International Journal of Thermal Sciences 139 (2019) 424–432

Contents lists available at ScienceDirect

International Journal of Thermal Sciences

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

A review of the models using the Cattaneo and Vernotte hyperbolic heat T
equation and their experimental validation
Denis Mailleta,b,∗
a
Université de Lorraine, LEMTA, Vandoeuvre-lès-Nancy, France
b
CNRS, LEMTA, Vandoeuvre-lès-Nancy, France

ARTICLE INFO ABSTRACT

Keywords: Models using Cattaneo and Vernotte hyperbolic heat equation or derived from it (Double Phase Lag model and
Non-Fourier their various versions) are very common in the present thermal literature, especially for simulating heat transfer
Cattaneo at the meso scale, such as in bio heat transfer. Such papers refer to so-called experimental validations made in
Vernotte several previous articles. We show in this short review that the corresponding experiments were biased, because
Double phase lag
of a deficient methodological approach based on too simple assumptions or on poor data reduction techniques.
Hyperbolic heat equation
Bio heat transfer

1. Introduction: non-Fourier models for heat conduction q

q + = k T
t (1)
There are presently many papers published in the heat transfer lit- k is the thermal conductivity of the medium while is an extra ther-
erature that deal with « Dual Phase Lag » models or using the Cattaneo- mophysical property of the medium, a relaxation time, specific to this
Vernotte term, an extra hyberbolic term to « correct » the Fourier heat model which is denoted CV hereafter. As a consequence, a local heat
equation. These models were originally designed for very short times balance on a homogeneous medium with no thermal dependance nor
and distances. They are based on a phenomenological adhoc correction velocity, yields the CV heat equation:
for the Fourier constitutive law, that is the definition of the heat flux.
They also seem to be very popular within the domain of bio heat
T
c t
= .q (2a)
transfer.
It will be shown here that papers dealing with this subject refer
( )=k
T 2T
c + 2T
either to models that are not relevant with the corresponding experi- t t2 (2b)
ments or to ill-designed parameter estimation methods. We will first where is the density and c the specific heat of the medium.
present a short review of these models, and then scrutinize in which Let us note that the classical Fourier model corresponds to a zero
conditions the authors claim they have validated them for use in bio relaxation time in equation (1), with a heat equation of parabolic
heat transfer, beginning with the experimental results. nature, that is of the diffusion type. As soon as departs from zero,
The final part of this review will concern a paper devoted to a equation (2b) becomes an hyperbolic partial differential equation, that
Fourier based model for one-dimensional thermal diffusion in materials is a wave equation:
with a non-homogeneous inner structure. This one uses time dependent
1 T 1 2T
conductivities and volumetric heat capacities. + = 2T
a t C2 t2 (3)
1.1. The Cattaneo-Vernotte model where celerity C depends on and on the thermal diffusivity of the
medium a = k / c :
We remind the definition of the heat flux q proposed by Cattaneo
[1] and Vernotte [2]: C = a/ (4)
The Laplace transform of equation (1) yields, for a zero initial
temperature field:

∗
Université de Lorraine, LEMTA, Vandoeuvre-lès-Nancy, France.
E-mail address: [email protected].

https://doi.org/10.1016/j.ijthermalsci.2019.02.021
Received 2 November 2018; Received in revised form 11 February 2019; Accepted 12 February 2019
Available online 27 February 2019
1290-0729/ © 2019 Elsevier Masson SAS. All rights reserved.
D. Maillet International Journal of Thermal Sciences 139 (2019) 424–432

q¯ =
k
T¯ restriction for the values of the two time lags T q , because, in the
1+p
opposite case, equation (7) would violate the causality principle with a
where ¯ (p ; r ) L [ (t , r )] 0
(t , r )exp( pt )dt (5) flux at a given time that would depend on a temperature at a later time.
In equation (5), the upper bar on any function of time t designates
its Laplace transform also noted L [ ], variable p being the Laplace 2. Review of experimental papers trying to validate non-Fourier
parameter and r being the vector of space coordinates. models for bio heat transfer
Returning to the time domain, a convolution product appears in the
CV definition of the heat flux q , which denotes a memory effect for the The experiment-based papers we have decided to review next tried
system: to validate a non-Fourier model for different classes of material sam-
ples. It will be shown that they suffer from serious flaws that do not
t
q (t , r ) = 0
f (t t) T (t , r )dt (6a) allow them to reach their objective.

with f (t ) =
k
exp ( ) t
(6b)
2.1. Kaminski [6]

