Wavelet spectral nite element for wave propagation in shear

deformable laminated composite plates

Dulip Samaratunga
a
, Ratneshwar Jha
a,
, S. Gopalakrishnan
b
a
Raspet Flight Research Laboratory, Mississippi State University, 114 Airport Road, Starkville, MS 39759, USA
b
Department of Aerospace Engineering, Indian Institute of Science, Bangalore 560 012, India
a r t i c l e i n f o
Article history:
Available online 25 September 2013
Keywords:
Wave propagation
Composite
Spectral nite element
Abaqus
Shear deformation theory
Structural health monitoring
a b s t r a c t
This paper presents a new 2-D wavelet spectral nite element (WSFE) model for studying wave propa-
gation in thin to moderately thick anisotropic composite laminates. The WSFE formulation is based on
the rst order shear deformation theory (FSDT) which yields accurate results for wave motion at high fre-
quencies. The wave equations are reduced to ordinary differential equations (ODEs) using Daubechies
compactly supported, orthonormal, scaling functions for approximations in time and one spatial dimen-
sion. The ODEs are decoupled through an eigenvalue analysis and then solved exactly to obtain the shape
functions used in element formulation. The developed spectral element captures the exact inertial distri-
bution, hence a single element is sufcient to model a laminate of any dimension in the absence of dis-
continuities. The 2-D WSFE model is highly efcient computationally and provides a direct relationship
between system input and output in the frequency domain. Results for axial and transverse wave prop-
agations in laminated composite plates of various congurations show excellent agreement with nite
element simulations using Abaqus

.
2013 Elsevier Ltd. All rights reserved.
1. Introduction
Wave propagation in elastic structures has been studied exten-
sively and applied for transient response prediction, mechanical
property characterization, and nondestructive evaluation (NDE)
[15]. Composite (elastic) structures are increasingly used in many
industries such as transportation (air, land, and sea), wind energy,
and civil infrastructure due to several advantages including higher
specic strength and modulus, fewer joints, improved fatigue life,
and higher resistance to corrosion. Lamb wave based structural
health monitoring (SHM), which aims to perform nondestructive
evaluation through integrated actuators and sensors, has been a
very active area of research in the past decade [610]. A validated
physics-based model for wave propagation combined with exper-
imental measurements is generally required for complete charac-
terization (presence, location, and severity) of damages.
The modeling of wave propagation in composites presents com-
plexities beyond that for isotropic structures [2,4]. Analytical solu-
tions for wave propagation are not available for most practical
structures due to complex nature of governing differential equa-
tions and boundary/initial conditions. The nite element method
(FEM) is the most popular numerical technique for modeling wave
propagation phenomena. However, for accurate predictions using
FEM, typically 20 elements should span a wavelength [11], which
results in very large system size and enormous computational cost
for wave propagation analysis at high frequencies. In addition,
solving inverse problems (as required for NDE/SHM) is very dif-
cult using FEM. Spectral nite element (SFE), which follows FEM
modeling procedure in the transformed frequency domain, is
highly suitable for wave propagation analysis [1214]. SFE models
are many orders smaller than FEM and highly suitable for efcient
NDE/SHM. Frequency domain formulation of SFE enables direct
relationship between output and input through system transfer
function (frequency response function). SFE has very high compu-
tational efciency since nodal displacements are related to nodal
tractions through frequency-wave number dependent stiffness
matrix. Mass distribution is captured exactly and the accurate ele-
mental dynamic stiffness matrix is derived. Consequently, in the
absence of any discontinuities, one element is sufcient to model
a beam or plate structure of any length.
Fast Fourier Transform (FFT) based Spectral Finite Element
(FSFE) method was popularized by Doyle [12], who formulated
FSFE models for isotropic 1-D and 2-D waveguides including ele-
mentary rod, Euler Bernoulli beam, and thin plate. Gopalakrishnan
and associates [13,15] extensively investigated FSFE models for
beams and plates-with anisotropic and inhomogeneous material
properties. The FSFE method is very efcient for wave motion anal-
ysis and it is suitable for solving inverse problems; however, FSFE
0263-8223/$ - see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compstruct.2013.09.027

Corresponding author. Tel.: +1 662 325 3797; fax: +1 662 325 3864.
E-mail addresses: dulip@raspet.msstate.edu (D. Samaratunga), jha@raspet.
msstate.edu (R. Jha), krishnan@aerospace.iisc.ernet.in (S. Gopalakrishnan).
Composite Structures 108 (2014) 341353
Contents lists available at ScienceDirect
Composite Structures
j our nal homepage: www. el sevi er . com/ l ocat e/ compst r uct
cannot model waveguides of short lengths. For 2-D problems, FSFE
are essentially semi-innite, that is, they are bounded only in one
direction [12,13]. Due to the global basis functions of the Fourier
series approximation of the spatial dimension, the effect of lateral
boundaries cannot be captured. In addition, FSFE requires assump-
tion of periodicity in time approximation resulting in wrap-
around problem for smaller time window, which totally distorts
the response.
The 2-D Wavelet based Spectral Finite Element (WSFE) pre-
sented by Gopalakrishnan and Mitra [14] overcomes the wrap-
around problem and can accurately model 2-D plate structures
of nite dimensions. WSFE uses orthogonal compactly supported
(localized) Daubechies scaling functions [16] as basis for both tem-
poral and spatial approximations. Gopalakrishnan and associates
have formulated WSFE elements for wave propagation in rods,
higher order beams, and plates with both isotropic and anisotropic
material properties [1720]. However, the 2-D WSFE plate formu-
lation presented in [14,19,20] is based on the classical laminated
plate theory (CLPT) [21]. The CLPT based formulations exclude
transverse shear deformation and rotary inertia resulting in signi-
cant errors for wave motion analysis at high frequencies, especially
for composite laminates which have relatively low transverse shear
modulus [22,23]. Wave propagation in composite laminates based
on the rst order shear deformation theory (FSDT) [21], which ac-
counts for transverse shear and rotary inertia, yields accurately re-
sults comparable with 3-D elasticity solutions and experiments
even at high frequencies [22,23]. For isotropic materials FSDT is
known to be exceptionally accurate down to wavelengths compara-
ble with the plate thickness h, whereas CLPT is of acceptable accu-
racy only for wavelengths greater than, say, 20h [24].
This paper presents a new 2-D WSFE based on FSDT for high fre-
quency analysis of waveguides with nite dimensions and aniso-
tropic material properties. Governing partial differential equations
(PDEs) for wave motion and their temporal approximation using
Daubechies compactly supported high-order scaling functions are
presented. An eigenvalue analysis is performed to decouple the re-
duced PDEs in spatial dimensions. The decoupled PDEs are then
approximated in one spatial dimension using Daubechies lower-or-
der scaling functions followed by an eigenvalue analysis similar to
the time approximation. The resulting ordinary differential equa-
tions (ODEs) are solved exactly in frequency-wavenumber domain
and the solution is used as shape function for the 2-D spectral ele-
ment. Numerical experiments are performed to highlight the differ-
ences between FSDT and CLPT in dispersion curves, provide
spectrum relationships, and present time domain responses. Results
for the new WSFE formulation are validated with Abaqus

