IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control , vol. 56, no.

9, September 2009 1955

Determination of Epoxy Film Parameters in a

Three-Layer Metal/Adhesive/Metal Structure
Naïma Alaoui Ismaili, Driss Chenouni, Zakia Lakhliai, Mounsif Ech-Cherif El-Kettani, Bruno Morvan,
and Jean-Louis Izbicki

Abstract—The aim of this work is to propose a method to layer (metal/water/metal). Their study was performed
determine the elastic parameters and the thickness of a thin at high frequencies (5.5 < f < 8.5 MHz) that allowed the
epoxy film located inside a 3-layer aluminum/epoxy/aluminum
structure, based on ultrasonic measurements. This study is
excitation of the resonances of the thin water layer. In
conducted at low frequencies, to allow the vibration of the our case the frequency range is around 1 MHz, allowing
whole structure. First, the direct problem is addressed. The the vibration of all the structures, without reaching the
sensitivity of the vibration modes to the parameters of in- first resonance of the thin layer. In addition, in a low-
terest is studied to select the most sensitive one for a given frequency range, the phenomenon of absorption can be
parameter to be determined. Second, the identification of the
parameters with the selected modes is obtained by a minimiza-
considered as negligible. Balthazar et al. [11] have solved
tion of the characteristic equation. This process is applied to the problem in the case of an isotropic layer embed-
experimental data: the longitudinal and shear wave velocities ded between 2 semi-infinite solids. Their study is based
and the thickness of the epoxy film are obtained within a 3% on the reflection coefficient spectra under normal and
error range. oblique incidence, but for high frequencies that produce
resonances in the thin layer.
For the best determination of the parameters of the
I. Introduction intermediate layer from the characterization of the modes
of the structure, it is interesting to determine first the

A dhesive-bonded components and structures have

become an important part of manufacturing in areas
such as the aerospace, the automobile, or marine indus-
sensitivity of these modes to the parameters. Then, the
most sensitive mode to determine the corresponding pa-
rameter can be selected. As it will be shown, it appears
tries. This is due to their low cost, low weight, and ease that the vertical modes [12] are more sensitive to the thin
of assembly. The improvement of this technique requires adhesive layer parameters. Consequently, these modes
a better knowledge of the adhesion process and more reli- are used in an original algorithm to extract the thick-
able techniques for their nondestructive characterization. ness and the longitudinal and shear wave velocities of the
The external constraints (e.g., pressure and tempera- thin adhesive layer. These data are introduced into the
ture) have a significant effect on the bonding performance, characteristic equation (CE) of the structure. The very
and therefore they have a direct impact on the product fast convergence observed for obtaining the parameters
life cycle. The degradation of the bonding layer leads results from 2 facts: 1) the selected modes are sensitive
to a change of elastic parameters, such as longitudinal to the parameter to be determined and 2) a systematic
and shear waves velocities. Conversely, the estimation of and simple method based on the averaged and the stan-
these parameters allows one to obtain information on the dard deviation values to limit the intervals of inspection
quality of adhesive joint. The problem of adhesive joint is used.
evaluation has been studied in several papers [1]–[9]. In a previous work [13], our method has been applied
This study concerns a 3-layer sandwich structure to the case of an aluminum plate. The longitudinal and
(metal/adhesive/metal). It is based on the use of guided shear wave velocities were obtained by solving an inverse
Lamb waves, because they have proven to be an effective problem, using experimental data and the theoretical dis-
tool for many nondestructive testing applications. persion equation of the plate [13], [14]. Here, this method
In previous works, several authors have proposed solv- is extended to the case of a 3-layer metal/adhesive/metal
ing the inverse problem to characterize a thin layer be- structure in the aim to determine the parameters of the
tween 2 solids. Kinra et al. [10] have solved an inverse adhesive layer: d the thickness, cL the longitudinal wave
problem based on the frequency spectra, but in the lim- velocity, and cT the shear wave velocity.
ited case of normal incidence and of a fluid intermediate Section II deals with the dispersion equation solution
for determining the couple’s wave number kx-frequency f,
Manuscript received February 13, 2009; accepted May 16, 2009. knowing the values of parameters d, cL, and cT (resolu-
N. A. Ismaili and D. Chenouni are with Laboratoire d’électronique tion of the direct problem). The sensitivity of some modes
signaux-systèmes et d’informatique LESSI, Fès, Morocco.
Z. Lakhliai is with Ecole Supérieure de Technologie de Fès, Fès, Mo- with respect to the parameters of interest is studied. In
rocco. Section III, the inverse problem method is presented and
M. Ech-Cherif El-Kettani, B. Morvan, and J.-L. Izbicki are with Labo- applied first on theoretical data slightly perturbed in the
ratoire Ondes et Milieux Complexes, University of Le Havre, Le Havre,
France (e-mail: [email protected]; [email protected]). range of ± 5% range to test the reliability of the method.
Digital Object Identifier 10.1109/TUFFC.2009.1271 The method is then applied to experimental data.