This CV heat equation has been introduced because of the ap- This paper deals with the CV model given in equations (3) and (6a),
parent « paradox » of an infinite value for the propagation velocity C , with an attempt to estimate the relaxation time from temperature
especially for « short times » after the initial thermal excitation of a measurements in materials with « nonhomogeneous inner structure »,
material, while the Fourier heat equation would remain valid for long without any clear definition of this term. The morphological structure
times only. of the tested samples seems to be granular porous beds of given void
To our knowledge, the CV heat equation has never been validated fraction (« H acid », sodium bicarbonate, sand, glass ballotini, « ion
experimentally on classical solid homogeneous materials. Some ex- exchanger »). The setup, see Fig. 1, consisted of a sample containing a
perimental works [3,4] showed that for very short times (10−12 to line electrical heater and a thermocouple located in a needle parallel to
10−13 s, that is for pico second laser excitations or shorter durations), it, at a distance xTC .
and consequently for very small distances, the Fourier heat equation is This distance was equal to 6.8 and 16.8 mm in the first and second
not valid any more and 2 temperatures (electrons and lattice) have to experiments respectively. In order to estimate the relaxation time, the
be introduced to describe the transfer of heat. author introduced an ill-defined penetration time tp , corresponding to
the time for the temperature signal to depart from its initial value be-
1.2. The dual phase lag model fore the start of the excitation, that was supposed to be a step thermal
power in time. Once this penetration time has been measured, the re-
Another non-Fourier constitutive equation, the dual phase lag laxation time was calculated using the celerity C of the « heat wave »
model (DPL), has been introduced in 1995 by J. Zhou [5]. Its con- and equation (4):
stitutive equation is:
xTC = C tp = a tp2/ xTC
2
(12ab)
q (t + q, r) = k T (t + T, r) (7)
This required the determination of the diffusivity that was estimated
It involves two time lags, one for the flux, q and the other for using the Fourier model for a line heat source, which is not consistent
temperature, T . In fact J. Zhou made a first order Taylor expansion with the purpose of the author: the two parameters and a should have
with respect to q in the left term of equation (7) and with respect to T been estimated using the solution of CV model corresponding to this
in its right term, to obtain the linearized DPL model: geometry and excitation only.
q T The authors measured values of ranging from 11 to 54 s for the
q (t , r ) + = k T (t , r ) + (t , r ) considered samples but did not show the corresponding measured
q
t
T
t (8)
temperature responses in this paper.
This development was made with the assumption that both q and T The major bias in these experiments results from the definition of
are small (and, implicitely, of the same order of magnitude) but Zhou the penetration time tp that is specific to the location of the
did not mention how small they should be: it should be small with a
reference quantity.
It is easy to show, using a Laplace transformation of (8):
1+p T
q¯ = k T¯
1+p q (9)
Returning to the time domain, a convolution product appears in the
constitutive equation of the DPL model:
q (t , r )
T t
= k T (t , r ) 0
g (t t) T (t , r )dt
q

T t
with g (t ) = k 1 exp
q q (10)
The corresponding DPL heat equation can be derived from the local
heat balance (2a) and from equation (8):

T 2T T
2T 2 Fig. 1. Setup used by Kaminski [6]: (1) resistance wire, (2) needle, (3) electric
c + q = k + k T
t t2 t (11) insulation, (4) needle, (5) thermocouple, (6) insulation, (7) support, (8) to
power supply and stabilizer, (9) to amplifier and recorder, (10) material.
Of course, the Fourier model is recovered if q = T = 0 , which Reprinted by permission from: [ASME] [Journal of Heat Transfer] [Hyperbolic
was already apparent in equation (7) while the CV model corresponds heat conduction equation for materials with a non-homogeneous inner struc-
to T = 0 . Let us remark that it should be natural to have the following ture, W. Kaminski] (1990).

