Ultrasonics 56 (2015) 308–317

Contents lists available at ScienceDirect

Ultrasonics
journal homepage: www.elsevier.com/locate/ultras

Ultrasonic broadband characterization of a viscous liquid:

Methods and perturbation factors
Nacef Ghodhbani, Pierre Marechal ⇑, Hugues Duflo
Laboratoire Ondes et Milieux Complexes, LOMC UMR 6294 CNRS, Université du Havre, 75 rue Bellot, 76600 Le Havre, France

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

Article history: The perturbation factors involved in ultrasonic broadband characterization of viscous fluids are analyzed.
Received 20 May 2014 Precisely, the normal incidence error and the thermal sensitivity of the properties have been identified as
Received in revised form 30 July 2014 dominant parameters. Thus, the sensitivity of the ultrasonic parameters of attenuation and phase velocity
Accepted 12 August 2014
were measured at room temperature in the MHz frequency range for two reference silicone oils, namely
Available online 6 September 2014
47V50 and 47V350 (Rhodorsil). Several methods of characterization were carried out: time of flight, cross-
correlation and spectral method. These ultrasonic parameters are measured at room temperature. For this
Keywords:
family of silicone oil, the dispersion of the attenuation spectrum is modeled by a power law. The velocity
Ultrasonic characterization
Perturbation factors
dispersion is modeled by two dispersion models: the quasi-local and the temporal causal. The impact of
Dispersion relation the experimental reproducibility of the phase velocity and acoustic attenuation was measured in the
Phase velocity MHz frequency range, using a set of ultrasonic transducers with different center frequencies. These mea-
Attenuation surements are used to identify the dispersion of the ultrasonic parameters as a function of the frequency.
A first experimental and descriptive approach is developed to assess the reproducibility of the normal
incidence between the acoustic beam and the viscoelastic material. Thus, the relative error on the mea-
surements of velocity and attenuation are directly related to the angular deviation of the ultrasonic wave,
as well as the sampling and signal-to-noise ratio. A second experimental and phenomenological approach
deals with the effect of a temperature change, typical of a polymerization reaction. As a result, the
sensitivity of the phase velocity of silicone oil 47V50 was evaluated around 2 m s1 K1.
Ó 2014 Elsevier B.V. All rights reserved.

1. Introduction
Ultrasonic techniques in non destructive testing (NDT) or non-
destructive evaluation (NDE) have proven to be useful in several
material evaluation areas. The measurements of velocity and
attenuation dispersions are the most important factors in the mon-
itoring of several properties such as ageing of a structure [1],
viscoelastic behavior [2], polymer cure [3–10] and porosity charac-
terization [11]. Reproducibility at normal incidence and tempera-
ture sensitivity has an important effect on dispersion acoustic
parameters. In this paper, a measurement method is proposed for
the ultrasonic characterization of a fluid layer by several methods:
time of flight, cross-correlation and spectral method. Their impact
is studied in the context of an application on a production monitor-
ing of composite plates by RTM. The dispersion in attenuation
characteristic is modeled by power law [12–14]. The velocity
dispersion is modeled by Szabo [15] and O’Donnell et al. [16]. In
normal incidence, the error of distance measurement, the relative
⇑ Corresponding author.
E-mail address: pierre.marechal@univ-lehavre.fr (P. Marechal).
error of time due to the sampling measurement and the relative
error of time due to the signal–noise ratio (SNR) present a major
impact on acoustic parameters. Furthermore, the sensitivity of
the phase velocity to temperature is evaluated as an essential
parameter. The article is divided into three parts: the first presents
the equations for measuring the acoustic properties; the second
part describes the proposed experimental setup and the results
of reproducibility of the normal incidence and the sensitivity to
temperature. In the third part, acoustics parameters were mea-
sured using several methods for two reference silicone oils by dif-
ferent center frequency transducers in the MHz range. Viscosity is
evaluated using the Cole–Cole model [17].
2. Theory
2.1. Velocity and attenuation calculations
In order to characterize ultrasonic velocity and attenuation in
the fluid samples, we placed the fluid in a tank. The measurements
were carried out at normal incidence which corresponds to the

http://dx.doi.org/10.1016/j.ultras.2014.08.013
0041-624X/Ó 2014 Elsevier B.V. All rights reserved.
 N. Ghodhbani et al. / Ultrasonics 56 (2015) 308–317 309

maximum acoustic energy signal. Considering the first round-trip 2.3. Cross-correlation method
signals s1(t) and s2(t), measured at two positions dref and dref + Ddoil
respectively (Fig. 1(a)), the arrival times and maximum amplitudes The cross-correlation technique is commonly used to calculate
of the first round-trip echoes are denoted as (t1, A1) and (t2, A2), time delay dtcor between two signals supposed to have the same
respectively. shape (no distortion). The cross-correlation function between the
two signals s1(t) and s2(t) is written as:
2.2. Time of flight method Z þ1
cors1 =s2 ðtÞ ¼ s1 ðsÞs2 ðs þ tÞds ð3Þ
1
Using the time of flight method, the velocity coil,t (Eq. (1)) is
expressed as a function of dttof = t2  t1 and the attenuation aoil,t The time of the maximum of this cross-correlation function (Eq. (3))
(Eq. (2)) is related to the echo amplitude ratio A1/A2: corresponds to the characteristic time delay dtcor between s2(t) and
s1(t). Using time cross-correlation technique, the phase velocity
2Ddoil coil,t,cor is written as:
coil;t ¼ ð1Þ
dt tof
2Ddoil
coil;t;cor ¼ ð4Þ
and dtcor
 