simula-
tions using shear exible shell elements [25].
2. Formulation of wavelet spectral nite element with shear
deformation
Two dimensional wavelet spectral nite element with shear
deformation is formulated here for anisotropic composite
laminates.
2.1. Governing differential equations for wave propagation
Consider a laminated composite plate of thickness h with the
origin of the global coordinate system at the mid-plane of the plate
and Z axis being normal to the mid-plane as shown in Fig. 1(a).
Fig. 1(b) shows the corresponding nodal representation with de-
grees of freedom (DOFs). Using FSDT, the governing partial differ-
ential equations (PDEs) for wave propagation have ve degrees
of freedom: u, v, w, /, and, w. The terms u(x, y, t) and v(x, y, t) are
mid-plane (z = 0) displacements along X and Y axes; w(x, y, t) is
transverse displacement in Z direction, and w(x, y, t) and /(x, y, t)
are the rotational displacements about X and Y axes, respectively.
The displacements w, / and w do not change along the thickness
(Z direction). The quantities (N
xx
, N
xy
, N
yy
) are in-plane force resul-
tants, (M
xx
, M
xy
, M
yy
) are moment resultants, and (Q
x
, Q
y
) denote
the transverse force resultants.
The FSDT displacement eld Eq. (1) represents a constant shear
strain in the transverse (thickness) direction.
Ux; y; z; t ux; y; t z/x; y; t
Vx; y; z; t vx; y; t zwx; y; t
Wx; y; z; t wx; y; t
1
The equations of motion based on the above displacement eld are
given by [21],
A
11
@
2
u
@x
2
2A
16
@
2
u
@x@y
A
66
@
2
u
@y
2
A
16
@
2
v
@x
2
A
12
A
66

@
2
v
@y@x
A
26
@
2
v
@y
2
B
11
@
2
/
@x
2
2B
16
@
2
/
@x@y
B
66
@
2
/
@y
2
B
16
@
2
w
@x
2
B
12
B
66

@
2
w
@y@x
B
26
@
2
w
@y
2
I
0
@
2
u
@t
2
I
1
@
2
/
@t
2
A
16
@
2
u
@x
2
A
12
A
66

@
2
u
@y@x
A
26
@
2
u
@y
2
A
66
@
2
v
@x
2
2A
26
@
2
v
@x@y
A
22
@
2
v
@y
2
B
16
@
2
/
@x
2
B
12
B
66

@
2
/
@y@x
B
26
@
2
/
@y
2
B
66
@
2
w
@x
2
2B
26
@
2
w
@x@y
B
22
@
2
w
@y
2
I
0
@
2
v
@t
2
I
1
@
2
w
@t
2
KA
55

@/
@x

@
2
w
@x
2
KA
45

@/
@y

@w
@x

@
2
w
@y@x
KA
44

@w
@y

@
2
w
@y
2

I
0
@
2
w
@t
2
B
11
@
2
u
@x
2
2B
16
@
2
u
@y@x
B
66
@
2
u
@y
2
B
16
@
2
v
@x
2
B
12
B
66

@
2
v
@y@x
B
26
@
2
v
@y
2
D
11
@
2
/
@x
2
2D
16
@
2
/
@y@x
D
66
@
2
/
@y
2
D
16
@
2
w
@x
2
D
12
D
66

@
2
w
@y@x
D
26
@
2
w
@y
2
KA
55
/
@w
@x
KA
45
w
@w
@y

I
1
@
2
u
@t
2
I
2
@
2
/
@t
2
B
16
@
2
u
@x
2
B
12
B
66

@
2
u
@y@x
B
26
@
2
u
@y
2
B
66
@
2
v
@x
2
2B
26
@
2
v
@y@x
B
22
@
2
v
@y
2
D
16
@
2
/
@x
2
D
12
D
66

@
2
/
@y@x
D
26
@
2
/
@y
2
D
66
@
2
w
@x
2
2D
26
@
2
w
@y@x
D
22
@
2
w
@y
2
KA
45
/
@w
@x
KA
44
w
@w
@y

I
1
@
2
v
@t
2
I
2
@
2
w
@t
2
2
where the stiffness constants A
ij
, B
ij
, D
ij
and the inertial coefcients
I
0
, I
1
and I
2
are dened as
A
ij
; B
ij
; D
ij

Np
q1
_
z
q1
zq
Q
ij
1; z; z
2
_
dz; I
0
; I
1
; I
2
f g

_
h=2
h=2
f1; z; z
2
gqdz 3
The term Q
ij
are the stiffnesses of the q
th
lamina in laminate coordi-
nate system, N
p
is the total number of laminae (plies), q is the mass
342 D. Samaratunga et al. / Composite Structures 108 (2014) 341353
density, and K is the shear correction factor. The associated natural
boundary conditions are,
N
nn
n
2
x
N
xx
n
2
y
N
yy
2n
x
n
y
N
xy
;
N
ns
n
x
n
y
N
xx
n
x
n
y
N
yy
n
2
x
n
2
y
_ _
N
xy
Q
n
Q
x
n
x
Q
y
n
y
M
nn
n
2
x
M
xx
n
2
y
M
yy
2n
x
n
y
M
xy
;
M
ns
n
x
n
y
M
xx
n
x
n
y
M
yy
n
2
x
n
2
y
_ _
M
xy
4
where n and s denote coordinates normal and tangential to the
plate edge, respectively; and n
x
and n
y
are unit normal into X and
Y directions, respectively. The boundary conditions at edges parallel
to Y axis can be derived by setting n
x
to 1 and n
y
to zero. Then N
nn
and N
ns
become the specied normal and shear forces into X and Y
directions, M
nn
and M
ns
are the specied moments about Y and X
axes, and Q
n
is the applied shear force in Z direction. Force and mo-
ment resultants are given by
N
xx
N
yy
Q
y
Q
x
N
xy
M
xx
M
yy
M
xy
_

