Composite Structures 215 (2019) 317–330

Contents lists available at ScienceDirect

Composite Structures
journal homepage: www.elsevier.com/locate/compstruct

Numerical analysis of elastic wave mode conversion on discontinuities T

⁎
T. Wandowski , P. Kudela, W.M. Ostachowicz
Polish Academy of Sciences, Institute of Fluid-Flow Machinery, 14 Fiszera St., 80-231 Gdansk, Poland

A R T I C LE I N FO A B S T R A C T

Keywords: In this paper results of numerical analysis of S0/A0 elastic wave mode conversion at discontinuities in glass fibre
Guided waves reinforced polymer (GFRP) plates are presented. Analysis of S0/A0 mode conversion effect at Teflon inserts
Mode conversion simulating delamination (circular Teflon inserts located at a different depth, Teflon inserts with different shapes
Spectral element method at the same depth) and impact damage with energies 5 J, 10 J and 15 J is conducted. Numerical analysis is based
Scanning laser Doppler vibrometry
on spectral element method (SEM) in the time domain. Numerical model includes a composite plate, a piezo-
Impact damage
electric transducer and bonding layer for transducer and damage. Numerical model takes also into account the
influence of material damping. Numerical results are compared with experimental results based on full wavefield
measurements using scanning laser Doppler vibrometry (SLDV). It is shown that based on effects of S0/A0 mode
conversion it is possible to detect the damage and determine its shape and size.

1. Introduction testing, [7,8] and guided wave based technique using: piezoelectric
transducer for wave excitation and sensing [9,10], air-coupled trans-
Fibre reinforced polymer (FRP) materials are very popular in dif- ducer [11], laser source for wave excitation [12] or scanning laser
ferent areas of industry like aerospace, maritime and automotive. This Doppler vibrometer (SLDV) for wave sensing [13–15] are utilised. The
is caused by numerous advantages of utilisation of fibre reinforced industry is looking for a very fast and reliable damage assessment
composite materials. Application of composite materials in the struc- methods. Conventional ultrasound testing is a very popular method,
tures allows reducing the weight simultaneously achieving very good e.g. for inspection of aircraft. However, this method must be utilised by
mechanical strength. Weight reduction is especially very important for highly qualified personnel able to interpret the results. The methods
aerospace structures because it allows reducing also the fuel con- based on thermography are very fast but there is still an issue with the
sumption and CO2 emission. Besides, composite materials can be easily interpretation of results [16]. Recently, a method based on the full
formed in desired shapes. There is also an important issue related to the wavefield of elastic guided waves based on laser vibrometry has be-
application of FRP composites: there are prone to impact damage. Such come very popular. Measurement of full wavefield data by the laser
damage is created as a source of transverse force applied to the struc- vibrometer takes much longer than thermography measurement on the
tural composite part. It could be bird strike, fall of the wrench by me- same area. However, there is research related to the improvement of the
chanic or collision. Impact damage not always is related to visual de- speed of testing using laser vibrometry based on compressive sensing
fects located on the surface, very often such damage is barely visible [17]. There are many works related to elastic wave propagation method
and called barely visible impact damage (BVID) or completely hidden in utilised for damage assessment of composite structures. Artificial neural
the internal structure of composite part. Detection of such damage is networks were utilised for the impact identification purpose on com-
crucial because it allows preventing failure of the structure with cata- posite stiffened panels based on signals registered by piezoelectric
strophic results. Therefore many research works are related to the de- sensors [18]. The time reversal focusing method was utilised for impact
velopment and improvement of non-destructive testing techniques imaging in a composite panel with stiffeners attached to wing box [9]
(NDT) which are utilised in order to assess the structural state of and in anisotropic structures [19]. Model-independent spatial wave-
composite structures. The aim of these techniques is not only to detect number filter could be also utilised for impact imaging in composite
and localise the damage related to operational/exploitation process of structures like fuel tank [10]. Moreover, there are many new problems
structure but also defects created at the manufacturing stage. Such in- with other types of damage in composite structures like thermal de-
spection methods like thermography [1,2], eddy current technique gradation, contaminations by hydraulic fluid and release agent con-
[3,4], X-ray tomography and micro-tomography [5,6], ultrasonic tamination, moisture or weak bonds due to poorly cured adhesives.

