
2 Transform for Time Resolved Thermal ferent ways 50
3 Characterization Experiment TM 共t兲 = L−1关H共M,p兲 ⫻ L关Q共t兲兴兴 51
4
or 52
5 J. Toutain T M 共t兲 = L关H共M,p兲兴 ⴱ Q共t兲 53
PR
6 J.-L. Battaglia where ⴱ denotes the convolution product. 54
7 e-mail: jean-luc.battaglia@ensam.bordeaux.fr It is obviously more efficient including the excitation in the 55
function to be inverted instead of using the convolution technique. 56
8 C. Pradere Indeed, the convolution method needs to sample the excitation 57
waveform, whereas the sampling interval can be chosen indepen- 58
OO
9 J. Pailhes dently from the excitation transient behavior when calculating the 59
inverse Laplace transform of the product. On the other hand, that 60
10 A. Kusiak is made possible since it is always possible to find the Laplace 61
transform of the excitation in the field of the thermal character- 62
W. Aregba
F
11 ization experiments. However, even if literature offers a broad 63
range of numerical inverse Laplace algorithms, not all of them fit 64
12 J.-C. Batsale with the design of the experiment in terms of the time-dependent 65
CO

13 disturbance. In this note, it is compared to three numerical inverse 66

14 Laboratoire inter établissement “TRansferts Ecoulements Laplace algorithms, representative of the three most important 67
15 Fluides Energétique”, sets of methods. 68
16 UMR 8508,
2 Numerical Inverse Laplace Algorithms 69
PY

17 Ecole Nationale Supérieure d’Arts et Métiers,

18 Esplanade des Arts et Métiers, Talence, There are dozens of algorithms available for the numerical in- 70
version of the Laplace transform. Some papers 关2,3兴 have pro- 71
19 Cedex 33405, France
posed comparative studies on algorithms, which can be split into 72
four main categories: Fourier series expansion, Laguerre function 73
expansion, combination of Gaver functionals, and deform of Bro- 74
[H

20 The aim of this technical brief is to test numerical inverse Laplace mwich contour. 75
21 transform methods with application in the framework of the ther- Three algorithms belonging to each main category are tested 76
22 mal characterization experiment. The objective is to find the most here. The first one is based on approximating the inversion inte- 77
T-

23 reliable technique in the case of a time resolved experiment based gral by an infinite Fourier series 关4兴. The second technique is also 78
24 on a thermal disturbance in the form of a periodic function or a widely known 关5,6兴 and based on the Post–Widder formula. The 79
25 distribution. The reliability of methods based on the Fourier se- last algorithm is due to Den Iseger 关7兴, which belongs to the class 80
10

26 ries methods is demonstrated. 关DOI: 10.1115/1.4002777兴 of Fourier series methods and uses the fast Fourier transform. This 81
algorithm was selected since it is very accurate 共see Appendix A兲 82
27 Keywords: thermal characterization, analytical solution, numeri- even in the case of function discontinuities or singularities. It is 83
-1

28 cal Laplace inversion, Fourier series also very easy to code and generates values on a regular grid of 84
time points, which is of great interest when numerical inversion is 85
aimed to be compared with experimental data usually obtained at 86
09

regular intervals. The full mathematical background of this 87

method is not presented here. It is based on a Gaussian quadrature 88
29 1 Introduction rule for the accurate approximation of the Poisson summation 89
formula 共see overview in Appendix A兲. The code in MATLAB is 90
3]

30 People who are concerned with the measurement of thermal presented in Appendix B. 91
31 properties always look for analytical solutions of the heat transfer
32 mathematical governing equations established for the studied ex- 3 Application 92
AQ: 33 perimental configuration. Indeed, they can be implemented easily
00

#1 34 in an optimization algorithm in order to find the values that lead to An application related to a classical experimental configuration 93
35 reaching the minimum of an objective function. One of the most in the framework of thermal characterization of materials is pre- 94
36 interesting ways to obtain an analytical solution of the heat diffu- sented. The front face of an infinite medium is subjected to a heat 95
31

37 sion equation is based on the use of integral transforms 关1兴. The flux from a photonics source as a laser. The penetration depth of 96
38 Laplace transform, denoted by L, is largely implemented due to the photons is calculated according to the extinction coefficient ␬ 97
39 its simplicity and powerfulness to treat a lot of transient excitation of the material, as 1 / ␹ = ␯ / 4␲␬, where ␯ is the wavelength of the 98
40 functions 共Dirac, Heaviside, periodic functions, etc.兲. It is gener- photons. Thus, this absorption will result as a heat source and heat 99
02