_
_

A
11
A
12
0 0 A
16
B
11
B
12
B
16
A
12
A
22
0 0 A
26
B
12
B
22
B
26
0 0 A
44
A
45
0 0 0 0
0 0 A
45
A
55
0 0 0 0
A
16
A
26
0 0 A
66
B
16
B
26
B
66
B
11
B
12
0 0 B
16
D
11
D
12
D
16
B
12
B
22
0 0 B
26
D
12
D
22
D
26
B
16
B
26
0 0 B
66
D
16
D
26
D
66
_

_
_

_
@u=@x
@v=@y
@w=@y w
@w=@x /
@u=@y @v=@x
@/=@x
@w=@y
@/=@y @w=@x
_

_
_

_
5
Without loss of generality in all essential aspects of the problem, a
laminate composed of an arbitrary number of orthotropic layers
such that the axes of material symmetry are parallel to the XY axes
of the plate (hence A
16
= A
26
= A
45
= B
16
= B
26
= D
16
= D
26
= 0) is con-
sidered for further analysis.
2.2. Temporal approximation
The governing PDEs (Eq. (2)) and boundary conditions (Eq. (4))
have three independent variables (x, y, and t), and derivatives with
respect to them, making it very complex to solve. Therefore,
Daubechies compactly supported scaling functions [16] are used
to approximate the time variable which reduces the set of
equations to PDEs in x and y only. Compactly supported scaling
functions have only a nite number of lter coefcients with
non-zero values, which enables easy handling of nite geometries
and imposition of boundary conditions. The use of Daubechies
compactly supported wavelets to solve partial differential wave
equations is explained in detail in [14].
Let the timespace variable u(x, y, t) be discretized at n points in
the time window (0, t
f
) and s = 0, 1, . . . , n 1 be the sampling
points, then t Dt s where Dt is the time interval between two
sampling points. The function u(x, y, t) can be approximated at an
arbitrary scale as
ux; y; t ux; y; s

k
u
k
x; yus k; k 2 Z 6
where u
k
(x, y) are the approximation coefcients at a certain spatial
dimension (x and y) and u(s) are scaling functions associated with
Daubechies wavelets. The other translational and rotational dis-
placements v(x, y, t), w(x, y, t), /(x, y, t) and w(x, y, t) are transformed
similarly. The next step is to substitute these approximations into
the Eq. (2). The approximated equation is then multiplied with
translations of the scaling function (us j; for j 0; 1; . . . ; n 1)
and inner product is taken on both sides of the equation. The
orthogonal property of Daubechies scaling function results in the
cancelation of all the terms except when j = k and yields n simulta-
neous equations. For example, the rst equation of Eq. (2)becomes
A
11
@
2
u
j
@x
2
_ _
A
66
A
12

@
2
v
j
@y@x
_ _
B
11
@
2
/
j
@x
2
_ _
B
12
B
66

@
2
w
j
@y@x
_ _
A
66
@
2
u
j
@y
2
_ _
B
66
@
2
/
j
@y
2
_ _
I
0
C
1
_ _
2
u
j
I
1
C
1
_ _
2
/
j
7
where C
1
is the rst-order connection coefcient matrix obtained
after using the wavelet extrapolation technique for non-periodic
extension. Connection coefcients are the inner product between
the scaling functions and its derivatives [26]. The wavelet extrapo-
lation approach of Williams and Amaratunga [27] is used for the
treatment of nite length data before computing the connection
coefcients. This method uses polynomials to extrapolate the coef-
cients lying outside the nite domain and it is particularly suitable
for approximation in time and the ease to impose initial values. This
(a) (b)
Fig. 1. (a) The plate element (b) nodal representation with DOFs.
D. Samaratunga et al. / Composite Structures 108 (2014) 341353 343
extrapolation technique addresses one of the serious problems with
Fourier based SFE method, namely, the wraparound problem
caused by treating the boundaries as periodic extensions. By solving
the wraparound problem, WSFE method is able to handle short
waveguides and smaller time windows efciently. Next, the cou-
pled PDEs are decoupled using eigenvalue analysis of C
1
following
the procedure in [14]. The nal decoupled form of the reduced PDEs
given in Eq. (7) can be written as
A
11
@
2
^ u
j
@x
2
A
66
A
12

@
2
^ v
j
@y@x
B
11
@
2
^
/
j
@x
2
B
12
B
66

@
2
^
w
j
@y@x
A
66
@
2
^ u
j
@y
2
B
66
@
2
^
/
j
@y
2
I
0
c
2
j
^ u
j
I
1
c
2
j
^
/
j
; j 0; 1; . . . ; n 1 8
where ^ u
j
is given by ^ u
j
U
1
u
j
, U is the eigenvector matrix of C
1
and ic
j
i

1
p
are the corresponding eigenvalues. Following ex-
actly similar steps, the other four governing PDEs (in Eq. (2)) and
the natural boundary conditions (Eq. (4)) are transformed to PDEs
in spatial dimensions (x and y) only. It should be mentioned here
that the sampling rate Dt should be less than a certain value to
avoid spurious dispersion in the simulation using WSFE. In Mitra
and Gopalakrishnan [14], a numerical study has been conducted
from which the required Dt can be determined depending on the
order N of the Daubechies scaling function and frequency content
of the load.
2.3. Spatial (Y) approximation
The next step is to further reduce each of the transformed and
decoupled PDEs given by Eq. (8)(and similarly for the other trans-
formed governing equations and boundary conditions) to a set of
coupled ODEs using Daubechies scaling function approximation
in one of the spatial (Y) dimension. Similar to time approximation,
the time transformed variable ^ u
j
is discretized at m points in the
spatial window (0, L
Y
), where L
Y
is the length in Y direction. Let
f = 0, 1, . . . , m 1 be the sampling points, then y = DYf where DY
is the spatial interval between two sampling points.
The function ^ u
j
can be approximated by scaling function u(f) at
an arbitrary scale as
^ u
j
x; y ^ u
j
x; f

l
^ u
lj
xuf l; l 2 Z 9
where ^ u
lj
are the approximation coefcients at a certain spatial
dimension x. The other four displacements ^ v
j
x; y; ^ w
j
x; y;
^
/
j
x; y; and
^
w
j
x; y are similarly transformed. Following similar
steps as the time approximation, substituting the above approxima-
tions in Eq. (8)and taking inner product on both sides with the
translates of scaling functions u(f i), where i = 0, 1, . . . , m 1
and using their orthogonality property (which results in the cancel-
ation of all the terms except when i = l), we get m simultaneous
ODEs. For example, the rst of the Eq. (2), which is temporally
reduced in Eq. (8), can be spatially reduced as
A
11
d
2
^ u
ij
dx
2
A
66
A
12