1 A1 In addition, the attenuation can be evaluated by the normalized
aoil;t ¼ log ð2Þ cross-correlation rs1/s2 between s1(t) and s2(t) (Eqs. (3) and (5)). If
2Ddoil A2
there is no frequency dependent attenuation, one can write
Some alternative are based on those time of flight methods. Partic- s2(t) = A2/A1s1(t + dtcor), and the maximum amplitude of the enve-
ularly, the time of flight dttof can be estimated using the rising time lope of the normalized cross-correlation rs1/s2,max directly gives the
dt10 or dt20, at 10% or 20% of the maximum of the envelope of the amplitude ratio A2/A1. This assumption corresponds to the case
round-trip echoes, respectively. In the case of the amplitude ratio where the attenuation is considered constant in the bandwidth of
A1/A2, it can be estimated using the ratio between the maximum the transducer:
of the rectified echoes or the ratio between the maximum of the
cors1 =s2 ðtÞ
envelope of the echoes. rs1 =s2 ðtÞ ¼   ð5Þ
max env ðcor s1 =s1 ðtÞÞ
(a) and
 
  max env ðcor s1 =s2 ðtÞÞ A
rs1 =s2 ;max ¼ max env ðr s1 =s2 ðtÞÞ ¼  ¼ 2 ð6Þ
max env ðcor s1 =s1 ðtÞÞ A1

ek ðtÞ ¼ env ðsk ðtÞÞ ¼ jsk ðtÞ þ j  Hilbertfsk ðtÞgj

where the envelope env(x(t)) of the signal x(t) is determined using
its Hilbert transform Hilbert{x(t)} as env(x(t)) = |x(t) + j.Hilbert{x(t)}|.
Thus, the attenuation in the oil aoil,t,cor estimated via the normalized
cross-correlation amplitude rs1/s2,max (Eq. (6)) is illustrated by
Fig. 3(b) and given by:
   
 log r s1 =s2 ;max 1 A1
aoil;t;cor ¼ ¼ log ð7Þ
2Ddoil 2Ddoil A2

2.4. Spectral method

The spectral method makes it possible to find the ultrasonic dis-

persion properties with frequency [14,18]. The complex spectra of
useful roundtrip signals s1(t) and s2(t), denoted S1(f) and S2(f) are
given by:
(b)
8 4pfd
> j c ðfrefÞ
< S1 ðf Þ ¼ S0 ðf Þe2aoil ðf Þdref e oil RAl=oil
ð8Þ
>
: 4pf ðdref þDdoil Þ
2aoil ðf Þðdref þDdh Þ j
S2 ðf Þ ¼ S0 ðf Þe e coil ðf Þ
RAl=oil

where S0(f) is the spectrum of the emitted signal, RAl/oil(f) is the

reflection coefficient of the aluminum to the oil, coil(f) is the phase
l1 l2 velocity as a function of frequency and aoil(f) is the attenuation as
d
θ a function of frequency. The transfer function between these two
positions is:

4pf Ddoil
S2 ðf Þ j
Tðf Þ ¼ ¼ e2aoil ðf ÞDdoil e coil ðf Þ ð9Þ
S1 ðf Þ

Fig. 1. Experimental setup (a) with the two positioning configurations, and (b) The argument of T(f) corresponds to the phase difference between
associated incidence discrepancy relative to the normal theoretical case. the signals s1(t) and s2(t) is:
310 N. Ghodhbani et al. / Ultrasonics 56 (2015) 308–317

(a) 0.3 s1(t) (a) 1

rs /s
env 0.7064 1 2
0.2 max écho env(rs )
0.5 /s
max env 1 2
0.1

(t) (a.u.)
10% max env rs /s ,max
s1 (t) (V)

20% max env 1 2

0
0

2
/s
-0.1

1
rs
-0.2 -0.5

-0.3
11.933
9 10 11 12 13 -1
10 11 12 13 14 15
t (µs)
t (µs)

(b) 0.3 s2(t)

env
(b) 200
Experiment
0.2 max écho Fit
max env
0.1 150 rs
10% max env /s ,max

(Np/m)
1 2
s2 (t) (V)

20% max env

0
100

oil,t,cor
-0.1

-0.2 α
50
28.97 0.7064
-0.3

21 22 23 24 25 0
0 0.2 0.4 0.6 0.8 1
t (µs)
rs /s ,max
(a.u.)
1 2
Fig. 2. First round-trip echo in the oil 47V50, f0 = 7.5 MHz and T = 20 °C and
associated arrival times at the maximum of the rectified echo (black), maximum Fig. 3. (a) Normalized cross-correlation function rs1/s2, and (b) attenuation changes
envelope of echo (red), 10% of the maximum of envelope (magenta), 20% of the versus the maximum value of the envelope of the normalized cross-correlation
maximum of envelope (green) (a) in position 1 and (b) in position 2. (For rs1/s2,max for the reference oil 47V50, with f0 = 7.5 MHz and Df6,r = 80%.
interpretation of the references to colour in this figure legend, the reader is referred
to the web version of this article.) viscous medium. The second is the divergence in the far-field zone
of the transducer, directly related to the diameter and the center
4pf Ddoil
ArgðTðf ÞÞ ¼  ð10Þ frequency of the transducer. The third is the diffusion of the wave
coil ðf Þ
at the meeting of heterogeneities. In our study, to characterize the
In practice, the phase Arg(T(f)) is discontinuous, ranging between silicone oil, we assumed that the propagation medium is homoge-
p and +p. Thus, the phase is unwrapped to obtain u(f). It is then neous and isotropic. We also neglect the diffraction phenomenon
fitted by a linear regression in order to identify the value at the ori- because the transducer is placed in the near-field [14]. The attenu-
gin +2pm (where m is an integer): ation associated to the Kelvin–Voigt model is proportional to the
square of the frequency but this model has a tendency to overesti-
ulin ðf Þ ¼ a0  f þ 2pm ð11Þ
mate the attenuation. The models of Maxwell and Zener take into
where a0 is the slope and +2pm is the intercept. The term a0f cor- account the relaxation time but also overestimate the attenuation
responds to the phase shift between two signals. The integer m is [17,21]. For a wide variety of materials (viscous fluids and visco-
first estimated using the time delay between the two echoes s1(t) elastic tissues), the attenuation can be modeled on a finite band-
and s2(t) throughout their correlation [18]. Then we can calculate width by a power law [12,14] of the following form:
the phase in the bandwidth by subtracting 2pm from the
aoil ðf Þ ¼ a0 f n ð14Þ
unwrapped phase [11,12,19,20]. The velocity coil(f) is written as:
where a0 is the attenuation coefficient and n is a positive real num-
uðf Þ  2pm a0
coil ðf Þ ¼   ð12Þ ber between 1 and 2.
4pf Ddoil 4p  Ddoil
The empirical Cole–Cole distribution is an analytical three
The attenuation is calculated from the modulus of the transfer parameter model of the relaxation spectra that provides accurate
function of Eq. (9): fits to experimental dynamic viscosity data for many viscoelastic
media. The viscosity varies with the frequency and written as:
1
aoil ðf Þ ¼  log ðjTðf ÞjÞ ð13Þ g0
2Ddoil gðxÞ ¼ ð15Þ
1b
1 þ ðjxsÞ
2.5. Broadband behavior of attenuation and phase velocity dispersions where g0 is the Newtonian dynamic viscosity at low frequency, s is
the relaxation time and b (between 0 and 1) is an indicator factor of
Ultrasound attenuation can be decomposed in three factors. The the width of the distribution of relaxation times. At high frequency,
first is the absorption due to the conversion of mechanical energy when xs  1, the Cole–Cole attenuation model [17,21] is written as
into heat caused by the interaction between the acoustic wave and (Appendix A):
 N. Ghodhbani et al. / Ultrasonics 56 (2015) 308–317 311

