Available online at www.sciencedirect.

com

ScienceDirect
Procedia Engineering 86 (2014) 427 439

1st International Conference on Structural Integrity, ICONS-2014

Advanced Signal Processing Algorithms in Structural Integrity

Monitoring
Gerges Diba,*, OleksiiKarpenkoa,Mahmoodul Haqb, Lalita Udpaa, and Satish Udpaa
a
Non-Destructive Evaluation Laboratory, Electrical and Computer Engineering Department,
Michigan State University, East Lansing, 48822, MI, USA
b
Composite Vehicle Research Center, Michigan State University, East Lansing, 48910, MI, USA
*
E-mail ID: dibgerge@msu.edu.

Abstract

Damage detection using Lead-Zirconate-Titanate (PZT) transducers for exciting guided waves was investigated. The analytical
model for guided wave excitation using PZT transducers was used in a Finite Element Model for simulating guided waves in a
thin aluminum plate with various defect depths. The fundamental symmetric (S0) mode was excited, and four sensors were placed
for detecting both the reflected and transmitted waves from defects. The model was used to conduct a parametric study of the
effect of the PZT adhesive bonding on the measured signal and hence probability of detection (POD).The model was also used
for optimizing signal analysis algorithms, particularly for the case of multiple overlapping modes due to mode conversion from
defects. A mode decomposition algorithm is proposed for separating the reflected/transmittedS0mode from the mode converted
antisymmetric mode (A0). The signal processing algorithms were optimized to maximize the POD.
2014 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
2014 The Authors. Published by Elsevier Ltd.
(http://creativecommons.org/licenses/by-nc-nd/3.0/).
Peer-review under responsibility of the Indira Gandhi Centre for Atomic Research.
Peer-review under responsibility of the Indira Gandhi Centre for Atomic Research
Keywords: Guided waves, probability of detection, structural health monitoring, mode decomposition, PZT.

1. Introduction

Guided waves are mechanical waves with a wavelength of the same order, or larger, than the thickness of
the structure they are travelling within. Guided wave inspection systems have shown the capability for detecting
defects throughout the thickness of thin structures[1-4]. Lead-Zirconate-Titanate (PZT) piezoelectric transducers
have been well studied for the actuation and sensing of guided waves for damage detection[5-7]. Guided waves have
shown their potential for in-situ structural health monitoring when their ability to interrogate large areas of a
structure is combined with the small size and efficiency of PZT transducers[1] [2].

PZT transducers are surface mounted on the structure using adhesive bonding. The phase and amplitude of

1877-7058 2014 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/3.0/).
Peer-review under responsibility of the Indira Gandhi Centre for Atomic Research
doi:10.1016/j.proeng.2014.11.056
428 Gerges Dib et al. / Procedia Engineering 86 (2014) 427 439

excited guided waves could be significantly afffected by the quality of the adhesive bonding, which co ould result in
false alarms or even miss the detection of a deffect [3] [4]. The adhesive layer is susceptible to degradatio
on over time
depending on the environmental conditions. Several studies have proposed impedance methods to monitor the
integrity of the PZT transducer and its bondinng[5] [6]. However, it is necessary to quantify the ability to detect
defects in the presence of variability in the PZZT actuators. This paper presents a model-based approach h to quantify
the effect of adhesive layer variations on the proobability of detection (POD) of cracks in an aluminum plate.

(a) (b)

Figure 1. (a) (a) Schematic for PZT bonded too the surface of a plate, b) The distribution of shear streess () at the
interface between the PZT and the structures, for
f varying adhesive shear modulus. The function was computed
c for
tb=40m, a=3.5mm, Ea=76 GPa, ta=0.2 mm, Es=71 GPa, and ts=1.9 mm.