1
DY

liN2
liN2
d^ v
lj
dx
X
1
il
B
12
B
66

1
DY

liN2
liN2
d
^
w
lj
dx
X
1
il
B
11
d
2
^
/
ij
dx
2
A
66
1
DY
2

liN2
liN2
^ u
lj
X
2
il
B
66
1
DY
2

liN2
liN2
^
/
lj
X
2
il
I
0
c
2
j
^ u
ij
I
1
c
2
j
^
/
ij
10
where N is the order of Daubechies wavelet and X
1
il
and X
2
il
are the
connection coefcients for rst- and second-order derivatives [26].
It can be seen from the ODEs given by Eq. (10) that similar to
time approximation, even here certain coefcients ^ u
ij
near the
vicinity of the boundaries (i = 0 and i = m1) lie outside the spatial
window (0, L
Y
) dened by i = 0, 1, . . . , m 1. These coefcients
must be treated properly for nite domain analysis. Unlike time
approximation, these coefcients are obtained through periodic
extension for free lateral edges, while other boundary conditions
may be imposed using a restraint matrix [14]. In the present study,
the lateral boundaries are unrestrained (freefree) and boundary
conditions have been imposed using periodic extension. In addi-
tion, it allows decoupling of the ODEs using eigenvalue analysis
and thus reduces the computational cost. Here, after expressing
the unknown coefcients lying outside the nite domain in
terms of the inner coefcients considering periodic extension, the
ODEs given by Eq. (10) can be written as a matrix equation of
the form as
A
11
d
2
^ u
ij
dx
2
_ _
A
66
A
12
K
1

d^ v
ij
dx
_ _
B
11
d
2
^
/
ij
dx
2
B
12
B
66
K
1

d
^
w
ij
dx
_ _
A
66
K
1

2
f^ u
ij
g B
66
K
1

2
^
/
ij
_ _
I
0
c
2
j
f^ u
ij
g I
1
c
2
j
^
/
ij
_ _
11
where K
1
is the rst-order connection coefcient matrix obtained
after periodic extension.
The coupled ODEs given by Eq. (11) are decoupled using
eigenvalue analysis similar to that done in temporal approxi-
mation. It should be mentioned here that matrix K
1
obtained
after periodic extension has a circulant form and its eigen
parameters are known analytically [14]. Let the eigenvalues
be ib
i
i

1
p
, then the decoupled ODEs corresponding to
Eq. (11) are given by
A
11
d
2
~ u
ij
dx
2
ib
i
A
66
A
12

d~ v
ij
dx
B
11
d
2
~
/
ij
dx
2
ib
i
B
12
B
66

d
~
w
ij
dx
b
2
i
A
66
~ u
ij
b
2
i
B
66
~
/
ij
I
0
c
2
j
~ u
ij
I
1
c
2
j
~
/
ij
12
where ~ u
ij
(and similarly other transformed displacements) are given
by ~ u
ij
W
1
^ u
ij
; W is the eigenvector and ib
i
are the eigenvalues of
connection coefcient matrix K
1
. Exactly the same procedure is fol-
lowed to obtain decoupled form of the other four ODEs in Eq. (2) to
obtain
A
66
d
2
~ v
ij
dx
2
B
66
d
2
~
w
ij
dx
2
ib
i
A
66
A
12

d~ u
ij
dx
b
2
i
A
22
~ v
ij
ib
i
B
66
B
12

d
~
/
ij
dx
b
2
i
B
22
~
w
ij
c
2
j
I
0
~ v
ij
c
2
j
I
1
~
w
ij
KA
55
d
2
~ w
ij
dx
2

d
~
/
ij
dx
_ _
KA
44
b
2
i
~ w
ij
ib
i
~
w
ij
_ _
c
2
j
I
0
~ w
ij
B
11
d
2
~ u
ij
dx
2
ib
i
B
66
B
12

d~ v
ij
dx
D
11
d
~
/
ij
dx
2
ib
i
D
66
D
12

d
~
w
ij
dx
b
2
i
B
66
~ u
ij
b
2
i
D
66
~
/
ij
KA
55
d~ w
ij
dx

~
/
ij
_ _
c
2
j
I
2
~
/
ij
c
2
j
I
1
~ u
ij
B
66
d
2
~ v
ij
dx
2
D
66
d
2
~
w
ij
dx
2
ib
i
B
66
B
12

d~ u
ij
dx
b
2
i
B
22
~ v
ij
ib
i
D
66
D
12

d
~
/
ij
dx
b
2
i
D
22
~
w
ij
KA
44
ib
i
~ w
ij

~
w
ij
_ _
c
2
j
I
2
~
w
ij
c
2
j
I
1
~ v
ij
13
The natural boundary conditions (Eq. (4)) after temporal and spatial
approximations are given as
344 D. Samaratunga et al. / Composite Structures 108 (2014) 341353

N
xx
A
11
d~ u
ij
dx
ib
i
A
12
~ v
ij
B
11
d
~
/
ij
dx
ib
i
B
12
~
w
ij

N
xy
A
66
ib
i
~ u
ij

d~ v
ij
dx
_ _
B
66
ib
i
~
/
ij

d
~
w
ij
dx
_ _

Q
x
KA
55
d~ w
ij
dx

~
/
ij
_ _

M
xx
B
11
d~ u
ij
dx
ib
i
B
12
~ v
ij
D
11
d
~
/
ij
dx
ib
i
D
12
~
w
ij

M
xy
B
66
ib
i
~ u
ij

d~ v
ij
dx
_ _
D
66
ib
i
~
/
ij

d
~
w
ij
dx
_ _
14
The nal decoupled ODEs given by Eqs. (12)(14)are used for 2-D
WSFE formulation.
2.4. Spectral nite element formulation
The 2-D WSFE has ve degrees of freedom per node,
^ u
ij
; ~ v
ij
; ~ w
ij
;
~
/
ij
; and
~
w
ij
, as shown in Fig. 1(b). The decoupled ODEs
presented above are solved for ^ u
ij
; ~ v
ij
; ~ w
ij
;
~
/
ij
;
~
w
ij
and the nal
displacements u(x, y, t), v(x, y, t), w(x, y, t), /(x, y, t) and w(x, y, t) are
obtained using inverse wavelet transform twice for spatial Y
dimension and time.
The ODEs given by Eqs. (12)(14)have forms similar to that in 1-
D WSFE. For further steps, the subscripts j and i are dropped for
simplied notations and all of the following equations are valid
for j = 0, 1, . . . , n 1 and i = 0, 1, . . . , m 1 for each j. Let us con-
sider the displacement vector
~
d as follows,
~
d
~ ux
~ vx
~ wx
~
/x
~
wx
_