g0 ð2pÞbþ1 sb1  p bþ1 transducer. The global relative error on the attenuation measure-
aðf Þ  3
cos ðb  1Þ f ð16Þ ment (Eq. (B16)) can be evaluated by Dar  Ddrt,r + DaSNR,r, and is
2qcL 2
maximized by:
Identification of the parameters of Eq. (16) with Eq. (14) allows us
to write: 2
Dar 6 h2 þ   ð24Þ
A1
 SNR  log
bþ1 b1
g0 ð2pÞ s p A2
a0  cos ðb  1Þ ð17Þ
2qc3L 2
and n  b þ 1 ð18Þ 3. Experiment

Dispersion models Szabo [15] and O’Donnell et al. [16] are used to
3.1. Configuration
connect the evolution of the phase velocity and ultrasonic attenua-
tion over a range frequency. Taking into account the Kramers–Kro-
In the studied configuration (Fig. 1(a)) we placed a transmitter/
nig relations [22,23] and the assumption of small variations of
receiver single element transducer in an oil tank located in a refer-
attenuation and phase velocity in a field of finite frequencies, Szabo
ence position dref via a stepper motorized unit to translate by a dis-
and O’Donnell obtain the following relations between the phase
tance Ddoil in parallel relative to the axis of the transducers. The
velocity c(f), the attenuation a0 and the power coefficient n
measurements were carried out with dref = 5 mm and Ddoil = 6 mm.
[16,24,25]:
Acoustic signals received in two different positions are used to cal-
8  
>
< cðf1 Þ  cðf1 Þ ¼ pa02 log ff culate the acoustic attenuation and the phase velocity to deduce the
if n ¼ 1
c

c
 ð19Þ viscosity. Two references of silicone oils, 47V50 and 47V350, are
>
: cðf1 Þ  cðf1 Þ ¼ n  f n1 n1
 fc if n – 1 analyzed. The manufacturer indicates for the oil referenced 47V50
c
that the Newtonian kinematic viscosity is m = 50 mm2 s1 at
where c(fc) is phase velocity at a reference frequency fc and n is a Tamb = 25 °C with a temperature sensitivity given at dm/dT 
constant defined by: 0.41 mm2 s1 K1. For the 47V350 oil reference, the Newtonian
8 a0 kinematic viscosity is m = 350 mm2 s1 at Tamb = 25 °C with a tem-
< n ¼ nO’D ¼ p2 ðn1Þ according to Orsquo; Donnell perature sensitivity evaluated around dm/dT  0.38 mm2 s1 K1.
  ð20Þ
a
: n ¼ nS ¼ 0 cot ðn1Þ p
according to Szabo For the characterization of the ultrasonic properties, we used
2p 2
longitudinal wave piezoelectric transducers. In order to identify
As highlighted by Eq. (20), these dispersion models are convergent the frequency dispersion of the ultrasonic parameters, in view to
with n close to unity. cover a large frequency bandwidth, we used four transducers with
adjacent center frequency (Krautkramer), i.e. f0 = 1, 2.25, 5 and
2.6. Reproducibility at normal incidence 7.5 MHz. All transducers have an active radius a = 6.5 mm. The
excitation signals for the transducers is a burst (1 sinusoidal cycle)
The estimation of the phase velocity is closely related to per- generated by a programmable arbitrary waveform generator
pendicularity between the transducer axis and the plane of the (Agilent 33220A). The round-trip signals are acquired by a digital
reflector (Fig. 1(b)). When the transducer is placed at the normal oscilloscope (Yokogawa DL9140-5G/s 1 GHz), on 125 k points, with
incidence of the reflector, the theoretical wave round trip distance a sampling frequency fs = 2.5 GHz and averaged on Nav = 1024
is drt,th = 2d. If the transducer axis is inclined relatively to the the- acquisitions to improve the signal-to-noise ratio. The length of
oretical normal incidence by h, the effective wave travel distance is the aluminum reflector was chosen to avoid overlapping echoes.
drt = l1 + l2, with l1 = d/cos(h) and l2 = l1/cos(2h). The difference
between the theoretical and effective travel distance is: 3.2. Attenuation and phase velocity

d d 3.2.1. Time of flight method

Ddrt ¼ drt;th  drt ¼ 2d   ð21Þ
cosðhÞ cosðhÞ cosð2hÞ The first round-trip echoes, referenced as s1(t) and s2(t) are
The quadratic Taylor series expansion near to zero of Eq. (21) gives: obtained in positions 1 (Fig. 2(a)) and 2 (Fig. 2(b)), respectively.
  The arrival times and amplitudes (tk, Ak), where k = 1 or 2 are defined
Ddrt 1 1 by a characteristic time of the echo and a maximum amplitude of the
Ddrt;r ¼ ¼1 1þ
drt;th 2 cosðhÞ cosð2hÞ echo. More precisely, the arrival time tk can be defined by: tk,abs the
2ð1  h2 Þ time of maximum of the rectified echo, tk,env the time of maximum
1 ð22Þ of the envelope echo (Eq. (25)), tk,10 the time at which 10% of the
ð2  h2 Þð1  2h2 Þ
maximum amplitude of envelope is reached, tk,20 the time at which
Thus, the relative error of distance measurement is approximately 20% of the maximum amplitude of envelope is reached (Fig. 2).
Ddrt,r  h2. The estimated phase velocity (Eq. (1)) and attenuation (Eq. (2))
Based on theoretical considerations (Appendix B), the maxi- are calculated on the basis of those estimators. These estimated
mum relative error of time can be decomposed in two parts, the results are associated to the transducers used at f0 = 1, 2.25, 5
one Ddts,r (Eq. (B2)) due to the sampling precision and the other and 7.5 MHz, operating at a centroid frequency, depending on
DdtSNR,r (Eq. (B10)) related to the signal-to-noise ratio (SNR). the broadband spectrum of the considered transducer, i.e. at
Therefore, the maximum relative error on the estimated time- fc = 1.0, 2.3, 4.5 and 6.0 MHz, respectively.
of-flight phase velocity (Eq. (1)) is around Dcr  Ddrt,r + Ddts,r +
DdtSNR,r, and can maximized by: 3.2.2. Cross-correlation method
2 1 Fig. 3(a) shows the normalized cross-correlation function rs1/s2
Dcr 6 h2 þ þ ð23Þ (Eqs. (3) and (5)) between s1(t) and s2(t), in the case where the max-
dt th  f s 2  f c  dt th
imum amplitude of the envelope is rs1/s2,max = 0.7064 at the time
where h is the inclination relatively to the theoretical normal inci- dtcor = 11.933 ls. The phase velocity (Eq. (4)) is thus evaluated
dence, dtth is the theoretical time delay between the two signals, at coil,t,cor = 1005.6 m s1. From Fig. 3(b), the estimate of the
fs is the sampling frequency and fc is the center frequency of the maximum amplitude of the envelope rs1/s2,max = 0.7064 gives
312 N. Ghodhbani et al. / Ultrasonics 56 (2015) 308–317

