Ultrasonics 40 (2002) 605–610

www.elsevier.com/locate/ultras

Modeling the propagation of ultrasonic waves in the interface

region between two bonded elements
a,* b,1
P.P. Delsanto , Sigrun Hirsekorn , V. Agostini a, R. Loparco a, A. Koka b

a
Dipartimento di Fisica, INFM – Dip. Fisica, Politecnico di Torino, Corso Duca degli Abruzzi, 24, 10129 Torino, Italy
b
Fraunnhofer Institut fuer zerstoerungsfreie Pruefverfahren (IZFP), Universitaet, Geb. 37, D-66123 Saarbruecken, Germany

Abstract

An important task in nondestructive materials evaluation is the development of techniques to characterize the bond quality of
adherent joints. Binding forces are nonlinear and cause a nonlinear modulation of transmitted and reflected ultrasonic waves. As a
consequence, the higher harmonics generated by an insonified monochromatic wave give information about the adhesive bonds. The
local binding forces in thin bonded interfaces can be obtained by the amplitudes of the ultrasonic waves of the insonified frequency
and its higher harmonics as transmitted through the interface. Additional phase measurements may enable one to obtain the
evaluation of the full hysteretic cycle of the interaction force. In order to gain a deeper understanding of the interface region and to
improve the technique, numerical simulations of the ultrasonic wave propagation through specimens of two bonded elements can be
used. A simple model based on the local interaction simulation approach (LISA) is described in this contribution, and a comparison
between the results of the simulations and the experimental data is presented. Besides its intrinsic relevance for NDE, the problem
considered in this paper may be very useful to analyze and test models for the simulation of ultrasonic wave propagation in
nonclassical nonlinear mesoscopic elastic materials. Ó 2002 Elsevier Science B.V. All rights reserved.

Keywords: Adhesion; Binding forces; Higher harmonics; Nonlinear ultrasound; Numerical simulations

"  2  2 #
1. Introduction o2 u S o ou ou ou
¼ þb þd þ    þ Aðe; e_Þ;
ot2 q ox ox ox ox
For traditional nonlinear materials the one-dimen-
sional wave equation governing the propagation of ð1Þ
acoustic (or ultrasonic) pulses may be written by simply where q is the density and S the elastic modulus. These
expressing the driving force for the local displacement u materials include many of interest in seismology, such as
as a power series in the strain e ¼ ou=ox. For these rock, sand, and soil, and others of relevance in industry
materials, called atomic elastic materials, elasticity arises and civil engineering, such as concrete and atomic elastic
from well known forces at the atomic or molecular level. materials with microscale damages [2]. All these mate-
These materials are well described by the classical rials share several important properties, such as a hys-
(Landau) theory of elasticity [1]. There are, however, teretic behavior and discrete memory in quasi-static
several materials, which we shall call nonlinear meso- experiments, and the appearance of effects in resonant
scopic elastic (NME), for which such a description is wave experiments, such as a shift of the resonance fre-
inadequate. For them it is necessary to add a nonana- quency when increasing the driving amplitude, the
lytical term Aðe; e_Þ in the wave equation: generation of higher harmonics at a prescribed rate in
the response, and the phenomenon known as slow dy-
namics. Since these effects may be very large they can be
*
advantageously exploited for NDE purposes [3,4].
Corresponding author. Tel.: +39-011-5647320; fax: +39-011- Nonclassical nonlinear effects are believed to be due to
5647399.
E-mail addresses: [email protected] (P.P. Delsanto), hirsek-
the presence of softer regions in hard materials. Typi-
[email protected] (S. Hirsekorn). cal examples are microcracks, flat pores, and soft
1
Tel.: +49-681-9302-3836; fax: +49-681-9302-5903. bonding regions between grains in granular matter. In