_
_

_
15
Since the coupled set of equations given by Eq. (13) is ordinary
equations with constant coefcients, we assume a solution of
~
d as
~
d
~
d
0
e
ijx
;
~
d
0

~ u
0
x
~ v
0
x
~ w
0
x
~
/
0
x
~
w
0
x
_

_
_

_
16
where j stands for the wavenumber. Substituting Eq. (16) into the
Eqs. (12) and (13), the set of equations can be posed in the Polyno-
mial Eigenvalue Problem (PEP) form [13]. PEP uses the concept of
the latent roots and right latent eigenvector of the system matrix
(also called the wave matrix) for computing wavenumber and
amplitude ratio matrix. PEP form of Eqs. (12) and (13) can be ex-
pressed as,
fA
2
j
2
A
1
j A
0
gf
~
d
0
g 0 17
where
Fig. 2. Uniform cantilever plate (length = L
X
, width = L
Y
, and thickness = 0.01 m).
Fig. 3. Dispersion relations for an 8-ply [0]
8
composite laminate; dashed line CLPT prediction, and solid line FSDT prediction.
Fig. 4. Dispersion relations for an Aluminum plate; dashed line CLPT prediction, and solid line FSDT prediction.
D. Samaratunga et al. / Composite Structures 108 (2014) 341353 345
A
2

A
11
0 0 B
11
0
0 A
66
0 0 B
66
0 0 KA
55
0 0
B
11
0 0 D
11
0
0 B
66
0 0 D
66
_

_
_

_
A
1

0 bA
12
A
66
0 0 bB
12
B
66

bA
12
A
66
0 0 bB
12
B
66
0
0 0 0 iKA
55
0
0 bB
12
B
66
iKA
55
0 bD
12
D
66

bB
12
B
66
0 0 bD
12
D
66
0
_

_
_

_
Fig. 5. The (a) real and (b) imaginary parts of the wavenumber of a plate with symmetric ply layup of [0
10
] based on WSFE + FSDT.
Fig. 6. The (a) real and (b) imaginary parts of the wavenumber of a plate with symmetric ply layup of [0
10
] based on WSFE + CLPT.
Fig. 7. The (a) real and (b) imaginary parts of the wave number of a plate with asymmetric ply layup of [0
5
/90
5
] based on WSFE + FSDT.
346 D. Samaratunga et al. / Composite Structures 108 (2014) 341353
A0
b
2
A66 c
2
I0 0 0 b
2
B66 c
2
I1 0
0 b
2
A22 c
2
I0 0 0 b
2
B22 c
2
I1
0 0 b
2
KA44 c
2
I0 0 ibKA44
b
2
B66 c
2
I1 0 0 b
2
D66 KA55 c
2
I2 0
0 b
2
B22 c
2
I1 ibKA44 0 b
2
D22 KA44 c
2
I2
_

_
_

_
The wavenumbers j are obtained as eigenvalues of the PEP given by
Eq. (17). Similarly, the vector
~
d
0
is the eigenvector corresponding to
each of the wavenumbers. The solution of Eq. (17) gives a 5 10
eigenvector matrix of the form
R fd
0
g
1
; fd
0
g
2
; . . . ; fd
0
g
10
18
and the solution,
~
d can be written as
f
~
dg RHfag 19
where [H] is a diagonal matrix with the diagonal terms
diagH fe
k
1
x
; e
k
1
L
X
x
; e
k
2
x
; e
k
2
L
X
x
; e
k
3
x
; e
k
3
L
X
x
; e
k
4
x
;
e
k
4
L
X
x
; e
k
5
x
; e
k
5
L
X
x
g
Fig. 8. The (a) real and (b) imaginary parts of the wave number of a plate with asymmetric ply layup of [0
5
/90
5
] based on WSFE + CLPT.
Fig. 9. (a) Unit impulse and (b) 3.5 cycle tone burst input excitations with their FFTs.
D. Samaratunga et al. / Composite Structures 108 (2014) 341353 347
Here, {a}
T
= {C
1
, C
2
, . . . , C
10
} are the unknown coefcients which can
be determined as described later. Since the procedure beyond this
step is similar to the 2-D FSFE technique [12], it is not repeated
here. Finally, the transformed nodal forces fF

e
g and transformed no-
dal displacements fu

e
g are related by
fF

e
g K

e
fu

e
g 20
where K

e
is the exact elemental dynamic stiffness matrix. The
solution of Eq. (20) and the assembly of the elemental stiffness
matrices to obtain the global stiffness matrix are similar to the
FEM. One major difference is that in the (conventional) FEM, time
integration of the equations of motion is performed using a suitable
nite difference scheme; however, in the SFEM performs dynamic
stiffness generation assembly and solution as a part of a double
do loop over frequency and horizontal wavenumber. Although this
procedure is computationally expensive, the problem size in SFEM
is so small that it does not increase overall computational cost. An-
other major difference is that, unlike FEM, SFEM deals with only one
dynamic stiffness matrix and hence matrix operation and storage
require minimum computations.
3. Numerical experiment results and discussion
The formulated 2-D WSFE is used to study axial and transverse
wave propagation in several composite laminates (graphite-epoxy,
AS4/3501) having symmetric and antisymmetric ply orientations.
Dispersion and spectrum relations are presented and compared
with CLPT based solutions. Simulation results are presented in time
domain for impulse and sinusoidal tone burst excitations.
Validation studies using FEM are performed using Abaqus