Moreover, the analysis of guided wavve signals and damage identification still remains a challenge, due to
the multimodal and dispersive nature of guidedd waves. At any frequency, there are at least two fundameental modes,
the symmetric mode (S0) and the antisymmetric mode (A0), propagating in the structure. To sim mplify signal
processing, a technique for tuning PZT actuatoors to excite a single mode in the structure has been prop posed in [7].
However, mode conversion from defects is inevvitable. If the sensor is not far enough from the defect, thee reflected or
transmitted wave which contains both A0 and S0 modes cannot separate in time. Overlapping modes mak ke it difficult
to extract correct signal features for damage deetection. Several algorithms have been proposed to separrate partially
overlapped modes, by representing the signal inn the time-frequency domain [8] [9], or by using mode deecomposition
algorithms [10] [11].

For proper feature extraction, a matchhing pursuit algorithm using a dictionary based on the guided
g wave
analytical model is proposed. Section 2 describes the analytical solution for guided waves excited by a PZT
P actuator.
Section 3 describes the finite element model and
a the parametric study. The matching pursuit mode deecomposition
algorithm is proposed in section 5. The modell-based POD study for both cases of reflected and transm mitted waves
from defects is shown in Section 6.

2. Analytical Solution for Guided Waves Excited

E By a PZT Actuator

Consider the two dimensional case whhere a PZT actuator is bonded to the surface of a plate as shown
s in Fig
1(a). The PZT actuator has a negligible mass compared to the structure and its dynamics can be decoup pled from the
structure dynamics. This allows the use of a staatic model for computing the effective shear stress () wh
hich the PZT
actuator applies on the plate surface. Then coould be used as a boundary condition on the plate to solvve the guided
wave equation. This problem has been addresseed by Giurgiutiu[7], and is outlined in the next section.
 Gerges Dib et al. / Procedia Engineering 86 (2014) 427 439 429

2.1 Surface Shear Stress Applied By A PZT Actuator

Consider a PZT actuator with a voltage difference V applied across its thickness.The resulting shear stress
at the interface between the PZT and the plate is given by[7]:

(a) (b)

Figure 2. The frequency response of (a) S0 mode, and (b) A0 mode, with varying Gb.

t sinh x
( x) = a Ea ISA a for | x | a
a + cosh a (1)
where,
Et G 1 + d V
= s s , 2 = b , and ISA = 31 ,
Eata Ea tatb ta

d31, Ea, ta, and a are the coupling coefficient, Youngs modulus, thickness, and half length of the PZT actuator,
respectively. Es and ts are the Youngs modulus and the thickness of the plate, respectively. Gb and tb are the shear
modulus and thickness of the bonding layer, respectively. The parameter depends on the strain and stress
distribution through the thickness of the plate. In eq. (1), only the shear-lag parameter depends on the bonding
properties, and it controls the effectiveness of shear transfer between the actuator and the plate. The function (x) is
shown in Fig. 1(b) for different values of Gb and when tb= 40 m. As the value for the shear modulus increases, the
shear stress is more concentrated at the edges of the PZT, which translates into more efficient shear transfer between
the PZT and the structure.

2.2 Wave Equation Solution or Guided Waves Excited By A PZT Actuator

If a harmonic excitation voltage with angular frequency is applied to the PZT, then the applied shear
it
stress can be expressed as ( x, t ) = ( x)e . Assuming only the A0 and S0modes are present, the guided wave
in-plane displacement field is expressed as [7]:

U x ( x, ) = U x 0 ( x , ) + U x 0 ( x, ) ei t
S A
(2)

430 Gerges Dib et al. / Procedia Engineering 86 (2014) 427 439

S 1 N ( s ) i S x
U x 0 ( x, ) =  ( s ) s e
2 D ' s ( s )
N s ( s ) = s q ( s 2 + q 2 ) cos ph cos qh , D s ( s ) = ( s 2 q 2 ) 2 cos ph sin qh + 4 s 2 pq sin ph
p cos qh ,

A 1 N ( A ) i A x
U x 0 ( x, ) =  ( A ) A e
2 D ' ( A )A

N A ( ) = q ( 2 + q 2 ) sin ph sin qh , D ' A ( A ) = ( A 2 q 2 ) 2 sin ph cos qh + 4 A 2 pq coss ph sin qh ,

A A A

Figure 3. Schematic foor the finite element model parametric study.