425
D. Maillet International Journal of Thermal Sciences 139 (2019) 424–432

thermocouple. As there is an asymptotic zero level for time going to Many questions arise, first concerning the choice of the initial time:
zero in the Fourier solution, tp is the time where the signal gets larger the experimental procedure of bringing two pieces of meat into contact
than its measurement noise, that is a threshold that can be quantified by was not described and the contact model used was too simple since the
its standard deviation. So the penetration time depends on the level of samples were soft materials whose thicknesses and structures probably
the forced excitation, the step of thermal power here: an increase of this varied once a pressure had been applied. In the same way the positions
power will allow the temperature signal to gets larger than the noise of the thermocouple junctions may have moved during this mechanical
level at an earlier time, with a corresponding decrease for tp . operation. Another question arises: the cold samples were removed
Other biases are the definition of the time t = 0 that depends on the from a cold chamber before the experiment. In this lapse of time an
inertia of both the heater and of the thermocouple sheaths and on the internal temperature gradient may have built up inside: this could have
uncertainty in their exact locations, together with the intrusive char- been checked by recording the temperatures before the experiment, in
acter of the thermocouples. These points have not been discussed in the order to see whether a corresponding relaxation appeared in the
paper, as well as other possible sources of error: thermal contact re- monitored samples.
sistances between both actuator and sensor with the porous bed, va-
lidity of the 1D model that is used for estimating the diffusivity (3D 2.3. Roetzel et al. [7]
conduction can take place because of the finite length of the resistance).
Several weaknesses of this paper were pointed out in Refs. [7,8]. The first two parts of this article are analyzed here: it contains first a
balanced review of 4 papers of the literature on the CV model and on
2.2. Mitra et al. [9] the attempts made for its experimental validation/invalidation. The
first two [6,9] are analyzed above, while the last two conclude on the
This paper also deals with an attempt to validate the CV model on validity of Fourier law and on the absence of hyperbolic effects after
processed meat samples. The principle of the first experiment consisted experiments on wet sand [10], and on dry sand and processed meat [8].
in bringing into contact 2 identical meat samples initially at two dif- We sum up here the main results of these last two papers before re-
ferent temperatures (room temperature and a cold temperature), with 2 turning to the analysis of the last part of [7].
thermocouples embedded in the cold sample at 2 different distances
from the contact plane, see Fig. 2a. 2.3.1. Graβman and Peters [10]
Three other similar contact experiments involving 3 meat samples Graβman and Peters [3] constructed an instrumented cylindrical cell,
or an aluminium plate have also been made, see Fig. 2a and b. So, with PVC walls and lids, containing the porous material for which the
contrary to Kaminski's experiment (forced thermal regime generated by validity of the Fourier law had to be tested. It was heated by Joule effect
a step in thermal power), the transient character of the recorded tem- in a metallic wire located on the axis of the cell and with NTC thermis-
perature signal stemmed from a relaxation of the temperature field (a tances located at different radii in the central horizontal plane in Fig. 3.
free regime) in each of the materials brought into contact. So, in this plane, a corresponding analytical model was found: it was 1D,
The relaxation times (of the order of 15 s for the 4 experiments) in the radial direction, because of angular symmetry, and transient. It
were measured exactly in the same way as in Ref. [6], that is using a took into account the inertia of the heated wire and linear heat losses with
penetration time. So, they suffer from the same bias: penetration times the lateral wall and the outside environment. It was verified using a finite
depend both on the measurement noise and on the initial temperature difference code. Experiments were made with wet sand, with calculation
difference between the two bodies that are brought into contact. of the volumetric heat capacity of the porous medium by a mixing law
The authors plotted together the thermal responses of the tem- and measurement of its equivalent conductivity by a 1D steady state
perature sensors and the Fourier theoretical responses valid for in- experiment in the same cell. The authors used a Joule heating composed
stantaneous contacts between semiinfinite media, with an important of a succession of door functions, with relaxation regimes in between (on/
time lag for the measurements. off switchings). The simulated temperature curves and the experimental
one overlap very well, once the previously estimated conductivity has
been modified, in order to take heat losses through the lids into account:
these could not be considered in the previous 1D steady experiment. No
phase lag appeared, even for the fastest temperature responses.

Fig. 2. Setup used by Mitra et al. [9] (a) Schematic of the experimental con-
ditions for Experiments I, II and II (b) Schematic of the experimental conditions
for Experiment IV Reprinted by permission from: [ASME] [Journal of Heat Fig. 3. Setup used by Graβman and Peters [10] Reprinted by permission from:
Transfer] [Experimental evidence of hyperbolic heat conduction in processed [Springer] [Heat and Mass Transfer] [Experimental investigation of heat con-
meat, K. Mitra, S. Kumar, A. Vedavarz, M.K. Moallemi] (1995). duction in wet sand, A. Grassman and F. Peters] (1999).

426
D. Maillet International Journal of Thermal Sciences 139 (2019) 424–432

Fig. 4. a,b – Setup used by Herwig and Becker [8]. (a) Sketch of the experimental arrangement. (b) Geometrical details of the experimental arrangement Reprinted by
permission from: [Springer] [Heat and Mass Transfer] [Experimental evidence about the controversy concerning Fourier or non-Fourier heat conduction in materials
with a nonhomoneous inner structure, H. Herwig, K. Beckert] (2000).

was approached in Mitra et al. [9] than to a uniform flux (or more
precisely a uniform volumetric generated power inside a thermal ac-
tuator) as in the preceding work [10]. Two thermocouples were located
in the material and an infrared camera was able to monitor the 2D
temperature field of its upper free surface, see Fig. 4b.
The authors used processed meat and dried sand as filling material
in the box. The thermocouple responses did not exhibit any time lag
once the water flow valve was open. The same was true for the tem-
perature measured by the infrared camera at 3 different axial locations
on the free surface. The thermal diffusivity was estimated, for both dry
sand and processed meat, in order to minimize the temperature re-
siduals (curve fitting) between the two thermocouple measurements
and the corresponding output of a numerical simulation code. The es-
timated diffusivities, for processed meat, were consistent with the dif-
fusivities used in Mitra's study [9].