0041-624X/02/$ - see front matter Ó 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 0 4 1 - 6 2 4 X ( 0 2 ) 0 0 1 8 3 - X
606 P.P. Delsanto et al. / Ultrasonics 40 (2002) 605–610

fact, when microscopic contacts are present, dislocations all kinds of interface mechanisms. Recently, the PM-
and static friction may modify the material strength [5]. space model of Guyer and McCall was implemented
Therefore, oscillations induced by an external forcing numerically in the LISA framework to simulate, at the
may result in the accumulation of energy in the proximity 1-D level, the influence of the local nonlinear elastic
of the contact areas. If the energy becomes large enough, properties on the wave propagation in nonclassical
some bonds may be broken or weakened, with noticeable materials [15,16]. To our knowledge, this is the only
consequences even at a macroscopic level [6]. study that explicitly incorporates the phenomenological
While the traditional nonlinear theory [1] is inade- PM-model in a macroscopic simulation of dynamic
quate to describe nonclassical nonlinearity, hysteresis nonlinearity and hysteresis. To be more specific, let us
and discrete memory have been successfully reproduced discretize a bar of a multi-grained material as a 1-D
by the model of Guyer and McCall [7]. Their model is lattice of ‘‘cells’’ representing the grains, separated by
based on a Preisach–Mayergoyz (PM) representation in other cells representing the interstices between contigu-
analogy with the treatment of magnetic hysteresis. It ous grains. It becomes then possible to devise an almost
provides a simple phenomenological representation of endless variety of ways in which the necessary ingredi-
the complex elastic behavior of an elementary elastic ents of the problem, i.e. linear and nonlinear elasticity,
unit (a grain, the cement paste between grains, a mi- nonclassical behavior, attenuation, etc., may be dis-
crocrack, etc.) in the composition of an arbitrary tributed among the two kinds of cells and implemented,
material. Each unit is described by an elementary con- e.g., just as in the original model of Guyer and McCall
stitutive law that accounts for effects such as nonlin- [7], in the model of [12] the interstice length was arbi-
earity, hysteresis, and memory. The collection of all trarily set to vary between two fixed values, except that,
units, each of them with its particular constitutive rela- taking advantage of the numerical nature of the treat-
tion, yields the so-called ‘PM-space’, which characterizes ment, the variation was continuous rather than abrupt.
the material specimen. Guyer and McCall have used the On the contrary, in a more recent version of the model
PM-space to predict the static mechanical behavior of [13] that restriction was released and a restoring force
rocks in agreement with macroscopic observations. included in order to keep the interstice from collapsing
For the implementation of the model, analytical ap- under compression or breaking apart under tension.
proaches usually require a drastic simplification of the Of course, of all the possible models only a few stand
problem and may not succeed in reproducing a whole a chance under the requirements of making sense from
set of observed phenomena. The use of numerical cal- a physical point of view and being able to reproduce,
culations can be a very advantageous alternative for a at least qualitatively, all the known phenomenology.
more complete theoretical analysis, including the ex- However, there does not seem to be, at present, a de-
tension of a basic one-dimensional model to higher di- finitive answer about which model is most realistic or,
mensions. Computer models based on a microscopic even, whether a unique model might fit all cases.
approach, such as ab initio calculations and molecular In order to investigate this very basic question we
dynamics techniques, are most commonly used and al- believe that it is expedient to analyze in great detail a
low to understand atomic-scale features such as the problem, in which only one soft region exists, i.e. the
formation of dislocations and their behavior under an characterization of an adhesive interface between two
applied stress [8]. However, these methods are usually linear elastic components, e.g. plates, using nonlinear
not practical because of the huge CPU time required, acoustics. This problem, of a great practical relevance
even for the simulation of a small collection of atoms. on its own, is discussed in the next section. In Section 3,
Therefore, a bridging of microscopic mechanisms to- the formalism for the determination of the relative
wards a macroscopic description is imperative. binding force in the interface region [17–21] is reviewed.
In wave propagation applications, Delsanto et al. The specimen and experimental results are briefly de-
[9,10] proposed a new approach for numerical simula- scribed in Section 4. Finally, in Section 5, the discreti-
tions of macroscopic wave phenomena in complex het- zation scheme and simulation approach are discussed,
erogeneous media. The approach is based on the local and a comparison between simulation and experimental
interaction simulation approach (LISA) [11–13] in con- data is presented.
junction with a spring model [14]. A very important
feature of LISA is, as its name implies, the capability of
implementing at the local level even very complex 2. Quality assessment of bonded interfaces by nonlinear
mechanisms, which would be extremely hard to include ultrasonic transmission
in partial differential equations. In fact, the method al-
lows full freedom in the choice of the interactions among The load capacity of composite materials is often
the nodes which represent the material cells. limited by the quality of the bonded interfaces. There-
By splitting the nodes at the interfaces between dif- fore, an important task in NDE is the development of
ferent material components, it is also possible to include techniques to characterize the bond quality or even to
 P.P. Delsanto et al. / Ultrasonics 40 (2002) 605–610 607