0885–3010/$25.00 © 2009 IEEE

1956 IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control , vol. 56, no. 9, September 2009

Fig. 1. Geometry of the tri-layer.

II. Theoretical Study of the Direct Problem

Fig. 2. Theoretical dispersion curves for a duraluminum/epoxy/duralu-
A. Dispersion Equation, Dispersion Curves minum structure. The epoxy parameters are d = 0.5 mm, ρ = 1160 kg/
m3, cL = 2622 m/s, cT = 1356 m/s.
In this section, we present briefly the formulation prob-
lem for a 3-layer structure composed of 2 identical metallic
layers (S1, S3) joined by an adhesive layer (S2) [15], [16],
as illustrated in Fig. 1.
The boundary conditions at the upper and lower in-
terfaces [z = 2h + d/2, −2h − d/2] are the free surface
conditions. Conditions of perfect contact between 2 solids
are used at the bonded interfaces (z = ± d/2). A linear
and homogeneous system of 12 equations is then obtained.
The determinant associated with this system is a function
of 2 variables, i.e., frequency f and wave number kx. The
dispersion equation is obtained by setting this determi-
nant to zero. The resolution of this equation is performed
by a minimization method based on the Newton-Raphson
algorithm [17]. The numerical resolution of this equation
gives the dispersion curves for a given set of material con-
stants d, cL, and cT.
Fig. 2 shows the dispersion curves obtained for an alu-
minum/epoxy/aluminum structure with the following pa-
rameters: cL0 = 6380 m/s, cT0 = 3100 m/s, ρ0 = 2800 kg/
m3, and 2h = 3 mm for the aluminum, cL = 2622 m/s, Fig. 3. Evolution of the resonance frequencies versus the longitudinal
wave velocity cL. The reference value is cL = 2622 m/s.
cT = 1356 m/s, ρ = 1160 kg/m3, d = 0.5 mm for the ep-
oxy. It has to be noticed that the velocities cL and cT in
the epoxy have been measured experimentally [12], with broad to appreciate the overall sensitivity. The following
a method based on the cutoff frequency determination. variation ranges have been explored: [0.7 cL, 1.8 cL] for cL,
Different families of modes exist. The A0 and S0 modes [0.4 cT, 1.2 cT] for cT and [0.6 d, 1.6 d] for d. The results
will not be considered in this work. The vertical modes are presented with the dimensionless wave number solu-
VL and VT (for longitudinal and transversal modes) are tion k x = 0.190 (a normalization with respect to a wave
typical modes related to the intermediate thin layer [16]. number in water k x = kx/kw is made, with the wave veloc-
These modes will be helpful as well as the other Lamb ity in water cW = 1500 m/s) in Figs. 3–5. This particular
modes (labeled as M modes) for the determination of the value allows us to obtain solutions close to the cutoff fre-
characteristics of the layer, as explained in Section III. quencies, which should be more sensitive to the parame-
ters variations. To solve the inverse problem, it is interest-
B. Modes Sensitivity to d, cL, and cT ing to get an a priori knowledge about the most sensitive
modes to the parameters of interest. Fig. 3 shows that the
In this section, the sensitivity of the modes to the vari- M6L mode is sensitive to cL while the M3T mode is not
ations of parameters cL, cT, and thickness d of the thin sensitive to this parameter. Fig. 4 shows that the M1T
adhesive layer is investigated. Two parameters are fixed mode is sensitive to cT while M2T is less sensitive. Most of
and the evolution of the resonance frequencies versus the the modes are sensitive to the thickness, as shown on Fig.
third parameter is studied. The domain of variation is 5. The VT mode is sensitive to the thickness and to the
ismaili et al.: determination of epoxy film parameters in a three-layer structure 1957

Fig. 4. Evolution of the resonance frequencies versus the shear wave

velocity cT. The reference value is cT = 1356 m/s.