Table 1
Phase velocity coil,t (m s1) and attenuation aoil,t (Np m1) measured by temporal methods for the reference oils 47V50 and 47V350, f0 = 7.5 MHz and Tamb = 20 °C.

Methods Rectified Envelope Rising 10% Rising 20% Cross-correlation Absolute error Relative error (%)
1
47V50 coil,t (m s ) 1005.4 1004.6 1006.1 1006.0 1005.6 1.5 0.15
aoil,t (Np m1) 27.8 28.4 – – 29.0 1.2 4.32
47V350 coil,t (m s1) 1015.8 1015.2 1016.9 1016.5 1016.0 1.7 0.17
aoil,t (Np m1) 33.9 34.4 – – 35.0 1.1 3.24

(a) 1006.5 Spectral (b) 100 Spectral

Rectified
Enveloppe Rectified
1006 80 Enveloppe
Rising 10%
coil (m/s)

Cross-correlation
Rising 20%
Cross-correlation 60
1005.5

(Np/m)
40

oil
1005

α
20

1004.5 0
4 5 6 7 8 9 10 4 5 6 7 8 9 10
f (MHz) f (MHz)

(c) 1017 Spectral

(d) 100 Spectral
Rectified Rectified
Enveloppe 80 Enveloppe
1016.5
Rising 10% Cross-correlation
coil (m/s)

Rising 20%
60
Cross-correlation
(Np/m)

1016
40
oil
α

1015.5
20

1015 0
4 5 6 7 8 9 10 4 5 6 7 8 9 10
f (MHz) f (MHz)

Fig. 4. Experimental characterization by spectral method and comparison with temporal methods, at f0 = 7.5 MHz and Tamb = 20 °C: (a), (c) phase velocity coil,f (m s1) and (b),
(d) attenuation aoil,t (Np m1), for the reference oils (a), (b) 47V50 and (c), (d) 47V350.

aoil,t,cor = 28.97 Np m1. Table 1 shows phase velocity and attenua-
tion measured by the different temporal methods for the reference
oils 47V50 and 47V350 with a center frequency transducer
f0 = 7.5 MHz and at ambient temperature Tamb = 20 °C. As a result,
the phase velocity and the attenuation at constant temperature
are greater for the reference oil 47V350 than for the reference oil
47V50 (i.e. increase with viscosity). The relative error evaluated
for different temporal methods and for both reference oils 47V50
and 47V350 do not exceed 0.2% for the phase velocity, and is around
4.5% for the attenuation.
3.3. Reproducibility and dispersion at normal incidence
To estimate the perpendicularity error, a range of 15 measure-
ments was performed in the same experimental configuration
and at the same temperature. For each measurement, the orienta-
tion of the transducer relative to the reflector was readjusted as
close as possible to the normal incidence. The average and stan-
dard deviation of the phase velocity (Fig. 5(a)) and attenuation
(Fig. 5(b)) are plotted for 15 measurements on the reference oil
47V50, at f0 = 7.5 MHz and Tamb = 20 °C.

3.2.3. Spectral method 3.3.1. Relative error on velocity and attenuation

Fig. 4 shows the phase velocity (Eq. (12)) and the attenuation
(Eq. (13)) are calculated by the spectral method in a frequency
band between 4 MHz and 10 MHz, obtained with a transducer at
f0 = 7.5 MHz, operating at a centroid frequency fc = 6.0 MHz. The
comparison between temporal and spectral methods for both ref-
erence oils 47V50 (Fig. 4(a) and (b)) and 47V350 (Fig. 4(c) and
(d)) shows that maximum absolute relative error is 0.12% for the
phase velocity and is around 9% for the attenuation.
The phase velocity and attenuation differences are evaluated
close to Dcoil  2 m s1 and Daoil  1.62 Np m1, respectively.
These error bars can be compared with those estimated on the
basis of the errors sources (Eq. (23)), i.e. Dcr  Ddrt,r + Ddts,r +
DdtSNR,r. In our experimental configuration, we evaluated
Dcr  0.1%, which is in the order of magnitude based on the
theoretical limit (Appendix C) with Ddts,r  6.67  103%,
DdtSNR,r  8.81  102%. It results in an evaluated error on the
 N. Ghodhbani et al. / Ultrasonics 56 (2015) 308–317 313

1010
(a) Experiment
Arrhenius fit
1007
1000 Linear fit

1006

coil (m.s-1)
990
coil (m/s)

1005
coil,f max
980
coil,f min
1004
coil,f moy
970
1003 O'Donnell
Szabo
2 4 6 8 10 960
20 25 30 35 40
f (MHz)
T (°C)

(b) 100 Experiment

40
Fig. 6. Phase velocity versus temperature of the reference oil 47V50, at f0 = 5 MHz
and T = 21–40 °C: experiment, Arrhenius law (Eq. (28)) and associated linear fit (Eq.
Power law
(29)).
2
θ + 2/(SNR*log(A1/A2))

These velocity dispersion laws can be fitted by a linear regression in

(Np/m)

the bandwidth of the transducer, i.e. around fc = 6 MHz, from f = 2 to

Δ α (%)

50 20 10 MHz:
r
oil

8
α

< cO’D ðf Þ  1006:2 þ f f c

 0:33
4  106
ð27Þ
: cS ðf Þ  1006:2 þ f f c
 0:12
4  106
Δ αr = 5.29 %
It is observed that the models of dispersion of phase velocity are
within the error bars of the reproducibility.
0 0
2 4 6 8 10
f (MHz) 3.4. Temperature sensitivity