2 2 ( + 2 ) 2
p2 = - 2 , q2 = - 2 , c2p = , cs = ,
c2p cs2

s A
Where and are Lame constants of the platee, is the plate density, and , are the wave numberrs for S0 and
A0 modes, respectively. The wave numbers couuld be calculated using the Rayleigh-Lamb dispersion eq quations[12].
 ( ) is the spatial Fourier transform of ( x ) , and it could be computed directly from eq. (1) as:
2i ta Ea ISA a
 ( ) = ( sinnh a cos a cosh a sin a ) (3)
a + 2 + 2 cosh a

Using eq. (2) and eq. (3), the frequency responnse of the guided wave could be computed as shown in Fig.
F 2. As the
shear modulus Gb decreases (lower quality adhhesive), the maximum amplitude of the frequency responsse decreases.
Also it is noted that the zero of the frequency reesponse shifts to a higher value.

3. Finite Element Model

A two-dimensional finite element model

m using Abaqus/Explicit was developed. Figure 3 shows the
schematic and material properties used in thhe finite element model. Utilizing the axisymmetric naature of this
problem, only half the PZT actuator and plate length were modeled. A zero displacement boundary co ondition was
imposed at the left boundary. An absorbing boundary on the right side of the plate was implemented d in order to
minimize reflections from the right edge. A nottch defect with thickness 0.2mm and variable depth d waas introduced
at 353.5mm from the actuator center. Two PZT T actuators with length 7mm, one at the top of the plate an
nd one at the
bottom were simulated, and both excited with the
t same voltage VBOT = VTOP , thus excitingtheS0 mode only.
o At low
 Gerges Dib et al. / Procedia Engineering 86 (2014) 427 439 431

frequencies, the S0 mode is less sensitive to PZT sensor degradation that the A0 mode, due to its broader frequency
response, as shown in Fig. 2.

Running a multi-physics simulation that includes the PZT actuator, the plate, and the adhesive layer would
require significant computing time. A parametric study with tens of simulations would be impractical. To avoid this
problem, eq. (1)was applied directly as a distributed surface traction load at the interface between the PZT and the
structure. Thus the PZT actuator itself need notbe meshed and simulated. The mesh element size was chosen to be
0.1 mm, and the time steps was 15 ns, which satisfies the requirement for wave propagation in FEM[13].

A 5 cycle Gaussian modulated tone burst with center frequency of 190 KHz is used for the excitation. The
axial displacement (x-direction) of four mesh nodes shown in Fig. 3 was measured. Two of the sensor nodes (Rx1
and Rx2) are placed such that the defect is outside the actuator and sensor path, and hence the sensors measure the
GW signal reflected from the defect. The other two sensor nodes (Tx1 and Tx2) are placed such that the defect is
between the actuator and the sensor, and thus the sensors measured the transmitted signal. Since in a practical
situation, the position of the defect is not known relative to the sensors, the sensors positions are chosen to model
two representative cases. In the first case, Rx1 Sensor and Tx1 Sensor are chosen far away from the defect, enough
for the S0 and A0 modes converted at the defect to separate temporally.

(a) (b)

(c) (d)

Figure 4. Simulated signals for tb= 40 m, and d = 0.8 mm, and two different adhesive shear moduli Gb. (a) Tx1
Sensor, (b) Rx1 Sensor, (c) Tx2 Sensor, and (d) Rx2 Sensor.

The signals measured at those sensors are used for the POD study in Section 5. Rx2 Sensor and Tx2 Sensor
are chosen to be close to the defect so that the two modes do not separate temporally. Section 4 describes a mode
decomposition algorithm to separate the S0 mode from the overlapping A0 converted mode.