2.3.3. Experimental study of Roetzel et al.’s paper [7]

In the second part of Roetzel et al.'s paper [7], the authors con-
Fig. 5. Setup used by Roetzel et al. [7]: the test cell for experiment.
structed a vertical cylindrical box, see Fig. 5.
This box was filled with the material to be characterized, and the
2.3.2. Herwig and Beckert [8] bottom face was stimulated by a Peltier element (4) in harmonic re-
Herwig and Beckert [8] tested the Fourier law in a box filled with gime. Its top surface was cooled by another Peltier cell (8), using the
the material of interest, se Fig. 4a. A copper pipe was set below the free same thermostatic bath as the cooling water of the Peltier cell. Parallel
upper surface. At time t = 0, a valve was open and water started to flow thermocouples (5) were inserted in planes normal to the axis, with hot
at a temperature lower than the initial temperature in the system. So, junctions at different axial positions. The authors constructed the ana-
the thermal excitation was closer to a uniform temperature jump, which lytical steady harmonic solution (complex temperature) of the CV heat

427
D. Maillet International Journal of Thermal Sciences 139 (2019) 424–432

equation model in a 1D semi infinite medium, where the two para-

meters, the relaxation time and the diffusivity a , were present in both
the amplitude ratio (attenuation) and in the phase shift together with
the angular frequency of the source. They used a data acquisition
system to measure the attenuation B and the phase shift G of the signal
and a closed form formula allowed them to retrieve both parameters
using one frequency and the output of two thermocouples. They got
values of close to 2 s for both dry sand and processed meat.
The following remarks can be made about these measurements and
their parameter estimation part:

1) The model used is based on a 1D heat transfer assumption in the

sample to estimate its two parameters, using the well known
Angström bar method [11] for measuring thermal diffusivity. This
method relies on a very strong assumption, that is the absence of
losses in the transverse direction, and is not used for estimating the
thermal diffusivity anymore. However, no thermocouple was in-
serted with off axis junctions to check the effect of a 2D transfer
caused by simultaneous diffusion in the radial direction. This could
have happened here because the metallic sheathed thermocouples
were set in a direction perpendicular to the main (axial) tempera- Fig. 6. Setup used by Andrä et al. [13] and investigated by Liu and Chen [14]
ture gradient, contrary to the arrangement made by Graβman and (a) Scheme of localized magnetic hyperthermia applied to a breast carcinoma.
Peters [10]. So they could have offered a preferential path for heat The tumor region contains injected magnetic particles, from Ref. [13]. (b) Il-
lustration for model comparison, from Ref. [14].
conduction in the radial direction. So this model bias has probably
happened in the case of process meat since the characteristic 1D
diffusion time based on the radius of the sample R = 20 mm is conductivity of the heated region. The authors concluded that there was
tc = 0.132 R2/ a = 400 s [12] for a period of the harmonic source a agreement between their temperature measurements and the outputs
shown in the paper close to this value (about 7 mn). of their model, which was based on the classical Fourier law.
2) No temperature residual plot, showing the difference between the Liu and Chen [14] revisited the measurements of Andrä et al., in
experimental signal and its recalculated values using the analytical order to compare them to the output of a DPL heat transfer model. Since
model fed with the estimated parameters has been shown: it would this model needed 2 lag times ( q and T ) per region, that is 4 para-
have been useful to check the 1D hypothesis at least. In a similar meters, a sound inverse technique would have consisted in im-
way, replication of the estimation for different pairs of sensors plementing a non linear least square parameter estimation technique to
would have allowed to check the relevance of the model. all the transient temperature measurement points of [13], that is the 44
3) The value of 2 seconds that was found for is two orders of magni- measurement made in the non heated region. However, the authors
tude smaller than the period of the excitation, so a sound error ana- used only 4 transient measurements per radius, to get one set of 4 re-
lysis should have been made to assess the values of its uncertainty. laxation time lags per radius.
The error analysis given in equations (20) and (21), whose numerical They used therefore what is called « exact matching » in statistical
values were not given in the paper, is deficient for two reasons: the parameter estimation [15,16], which means that they solved 4 equa-
relative errors in B and G are correlated since they stem from the tions with 4 unknowns per radius. This resulted in residuals equal to
same measurements and the authors forgot to transform the minus zero for each radius: there was no room left for the effect of measure-
sign into a plus sign when they passed from the logarithmic differ- ment noise, for the error caused by uncertainty in the « clas-
ential of a ratio to the corresponding relative error (if z = y / x , then sical » thermophysical properties of both materials as well as for the
z /z = y / y + x / x and not z /z = y / y x / x ). error on the precise location of the thermocouples. Under these con-
ditions, no wonder the DPL model outputs went precisely through the
2.4. Interpretation of the experiments of Andrä et al. (13) by Liu and Chen experimental points.
[14] The authors found q between 7.4 and 8.9 s and T between 14.5 and
21.4 s. Let us note that these values violate the causality principle
Andrä et al. [13] made transient temperature measurements in a (T q ) already mentioned at the end of section 1 above.
medium composed of two regions: a cylindrical region (carrageenan, a
sulfated polysaccharide) containing magnetic particles heated by an 2.5. Jaunich et al. [17]
alternating magnetic field, that was embedded in an extended muscle
tissue from cow. This system was used to model magnetic hyperthermia These researchers, which were interested in laser induced hy-
therapy for small breast carcinomas, see Fig. 6a. Thermocouples, whose perthermia made experiments with single- and multi-layer tissues
junctions were located in both regions, allowed transient temperature phantoms that simulated skin, with inhomogeneities simulating a
measurements, see Fig. 6b. tumor, and freshly excised mouse skin tissue samples. These phantoms
A corresponding Fourier type temperature field has been modeled, were composed of different polymer resins with India ink as a radiation
where the heated region (a step heating) was given a spherical shape. absorber and Titanium dioxide particles as scatterers. Inhomogeneities
Comparison between the measured and modeled 1D (radial) transient have been drilled in the tissue phantoms, but no information was given
temperatures were made, with small differences during the transient about their composition in the paper. A focused laser beam, as well as a
phase only. The authors explained the systematic deviations (measured collimated beam, was used in order to obtain a higher required tem-
temperatures lower than simulated ones) by errors in the thermocouple perature rise at the region of interest for the former type of excitation,
positions. Another source of error, apart from the spherical shape as- see Fig. 8.
sumption that can modify the temperature field for short times, can The temperature was measured axially by embedded thermocouples
stem from the different thermopysical properties of the two materials while an infrared camera was said to give some informations about its
that were taken from the literature, with a series model for the radial distribution.