Fig. 5. Reproducibility results in terms of average and standard deviation obtained on
15 measurements for the reference oil 47V50, at f0 = 7.5 MHz and Tamb = 20 °C: (a)
phase velocity coil (m s1) compared with the O’Donnell (magenta) and Szabo (green)
models identified from the power law dispersion of the averaged attenuation
(a0  15.7  1012 Np m1 Hzn and n  1.81); (b) averaged experimental attenuation
aoil (Np m1), associated power law dispersion fit with a0  15.7  1012 Np m1 Hzn
and n  1.81 (magenta); error of attenuation measurement Dar (%) (green).
round-trip distance Ddrt,r  5.2  103%, i.e. h  7.2 mrad  0.41°.
The error of attenuation Dar is evaluated from Eq. (24) with A1/
A2  1.46 and SNR  100, i.e. Dar  5.29% (Fig. 5(b)). It is observed
that the experimental error of attenuation is less than Dar between
3.5 and 10 MHz. Between 2 and 3.5 MHz, the experimental error of
attenuation is greater than evaluated attenuation error Dar and the
power law tendency underestimates the experimental values.
3.3.2. Dispersion tendencies of velocity and attenuation
In our reference oil 47V50, a fit of the experimental attenuation
spectrum to a power law model (Eq. (14)) gives a set of parameters
a0  15.7  1012 Np m1 Hzn and n  1.81.
The dispersion models of Szabo and O’Donnell (Eqs. (19) and
(20)) are fitted (Fig. 5(a)) using the parameters (a0, n) identified
for the power law dispersion of attenuation (Fig. 5(b)). This implies
the choice of a center frequency, i.e. fc = 6 MHz and a velocity at
this center frequency, i.e. c(fc) = 1006.2 m s1:
a47V50 ðf Þ  15:7  1012 f 1:81 ð25Þ
The temperature dependence of the phase velocity of an epoxy
resin was measured by Wada and Yamamoto [26]. A correction as a
function of temperature is proposed by extrapolating that obtained
for the Young modulus [27,28] or for the viscosity [29]. The
temperature sensitivity can be written in an exponential form,
the so called Arrhenius law [9,10]:
 
B 1 1
T K T amb;K
cth;T ðT K Þ ¼ coil;T amb e ð28Þ
where TK (K) = T (°C) + T0 is the temperature of the reference oil
47V50 during cooling, with T0 = 273 °C, and Tamb,K = Tamb
(°C) + 273 = 296 K, is the ambient temperature, and B (K) is a con-
stant. In order to investigate the influence of temperature, the phase
velocity was monitored during a cooling of the reference oil 47V50.
The temperature is measured using a Pt100 probe associated to a
specific instrumentation. The variations of the experimental results
can be explained by the precision of the Pt100 instrumentation
setup. As illustrated by Fig. 6, a fit of the Arrhenius law (Eq. (28))
to the experimental results gives two constants: coil,Tamb = 998 m s1
with Tamb = 23 °C and B = 197 K. For small temperature variations,
the phase velocity is nearly linearly sensitive to the temperature.
Therefore, a fit of the experimental measurements can be obtained
by linearization (Appendix B) of the Arrhenius formula (Eq. (28))
gives:
coil;T;lin ðTÞ  1044  2  T ð29Þ

and where T (°C) is the oil temperature during cooling, in the range from
8   1 21 to 40 °C.
0:81
> 1
< cO’D ðf Þ ¼ 1006:2  15:7  1012
 f
0:81
 ð6  106 Þ According to this linear fit, the derivative of the phase velocity
0:81p2
 relative to the temperature dcoil,T,lin/dT is constant and negative in
>
: c ðf Þ ¼ 12    0:81 0:81
1
S
1
1006:2
 15:7 2 p10 cot 0:81 p2  f  ð6  106 Þ the range from 20 to 40 °C. As a result, this derivative dcoil,T/dT
around the ambient temperature Tamb = 23 °C can be developed
ð26Þ on the basis of the Arrhenius formula (Eq. (28)):
314 N. Ghodhbani et al. / Ultrasonics 56 (2015) 308–317

(a) 1007 (b) 35 Spectral

Rectified
30 Envelope
1006 Cross-correlation

25

(Np/m)
coil (m/s)
1005
20

oil
1004
15

α
Spectral
Rectified
Envelope
1003 Rising 10% 10
Rising20%
Cross-correlation
1002 5
1 2 3 4 5 6 1 2 3 4 5 6
f (MHz) f (MHz)

(c) 1017 (d) 40 Spectral

Rectified

1016 35 Envelope
Cross-correlation

30
1015

(Np/m)
coil (m/s)

25
1014
20

oil
1013 Spectral

α
Rectified 15
Envelope
1012 Rising 10%
10
Rising20%
Cross-correlation
1011 5
1 2 3 4 5 6 1 2 3 4 5 6
f (MHz) f (MHz)

Fig. 7. (a), (c) Phase velocity coil,t (m s1); (b), (d) attenuation aoil,t (Np m1) measured by several methods for the reference oils (a), (b) 47V50; (c), (d) 47V350, using a set of
broadband transducers having centroid frequencies at fc = 1.0, 2.3, 4.5, 6.0 MHz, at room temperature Tamb = 20 °C.