The finite element model is used for conducting a parametric study of the effects of variations in PZT
actuator bonding layer on the resulting signal. The defect depth (d) is varied between 0 mm (no defect) to 1.8 mm in
432 Gerges Dib et al. / Procedia Engineering 86 (2014) 427 439

steps of 0.1 mm, resulting in a total of 19 meshed geometries. For each defect, the excitations TOP(x,t) = BOT(x,t)are
computed from eq.(1). The two parameters in eq. (1)that are controlled by the adhesive are Gb and tb. The bonding
thickness tb is usually known during manufacturing, and Gb is the parameter that mostly represents the bonding
quality. The adhesive shear modulus (Gb) is varied from 0.02 GPa to 3 GPa. A simulation was run for each
combination of (Gb, d). Figure 4 shows the signals from the four different sensors for two values of Gb. When the
defect is between the actuator and sensor (Fig. 4(a),(c)), there are two wave packets: the transmitted S0 mode and
then the A0 mode that is converted at the defect. Since the group velocity of A0 is much lower than S0, it needs more
time to reach the sensor. However, it can be seen that when the defect is closer to the sensor (Fig. 4(c)), the two
modes are partially overlapping. When the defect is not between the actuator and sensor (Fig. 4(b),(d)), there are
three wave packets: the incident S0 mode, the reflected S0 mode and then the A0 mode that is converted at the defect.

(a) (c)

(b) (d)

Figure 5. Dictionary atoms for (a) S0 mode, (b) A0 mode, at x=250 mm; (c) bimodal atom at x=30mm, and (d)
bimodal atom at x=80mm.

4. Mode Decomposition Algorithm

Reliable feature extraction in guided wave structural health monitoring requires the implementation of a
mode decomposition algorithm. In the present parametric study, mode decomposition is particularly useful in
identifying the exact amplitudes of transmitted/reflected modes from defects in the plate.

In this section, an iterative algorithm based on matching pursuit (MP) is used to separate temporally
overlapping modes and extract the amplitude features required for the calculation of POD. MP decomposes a time
signal into a weighted sum of functions with limited time support, called atoms. The set of atoms forms a redundant
dictionary Dof size M that spans the signal space. The result of decomposition is represented as:
 Gerges Dib et al. / Procedia Engineering 86 (2014) 427 439 433

N
x (t ) = n d n (t ) , (4)
n =0
where x(t ) is the input signal, n are weight coefficients, d n (t ) are the over complete dictionary atoms, and
(N+1) is number of atoms required to represent x(t) with a given accuracy.
MP iteratively projectsx(t ) onto the elements of the dictionary in order to find the sparsest combination in eq.(4). If
we denote the residual of the signal at ith iteration as Ri , the key steps of matching pursuit can be described as
follows:

(1) Specify initial condition: R0 (t ) = x (t ) .

(a) (b)

(c) (d)

(e) (f)

Figure 6. The results for the mode decomposition algorithm when the defect size is 0.9 mm: (a) Detected signal at
Tx1 sensor, (b) Decomposed transmitted S0 mode at Tx1, (c) Decomposed transmitted A0 mode at Tx1; (d) Detected
signal at Rx1 sensor, (e) Decomposed reflected S0 mode at Rx1, and (f) Decomposed reflected A0 mode at Rx1.
(2) Compute the inner product of the residual with each atom in the dictionary, and select the atom that results in
434 Gerges Dib et al. / Procedia Engineering 86 (2014) 427 439

the largest product: i = arg max Ri 1( t ), d k ( t ) ,for k=1M.

(3) Subtract selected atom from the current ressidual: Ri = Ri 1 i .d k max (t ) .

The decomposition is completed whenn the energy of the residue is sufficiently small or the allowed number
of iterations has been exceeded. Selection of the MP dictionary atoms determines the qualitty of mode
decomposition. The result will be the most acccurate if the shape of the atoms match the shape of wav ve packets in
simulated or experimental signals. On the othher hand, the dictionary should be easy enough to consttruct without
time consuming finite modeling or conductingg experiments. Therefore, the dictionary is constructed by y generating
atoms based on eq. (2), which describe the PZ
ZT excited guided wave propagation. Equation(2) describ bes the wave

Figure 7. Schematic describbing the procedure for computing the POD curves.