measure the bond strength of adherent joints. Binding additional distortion of the interface, which is neglected
forces are nonlinear and cause a nonlinear modulation of as a higher order effect. The distortion of the interface
transmitted and reflected ultrasonic waves. As a conse- by the transmitted wave of fundamental frequency keeps
quence, the generated higher harmonics of an insonified the sinusoidal form of the interface vibration but only
monochromatic wave will yield informations about the causes a phase shift and change in amplitude. It is to be
adhesive bonds. expected that, because of the nonlinearity of the inter-
Thin bonded interfaces in composite materials can be action force F ðaÞ, the strain amplitude eBI in the inter-
described by two surfaces very close to each other held face is nonlinearly related to its cause, the strain
together by adhesive forces. These binding forces in- amplitude ein of the incident wave in the plates.
crease with interface distance for strong joints first Assuming for the time being only classical nonlin-
linearly and then nonlinearly. For weak joints the non- earity and comparing Eq. (3) with a power series ex-
linear behavior sets in at an earlier stage reaching a pansion of F ðaðtÞÞ at the static equilibrium distance ae
maximum at a distance where the joints break. Because [18,19,21], we obtain for the phases un
the distance in the interface is modulated by ultrasonic
waves, their reflection and transmission is characteris- sin u2n ¼ ð1Þnþ1 ; cos u2n ¼ 0; n ¼ 1; 2; 3; . . . ; ð4aÞ
tically influenced by the nonlinearity of the binding n
cos u2nþ1 ¼ ð1Þ ; sin u2nþ1 ¼ 0; n ¼ 0; 1; 2; . . .
forces, e.g. in the case of a monochromatic incident
wave higher harmonics are generated. ð4bÞ
The amplitudes of the transmitted harmonics and Hence,
combinations of them are measures of the nonlinearity 1
of the interaction forces and, as a consequence, of the
X
F ðaðtÞÞ ¼ F0 þ F2nþ1 ð1Þn sin½ð2n þ 1Þxt
bond strength of the interface. Conventional measures n¼0
are e.g. the so-called distortion factor K or the nonlin- 1
X nþ1
earity parameter b (see e.g. [20]). Another possibility is þ F2n ð1Þ cos½2nxt : ð5Þ
to investigate the binding force curve itself and evaluate n¼1
the interaction forces as function of the interface dis- The consideration of the special times within a vi-
tance by the measured amplitudes and phases of the bration cycle for which xte ¼ pm and xtmax ¼ p=2 þ 2pm,
ultrasonic waves transmitted through the interface. where m is an integer, yields the series of force amplitudes
representing the static equilibrium binding force F ðae Þ
and the interaction force at the maximum interface
3. Theory distance ae ð1 þ eBI Þ within the same vibration cycle.
Their difference
A monochromatic compressional wave of circular
frequency x incident perpendicularly onto a bonded DF ¼ F ðae ð1 þ eBI ÞÞ  F ðae Þ ¼ F1  2F2 þ F3  ð6Þ
interface modulates its static equilibrium distance ae si- represents the corresponding relative binding force
nusoidally, so that the interface distance becomes (RBF). In Eq. (6), Fn ðn ¼ 1; 2; . . .Þ can be computed
aðtÞ ¼ ae ð1 þ eBI sin xtÞ; ð2Þ simply as Fn ¼ Cen , where C is the elastic modulus of the
where eBI is the interface strain amplitude. The resulting plates in the direction perpendicular to the adhesive
dynamic interaction force can be represented as a sum of interface. The strain amplitude in the interface eBI as
sinusoidal dynamic forces [17] well as the strain amplitudes of the transmitted ultra-
sonic waves en and as a consequence DF are functions of
1
F ðaðtÞÞ ¼ F0 þ
X
Fn sinðnxt þ un Þ; ð3Þ the incident strain amplitude ein . By plotting DF versus
n¼1 ein one obtains a curve, whose maximum represents the
dynamic bond strength of the interface [18,19,21].
where Fn and un are the amplitudes and phases of the
various force components. The constant part F0 is due to
the mean interface distance during insonification, which
in the case of nonlinear interaction forces differs from 4. Experiments
the static equilibrium interface distance ae . The dynamic
parts generate the transmitted (and reflected) ultrasonic The experimental set-up to measure ultrasonic
waves, so that the phases un are transferred to the transmission through bonded interfaces is described
transmitted waves of fundamental frequency and higher elsewhere [20]. Therefore, we limit ourselves to sketch
harmonics. Their strain amplitude en multiplied by the in Fig. 1 the specimen, which is a structure consisting
elastic constant relevant for the propagation of the of two 4 mm thick Al plates with an interface of epoxy
waves in the linear elastic plates are equal to the force of 0.1 mm thickness. The specimen is insonified per-
amplitudes Fn . The generated higher harmonics cause an pendicularly to the interface by a compressional
608 P.P. Delsanto et al. / Ultrasonics 40 (2002) 605–610