428
D. Maillet International Journal of Thermal Sciences 139 (2019) 424–432

Both the Fourier and Cattaneo Vernotte forms of the heat equation
were solved numerically, with the volumic source term calculated
through the solution of the radiative transfer equation (transient dis-
crete ordinate method), with absorption and scattering, but without
emission, for a given distribution of the absorbed light of the laser
beam.
The authors showed simulated radial temperature and axial dis-
tributions and compared them with several in and off-axis point mea-
surements. This was made for discrete values of the unique relaxation
time τ of the CV model (from 10 to 20 s) or of the Fourier model
(τ = 0 s). However, the way the off-axis points have been measured is
unclear: the presence of corresponding thermocouple positions is not
given and, transformation of the infrared camera signal into a local
inner, and not surface temperature, for rather absorbing materials in its
spectral interval [8–14 μm] is not explained, even if this camera seems
to have been used for measuring temperatures in the top circular sur-
face of the tissue. The corresponding temperature histories are also
given for different values of the irradiation times (8–17 s). In all pre-
sented figures, the temperatures output of the Fourier model ly below
the CV corresponding simulations and below the experimental points.
However, the fit (in terms of shape of the space or time distributions)
between CV and measured temperatures is not really good, which is not
surprising because of the large number of not precisely known para-
meters present in the numerical model.