field for a single harmonic frequency. For a geeneral band limited excitation signal with a Fourier transfform F ( ) ,
three sets of atoms are computed as:

d k (t ) = ) 1 F ( )U 0 ( xk , ) , ) 1 F ( )U 0 ( xk , ) , ) 1 F ( )U ( xk , ) (5)
A S
xk xk xk

representing a pure A0 mode, pure S0 mode, annd the bimodal case for when the wave packets are overlaapping in the
vicinity of a defect. Since the distance betweeen the defect and the sensor is not known, atoms were generated
g for
different values of the propagation distance xk ranging from x = 0mm to the farthest sensor position Tx1, where
x=533.7 mm, with 2 mm steps. The time spann of the atoms was set to 200 s to match the length of FE modeled
waveforms. Figure 5demonstrates typical S0, A0, and S0+A0 wave packets included in the dictionary to represent
transmission/reflection from the defect. Sincee the relative reflection and transmission coefficients off the S0 and
A0modes are unknown, the generated bimodal atoms should also take into account relative amplitudes of S0 and A0
wave packets. Therefore, dictionary components corresponding to this set consisted of two blocks. In thhe first block
the energy of S0 mode was fixed and the energyy of A0 was varied from 1% to 99% of the S0 energy. Con nversely, the
second block included atoms with fixed energyy of A0 wave packets, while the energy of S0 mode was vaaried. Atoms
in the blocks were also allowed to be flipped in polarity. The result for mode decomposition algorithm for the case
when defect depth is 0.8mm is shown in Fig. 6ffor the signals detected at Tx1 and Rx1 sensors.
 Gerges Dib et al. / Procedia Engineering 86 (2014) 427 439 435

5. Effects of PZT adhesive on POD

The procedure for estimating the POD curve is described in Fig. 7. The implementation of this procedure is
described next:
(1) The envelop of the time signal was computed by utilizing the Hilbert transform. Then the peak amplitude of
the first reflected/transmitted wave packet was computed, and used as the feature for damage indication. The
mode decomposed algorithm was used for the case when two or more modes are overlapping. In the case when
the defect is between sensor and actuator, the peak amplitude was subtracted from the no defect peak
amplitude. Figure 8 shows the peak amplitudes with varying shear modulus for four defect depths.

1.6mm defect 1.6mm defect

40 1.1mm defect 1.1mm defect
40
0.6mm defect 0.6mm defect
20 30
SNR (dB)

SNR (dB)
20
0
10
-20 0.1mm defect
0.1mm defect 0
-40
-10
0.5 1 1.5 2 2.5 3 0.5 1 1.5 2 2.5 3
Shear Modulus (GPa) Shear Modulus (GPa)

(a) (b)

Figure 8. Variation of (a) transmitted, (b) reflected, S0 mode peak amplitude with varying shear modulus, for four
different defect depths. The circular markers represent the peak values extracted from the FEM data, and the solid
lines represent the fitted function plotted using eq. (6).

(2) The Displacement-Shear Modulus curves (Fig. 8) for each defect depth, are fitted by a fourth order log-
polynomial function yi = fi(x),where yi is the reflection/transmission amplitude for defect depth i, and x
represents the Shear Modulus value (x = Gb).
yi = Ai ln 4 x + Bi ln3 x + Ci ln 2 x + Di ln x +Ei (6)
Where Ai, Bi, Ci Di, and Ei are the polynomial coefficients for the function at each defect depth i. The adhesive
layer shear modulus could vary during manufacturing, but mostly it varies during operation, in which case its
value will decrease. Moreover, the value for the shear modulus should be positive. This is best modeled using
the unilateral Gamma distribution. The bonding layer shear modulus (x=Gb) was assigned a Gamma
probability density function (Fig. 9) with equation:
x k 1e x
p X ( x )= (7)
k ( k )
where the scale parameter = 0.25e9, and the shape parameter k = 3.5, and (k ) is the gamma function. The
mean shear modulus is x = k = 0.75 GPa .
(3) The probability density function of yi, which is itself a function of the random variable x, can be computed
from eq. (6) and eq. (7) using change of variable which gives:
436 Gerges Dib et al. / Procedia Engineering 86 (2014) 427 439

dx
pY ( yi ) = PX ( x ) (8)
dyi
dx
where the term is the derivate of the inverse of the function yi. From Fig. 8, it can be seen that the function
dyi
yi is monotonous, and thus the derivate of its inverse could be calculated as:
dx 1 x
= = (9)
dyi dyi 4 A ln3 x + 3B ln 2 x + 2C ln x + D
i i i i
dx