The measured transmitted signals are fast Fourier

transformed (FFT). The strain amplitudes of the fun-
damental frequency (e1 ) and its higher order harmonics
ðe2 ; e3 ; . . .Þ are evaluated as function of the strain am-
plitude ein of the incident wave. From the amplitudes of
the transmitted harmonics the relative binding force at
maximum interface distance during a vibration cycle is
then evaluated by means of Eq. (6). Fig. 2 shows the
resulting RBF curves for the five locations P1 to P5 . As
expected the curves are first linear, then the nonlinearity
becomes important and bends the curves, which reach a
Fig. 1. Sketch of the specimen, which consists of two Al plates joined maximum corresponding to the dynamic bond strength.
together by an epoxy layer. The pairs of transmitters and receivers are We observe that not only the bond strength increases
also indicated.
markedly when one moves to the least damaged mid-
section of the specimen, but also that the corresponding
input strain amplitude ein increases, i.e. a larger strain
can be tolerated.

5. The simulation approach

As mentioned in Section 1, for the interpretation of

the experimental results and optimization of the pro-
posed technique we have adopted the LISA [9–13].
Following the discretization scheme shown in Fig. 3, the
specimen is discretized into a very large number N of
tiny segments (or ‘‘cells’’) of equal length d, each rep-
resented by a grid-node in its center. The problem of
ultrasonic wave propagation in the bar is equivalent to
Fig. 2. Measured relative binding forces versus input strain amplitude the setting in motion of a system of elastic springs
for the five interface locations.
connecting each node to its two neighbors. The nodes
within the Al plates are connected by linear springs Fn ,
quasi-monochromatic ultrasonic wave (30 cycle tone- representing the force due to the node n 1 acting on
burst of frequency f ¼ 2:25 MHz). We have considered the node n.
five different positions for the transducer/receiver pairs We assume that the interstice region, also of length d,
corresponding to five different interface locations ðP1 ; is delimited by the nodes i and i þ 1, with an interaction
þ 
. . . ; P5 Þ characterized by an increasing level of defec- force FBI and FBI , respectively, given by the RBF (see
tiveness (due to water infiltration from the edges). Eq. (6)) represented for the five interface locations in

Fig. 3. Schematic representation of the 1-D discretization of the specimen.