Fig. 6. Sketch of the method.

f) given from the direct problem solutions. In a second

phase, the (kx, f) values are perturbed with a white noise
to evaluate the reliability of the method, as the experi-
mental values of (kx, f) are expected to be slightly different
than the theoretical ones. In the last part, experimental
results will be used.
Fig. 5. Evolution of the resonance frequencies versus the thickness d. The
reference value is d = 0.5 mm. A. Principle of the Procedure

The procedure is schematically shown in Fig. 6. Three

steps are necessary to obtain the most accurate set of
parameters.
shear wave velocity cT (Fig. 4), while the longitudinal ver-
The unknown parameters are relative to an adhesive
tical mode VL is sensitive both to the thickness and to the
layer that is a polymer. Therefore, the range of the possi-
longitudinal wave velocity cL (Fig. 3). These modes will be
ble values is quite large, so broad intervals for the param-
used to get a first estimation of the parameters for which
eters are considered. From these initial values, a first rough
they are sensitive, as it will be shown in Section III-A.
estimation (dˆ1, cˆL1, cˆT 1) is obtained from a collection of
data (kx, f) relative to the Lamb modes. The algorithm is
described below. In the second step, the vertical modes are
III. Proposed Method for the Inverse Problem used [16]. The VL mode is used for a better estimation of
the thickness and longitudinal wave velocity (dˆ21 and cˆL2),
In the direct problem, from the given set (d, cL, cT) while the VT mode is used for a better estimation of the
relative to the adhesive layer the values of kx and f were thickness and the transverse wave velocity (dˆ22, cˆT 2). As
found. Here the problem is to obtain the set (d, cL, cT) previously mentioned, the roughly estimated value of cL
from given values of (kx, f) by simulations or experiments. (cT) for the longitudinal (transverse) vertical mode is suf-
First, this problem is solved by the use of exact values (kx, ficient because this mode is practically independent of the
1958 IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control , vol. 56, no. 9, September 2009

TABLE I. Values of the Three Parameters for Different Perturbations.

σ(cL) σ(cT)
Mode Noise d (mm) Δd/d (%) σ(d) (mm) cL (m/s) ΔcL/cL (%) (m/s) cT (m/s) ΔcT/cT (%) (m/s)
VL 1% 0.511 2.2 0.0000 2612 −0.4 2 1340 −1.2 9
3% 0.515 3.0 0.0010 2602 −0.8 2.5 1316 −2.9 11
5% 0.506 1.2 0.0030 2600 −0.8 7.1 1376 1.5 8.4
VT 1% 0.515 3.0 0.0010 2666 1.6 5 1354 −0.15 0
3% 0.517 3.4 0.0021 2670 1.8 8.7 1350 −0.4 5.6
5% 0.502 0.04 0.0033 2690 2.5 11 1347 −0.7 20
M6L 1% 0.500 0 0.0000 2650 1.1 10 1346 −0.7 8
3% 0.503 0.6 0.0030 2676 2.1 12 1366 0.7 7
5% 0.516 3.2 0.0045 2690 2.6 10 1372 1.2 17
M2T 1% 0.500 0 0.0000 2672 1.9 19 1353 −0.2 13
3% 0.519 3.8 0.0015 2710 3.4 21 1348 −0.6 15
5% 0.501 0.2 0.0015 2736 4.3 28 1338 −1.3 6
Reference values: d = 0.5 mm, cL = 2622 m/s, cT = 1356 m/s. Range variations of each parameter pi: [0.7 pi,
1.3 pi].

value of cT (c­L). For this second step, the range of the

parameters used in the algorithm is (0.7 pi, 1.3 pi) with
pi = dˆ1, cˆL1, cˆT1.
As 2 values of the thickness are obtained, the estimated
value for the thickness dˆ2 is taken as the average value of
dˆ21 and dˆ22. Indeed, the 2 vertical modes are sensitive to
this parameter.
The last step is performed using a Lamb mode with an
interval built from the previous estimation. The interval is
(0.7 pi, 1.3 pi) with pi = dˆ2, cˆL2, cˆT2. The final values are
dˆ3, cˆL3, cˆT 3 and must be compared with the initial known
values. An error of less than 0.5% is obtained on the param-
eters. This error is attributed to the numerical process.
Fig. 7. Experimental dispersion curves (in gray level) and fitted curves
B. Algorithm for the Determination of the Parameters obtained using the estimated parameters.