!
dcth;T  B B þB
¼ coil;T amb  e T amb;K :  2  e T K
dT TK
 
B
T amb T
coil;T amb  B T2 coil;T amb  B
 e 0 
T 2K T 2K
The derivative of the phase velocity relative to the temperature
dcoil,T/dT appears nearly constant and negative (Fig. 6). It can be
expressed as a function of the temperature variation relative to
the ambient temperature DT = T  Tamb. As an illustration, a
temperature increase DT = +1 K involves a decrease by dcoil,T/
dT  2 m s1 K1, relative to the reference phase velocity, i.e.
Dcoil,T/DT  2 m s1 K1 for the reference oil 47V50.
3.5. Broadband characterization of viscous liquids
As illustrated in Fig. 7, the attenuation and phase velocity were
measured at room temperature Tamb = 20 °C using several methods
for the reference oils 47V50 and 47V350. These characteristics are
evaluated by different transducers having center frequencies at
f0 = 1, 2.25, 5 and 7.5 MHz, but centroid frequencies at fc = 1.0,
2.3, 4.5 and 6.0 MHz, respectively. The curve tendencies are consis-
tent with those observed previously (Figs. 4 and 5). The errors
obtained between several estimators are linked to the distortions
of echoes and the type of materials. This difference is important
for dissipative materials. In our case, the echoes were not distorted,
which explains the low error between the different methods. As
illustrated in Fig. 7, the results given by various estimators are in
agreement for both reference oils 47V50 and 47V350. These results
are summarized in Table 1. The temperature variations and angular
deviation between the acoustic beam and the reflector can explain
the changes during the measurement campaign.
The broadband dispersion of the attenuation is calculated by
different transducers using the spectral method (Eq. (13)) with
the previously described set of transducers for the reference oils
47V50 (Fig. 8(a)) and 47V350 (Fig. 8(b)). These spectral attenuation
results are in good agreement and fit to a power tendency curve.
Thus the power law model (Eq. (14)) is fitted with the experimen-
ð30Þ tal curves. This model is able to estimate the trend of attenuation
at high frequencies. At low frequency, the power law model under-
estimates the attenuation, above 3 and 2 MHz for the references
47V50 and 47V350, respectively. At higher frequency, the power
law is more suitable. When the viscosity increases, n tends to 2
and the power law model is equivalent to the Kelvin–Voigt model.
Alternatively, the Cole–Cole model (Eq. (15)) can also model the
attenuation dispersion accurately. Using the least-mean-square
method, the parameters for the Cole–Cole model (Eq. (15)) are
identified to fit the spectral results (Fig. 8). The fitted power law
and Cole–Cole parameters for the reference oils and 47V50 and
47V350 are summarized in Table 2. The Newtonian dynamic vis-
cosity g0 and the relaxation time s increases with viscosity. The
relaxation time confirms that xs  103  1, which validates Eq.
(16). The power law parameter n and the Cole–Cole power factor
b + 1 are very close, which confirm the relation between the ultra-
sonic and viscoelastic parameters (Eq. (18)).
4. Conclusion
In the aim of achieving accurate and reproducible ultrasonic
measurements, an experimental setup has been developed, making
it possible to characterize the ultrasonic phase velocity and attenu-
ation of a viscous fluid in the MHz frequency range. Several signal
processing characterization methods were compared, investigating
the perturbation effects of distance error, time error and amplitude
error. This first approach is at constant temperature and neglects the
diffraction effect. The identification of attenuation to a power law
allows estimating ultrasonic dispersion properties related to the
phase velocity dispersion. The reproducibility of the measurements
 N. Ghodhbani et al. / Ultrasonics 56 (2015) 308–317 315

relationship between the empirical Cole–Cole model and the ultra-

(a) 100
7.5 MHz sonic power law model in high frequency is verified experimentally.
5 MHz
80 Appendix A
2.25 MHz
1 MHz
A.1. Cole–Cole model
α (Np/m)

60 Power law
Cole-Cole Based on the Cole–Cole model, the attenuation can be written
as follows:
L

40 ( )
x2 g0
aoil ðxÞ ¼  Re ðA:1Þ
2qc3oil 1 þ ðjxsÞ
1b

20
In high frequency and if xs  1, the Cole–Cole attenuation model is
written:
0 ( )
0 2 4 6 8 10 x2 g0 1
f (MHz) aoil ðxÞ   Re
2qc3oil ðjxsÞ
1b

(b)100 x2 g0 ðxsÞb1 n  p  po

aoil ðxÞ  Re cos ðb  1Þ þ j sin ðb  1Þ
7.5 MHz 2qcoil
3 2 2
5 MHz x2 g0 ðxsÞb1  p
80 2.25 MHz aoil ðxÞ  cos ðb  1Þ
2qc3oil 2
1 MHz
ðA:2Þ
α (Np/m)

60
Power law
Cole Cole If we write the attenuation versus frequency:

g0 ð2pÞbþ1 sb1  p bþ1

L

40 aoil ðf Þ  cos ðb  1Þ f ðA:3Þ

2qcoil
3 2

By identification with the power model (Eq. (14)) we can write:

20
g0 ð2pÞbþ1 sb1  p
a0  cos ðb  1Þ ðA:4Þ
2qc3oil 2
0
0 2 4 6 8 10
and n  b þ 1 ðA:5Þ
f (MHz)

Fig. 8. Experimental characterization of the attenuation spectrum aoil,f (Np m1) for
two references of silicon oils (a) 47V50 and (b) 47V350, using a broadband set of Appendix B. Time error
transducers having center frequencies f0 = 7.5, 5, 2.25 and 1 MHz and associated
power law and Cole–Cole fit , at room temperature Tamb = 20 °C. B.1. Sampling time error

We consider t1 and t2 the arrival times of the maximum of ech-

Table 2 oes s1(t) and s2(t), respectively. As illustrated by Fig. B.1, the time of
Fitted parameters for the Cole–Cole and power-law models fit for two references of
the maximum value of a sampled signal tn (n = 1 or 2) is evaluated
silicon oils 47V50 and 47V350.
with an absolute time error ±dt/2, where dt = 1/fs is the sampling
Cole–Cole Power law period and fs is the sampling frequency. Thus, the estimate of tn
g0 (mPa s) s (ns) b a0 (1012 Np m1 Hzn) n depends on how the signal is sampled and the resulting maximum
47V50 50 1.6 0.79 15.7 1.81 absolute error is dt. As a result, the measured dtmeas depends on the
47V350 349 17.6 0.71 38.0 1.71 theoretical time delay dtth = t2  t1 and on the sampling period dt:
dt meas ¼ t1  t 2  2dt ðB1Þ
The relative time error associated to the sampling Ddts,r is maxi-
at normal incidence was assessed on the phase velocity and attenu- mized by:
ation in the silicone oil. From experimental relative phase velocity  
dt th  dt meas 
Ddt s;r ¼   6 Ddts;r;max ¼ 2: dt ¼ 2 ðB2Þ
error, the angular deviation h to the normal incidence between the dtth  dt th dtth  f s
reflector plane and the acoustical wave is calculated. The relative
attenuation error is then identified from h and compared with i.e. for dtth = t2  t1 = 12 ls and fs = 2.5  103 MHz, we get
experimental values. An other perturbation factor which has a sig- Ddts,r,max  6.67  105  6.67  103%.
nificant incidence on the measurement results is the temperature.
On the studied reference silicone oils, the temperature sensitivity B.2. Signal-to-noise ratio (SNR) time error
was evaluated at dcoil/dT  2 m s1 K1 from T = 21 to 40 °C. The
ultrasonic parameters are finally measured at room temperature In this sub-section, we suppose that the maximum value of the
using several methods and several center frequency transducers. signal is located on a sinusoidal waveform. Thus, the maximum
The spectral attenuation results obtained for the several center fre- amplitude is reached at the time tn, with extreme values denoted
quency transducers have a power tendency. The power law and as tn,min and tn,max (Fig. B.2). These values can be determined by
Cole–Cole models are fitted with experimental curves. The solving the following equation:
316 N. Ghodhbani et al. / Ultrasonics 56 (2015) 308–317