 P.P. Delsanto et al. / Ultrasonics 40 (2002) 605–610 609

for best convergence (s is the time step), we obtain the

iteration equations
ui;tþ1 ¼ uiþ1;t þ ui1;t  ui;t1 : ð10Þ
For the two interface nodes (i and i þ 1) we have
q0 o€ui ¼ FBI
þ
 Fi ; ð11aÞ

q0 o€uiþ1 ¼ Fiþ1
þ 
 FBI ; ð11bÞ
0
where q is the average density in the cells around the
interface nodes. From Eqs. (11a) and (11b) by applying
Eq. (7) one then obtains the corresponding iteration
equations.
As an application of this simple-minded model we
Fig. 4. An example of fitting the force curves (RBF versus ein ) by
can simulate the propagation of an arbitrary pulse
means of fourth order polynomial.
through the specimen. Then, from a FFT of the re-
sponse at the receiver, we can evaluate the transmitted
Fig. 2. In order to perform the numerical calculations we
harmonics. By considering the same pulse as in the ex-
fit the five curves separately by means of fourth order
periment leading to Fig. 2, we can compare the results of
polynomials
the simulations with the experimental harmonics. Fig. 5
DF ¼ ae þ be2 þ ce3 þ de4 ; ð7Þ displays such a comparison for one out of the five
where e is the incident strain amplitude. An example of chosen interface locations (P2 ). Similar results have been
the fitting for one interface location (P2 ) is reported in obtained for the other four points P1 , P3 , P4 , and P5 (not
Fig. 4. It is important to notice that only the points up shown for brevity). The agreement is qualitatively sat-
to the peak are relevant, since beyond that Eq. (6) may isfactory particularly for e1 and e3 . It must be remarked
be no longer valid. that, as mentioned before, the comparison is meaningful
Using the discretization scheme of Fig. 3, from Eq. only up to the strain corresponding to the peak of the
(1) we obtain the linear wave equation RBF, so that the last 10 points should be disregarded.
While we consider the results of Fig. 5 encouraging,
un ¼ Fnþ  Fn
qo€ for 1 6 n < i and i þ 1 < n 6 N ; we expect that, in order to achieve a better agreement
ð8Þ and be able to apply the method for practical purposes, a
more sophisticated discretization scheme should be
i.e. for the nodes in the Al-plates. From Eq. (8), ap-
adopted, e.g. as provided by the sharp interface model
plying the basic LISA formalism [22] and choosing
(SIM) [22]. Another probably more advantageous al-
s2 S ¼ qd2 ; ð9Þ ternative can be to consider the two interface nodes i and

Fig. 5. Comparison between the experimental amplitudes of the transmitted harmonics and the corresponding amplitudes obtained from the sim-
ulations.
610 P.P. Delsanto et al. / Ultrasonics 40 (2002) 605–610

i þ 1 as two subnodes belonging to the same node, as in a Acknowledgements

Spring Model treatment [11]. This procedure allows us to
tailor at will the interface properties. Also, a finer grid This work has been supported by the INFM Parallel
could be adopted with a large number of cells used to Computing Initiative and by the ESF-PESC program
describe the interface region, which are necessary if thick NATEMIS. The experimental work at IZFP was carried
interfaces are considered. It is important to remark that, out within a common project with the Department of
in performing the numerical simulations, a special cau- Mechanical Engineering, Emperial College, London
tion must be exercised lest convergence be lost. This is a (Prof. Peter Cawley). The authors are also grateful to
tricky proposition in nonlinear problems, since for them P.A. Johnson, K. Van Den Abeele, M. Scalerandi, and
no sure-fire convergence criterion exists [23]. W. Arnold for fruitful discussions.