41 ally applied on the time variable, whereas other transforms, such will diffuse inside the medium. A one dimensional heat transfer 100
42 as the Hankel or Fourier transforms, can be applied on the space inside the medium is assumed. Considering the excitation on the 101
43 variables. Applying the Laplace transform on the heat diffusion form of a Dirac function and applying the Laplace transform on 102
JH

44 equation with associated boundary condition and considering the the time variable of the heat diffusion equation with associated 103
45 excitation in the form of a Dirac function, the transformed im- boundary conditions, an analytical expression of the transformed 104
46 pulse response H共M , p兲 is obtained at each point M of the medium impulse response is found as 关8兴 105
47 according to the Laplace variable p. Thereby, assuming the linear ␹
冉 ␹ −␥z −␹z
冊 冑 p
R

H共z,p兲 = e −e , with ␥ = 共1兲

k共␹2 − ␥2兲 ␥ a 106
Contributed by the Heat Transfer Division of ASME for publication in the JOUR- where p denotes the Laplace variable, k is the thermal conductiv- 107
NAL OF HEAT TRANSFER. Manuscript received March 11, 2010; final manuscript re-
ceived September 17, 2010; published online xxxxx-xxxxx-xxxxx. Assoc. Editor: ity of the medium, and a is the thermal diffusivity. If pulse dura- 108
Ofodike A. Ezekoye. tion ␶ is considered, the expression of the new solution is 109

Journal of Heat Transfer Copyright © 2011 by ASME APRIL 2011, Vol. 133 / 1-1

PROOF COPY [HT-10-1093] 003102JHR

PROOF COPY [HT-10-1093] 003102JHR

PR
OO
F

Fig. 1 Comparison of the three numerical techniques for the

function h„z = 0 , t… Fig. 3 Comparison of the three numerical techniques for the
function h␶,T„z = 0 , t…
CO

1 − e −␶ p
H␶共z,p兲 = H共z,p兲 共2兲 algorithm. Finally, the simulation for the function H␶,T共z = 0 , p兲 is 129
110 p
PY

represented in Fig. 3. The number of sampling intervals for this 130

111 Finally, let us consider the excitation on the form of a pulse simulation is 105. The de Hoog and Stehfest algorithms lead to the 131
112 train, each pulse duration being denoted by ␶. If T denotes the average value of the function but they are not able to deal with the 132
113 time interval between two successive pulses, the solution is de- periodic behavior. On the contrary, we obtained a very good 133
114 rived from the previous one as agreement between the function obtained using the Den Iseger 134
algorithm and the exact solution given from relation 共4兲 共as dem-
[H

1 135
H␶,T共z,p兲 = H␶共z,p兲 共3兲 onstrated in Fig. 2, the function h␶共t兲 can be calculated accurately 136
115 1 − e−Tp
using one of the three numerical techniques for the computation of 137
116 As demonstrated in Appendix C, the “exact” solution for this the exact solution兲. In Fig. 4, the relative gap between the exact 138
T-

117 last configuration is solution and that simulated from the Den Iseger technique is rep- 139
⬁ resented. The gap over one period of time T for the x-axis and the 140
兺 h␶共z,t + nT兲, 0ⱕtⱕT 共4兲 period number on the y-axis is considered. This kind of represen- 141
10

S共t + nT兲 =
118 n=0 tation demonstrates that the relative gap is highest 共10−2兲 at the 142
time of the discontinuity. Furthermore, it is also demonstrated that 143
119 where h␶共z , t兲 is the inverse Laplace transform of H␶共z , p兲. this gap remains roughly constant for all the periods. The compu- 144
-1

120 The numerical values for the simulations are k tation time using the Stehfest algorithm is 0.031 s, it is 0.11 s with 145
121 = 160 W m−1 K−1, a = 8.163⫻ 10−5 m2 s−1, ␹ = 20 nm, ␶ = 0.1 ns, the de Hoog algorithm and 1.61 s with the Den Iseger one 共two 146
122 and T = 1 ns. In Fig. 1, the comparison between the three methods core CPU, 2.33 GHz兲. 147
09

123 considering the function H共z = 0 , p兲 from relation 共1兲 is presented. AQ:
124 A very good agreement is obtained between the three techniques. #2
125 The same simulation is carried out for H␶共z = 0 , p兲. Again, as rep-
Appendix A: Principles of Presented Algorithms 148
resented in Fig. 2, a satisfactory agreement is found for the three
3]