ðA þ aÞ  sinð2  p  f c  tÞ ¼ A  a ðB3Þ
8   A+a
>
< 2:p  f c  t n;min ¼ Arcsin Aa a
Aþa A
i:e:   ðB4Þ a
>
: 2:p  f c  t n;max ¼ p  Arcsin Aa A−a
Aþa

Rewriting the Eq. (B4) using the signal-to-noise ratio SNR = A/a, we
obtain:
8 tn,min tn tn,max
< t n;min ¼ ArcsinðSNRþ1Þ
SNR1
>
2pf c
ðB5Þ
>
: t n;max ¼ pArcsinðSNRþ1Þ
SNR1
Fig. B.2. Signal-to-noise ratio (SNR = A/a) time error estimate of the maximum
2pf c amplitude at the arrival time tn.

The theoretical time tn = 1/(4fc) is given by Eq. (B5), when SNR  1.

1
tn ¼ ðB6Þ
4f c Taking into account the maximum noise amplitude a (Fig. B.2),
the maximum attenuation aoil,max and minimum attenuation
The absolute time error dtSNR associated to the SNR is given by:
aoil,min are written as follows:
 
Arcsin SNR1 8  
1 SNRþ1 >
dt SNR ¼ tn  t n;min ¼  ðB7Þ < aoil;max ¼ 2D1d log AA1 þa
a
4f c 2pf c oil

2
 ðB12Þ
>
: aoil;min ¼ 2D1d log AA1 a
The relative time error Ddtr,SNR associated to the to-noise ratio SNR 2 þa oil

is:
  Taking into account the signal to noise ratio SNR1 = A1/a and
dtth  dtmeas;SNR 
Ddt r;SNR ¼  
 ðB8Þ SNR2 = A2/a, it becomes:
dtth
8  
where dtmeas,SNR = t2  t1 ± dtSNR. Eq. (B8) becomes:
>
< aoil;max ¼ 2D1d log SNR 1 þ1
oil SNR2 1
  ðB13Þ
2  dtSNR >
: aoil;min ¼ 2D1d log SNR 1 1
Ddt r;SNR ¼ ðB9Þ SNR2 þ1 oil
dt th
i.e. for dtth = t2  t1 = 12 ls, SNR = 100 and fc = 6 MHz, we get For simplification, we note SNRn = min{SNR1, SNR2} = SNR2. Then,
DdtSNR,r  8.81  102%. taking into account the sampling signal-to-noise ratio SNRs, we
The most severe case is SNR = 1, resulting in (Eq. (B7)) get SNR = SNRn  SNRs/(SNRn + SNRs), which is a priori the worst case.
dtSNR 6 tn = 1/(4fc) and the time error related to the signal noise Therefore, Eq. (B13) becomes:
ratio (Eq. (B8)) becomes:    

aoil;max  aoil;th  aoil;th  aoil;min 

1 DaSNR;r ¼ max  ; 
 


Ddt r;SNR 6 ðB10Þ aoil;th aoil;th
2f c  dtth SNRþ1
log SNR1
6 DaSNR;r;max ¼   ðB14Þ
i.e. for dtth = t2  t1 = 12 ls and fc = 6 MHz, we get log AA12
Ddtr,SNR 6 69.4  102%.
The influence of signal noise ratio SNR on the maximum relative
As an illustration, Fig. B.4 shows the variation of attenuation error
time error Ddtr,SNR,max is illustrated by Fig. B.3.
depending on the signal- to-noise ratio SNR, neglecting the error
on the round-trip distance Ddrt,r.
B.3. Sampling and signal-to-noise ratio (SNR) amplitude errors
The global relative error on the attenuation is:

The theoretical ultrasonic attenuation, i.e. without noise, is: Dar 6 Ddrt;r þ DaSNR;r ðB15Þ
 
1 A1 This expression can be approximated by:
aoil;th ¼ log ðB11Þ
2Ddoil A2
1
where A1 and A2 are the amplitudes of echoes s1(t) and s2(t)
respectively.
0.8
Δ δ tr,SNR,max (%)

0.6

0.4

0.2

0
0 20 40 60 80
SNR (dB)
Fig. B.1. Sampling time error estimate |tn,samp.1  tn,samp.2| of the maximum ampli-
tude located at tn = tn,samp.1 or tn,samp.2, depending on how the signal is sampled, Fig. B.3. Relative time error Ddtr,SNR,max as a function of the signal to noise ratio SNR
either by sampling 1 or 2. (dB).
 N. Ghodhbani et al. / Ultrasonics 56 (2015) 308–317 317

60 References

10
50 [1] Y. Gélébart, Evaluation non destructive par ultrasons du vieillissement

20
thermique d’une structure composite, PhD thesis, Université du Havre, 2007.

1
[2] A. El Mouhtadi, Caractérisation ultrasonore de plaques viscoélastiques
50100
20*log10 (A1 /A2 )

40 homogènes et composites, PhD thesis, Université du Havre, 2011.

[3] S. Rokhlin, D. Lewis, K. Graff, A. Laszlo, Real-time study of frequency
dependence of attenuation and velocity of ultrasonic waves during the
30 curing reaction of epoxy resin, J. Acoust. Soc. Am. 79 (1986) 1786–1793.
10

[4] A. Maffezzoli, E. Quarta, V.A.M. Luprano, G. Montagna, L. Nicolais, Cure

20

1
20 monitoring of epoxy matrices for composites by ultrasonic wave propagation,
J. Appl. Polym. Sci. 73 (10) (1999) 1969–1977.
50

[5] M. Rath, J. Döring, W. Stark, G. Hinrichsen, Process monitoring of moulding

100

10 10 compounds by ultrasonic measurements in a compression mould, NDT&E Int.