2.6. Sahoo et al. [18]

These authors made an experiment where a basin containing a na- Fig. 7. Setup used by Jaunich et al. [17]: Schematic of tissue phantom con-
noparticle (gold « mesoflowers ») embedded collagen gel was heated by taining inhomogeneity (focused beam technique.
a near infrared laser, see Fig. 8.
Several 1 mm sheathed thermocouples were inserted in- and off-axis
and the step heat source space distribution in the simulation model took
into account absorption and scattering of the laser beam through a
simple Beer's type law. The transient temperature signal of the ther-
mocouples were compared to the corresponding output of the numer-
ical solution of different versions of the DPL heat transfer model, see
equation (2a) with an extra volumetric heat source and (7). Its first
order linear version (8) was implemented, as well as higher order series
expansion of the two terms of (7), the values of the two lag times q and
T being derived from Ref. [14].
It should be noted that this model seems to have been solved in
rectangular 2D coordinates, while the axis of symmetry of the laser
beam should have defined the axisymmetrical cylindrical system of
coordinates of the right numerical solution of direct problem.
Fig. 8. Setup used by Sahoo et al. [18]: Schematic diagram of experimental
The authors found that the Fourier model underestimated the ex- setup (the full arrows of the original figure were connected to photographs of
perimental points more than the simpler linearized DPL model. the basin, of the laser and of its controller).
However higher order versions of the DPL model, with different forms
of the laser heat absorption and scattering model and different values of
3. Further remarks on the validity of the Fourier law for heat
the 4 penetration times could better fit the experimental points.
diffusion at the meso-scale
Regarding the experiment, a possible bias was derived from the semi
transparent character of the medium used, with a possible different ab-
The above review has shown that non-Fourier models have not been
sorption of the laser radiation by the thermocouple junctions that may
validated at the meso-scale for heat transfer in heterogeneous materials.
create a local temperature difference with the medium at short times.
One can wonder whether Fourier's law applied to each of its constituents
Regarding the model output, there are many « supposed to be
could lead to a global behaviour of the corresponding system that differs
known » parameters in the different forms of the DPL model used: the
from that of an equivalent classical equivalent homogeneous medium
heat exchange coefficient at the free surface of the basin, the thermal
having a given uniform volumetric heat and conductivity. A response is
diffusivity and volumetric heat of the homogenized composite medium,
possible for transient heat transfer in a multilayer stack.
the exact location of the thermocouples, the 2 DPL time lags, the 3
If the material is not homogeneous, but 1D periodical, it has been
optical parameters of the heat generation term, the intensity of the
shown that it can be replaced by an equivalent homogeneous material
laser, whose increase would bring the « classical Fourier based Pennes
that obeys Fourier's law, as soon as the number of unit cells is large
Model » at the same location as the experimental points in Fig. 7 of the
enough [19].
paper. In a similar way, the very simple solution of the coupled con-
If it is not the case, a non local approach has shown analytically, in
duction-radiation problem, that does not use any radiative transfer
the same paper, that a stack composed of a limited number of different
equation while some scattering is present, could be questioned.
materials can be replaced by an equivalent material system, in terms of
So, evidence of a non-Fourier behaviour is impossible to be brought
inlet and outlet temperatures and fluxes.
to the fore by this paper.

429
 D. Maillet

Table 1
Synthesis of experiment-based papers that claim to validate a hyperbolic model (CV or DPL).
First author, paper number and Tested materials Excitation and temperature measurement Principle of validation test of CV or DPL model Main remarks
year

Kaminski [6] 1990 sand and various materials linear Joule heating and transient thermocouple observation of the time lag in the measured The observed lag time depends on the levels of the
responses at an off-axis location temperature versus time curve to retrieve the CV excitation and on the measurement noise.
relaxation time
Mitra [9] 1995 processed meat Instantaneous contact between materials at observation of the time lag in the measured Same remarks as in Ref. [6]. The initial temperature field in
different initial temperatures with thermocouple temperature versus time curve to retrieve the CV each material is not controlled and the latter deformations
measurements relaxation time of the system are not taken into account.
Roetzel [7] 2003 processed meat, synthetic sand, Angström bar type experiment with planar Peltier non linear estimation of both CV relaxation time The 2D effects are not studied nor taken into account in the
sodium bicarbonate and various heating and cooling and thermocouple and thermal diffusivity using amplitude ratio model.
porous materials under variable gas measurements (harmonic regime) and phase shift The error analysis is deficient.

430
pressure
Liu [14] 2010 (discussion of muscle tissue from cow heating by a radio frequency field in a spherical estimation of the 2 phase lags per region of the Non consistent estimation technique: the phase lags
Andrä [13] 1999 region containing magnetic particles and transient 2D (radial and axial) DPL model to get the best depend on the chosen measurement radii (exact matching)
experiments) off- axis thermocouple responses in outer region fit for the measured temperatures
Jaunich [17], skin phantoms and mouse skin tissues laser beam heating and in and off-axis adjustment of the relaxation time of the 2D The final temperature solution of the CV heat equation uses
2008 thermocouple and infrared camera measurements (radial and axial) CV model to get the best fit for 2 other optical models (laser beam time and space
the measured temperatures variation, radiative transfer equation) for 2 or 4 regions
(case of the 3 layer phantom with 1 embedded
heterogeneity) which depend on many not precisely known
parameters that are retrieved from the literature. So, an
extensive sensitivity study is lacking.
Sahoo [18] collagen gel with embedded gold laser beam heating and in and off-axis adjustment of the 2 phase lags of various order Same type of remark as for [17]. A possible different
particles thermocouple measurements expansions of the 2D (radial and axial) DPL absorption of the laser beam by the thermocouple
model to get the best fit for the measured junctions, leading to biased temperature measurements, is
temperatures not considered.
International Journal of Thermal Sciences 139 (2019) 424–432
D. Maillet International Journal of Thermal Sciences 139 (2019) 424–432