Figure 9. The Gamma probability density function for =0.25e9 and k=3.5.

(a) (b)

Figure 10. The PDF of the (a) transmitted, and (b) reflected signal amplitude with noise for the no defect case
(dotted line), and the defect depths from 0.1mm to 0.7mm (solid lines).

The PDF pY(yi) is then computed for each defect depth.

 Gerges Dib et al. / Procedia Engineering 86 (2014) 427 439 437

(4) The effect of measurement noise is modeled by an additive Gaussian distributed noise to the peak amplitude
measurements yi such that:
zi = yi + N (0, ) (10)
2
Where N(0, ) is a number drawn from a zero mean Gaussian distribution, with a variance . The variance was
chosen such that = 10pm.The measurement zi is the sum of two independent random variables, and hence its
probability density function could be calculated by the convolution of the probability density functions of the
two variables:
pZ ( zi ) = ( pY * N )( zi ) (11)
i i

The PDF for the noisy measurement for each defect depth are shown in Fig. 10.
(5) To calculate the POD from the PDF calculated in eq. (11), a threshold needs to be selected based on
requirements on the maximum allowed percentage of false alarms, or the minimum defect to be detected. After
selecting a threshold T, the POD for each defect size was calculated from the cumulative density function:

(a) (b)

Figure 11. The POD curves for (a) transmitted and (b) and reflected signals.

PODi = pZ ( zi ) dzi (12)
T
Using a 5% false alarm rate criterion, the threshold is T = 17 pm. Then for each pZ ( zi ) the PODiwas calculated,
i

and the resulting POD curves for the transmission and reflection are shown in Fig. 11. It shows that if the defect is
between the actuator and sensor (transmission), defect depths more than 30% of total thickness have over 99%
probability of detection. When the defect is not between the actuator and sensor (reflection), defect depths more
than15% of total thickness have over 99% probability of detection. Those results show that the POD is higher for the
case of pulse-echo, where reflections from the defect are measured. However, the detection results are also affected
by the selected wave mode, and frequency.

6. Conclusion

The use of PZT actuator for guided wave structural health monitoring was studied in this paper. A mode
decomposition algorithm based on matching pursuit is used for accurate feature extraction of transmitted/reflected
S0 mode amplitudes. A dictionary of atoms based on the actual guided wave equation solutions were used for
obtaining accurate decomposed modes. A parametric study using a finite element model was conducted to find the
POD when the quality of the adhesive bonding degrades. Experimental verification of the obtained results is
currently being conducted, and will be published in a future work. Moreover, a study of the POD using different
wave modes (symmetric and antisymmetric modes), and different excitation frequency would be helpful to
quantitatively optimize defect detection.
438 Gerges Dib et al. / Procedia Engineering 86 (2014) 427 439