standard implicit simulations employing nine-node shell elements

with reduced integration (S9R5), which is shear exible and able to
model multiple layers.
Consider a uniform cantilever plate (Fig. 2) xed at one edge CD
and free at the other edge AB (along the Y-axis) with the other two
edges along X-axis in freefree condition. The in-plane dimensions
are L
X
and L
Y
along X and Y axes, respectively, while the thickness
(h) is constant at 0.01 m. The material properties are as follows:
E
1
= 144.48 GPa, E
2
= E
3
= 9.63 GPa, G
23
= G
13
= G
12
= 4.128 GPa,
m
23
= 0.3, m
13
= m
12
= 0.02, and q = 1389 kg m
3
.
3.1. Dispersion relations
Several authors have investigated the effect of shear deforma-
tion on dispersion relations for both composite laminates and iso-
tropic plates. We briey present a comparison of FSDT versus CLPT
dispersion curves for bending modes, before proceeding to the
application of WSFE for numerical experiments. Readers are re-
ferred to [23,28] for a detailed presentation of the equations used
for computing dispersion relations. Fig. 3 shows the dispersion
relations for an 8-ply [0]
8
laminate with the lamina properties gi-
ven above. Both theories give similar results in the low frequency
range (less than 30 kHz-mm) for the rst anti-symmetric mode
(A
0
), but CLPT over estimates the velocities at higher frequencies.
At 2000 kHz-mm, phase and group velocities based on CLPT are
3.9 and 7.7 times higher than FSDT predictions. Further, FSDT pre-
dicts second anti-symmetric mode (A
1
) and rst shear mode (SH
1
)
Fig. 10. (a) Axial and (b) transverse velocity response measured at the midpoint of edge AB (tip) of the [0]
8
cantilever plate to a 200 kHz-mm tone burst input load applied
along AB.
348 D. Samaratunga et al. / Composite Structures 108 (2014) 341353
due to the allowed shear exibility. It is to be mentioned that the
shear exibility included in FSDT comes at the expense of
increased number of degree of freedoms in the governing PDEs
causing increased computation time. While it is possible for
higher-order theories to provide even more accurate predictions,
computational effort becomes critical. Of the approximate plate
theories available, the FSDT with transverse shear exibility
appears to provide best compromise of solution accuracy, econ-
omy, and simplicity [21]. Fig. 4 shows dispersion relations for an
isotropic (aluminum) plate with the dimensions given in Fig. 2
and Youngs modulus and Poissons ratio as 68.9 GPa and 0.33,
respectively. At 2000 kHz-mm, phase and group velocities based
on CLPT are 1.7 and 3 times higher than FSDT predictions. The large
differences in group velocity predictions underscore the superior-
ity of FSDT (over CLPT) for high frequency wave analysis, especially
in composite laminates which have relatively lower transverse
shear modulus. Results presented in [1,17,18] show that FSDT pre-
dictions compare very well with 3-D elasticity theory and experi-
mental measurements for both composite and isotropic materials.
3.2. Spectrum relations
The spectrum relation for a laminate with L
X
= 0.5 m,
L
Y
= 0.25 m and symmetric ply lay up of [0
10
] is presented based
on the newly formulated element. Eq. (17) is used for obtaining
WSFE + FSDT spectrum relations, while readers are referred to
the formulations in [15,20] for WSFE + CLPT spectrum relations.
Fig. 5 shows real and imaginary parts of the wavenumbers for a
Y wave number of 50. The spectrum relation predicted by
WSFE + CLPT for the same laminate is shown in Fig. 6 for compar-
ison purposes. Similarly, spectrum relations for a composite
laminate with asymmetric ply sequence [0
5
/90
5
] are shown in
Figs. 7 and 8. It is observed that the wavenumbers have signicant
real and imaginary parts, implying their inhomogeneous nature
(attenuation while propagation). The wavenumbers have been
obtained with a sample time of Dt = 2 ls, giving the Nyquist fre-
quency of f
nyq
= 250 kHz. WSFE predicts accurate wave numbers
up to only a certain fraction of the Nyquist frequency [14]. This
fraction depends only on the order of Daubechies scaling function
(N) used for approximating the original PDEs in time. That fraction
is approximately 0.6 with N = 22 [29]; therefore, results are accu-
rate up to 150 kHz (0.6 * Nyquist frequency) only.
As observed in Figs. 3 and 5, there are ve modes corresponding
to each degree of freedom present in the FSDT based governing
equations. The numbers from 1 to 5 represent the cutoff frequen-
cies for each DOF where the wavenumbers change from imaginary
to real. These cutoff frequencies denote the arrival of propagating
modes and appear in the sequence of w, v, u, / and w. When the
axial-exural coupling is not present (B
ij
= 0), the in-plane modes
(u, v) behave the same way as they appear in the spectrum relation
for CLPT and the same is true for the exural mode (w) at low
frequencies. However, the wavenumber estimation for the exural
Fig. 11. Transverse velocity response to broadband impulse input at tip with measurement points at (a) tip and (b) center of the cantilever laminate with ply sequence of
[0]
10
.
Fig. 12. Transverse velocity response to 20 kHz 3.5 cycles tone burst input measured at (a) tip and (b) center of the cantilever laminate with ply sequence of [0]
10
.
D. Samaratunga et al. / Composite Structures 108 (2014) 341353 349
mode is under predicted by CLPT at high frequencies in the absence
of rotary inertia and transverse shear exibility terms in the for-
mulation. First shear mode (/) becomes real after the fourth cutoff
frequency whereas the corresponding mode for CLPT (dw/dx) is
completely imaginary at high frequencies. The second shear mode
(w) is an additional mode appearing in FSDT results only, which
starts propagating after the fth cutoff frequency.
Moreover, the imaginary parts of the shear wave numbers have
higher magnitudes compared to the imaginary parts of the fourth
wave number in CLPT. This indicates comparatively higher modal
damping obtained in the FSDT simulations. As happened in CLPT,
both w and / wave numbers initially contain nonzero real and
imaginary parts denoting the propagation of inhomogeneous
waves. When considering the laminate with asymmetric ply
sequence (Figs. 7 and 8), the wavenumber patterns remain qualita-
tively same. However, a close observation reveals increase in cutoff
frequencies in modes w, u and w, indicating that the asymmetry
prolongs propagation of these modes.
3.3. Response analysis in time domain
Time domain responses are studied using two types of input
excitations, namely sinusoidal tone burst and impulse (Fig. 9). Tone
burst is essentially a windowed sinusoidal signal with a limited
number of frequencies around the center frequency and it is nar-
row banded in the frequency domain. This type of input is used
when a minimal number of frequencies are desired in the response
to minimize wave dispersion. On the other hand, impulse
Fig. 13. Axial wave propagation over [0/90]
2s
laminate with dimensions L
X
= 1.0 m, L
Y
= 0.5 m, thickness = 0.01 m at time instants (a) 150 ls and (b) 250 ls for an impulse
input load.
350 D. Samaratunga et al. / Composite Structures 108 (2014) 341353
excitation consists of multiple frequencies (broadband in
frequency domain) which cause multiple modes being propagated
through a waveguide. In this study, axial and transverse loads are
applied at the tip of the cantilever plate along edge AB (Fig. 2). The
spatial distribution of load along the Y-axis is given by
FY e
Y=a
2
21
where a (= 0.030.05, as specied for each analysis) is a constant
and can be varied to change Y distribution of the load. The 2-D
WSFE is formulated with Daubechies scaling function of order
N = 22 for temporal and N = 4 for spatial approximations. The time
sampling rate is Dt = 1 2 ls (as specied for each analysis), while
the spatial sampling rate DY is varied according to DY = L
Y
/(m 1)
depending on L
Y
and the number of spatial discretization m. A
higher order scaling function is used in the temporal approximation
due to large input load frequencies. For a sampling time of 1 ls
(Nyquist frequency f
N
= 500 kHz) and Daubechies scaling function
with N = 22, the analysis results are considered accurate up to a
frequency of 500 * 0.6 = 300 kHz. The factor of 0.6 holds for Daube-
chies scaling function of order N = 22; readers may refer to [14] for
further details of this relationship. On the other hand, scaling func-
tion of order N = 4 for spatial approximation is sufcient consider-
ing the wavenumbers present in the simulations.
The time domain responses for WSFE + FSDT and WSFE + CLPT
are compared for both axial and transverse velocities. The laminate
dimensions considered are L
X
= L
Y
= 1.0 m and thickness = 0.01 m.
Time sampling rate is set at Dt = 2 ls with the number of spatial
discretization points m = 64 (DY = 1/64 m). The 200 kHz-mm tone
burst input load is applied at the edge AB with a spatial variation
given by Eq. (21) with a = 0.03. Fig. 10(a) shows that the axial
responses predicted by both laminate theories are nearly identical;
however, the transverse response prediction (Fig. 10(b)) shows
large difference between the two theories. The reection from
xed boundary (edge CD in Fig. 2) for CLPT prediction arrives
approximately twice as fast as FSDT based response. This is consis-
tent with the difference in group velocities predicted by the two
Fig. 14. Transverse wave propagation over [0/90]
2s
laminate with dimensions L
X
= 1.0 m, L
Y
= 0.5 m, thickness = 0.01 m at time instants (a) 250 ls and (b) 512 ls for an
impulse input load.
D. Samaratunga et al. / Composite Structures 108 (2014) 341353 351
theories as shown in Fig. 3(b). Another signicant difference is that
the group velocity based on FSDT is non-dispersive at this fre-
quency-thickness, whereas CLPT prediction is in the dispersive re-
gion. Therefore, the xed boundary reection using FSDT retains
the shape of the excited wave packet, but CLPT shows a highly dis-
persed wave. In summary, transverse shear and rotary inertia have
highly signicant effect on the wave motion at high frequencies
which necessitates the use of FSDT for accurate predictions.
3.4. Validation with conventional FEM
WSFE + FSDT results are compared with nite element simula-
tions using Abaqus