This equivalent material is characterized by a heat flux that is de- between model outputs and measurement points deserve to be analyzed
fined through a convolution product, different from equations (6) and with a very critical hindsight.
(10), between a time dependent conductivity and a time derivative of We also note that two experimental studies detailed above, Herwig
the temperature gradient if the stack of thickness e presents a material and Beckert [8], for processed meat and dried sand, and Graβman and
symmetry with respect to the x = e/2 plane, that is if inlet (x = 0 ) and Peters [10], for wet sand, with two completely different types of
outlet ( x = e ) can be switched. In that case, the constitutive law, thermal excitation and geometry, and where no obvious model or
purely based on Fourier heat transfer in each local internal layer of the measurement bias appeared, concluded that the Fourier law explained
stack is: their measurements in a satisfactory way.
t 2T For a more general point of view, a first conclusion of this review is
q (t , x ) = k (t t) ( t , x ) dt that before proposing a new heat transfer model and to « measure » its
0 x t (13)
coefficients, it is compulsory to prove that the classical model cannot
with the conductivity function k (t ) depending not only on time t but explain the experimental measurements.
also on the thermopysical properties and thicknesses of each layer of Another conclusion is that the experiment allowing to compare
the stack. these different models on a sound basis remains to be made and the
The related heat equation is: points presented above can be used as a roadmap for following a rig-
t 2T t 3T orous procedure.
c (t t ) (t , x ) dt = k (t t ) (t , x ) dt Let us finally note that Fourier's constitutive law has proved to be
0 t2 0 x2 t
remarkably robust in many fields of application, provided that the right
(14)
measurements and their critical analysis have been made.
with the volumetric heat capacity function c (t ) depending on the
same quantities as k (t ). Acknowledgements
Let us note that these equations only govern the temperature of the
equivalent material and yield the exact temperatures and fluxes at the The author wants to thank the American Society of Mechanical
boundaries of the domain ( x = 0 and outlet x = e ) only. Engineers for its permission to reproduce Fig. 1 from [Hyperbolic heat
In the case of a non symmetrical stack of layers, an additional conduction equation for materials with a non-homogeneous inner
convolutive term has to be added to the constitutive equation: structure, W. Kaminski, Journal of Heat Transfer 112 (1990)] and
t 2T Fig. 1a and b from [Experimental evidence of hyperbolic heat con-
q (t , x )= 0
k (t t) x t
(t , x )dt duction in processed meat, K. Mitra, S. Kumar, A. Vedavarz, M.K.
t
(t t)
T
(t , x )dt Moallemi Journal of Heat Transfer 117 (1995)].
0 t (15)
The author wants to thank Springer Nature for its permission to
where function (t ) plays the role of a pseudo conduction velocity. reproduce Fig. 1 from [Experimental investigation of heat conduction in
wet sand, A. Grassman and F. Peters, Heat an Mass Transfer 35 (1999)]
4. Synthesis of the experimental results and conclusion and Fig. 2 an 3 from [Experimental evidence about the controversy
concerning Fourier or non-Fourier heat conduction in materials with a
In the above review, we have shown that all the experiment-based nonhomoneous inner structure, H. Herwig and K. Becker, Heat and
papers that aimed at validating a non-Fourier model (CV or DPL) for Mass Transfer 36 (2000)].
different classes of material samples presented many flaws and meth-
odological biases. The main features of these papers are summed up in Appendix A. Supplementary data
Table 1 below.
These flaws are listed below, in a non exhaustive way: Supplementary data to this article can be found online at https://
doi.org/10.1016/j.ijthermalsci.2019.02.021.
- experimental boundary conditions that were not validated by direct
simulation (neglected losses with unvalidable 1D assumptions), and References
uncertain quantification of the experimental source term,
- origin of time (start of the thermal source creating the transient [1] C. Cattaneo, Sur une forme de l’équation de la chaleur éliminant le paradoxe d’une
transfer) not well documented for the experimental configuration, propagation instantanée (A form of heat conduction equation which eliminates the
paradox of instantaneous propagationh), Comp. Rend. Hebd. Séances Acad. Sci.
- assumption of a uniform initial temperature field that was not Paris 247 (4) (1958) 431–433.
confirmed by a temperature recording before time t = 0 , [2] P. Vernotte, Les paradoxes de la théorie continue de l’équation de la chaleur
- errors in the « imported parameters », derived from the values of the (Paradoxes in the continuous theory of the heat equation), Comp. Rend. Hebd.
Séances Acad. Sci. Paris 246 (1958) 3154–3155.
literature (thermophysical properties) that were not taken into ac- [3] J.-L. Battaglia, A. Kusiak, C. Rossignol, N. Chigarev, Thermal diffusivity and effu-
count, sivity of thin layers using time-domain thermoreflectance, Phys. Rev. B 76 (12
- ignorance of modern parameter estimation techniques, based on November 2007) 184110.
[4] S.I. Anisimov, B.L. Kapeliovich, T.L. Perel’man, Electron emission from metal sur-
least squares and analysis of the sensitivity coefficients, with ab-
faces exposed to ultrashort laser pulses, Zh. Eksp. Teor. Fiz. 66 (1974) 776 [Sov.
sence of calculation of the matrix of variance-covariance of the Phys. JETP 39, 375 ,1975].
parameters that are looked for [20,21]. This can lead to excellent [5] D.Y. Tzou, A unified field approach for heat conduction from macro to Microscales,
ASME J. Heat Tran. 117 (1995) 8–16.
residuals but to biased estimations because of non linearly in-
[6] W. Kaminski, Hyperbolic heat conduction equation for materials with a non-
dependent sensitivity coefficients, especially if the number of homogeneous inner structure, ASME J. Heat Tran. 112 (1990) 555–560.
parameters that are looked for is numerous, [7] W. Roetzel, N. Putra, S.K. Das, Experiment and analysis for non-Fourier conduction
- non-existent or deficient analysis of the errors made in the reduction in materials with non-homogeneous inner structure, Int. J. Therm. Sci. 42 (2003)
541–552.
of the experimental measurements. [8] H. Herwig, K. Beckert, Experimental evidence about the controversy concerning
Fourier or non-Fourier heat conduction in materials with a nonhomoneous inner
So, all the above points have to be checked, in order to make structure, Heat Mass Transf. 36 (2000) 387–392.
[9] K. Mitra, S. Kumar, A. Vedavarz, M.K. Moallemi, Experimental evidence of hyper-
thermal measurements and corresponding models agree using unbiased bolic heat conduction in processed meat, ASME J. Heat Tran. 117 (1995) 568–573.
data reduction techniques. This is particularly true for bio heat transfer, [10] A. Graβmann, F. Peters, Experimental investigation of heat conduction in wet sand,
where measurements are difficult because of the specific nature of the Heat Mass Transf. 35 (1999) 289–294.
[11] H.S. Carslaw, J.C. Jaeger, Conduction of Heat in Solids, second ed., Oxford Science
tissues involved. As a consequence, the analysis of the deviations