20 1
33 (2) (2000) 123–130.
5
1000 [6] J. Mc Hugh, Ultrasound technique for the Dynamic Mechanical Analysis (DMA)
0 10
100
50
20 100
50
20
10
1
0 20 40 60 80 of polymers, PhD thesis, Bundesanstalt für Materialforschung und -prüfung
(BAM), 2008.
SNR (dB) [7] R. Challis, R. Freemantle, R. Cocker, D. Chadwick, D. Dare, C. Martin, A.
Mahendrasingam, W. Fuller, Ultrasonic measurements related to evolution of
Fig. B.4. Relative error on the attenuation DaSNR,r (%) as a function of the SNR (dB) structure in curing epoxy resins, Plast., Rubber Compos. 29 (3) (2000) 109–118.
and the theoretical amplitude ratio 20log10(A1/A2) (dB). [8] F. Lionetto, A. Maffezzoli, Monitoring the cure state of thermosetting resins by
ultrasound, Materials 6 (9) (2013) 3783–3804.
[9] N. Samet, E. Ndiaye, P. Marechal, H. Duflo, Polymerization monitoring using
ultrasound, in: 13th International Symposium on Nondestructive
2
Dar 6 h2 þ   ðB16Þ Characterization of Materials (NDCM-XIII), <www.ndt.net>, 2013.
A1 [10] N. Samet, Evaluation ultrasonore des propriétés d’une résine et détection de
SNR  log A2 porosité à l’aide d’un transducteur multiéléments. Application aux matériaux
composites au cours de fabrication. PhD thesis, Université du Havre, 2012.
[11] H. Jeong, D. Hsu, Experimental analysis of porosity-induced ultrasonic
attenuation and velocity change in carbon composites, Ultrasonics 33 (3)
Appendix C. Linearization of the Arrhenius model (1995) 195–203.
[12] P. He, Experimental verification of models for determining dispersion from
attenuation, IEEE Trans. Ultrason. Ferroelectr. Freq. Control 46 (3) (1999) 706–
Arrhenius law is written in the following form: 714.
[13] B.E. Treeby, B.T. Cox, Modeling power law absorption and dispersion for
  acoustic propagation using the fractional laplacian, J. Acoust. Soc. Am. 127 (5)
B: 1  1
T K T amb;K
(2010) 2741–2748.
coil;T ðT K Þ ¼ coil;T amb e ðC1Þ [14] N. Samet, P. Maréchal, H. Duflo, Ultrasonic characterization of a fluid layer
( using a broadband transducer, Ultrasonics 52 (3) (2012) 427–434.
TK ¼ T0 þ T [15] T.L. Szabo, Causal theories and data for acoustic attenuation obeying a
where ðC2Þ frequency power law, J. Acoust. Soc. Am. 97 (1) (1995) 14–24.
T amb;K ¼ T 0 þ T amb [16] M. O’Donnell, E. Jaynes, J. Miller, Kramers–Kronig relationship between
ultrasonic attenuation and phase velocity, J. Acoust. Soc. Am. 69 (1981) 696–
8 701.
>
> T 0 ¼ 273 K
>
< [17] R.E. Challis, F. Blarel, M.E. Unwin, X. Guo, On the modelling of ultrasonic bulk
wave propagation in epoxies, J. Phys: Conf. Ser. 269 (1) (2011) 012001.
and T amb ¼ 23  C ðC3Þ [18] F. Peters, A broad band spectroscopy method for ultrasound wave velocity and
>
>
>
: attenuation measurement in dispersive media, Ultrasonics 41 (2003) 357–363.
T ¼ f20  C; 40  Cg [19] W. Sache, Y. Pao, On the determination of phase and group velocities of
dispersive waves in solids, J. Appl. Phys. 49 (1978) 4320–4327.
[20] J. Mobley, J.N. Marsh, C.S. Hall, M.S. Hughes, G.H. Brandenburger, J.G. Miller,
The linearization of the Eq. (C1) gives: Broadband measurements of phase velocity in AlbunexÒ suspensions, J.
Acoust. Soc. Am. 103 (4) (1998) 2145–2153.
   [21] R. Challis, F. Blarel, M. Unwin, J. Paul, X. Guo, Models of ultrasonic wave
1 1
coil;T;lin ðT K Þ  coil;T amb  1 þ B   ðC4Þ propagation in epoxy materials, IEEE Trans. Ultrason. Ferroelectr. Freq. Control
T K T amb;K 56 (6) (2009) 1225–1237.
[22] K.R. Waters, M.S. Hughes, J. Mobley, J.G. Miller, Differential forms of the
!! Kramers–Krönig dispersion relations, IEEE Trans. Ultrason. Ferroelectr. Freq.
B 1 1 Control 50 (1) (2003) 68–76.
i:e: coil;T;lin ðT K Þ ¼ coil;T amb : 1 þ   ðC5Þ [23] K.R. Waters, J. Mobley, J.G. Miller, Causality-imposed (Kramers–Kronig)
T0 1 þ TT0 1 þ TTamb relationships between attenuation and dispersion, IEEE Trans. Ultrason.
0
Ferroelectr. Freq. Control 52 (5) (2005) 822–833.
[24] A. El Mouhtadi, J. Duclos, H. Duflo, Simultaneous determination of physical and
! acoustical properties of a homogeneous plate in air, in: Ultrasonics
B Symposium (IUS), 2009 IEEE International, pp. 1–4, September 2009.
and coil;T;lin ðT K Þ  coil;T amb  1 þ  ðT amb  T Þ ðC6Þ
T 20 [25] A. El Mouhtadi, J. Duclos, H. Duflo, Ultrasonic Modeling of a Viscoelastic
Homogeneous Plate, in: Acoustics 2012 Nantes, April 2012.
[26] Y. Wada, K. Yamamoto, Temperature dependence of velocity and attenuation
By identification of the Eqs. (29) and (C6), we can write: of ultrasonic waves in high polymers, J. Phys. Soc. Jpn. 11 (8) (1956) 887–892.
[27] R. Bruls, H. Hintzen, G. de With, R. Metselaar, The temperature dependence of
8 the Young’s modulus of MgSiN2, ALN and Si3N4, J. Eur. Ceram. Soc. 21 (3)
> Bc
< coil;T amb þ oil;T2 amb T amb  1059 m s1 (2001) 263–268.
T 0
ðC7Þ [28] J.B. Wachtman, W.E. Tefft, D.G. Lam, C.S. Apstein, Exponential temperature
>
:  Bcoil;T amb  2:64 m s1 K1 dependence of Young’s modulus for several oxides, Phys. Rev. 122 (1961)
T0
1754–1759.
[29] Q. Fontana, Viscosity: thermal history treatment in resin transfer moulding
We find the expected orders of magnitude (Eq. (29)). process modelling, Compos. A Appl. Sci. Manuf. 29 (1998) 153–158.