7. References

[1]Ajay Raghavan and Carlos E. S. Cesnik, "Review of guided-wave structural health monitoring," The Shock and
Vibration Digest, vol. 39, no. 2, 2007.
[2]Zhigang Sun, Bruno Rocha, Kuo-Ting Wu, and Nezih Mrad, "A methodological review of piezoelectric based
acoustic wave generation and detection techniques for structural health monitoring," International Journal of
Aerospace Engineering, 2013.
[3]Sungwon Ha and Fuo-Kuo Chang, "Adhesive interface layer effects in PZT-induced lamb wave propagation,"
Smart Materials and Structures, vol. 19, no. 2, 2010.
[4]Xinlin P. Qing, Hian-Leng Chan, Shawn J. Beard, Teng K. Ooi , and Stephen A. Marotta, "Effect of adhesive
on the performance of piezoelectric elements used to monitor structural health," International Journal of
Adhesions and Adhesives, vol. 26, no. 8, 2006.
[5]Seunghee Park, Gyuhae Park, Chung-Bang Yun, and Charles R. Farrar, "Sensor self-diagnosis using a modified
impedance model for active sensing-bases structural health monitoring," Structural Health Monitoring, vol. 8,
no. 1, 2008.
[6]Stuart G. Taylor, Gyuhae Park, Kevin M. Farinholt, and Michael D. Todd, "Diagnostics for piezoelectric
transducers under cyclic loads deployed for structural health monitoring applications," Smart Materials and
Structures, vol. 22, no. 2, 2013.
[7]Victor Giurgiutiu, "Tuned Lamb wave excitation and detection with piezoelectric wafer active sensors for
structural Health monitoring," Journal of Intelligent Material Systems and Structures, vol. 16, no. 4, 2005.
[8]L. De Marchi, A. Marzani, S. Caporale, and N. Speciale, "Ultrasonic guided-waves characterization with
warped frequency transforms," IEEE Transactions on ultrasonics, Ferroelectrics and Frequency Control, vol.
56, no. 10, 2009.
[9]W. Prosser, N. Seale, and B. Smith, "Time-frequency analysis of the dispersion of Lamb modes," Journal of
Acoustical Society of America, vol. 105, no. 5, 1999.
[10]Ajay Raghavan and Carlos Cesnik, "Guided-wave signal processing using chirplet matching pursuits and mode
correlation for structural health monitoring," Smart Materials and Structures, vol. 16, no. 2, 2007.
[11]J. C. Hong, K. H. Sun, and Y. Y. Kim, "Waveguide damage detection by the matching pursuit approach
employing the dispersion-based chirp functions," IEEE Transactions on Ultrasonics, Ferroelectrics and
Frequency Control, vol. 53, no. 3, 2006.
[12]Igor Viktorov, Rayleigh and Lamb Waves. New York: Plenum Press, 1967.
[13]Friedrich Moser, Laurence J. Jacobs, and Jianmin Qu, "Modeling elastic wave propagation in waveguides with
the finite element method," NDT&E, vol. 32, no. 4, 1999.
[14]David N. Alleyne and Peter Cawley, "The Interaction of Lamb Waves with defects," IEEE Transactions on
Ultrasonics, Ferroelectrics, and Frequency Control, vol. 39, no. 3, 1992.
[15]Jayant Sirohi and Inderjit Chopra, "Fundamental Understanding of Piezoelectric Strain Sensors," Journal of
Intelligent Material Systems and Structures, vol. 11, 2000.
[16]Xiang Zhao, Roger L. Royer, Steven E. Owens, and Joseph L. Rose, "Ultrasonic Lamb wave tomography in
structural health monitoring," Smart Materials and Structures, vol. 20, no. 10, 2011.
[17]Chao Zhou, Zhongqing Su, and Li Cheng, "Probability-based diagnostic imaging using hybrid features
extracted from ultrasonic Lamb wave signals," Smart Materials and Structures, vol. 20, no. 12, 2011.
[18]Zhongqing Su, Lin Ye, and Ye Lu, "Guided Lamb waves for identification of damage in composite structures: A
review," Journal of Sound and Vibration, vol. 295, no. 3-5, 2006.
 Gerges Dib et al. / Procedia Engineering 86 (2014) 427 439 439

[19]Jayant Sirohi and Inderjit Chopra, "Fundamental behavior of piezoceramic sheet actuators," Journal of
Intelligent Material Systems and Structures, vol. 11, no. 1, 2000.
[20]Jeong-Beom Ihn and Fu-Kuo Chang, "Pitch-catch Active Sensing Methods in Structural Health Monitoring for
Aircraft Structures," Structural Health Monitoring, vol. 7, no. 1, 2008.