431
D. Maillet International Journal of Thermal Sciences 139 (2019) 424–432

Publications, Clarendon Press, Oxford, 1959, pp. 136–139. of short pulse laser interaction with tissue phantoms containing inhomogeneities,
[12] W.J. Parker, R.J. Jenkins, C.P. Butler, G.L. Abbot, Flash method for determining Int. J. Heat Mass Transf. 51 (2008) 5511–5521.
thermal diffusivity, heat capacity and thermal conductivity, J. Appl. Phys. 32 [18] N. Sahoo, S. Ghosh, A. Narasimhan, S.K. Das, Investigation of non-Fourier effects in
(1961) 1679–1684. bio-tissues during laser assisted photothermal therapy, Int. J. Therm. Sci. 76 (2014)
[13] W. Andrä, C.G. d'Ambly, R. Hergt, I. Hilger, W.A. Kaiser, Temperature distribution 208–220.
as function of time around a small spherical heat source of local magnetic hy- [19] C. Moyne, A. Degiovanni, Représentation non locale d’un milieu hétérogène en
perthermia, J. Magn. Magn. Mater. 194 (1999) 197–203. diffusion pure (Non local representation of pure diffusion in a heterogeneous
[14] K.-C. Liu, H.-T. Chen, Investigation for the dual phase lag behavior of bio-heat medium), Int. J. Therm. Sci. 42 (2003) 931–942.
transfer, Int. J. Therm. Sci. 49 (2010) 1138–1146. [20] Société Française de Thermique, Metti 5, Thermal measurements and inverse
[15] J.V. Beck, K.J. Arnold, Parameter Estimation in Engineering and Science, Wiley, techniques, Advanced Spring School, CNRS & Eurotherm Seminar 94, Roscoff, June
New York, 1977. 13-18, 2011 Proceedings (Open Access): http://www.sft.asso.fr/document.php?
[16] B. Rémy, S. André, Nonlinear estimation problems, in: Helcio R.B. Orlande, pagendx=12299.
Olivier Fudym, Denis Maillet, Renato M. Cotta (Eds.), Thermal Measurements and [21] Société Française de Thermique, Metti 6, Thermal measurements and inverse
Inverse Techniques, Publisher: CRC Press, Taylor & Francis Group, Bocca Raton, techniques, Advanced Winter School, CNRS & Eurotherm Seminar 104, Biarritz,
USA, May 09, 2011, pp. 315–354. France, March 1-6, 2015 Proceedings (Open Access): http://www.sft.asso.fr/metti-
[17] M. Jaunich, S. Raje, K. Kim, K. Mitra, Z. Guo, Experimental and numerical analysis 6.html.

432