standard implicit direct solver [25]. The lam-

inate has dimensions of L
X
= L
Y
= 0.5 m, time sampling rate
Dt = 1 ls with the number of spatial discretization points m = 64
(DY = 1/64 m). Both unit impulse and tone burst input loads are
applied at the edge AB with a spatial variation given by Eq. (21)
with a = 0.05. In the absence of any discontinuity, it is sufcient
to model the uniform laminate structure shown in Fig. 2 using just
one element in WSFE. Abaqus

model required a very rened mesh

with 7688 nine-node shell elements (S9R5) for high frequency sim-
ulations. As a result, WSFE analysis takes only 44s, whereas Aba-
qus

simulation takes 4379s, which is two orders of magnitude

higher than WSFE. Figs. 11 and 12 show transverse velocities mea-
sured at two points on the laminate (mid-point of tip and center of
the laminate) for impulse and tone burst inputs, respectively. It is
observed that WSFE + FSDT and Abaqus

results have excellent

correlation.
The rst peak of tip response shown in Fig. 11(a) is the incident
impulse which lasts for only a short time. For an impulse (broad
band) input, each frequency component propagates at its group
velocity and the xed boundary reection has multiple wave com-
ponents with the fastest one reaching the tip at approximately
700 ls. Responses measured at the center of the laminate
(Fig. 11(b)) show several reections, including the rst one arriving
soon after the incident impulse. For a sinusoidal tone burst input
(Fig. 12), both forward propagating and reected wave packets
are concentrated in time as well as frequency domains. As noted
earlier, the wave motion is non-dispersive at this frequency-thick-
ness (200 kHz-mm) thus preserving the shape of the wave packet.
Most of the Lamb wave based SHM techniques employ sinusoidal
tone burst input which makes signal interpretation less difcult
in the non-dispersive frequency range [7].
3.5. Axial and transverse wave visualization
The wavenumbers and amplitude ratios (j and d
0
)obtained by
solving Eq. (17), are used along with the relation between trans-
formed nodal forces and displacements (Eq. (20)) to determine
the unknown coefcients {a} in Eq. (19). Then Eq. (19) gives the
displacement at any X 2 [0, L
X
] in the transformed frequency-wave-
number domain. The wave velocities at sampling points along
Y-axis and at any point along X-axis can then be obtained from a
single simulation. For wave visualization, the laminate dimensions
considered are L
X
= 1.0 m, L
Y
= 0.5 m and thickness = 0.01 m with a
symmetric ply layup of [0/90]
2s
. Time sampling rate is set at
Dt = 2 ls with the number of spatial discretization points m = 64
(DY = 1/64 m). Fig. 13 shows the axial wave propagation in the
laminate at two different time instants for an unit impulse input
load applied at edge AB (Fig. 2) with a spatial variation given by
Eq. (21) with a = 0.03. Since the axial waves travel non-
dispersively, time domain response preserves the shape of the ex-
cited wave packet for any propagated distance. Further, Fig. 13(b)
shows the directional dependence of the wave velocity. For the
cross-ply laminate conguration, the stiffness is highest in 0 and
90 directions, resulting in the highest wave velocity along them.
Fig. 14 shows the exural wave response of the plate for an im-
pulse input load applied in the transverse direction. It is observed
that the exural (anti-symmetric) wave propagation velocity is
much smaller than the axial (symmetric) wave propagation veloc-
ity (Fig. 13) for this input load bandwidth. Further, Fig. 14(b) shows
the dispersive nature of the anti-symmetric mode. The wave is dis-
persed all over the plate due to multiple wave components with
different group velocities.
4. Concluding remarks
A new 2-D wavelet spectral nite element (WSFE) with shear
deformation was presented for accurate modeling of high fre-
quency wave propagation in anisotropic composite laminates.
The governing PDEs based on FSDT were used to account for trans-
verse shear strains and rotary inertia. Daubechies compactly sup-
ported, orthonormal, scaling functions were employed for
temporal and spatial approximations to reduce the governing PDEs
to ODEs in one spatial variable only. An eigenvalue analysis was
used to decouple ODEs which were solved exactly to obtain the
shape functions used in the element formulation. Since the devel-
oped element captured the exact inertial distribution, a single ele-
ment was sufcient to model a laminate of any dimension in the
absence of discontinuities. Further, due to the compact nature of
Daubechies scaling functions, the WSFE was able to model nite
waveguides without the signal wraparound problem (associated
with Fourier transform based spectral nite element).
Group/phase velocity dispersion relations for fundamental anti-
symmetric mode were compared between FSDT and CLPT showing
large discrepancy at high frequencies, especially in the case of
composite laminates which have lower transverse shear modulus
(compared to isotropic materials such as aluminum). Spectrum
relations describing the relationship between wavenumber and
frequency were presented and compared for the two theories
(FSDT and CLPT). The wavenumbers showed both real and imagi-
nary parts indicating attenuation while propagating. Comparison
between different ply sequences showed that the asymmetric ply
sequence ([0
5
/90
5
]) prolonged the cutoff frequency, the critical fre-
quency where the wavenumber changed from imaginary to real.
Time domain analysis was performed for two types of input loads,
narrowband tone burst and broadband impulse. Axial wave prop-
agation showed similarity between FSDT and CLPT, but exural
wave velocity was much faster for CLPT resulting in spurious
boundary reections. The developed WSFE was validated with
Abaqus