6. Conclusions References

[1] L.D. Landau, E.M. Lifshitz, Theory of Elasticity, Pergamon

The quality assessment of adhesive joints is an in- Press, Oxford, 1986.
trinsically very important NDE problem, both during [2] R.A. Guyer, P.A. Johnson, Phys. Today April (1999) 30.
their production and life cycle. In addition, since the [3] K. Van Den Abeele, P.A. Johnson, A. Sutin, et al., Res.
interface between two components (e.g. two Al plates) Nondestruct. Eval. 12/1 (2000) 17–30.
[4] K. Van Den Abeele, J. Carmeliet, J.A. TenCate, P.A. Johnson,
can be considered as a soft region in a hard material, it is
Res. Nondestruct. Eval. 12/1 (2000) 31–42.
expected that nonclassical nonlinear effects arise from [5] F. Heslot, T. Baumberger, B. Perrin, B. Caroli, C. Caroli, Phys.
the propagation of an ultrasonic pulse in such a bilay- Rev. E 49 (1994) 4973.
ered specimen. Since a very large number of interstices [6] J.A. Tencate, E. Smith, R.A. Guyer, Phys. Rev. Lett. 85 (2000)
are found in most other materials, which display a 1020–1024.
[7] K.R. McCall, R.A. Guyer, Nonlinear Process. Geophys. 3 (1996)
nonclassical nonlinear behavior, such a specimen with
89–101.
only one soft region can be used as a ‘‘prototype’’ for [8] F. Clery, S. Yip, D. Wolf, S.R. Phillpot, Phys. Rev. Lett. 79 (1999)
studying these effects in the simplest possible case, in 1309–1312.
which one can afford to test and compare many different [9] P.P. Delsanto, R. Mignogna, M. Scalerandi, R. Schechter, in: P.P.
models of the soft region. This is particularly timely at Delsanto, A.W. Saenz (Eds.), New Perspectives on Problems in
Classical and Quantum Physics, Gordon & Breach, New Delhi,
present, due to the current debate about which is the
1998, pp. 51–74.
most suitable model for performing numerical simula- [10] P.P. Delsanto et al., Wave Motion 20 (1994) 295–314.
tions to reproduce the nonclassical behavior in nonlin- [11] M. Scalerandi, A. Romano, C.A. Condat, Phys. Rev. E 58 (1998)
ear mesoscopic elastic materials. 4166–4173.
The problem of simulating the ultrasonic wave [12] M. Scalerandi, Il Nuovo Cimento B 114 (1999) 523–534.
[13] B. Capogrosso Sansone, M. Scalerandi, C.A. Condat, Phys. Rev.
propagation in a two bounded elements specimen and of
Lett. 87 (2001) 128102.
calculating the relative binding force at the interface [14] P.P. Delsanto, M. Scalerandi, J. Acoust. Soc. Am. 104 (1998)
between them has been considered in this contribution. 2584–2591.
A simple model, based on the LISA, has been adopted [15] M. Scalerandi, E. Ruffino, P.P. Delsanto, P.A. Johnson, K. Van
yielding an agreement with the corresponding experi- Den Abeele, in: D.O. Thompson, D.E. Chimenti (Eds.), Review of
mental data, which is encouraging, but not yet sufficient Progress in Quantitative Non Destructive Evaluation, vol. 19B,
2000, pp. 1393–1399.
to warrant a practical application. Alternative discreti- [16] E. Ruffino, M. Scalerandi, Il Nuovo Cimento B 115 (2000) 645–
zation schemes have also been proposed. By considering 652.
the oscillations between a minimum and a maximum [17] S.U. Fabender, W. Arnold, in: D.O. Thompson, D.E. Chimenti
of the interface distance in the dynamic interaction (Eds.), Rev. Progr. QNDE, vol. 15, Plenum Press, New York,
1996, pp. 1321–1374.
force, the model might generate an intrinsic nonanalyt-
[18] S. Hirsekorn, S.U. Fabender, A. Wegner, W. Arnold, Fortsch-
ical (hysteretical) mechanisms if no assumptions con- ritte der Akustik (1998) 518–519.
cerning the phases of the transmitted waves are made [19] S. Hirsekorn, A. Koka, A. Wegner, W. Arnold, in: D.O.
(as indicated in the last chapter of Section 3). A very Thompson, D.E. Chimenti (Eds.), Rev. Progr. QNDE, vol. 19B,
relevant question to ascertain is whether this mechanisms Plenum Press, New York, 1999, pp. 1367–1374.
[20] A. Wegner, A. Koka, K. Janser, U. Netzelmann, S. Hirsekorn, W.
may be sufficient to describe the dynamic behavior of
Arnold, Ultrasonics 38 (2000) 316–321.
adherent interfaces or whether it is necessary to include [21] S. Hirsekorn, Ultrasonics 39 (2001) 57–68.
some other ad hoc protocol, as in a treatment based on a [22] P.P. Delsanto et al., Wave Motion 16 (1992) 65–80.
PM-space model. [23] P.P. Delsanto et al., Math. Comput. Model. 25 (1992) 19–29.