126
127 techniques. However, as it can be seen at the discontinuity of the The widely spread Gaver–Stehfest algorithm is derived from 149
128 function, a very small deviation occurs when using the Stehfest the Post–Widder formula. Thus, given the Laplace transform F共s兲, 150
the approximation of the time domain solution is 151 AQ:
00

#3
31
02
JH
R

Fig. 2 Comparison of the three numerical techniques for the Fig. 4 Relative gap between the exact response and that cal-
function h␶„z = 0 , t… culated from the Den Iseger technique

1-2 / Vol. 133, APRIL 2011 Transactions of the ASME

PROOF COPY [HT-10-1093] 003102JHR

PROOF COPY [HT-10-1093] 003102JHR

兺 V F冉 冊
i=N
ln 2 ln 2
f共t兲 = i i 共A1兲
152 t i=i t % the following line may be replaced by a loop on 198
n = size共li兲 in order 199
153 where the elements of vector V are calculated by % to avoid memory failure for huge values of nt 200
= size共t兲共⬇106兲 201
min共i,N/2兲
kN/2共2k兲! 关k , li兴 = meshgrid共k , li兲; 202
Vi = 共− 1兲 i+N/2
兺 共N/2 − k兲!k!共k − 1兲!共i − k兲!共2k − i兲!
s = 共li+ cⴱk兲 / dt; 203
k=i+1/2 ft= real共Lf共s兲兲; 204
PR
154 共A2兲 ft= 4ⴱbiⴱft/ dt; 205
ft共1兲 = 0.5ⴱ共ft共1兲 + ft共N + 1兲兲; 206
155 The de Hoog and Den Iseger algorithms are derived from the % discrete Fourier’s inversion 207
156 inverse Laplace transform definition ft= ifft共ft共1 : N兲兲; 208

冕
ft= real共exp共a.ⴱ共0 : nt− 1兲兲.ⴱ共ft共1 : nt兲兲兲; 209
OO
␥+i⬁
1 return 210
f共t兲 = exp共pt兲F共p兲dp 共A3兲
157 2␲i ␥−i⬁ Appendix C: Analytical Solution for a Pulse Train Heat 211
158 where ␥ is such that the contour of integration is to the right-hand Flux Waveform 212
159 side of any singularities of F. After algebraic manipulation, Eq. The heat flux is assumed to be on the following form: ␾共t兲
F
213
160 共A2兲 may be expressed as follows: ⬁
= 兺n=−⬁ ␦共t − nT兲, where T is the period between two successive 214

冕
+⬁ pulses. The temperature at z = 0 is calculated from the convolution 215
1
CO

f共t兲 = exp共␥t兲 Re兵F共p兲exp共i␻t兲其d␻ 共A4兲 product S共t兲 = h␶共t兲 ⴱ ␾共t兲, where h␶共t兲 is the impulse response at 216
2␲
161 0 time t for a pulse duration ␶. Thus 217

冕 冕
162 where p = ␥ + i␻. From a trapezoidal rule with a step ␲ / T, it fol- ⬁ ⬁ ⬁
163 lows that the time domain solution can be approximated by S共t兲 = h␶共t⬘兲␾共t − t⬘兲dt⬘ = h␶共t⬘兲 兺 ␦共t − nT − t⬘兲dt⬘
冋 再 冉 冊 冉 冊冎册
PY

+⬁ −⬁ −⬁ n=−⬁ 218
F共␥兲 ik␲ ik␲t
兺
冕
1 ⬁
f共t兲 ⬇ exp共␥t兲 + Re F ␥ + exp ⬁
2␲ 2 k=1 T T = 兺 h␶共t⬘兲␦共t − nT − t⬘兲dt⬘ 共C1兲
219
164 共A5兲 n=−⬁ −⬁
[H

165 Unfortunately, the infinite series in Eq. 共A4兲 converges very A fundamental property of the Dirac function is that 220

冕 冕
166 slowly. De Hoog et al. used a Padé-type accelerator to approxi- ⬁ ⬁
167 mate the series under the form of a continued fraction whose ␦共x − a兲f共x兲dx = ␦共a − x兲f共x兲dx = f共a兲 共C2兲
T-

168 numerators and denominators are defined by recurrence. −⬁ −⬁ 221

169 The approach of Den Iseger is quite different since he proposed ⬁
170 the Gaussian quadrature rule for an accurate approximation of Eq. It, thus, leads to S共t兲 = 兺n=−⬁ h␶共t − nT兲. The causality 共h␶共t兲 = 0 222
when t ⬍ 0兲 leads to the final result
10