FEM simulations using shear exible elements. Excellent

correlation was observed in the results and WSFE computation
time was less than two orders of magnitude compared to Abaqus

.
Finally, a few snapshots at different time instants were presented
to visualize the wave motion in composite laminates using WSFE
simulations.
Acknowledgments
The authors gratefully acknowledge partial funding for this re-
search through AFOSR Grant Number FA9550-09-1-0275 (Program
Managers: Dr. Victor Giurgiutiu and Dr. David Stargel).
References
[1] Graff KF. Wave motion in elastic solids. Mineola (NY): Dover Publications Inc.,;
1991 [11501].
[2] Rose JL. Ultrasonic waves in solid media. New York: Cambridge University
Press; 2004.
[3] Nayfeh AH. Wave propagation in layered anisotropic media: with applications
to composites, vol. 39. North Holland; 1995.
[4] Datta SK, Shah AH. Elastic waves in composite media and structures: with
applications to ultrasonic nondestructive evaluation. CRC Press; 2008.
352 D. Samaratunga et al. / Composite Structures 108 (2014) 341353
[5] Chimenti DE. Guided waves in plates and their use in materials
characterization. Appl Mech Rev 1997;50:24784.
[6] Raghavan A, Cesnik C. Review of guided-wave structural health monitoring.
Shock Vib Dig 2007;39:91114.
[7] Giurgiutiu V. Structural health monitoring: with piezoelectric wafer active
sensors. MA: Elsevier; 2008.
[8] Su Z, Ye L, Lu Y. Guided lamb waves for identication of damage in composite
structures: a review. J Sound Vib 2006;295:75380.
[9] Boller C, Chang F-K, Fujino Y. Encyclopedia of structural health
monitoring. West Sussex (UK): John Wiley & Sons Ltd.; 2009.
[10] Diamanti K, Soutis C. Structural health monitoring techniques for aircraft
composite structures. Prog Aerosp Sci 2010;46:34252.
[11] Ochoa OO, Reddy JN. Finite element analysis of composite laminates, vol.
7. Springer; 1992.
[12] Doyle JF. Wave propagation in structures: spectral analysis using fast discrete
Fourier transforms. 2nd ed. New York: Springer-Verlag; 1997.
[13] Gopalakrishnan S, Roy Mahapatra D, Chakraborty A. Spectral nite element
method. London (United Kingdom): Springer-Verlag; 2007.
[14] Gopalakrishnan S, Mitra M. Wavelet methods for dynamical problems. Boca
Raton (FL): CRC Press; 2010.
[15] Chakraborty A, Gopalakrishnan S. A spectrally formulated plate element for
wave propagation analysis in anisotropic material. Comput Methods Appl
Mech Eng 2005;194:442546.
[16] Daubechies I. Ten lectures on wavelets, vol. 61. Philadelphia: Society for
Industrial and Applied Mathematics; 1992.
[17] Mitra M, Gopalakrishnan S. Spectrally formulated wavelet nite element for
wave propagation and impact force identication in connected 1-D
waveguides. Int J Solids Struct 2005;42:4695721.
[18] Mitra M, Gopalakrishnan S. Wavelet based spectral nite element for analysis
of coupled wave propagation in higher order composite beams. Compos Struct
2006;73:26377.
[19] Mitra M, Gopalakrishnan S. Wavelet based 2-D spectral nite element
formulation for wave propagation analysis in isotropic plates. Comput Model
Eng Sci 2006;1:1129.
[20] Mitra M, Gopalakrishnan S. Wave propagation analysis in anisotropic plate
using wavelet spectral element approach. J Appl Mech 2008;75:014504-14-6.
[21] Reddy JN. Mechanics of laminated composite plates and shells. Boca Raton:
CRC Press; 2004.
[22] Prosser WH, Gorman MR. Plate mode velocities in graphite/epoxy plates. J
Acoust Soc Am 1994;96:9027.
[23] Tang B, Henneke II EG, Stifer RC. Low frequency exural wave propagation in
laminated composite plates. Plenum Press; 1988.
[24] Rose LRF, Wang CH. Mindlin plate theory for damage detection: source
solutions. J Acoust Soc Am 2004;116:15471.
[25] Abaqus Theory Manual. Version 6.12-1. Providence (RI): Dassault Systmes
Simulia Corp.
[26] Beylkin G. On the representation of operators in bases of compactly supported
wavelets. SIAM J Numer Anal 1992;6:171640.
[27] Williams JR, Amaratunga K. A discrete wavelet transform without edge effects
using wavelet extrapolation. J Fourier Anal Appl 1997;3:43549.
[28] Duke Jr JC, Henneke II EG, Stinchcomb WW, Ultrason. Stress wave charact of
composite mater. NASA contractor report 3976; 1986.
[29] Mitra M, Gopalakrishnan S. Extraction of wave characteristics from wavelet
based spectral nite element formulation. Mech Syst Signal Process 2006;20:
204679.
D. Samaratunga et al. / Composite Structures 108 (2014) 341353 353