⁎
Corresponding author.
E-mail address: [email protected] (T. Wandowski).

https://doi.org/10.1016/j.compstruct.2019.02.076
Received 9 October 2018; Received in revised form 18 January 2019; Accepted 18 February 2019
Available online 19 February 2019
0263-8223/ © 2019 Elsevier Ltd. All rights reserved.
T. Wandowski, et al. Composite Structures 215 (2019) 317–330

Therefore researchers all over the world are looking for new so-called to A0 mode which is called A0′ [37]. However, there is still a small
extended-non-destructive testing (ENDT) methods [20]. One such pro- number of works including numerical as well as experimental results
mising method is based on elastic wave propagation and SLDV. There is related to the analysis of mode conversion phenomenon for Lamb
a lot of the research performed to make the process of results inter- waves. This phenomenon is related to the conversion of one type of
pretation for SLDV method automatic. SLDV is based directly on the wave mode in the other type due to interaction with discontinuities.
elastic wave propagation method which utilizes the fact that damage is Mode conversion phenomenon was utilised for estimation of rectan-
a source of the anomaly of elastic wave propagation. Anomalies in gular notch parameters in the metallic plate [38]. The phenomenon of
elastic wave propagation manifest as reflection, scattering or mode Rayleigh wave mode conversion to Lamb wave mode in the case of the
conversion. Based on the analysis of elastic wave propagation in the thick metal plate due to crack was investigated numerically and ex-
structure it is possible to detect and localize damage. In the approach perimentally [39]. Moreover, the problem of shear horizontal (SH)
based on SLDV, elastic waves are generated by piezoelectric transducer mode conversion in the tapered aluminium plate was also considered
permanently bonded on the surface of the structure. Wave sensing [40]. Besides, the phenomenon of torsional mode conversion in pipes
process is conducted in a non-contact manner using laser vibrometer. was investigated in the literature [41]. The effect of continuous mode
Numerical simulations related to elastic wave propagation method conversion in CFRP plate was noticed [28]. 3D scanning laser vi-
are very important. Results of numerical simulations give information brometry and mode conversion effect could be utilized for damage
about wave propagation phenomenon in complex structures with da- detection in the composite plate [13]. It is possible to develop an al-
mage. A numerical model that gives results that agree with experi- gorithm for isolate anomalous waves that are mode-converted from
mental results could be utilized in model assisted signal processing incident waves from damage interaction [42]. Their amplitudes could
methods in structural health monitoring (SHM). This allows improving be transformed into an image in which the existence and location of
its sensitivity and efficiency (automatic data interpretation). Many re- defects can be visualized. The boundaries of defects can be identified by
search have been done in the field on numerical modelling of elastic utilizing the statistical damage threshold calculation component, which
wave propagation in materials with isotropic properties [21,22], or- facilitates accurate damage size evaluation. The problem of non-sym-
thotropic properties (composites reinforced by long fibres) [23–26] and metrical location in respect to the thickness of Teflon insert in com-
anisotropic properties (composites reinforced by randomly oriented posite panels was investigated in [37]. S0/A0′ mode conversion occurs
short fibres) [27]. for all non-symmetrical locations of Teflon inserts what is obvious.
There are many methods utilized for the purpose of modelling the Results indicated that even for Teflon insert symmetrically located in
phenomenon of elastic wave propagation. The first method utilized for respect to the thickness of composite panel mode conversion S0/A0′
this purpose is the finite element method (FEM). This method was was noticed. It proved that perfect symmetry was not achieved in real
utilized for the problem of wave propagation in the composite struc- conditions.
tures [28,29]. Simulations of guided wave propagation were performed The main focus of the paper is the analysis of S0/A0′ mode con-
based on the commercial finite element modelling (FEM) packages in- version in thin composite panels caused by discontinuities. It is reported
cluded in ABAQUS, ANSYS, and COMSOL and a custom developed code that this phenomenon can be potentially used for damage detection
based on the elasto-dynamic finite integration technique (EFIT) [14]. purposes. Hence, extensive numerical studies were performed. Such
Hybrid boundary element method was also utilised for simulation of discontinuities like circular Teflon inserts located at a different depth of
wave propagation for example in a waveguide with varying thickness the composite specimen, Teflon inserts with different shapes located at
[30]. Moreover, a semi-analytical finite element (SAFE) method for the same depth and impact damage are investigated both numerically
modelling of wave propagation in waveguides even with arbitrary and experimentally.
cross-section [31]. The SAFE method requires the finite element dis-
cretization of the cross-section of the investigated waveguide. 2. Composite specimens
One of the best methods for wave propagation modelling excited by
a piezoelectric transducer seems to be the spectral element method Research presented in this paper was related to three specimens
(SEM) in the time domain [32]. This method utilizes high-order shape made of glass fibre reinforced polymer (GFRP) in the form of panels
functions, namely Lagrange polynomials, spanned over Gauss-Lobatto- with dimensions: 500 mm × 500 mm and thickness about 2 mm. Each
Legendre points. Nodes of spectral element coincide with quadrature investigated specimen consisted of twelve layers of VV192T/202 pre-
points used in numerical integration. The method is suitable for wave pregs with orientations: [0/90/0/90/0/90]S.
propagation modelling in composite materials and has the same flex- The first specimen contained four circular Teflon inserts with dia-
ibility as the conventional finite element method, but it is much more meters 20 mm located at different depths. The Teflon inserts are de-
computationally efficient. Efficiency comes mostly from the diagonality noted by T1, T2, T3 and T4 in Fig. 1. T1 depth is 1.66 mm (between
of mass matrix and application of explicit solver. This method can be 10th and 11th layer counting from the top surface – see layer nu-
realised based on parallel computation scheme on cards including meration in Fig. 1), T2 depth is 1.33 mm (between 8th and 9th layer),
graphics processing units (GPU) [33]. This method was originally in- T3 depth is 1 mm (between 6th and 7th layer) and T4 depth is 0.66 mm
troduced by Patera in 1984 for the solution of laminar flows [34]. Next, (between 4th and 5th layer). Teflon inserts are placed in the middle
this method was utilized for solving elastic wave propagation problems. between the piezoelectric transducer (PZT) and plate edge.
A layered 3D spectral element for improved laminate modelling accu- The second specimen contained Teflon inserts with different shapes:
racy was proposed [24]. A layerwise theory for thick laminated com- circular (T1), square (T2), rectangular (T3), and elliptic (T4). They are
posite strips with piezoelectric actuators and sensors and an associated located at the same depth, i.e. 1.66 mm (between 10th and 11th layer
explicit SEM with the aim to provide simulations of the transient cou- counting from the top surface) as it is shown in Fig. 2.
pled electromechanical wave response was developed recently [35]. It The third specimen contained impact damage introduced by a steel
should be emphasized here that despite the significant process of im- ball drop with energies 5 J, 10 J and 15 J (Fig. 3). The diameter of the
provement of SEM in time domain it is still challenging to model high- steel ball was 20 mm. During initiation of impact damage, the whole
frequency multimodal character of guided waves in thin composite surface of the sample was supported by the wooden block.
plates especially achieving high accuracy. It needs to be underlined that in the case of circular Teflon inserts
Lamb waves interacting with delamination are subjected to mode with diameter 20 mm its shapes and locations between layers of the
conversion phenomenon. Mode conversion phenomenon means that composite were checked using terahertz spectroscopy (THz). The same
excited wave mode is not only reflected or scattered from delamination technique was utilised in the GFRP sample with impact damage cases
[36] but also converts to another mode. For example, S0 mode converts with energies 10 J and 15 J. Results of THz spectroscopy for cases of the