The adopted algorithm is described in the following. At

the step n (n being an integer), the parameter needed to C. Reliability of the Method
build the interval that will be investigated is (0.7 pi, 1.3 pi) To test the reliability of the method in the perspec-
where pi corresponds to the parameters obtained at step tive of using experimental data that are slightly different
n-1. For each couple (kx, f) of a given Lamb mode, 3 loops from the theory, the wave number solution of the direct
on the set of parameters are performed, in the aim to set problem is randomly perturbed with a white noise in a
the characteristic equation to zero, using a classical New- ± 5% range. The method described above is then applied
ton-Raphson method [17]. A better refinement of param- on these perturbed solutions, each parameter pi varying in
eters is obtained for the successive loops. For a given pa- the range [0.7 pi, 1.3 pi]. The obtained parameters remain
rameter, its averaged value p̂ and its standard deviation in the range ± 3.8% for the thickness d, ± 4.3% for cL,
ŝ p are calculated. The following criterion is then intro- and ± 2.9% for cT, around the used values in the direct
duced: if sˆ p/pˆ £ 3% then the value of the parameter is problem.
assumed to be correct. Otherwise, the same procedure is As an example, Table I gives some values of the 3 pa-
repeated in the limited interval [ pˆ - 3/2sˆ p, pˆ + 3/2sˆ p ] un- rameters, for different perturbation values and for several
til the criterion is met. The number of iterations is limited modes. The values used in the direct problem resolution
to 20. The possible need to change the initial interval had are: cL = 2622 m/s, cT = 1356 m/s, and d = 0.500 mm.
been anticipated, but in fact this case never occurred.
For the initial values, before the first estimation, rather D. Application to Experimental Data
broad intervals are considered to be sure that the correct
final values were included. Here, the choices were for cL This experiment is briefly described below and more
(2000 m/s; 3600 m/s), for cT (300 m/s; 1400 m/s)—the details can be found in [18]. It is performed with a contact
condition cL > 2c T is fulfilled in the computer program— transducer of 1 MHz central frequency, fixed on a Plexi-
and for d (0.1 mm, 0.7 mm). glas wedge of 38.5° degrees geometrical angle. The trans-
ismaili et al.: determination of epoxy film parameters in a three-layer structure 1959