171 共A3兲. Without going into details, the time domain solution ap- 223
172 proximation is performed via a linear combination of inverse Fou- ⬁
173
174
rier transforms involving so-called Christophel numbers that de-
fine the ␭ and ␤ parameters 共see Appendix B below兲. Moreover,
S共t兲 = 兺 h␶共t + nT兲 共C3兲
-1

n=0 224
175 this algorithm avoids the location of the singularities, unlike de
However, if one wants to represent the solution on the broad 225
176 Hoog’s approach that needs an estimation of ␥.
time range, it is better to reformulate the previous relation as 226
09

177 Appendix B: MATLAB’s Code for the Iseger Numerical ⬁

178 Inverse Laplace Transform Algorithm S共t + nT兲 = 兺 h␶共z = 0,t + nT兲,
n=0
0ⱕtⱕT 共C4兲
227
179
3]

References 228
180 function 关ft兴 = Den Iseger共Lf, t兲 关1兴 Maillet, D., André, S., Batsale, J.-C., Degiovanni, A., and Moyne, C., 2000, 229
00

181 % Lf: anonymous function for the Laplace transform Thermal Quadrupoles: An Efficient Method for Solving the Heat Equation 230
182 calculation Through Integral Transforms, Wiley, New York. 231
183 % t: column vector of times from 0 to tmax 关2兴 Davies, B., and Martin, B., 1979, “Numerical Inversion of the Laplace Trans- 232
184 % ft: inverse transform 共same size as t兲 form: A Survey and Comparison of Methods,” J. Comput. Phys., 33, pp. 1–32. 233
关3兴 Duffy, D. G., 1993, “On the Numerical Inversion of Laplace Transform: Com- 234
nt= numel共t兲 ; dt= max共t兲 / 共nt− 1兲;
31

185 parison of Three New Methods on Characteristic Problems From Applica- 235
186 % parameter settings tions,” ACM Trans. Math. Softw., 19, pp. 333–359. 236
187 N = 8ⴱnt; a = 44/ N; 关4兴 de Hoog, F. R., Knight, J. H., and Stokes, A. N., 1982, “An Improved Method 237
188 % Numerical values of ␭ and ␤ 共see Ref. 关7兴 p. 29兲 for Numerical Inversion of Laplace Transforms,” SIAM 共Soc. Ind. Appl. 238
02

189 li= 关0, 6.28318530717958, 12.5663706962589, … Math.兲 J. Sci. Stat. Comput., 3, pp. 357–366. 239
190 18.8502914166954, 25.2872172156717, … 关5兴 Gaver, D. P., Jr., 1966, “Observing Stochastic Processes, and Approximate 240
191 34.2969716635260, 56.1725527716607, Transform Inversion,” Oper. Res., 14, pp. 444–459. 241
关6兴 Stehfest, H., 1970, “Algorithm 368: Numerical Inversion of Laplace Trans- 242
192 170.533131190126兴; forms,” Commun. ACM, 13, pp. 47–49. 243
JH

193 bi= 关1, 1.00000000000004, 1.00000015116847, … 关7兴 Den Iseger, P., 2006, “Numerical Transform Inversion Using Gaussian 244
194 1.00081841700481, 1.09580332705189, … Quadrature,” Probability in the Engineering and Informational Sciences, 20, 245
195 2.00687652338724, 5.94277512934943, pp. 1–44. 246
196 54.9537264520382兴; 关8兴 Battaglia, J.-L., Kusiak, A., and Batsale, J.-C., 2007, “Thermal Diffusivity 247
c = 2ⴱiⴱpi/ N ; li= a + i.ⴱli; k = 0 : N ; %i = sqrt共−1兲! Estimation in a Picosecond Photoreflectance Experiment,” ASME J. Heat 248
R

197
Transfer, 129, pp. 756–758. 249

Journal of Heat Transfer APRIL 2011, Vol. 133 / 1-3

PROOF COPY [HT-10-1093] 003102JHR

PROOF COPY [HT-10-1093] 003102JHR
NOT FOR PRINT! FOR REVIEW BY AUTHOR NOT FOR PRINT!

AUTHOR QUERIES — 003102JHR

#1 Au: Please check grammar changes #3 Au: Please check our renumbering of “Eqs.
throughout. 共A1兲–共A5兲 and 共C1兲–共C4兲” as equations must
#2 Au: Please provide the definition of CPU if be cited in numerical order in text.
possible.
PR
OO
F
CO
PY
[H
T-
10
-1
09
3]
00
31
02
JH
R

PROOF COPY [HT-10-1093] 003102JHR

View publication stats