ducer is excited by a short pulse with a −300 V amplitude. with those provided by the characteristic equation by us-
A laser vibrometer measures the normal displacement ver- ing the estimated parameters, even if the modes are not
sus time on the plate surface in the propagation direction used for the resolution of the inverse problem.
and is translated by steps of 0.1 mm over 60 mm. The
measurements are recorded on a microcomputer and ar-
ranged in a time-space matrix. A double space-time FFT References
is applied and allows a 3-D representation: gray level that
is the double FFT modulus, versus the frequency and the [1] K. Hello, L. J. Jacobs, and J. Qu, “Characterization of adhesive
wave number. An example is shown in Fig. 7 for an alumi- bond properties using Lamb waves,” NDT Int., vol. 33, pp. 555–563,
2000.
num/epoxy/aluminum structure. [2] V. Vlasie, S. de Barros, M. Rousseau, and L. Champaney, “Ultra-
The experimental data (kx, f) are collected, and the sonic rheological model of cohesive and adhesive zones aluminium
process described in Section III-A and III-B is applied on joint: Validation by mechanical tests,” Arch. Appl. Mech., vol. 755,
no. 4–5, pp. 220–234, 2006.
the different experimental dispersion curves corresponding [3] D. C. Worlton, “Ultrasonic testing with Lamb waves,” Non-destruct
to different modes. The higher frequency modes (that cor- Test., vol. 15, pp. 218–222, 1957.
respond to M Lamb modes) are first used for a first rough [4] M. Lowe and P. Cawley, “The applicability of plate wave techniques
estimation (dˆ1, cˆL1, cˆT 1 in Fig. 6). Then the lower frequency for the inspection of adhesive and diffusion bonded joints,” J. Non-
destruct Eval, vol. 13, pp. 185–200, 1994.
modes (that correspond to vertical modes) are used for a [5] M. Castaings and B. Hosten, “Guided waves propagating in sand-
better estimation [(dˆ21 + dˆ22)/2, cˆL2, cˆT 2 in Fig. 6], and the wich structures made of anisotropic, viscoelastic, composite materi-
als,” J. Acoust. Soc. Am., vol. 113, no. 5, pp. 2622–2634, May 2003.
refinement is processed using any other Lamb modes, that [6] A. Jungman, P. Guy, and G. Quentin, “Characterization of glue
are here the lowest frequency modes (dˆ3, cˆL3, cˆT 3 in Fig. bonds using ultrasonic reflected beam,” Rev. Prog. Quant. Nonde-
6). struct. Eval., vol. 10B, pp. 1319–1327, 1991.
The longitudinal and shear wave velocities and the [7] C. C. Guyott and P. Cawley, “The ultrasonic vibration character-
istics of adhesive joints,” J. Acoust. Soc. Am., vol. 83, no. 2, pp.
thickness of the epoxy film are then determined and give 632–640, 1988.
ĉ L = 2660 m/s, ĉ T = 1300 m/s, and dˆ = 0.511 mm. These [8] K. Heller, L. J. Jacobs, and J. Qu, “Characterization of adhesive
values are given with a maximum error of 3%. The com- bond properties using Lamb waves,” NDT Int., vol. 33, pp. 555–563,
2000.
putation time for the estimation of the parameters is [9] P. B. Nagy and L. Adler, “Nondestructive evaluation of adhesive
about 5 min using a Pentium 4 CPU at 1 GHz. Other joints by guided waves,” J. Appl. Phys., vol. 66, no. 10, pp. 4658–
values were obtained previously [15] by the determination 4663, 1989.
[10] V. K. Kinra, P. T. Jaminet, and C. Zhu, “Simultaneous measure-
of the cutoff frequencies. These experimental values are ment of the acoustical properties of a thin-layered medium—The
close to those given here. The advantage of the proposed inverse problem,” J. Acoust. Soc. Am., vol. 95, no. 6, pp. 3059–3074,
method is that the determination of the parameters is 1994.
[11] A. Baltazar, L. Wang, and B. Xie, “Inverse ultrasonic determina-
performed using much more experimental data, thus im- tion of imperfect interfaces and bulk properties of a layer between
proving the precision on the final parameters. two solids,” J. Acoust. Soc. Am., vol. 114, no. 3, pp. 1424–1434, Sep.
In the last step, we reintroduce in the characteristic 2003.
[12] Y. B. Mostafa, “Etude de la propagation des ondes ultrasonores dans
equation the estimated parameters obtained to fit all the un milieu multicouche application à l’évaluation non destructive des
experimentally generated Lamb modes, even those that collages,” Ph.D. dissertation, University of Compiegne, Compiegne,
were not used for the determination of the parameters as France, 1990. Because this document is not available in English, see
[16].
S0 and M3T modes. The computation results are repre- [13] N. Alaoui Ismaili, D. Chenouni, Z. Lakhliai, M. Echcherif Elkettani,
sented in Fig. 7. One can notice the good superposition of J.-L. Izbicki, and B. Morvan, “Détermination des paramètres élas-
the fitted curves with the experimental data. tiques d’un matériau par problème inverse,” in Proc. Congrès Médi-
terranéen de Télécommunications et Exposition, Tangier, Morocco,
2008, pp. 272–275.
[14] N. Alaoui Ismaili, D. Chenouni, Z. Lakhliai, M. Echcherif Elket-
tani, J.-L. Izbicki, and B. Morvan, “Inverse problem to determine
IV. Conclusion an epoxy film parameters in a three layer metal/adhesive/metal
structure,” presented at Acoustics 08, Paris, June 26–July 4, 2008.
[15] L. M. Brekhovskikh, Waves in Layered Media. London: Academic
This paper proposes a simple method for determining Press, 1960.
[16] V. Vlasie, S. de Barros, M. Rousseau, L. Champaney, H. Duflo,
the elastic parameters and the thickness of a thin epoxy and B. Morvan, “Mechanical and acoustical study of a structural
film of a 3-layer aluminum/epoxy/aluminum structure bond: Comparison theory/numerical simulations/experiment,” Eur.
from the results of ultrasonic measurements. A prelimi- J. Mech. A, Solids, vol. 25, no. 3, pp. 464–482, 2006.
[17] W. H. Press, S. A. Teukkolsky, W. T. Vetterling, and B. P. Flan-
nary study of sensitivity allows the appropriate mode to nery, Numerical Recipes in Fortran: The Art of Scientific Computing,
be selected for each given parameter. This fact, as well 2nd ed. Cambridge: Cambridge University Press, 1992.
as the limited intervals of study based on the averaged [18] P. Marical, M. E. Kettani, and M. V. Predoi, “Guided waves in
an elastic plate with Gaussian section variation: Experimental and
and the standard deviation of each parameter, allows a numerical results,” Ultrasonics, vol. 47, pp. 1–9, Dec. 2007.
fast convergence of the algorithm. A good agreement is
observed with previous works, and an improvement on the
determination error is obtained. In addition, good agree- The authors’ photographs and biographies not available at time of pub-
ment is shown between the experimental dispersion curves lication.