318
T. Wandowski, et al. Composite Structures 215 (2019) 317–330

Fig. 2. GFRP specimen with Teflon inserts with different shapes.

utilized – measurements were taken along laser beam approximating

out-of-plane guided elastic wave velocities of composite material par-
ticles. Measurements were taken at the grid of size 333 × 333 points
Fig. 1. GFRP specimen with circular Teflon inserts. (1.5 mm grid spacing).

sample with Teflon inserts and sample with impact damage were 3. Numerical model
published in [43]. The THz spectroscopy technique is based on the
phenomenon of electromagnetic wave propagation. Electromagnetic Each case was modelled by using the time domain spectral element
waves in terahertz frequency band are utilised. This technique could be method (SEM) [21,32]. SEM algorithm was prepared for parallel cal-
utilised for dielectric materials like for example GFRP. culations on GPU [33]. Very dense meshes were applied – each ply of
The THz spectroscopy is based on the differences of the real part of the composite laminate was modelled by one layer of 3D spectral ele-
refractive indexes of the interface of material and defect. In the case of ments. However, only 6 layers were modelled by averaging properties
research presented here differences of refractive indexes between at angles 0° and 90° instead of 12 layers as it was in experimental
GFRP/Teflon (for the case of inserts) and between GFRP/air (in the case specimens. It was dictated by the amount of GPU memory available,
of impact damage) are utilised. At the interface of media with different namely 6 GB (the number of degrees of freedom was about 8 million).
refractive indexes, refraction and reflections of electromagnetic waves The total thickness of the plate was the same as in the experiment,
occur. This technique could be utilised for the detection and localisa- namely 2 mm. It should be noted that 6 layers of composite laminates
tion of delaminations in dielectric composite materials [43–45]. allow modelling of the same delamination depths as it was in the ex-
Each GFRP specimen was equipped with a piezoelectric transducer periment. Therefore, similar elastic wave propagation behaviour was
of diameter 10 mm and thickness 0.5 mm which was placed at the expected which includes phenomena such as mode conversion and
centre of the specimen. The piezoelectric was made of NCE51 piezo- wave entrapment at delamination. However, certain discrepancies in
electric material with wrap-around electrodes (manufacturer: CTS, amplitude and velocity were expected due to a lower number of layers.
former NOLIAC). It was transversely polarized. The piezoelectric Piezoelectric transducer of thickness 0.5 mm was modelled at the
transducer was bonded to the top surface of the plate by cyanoacrylate centre of the plate. Also, a bond layer made of epoxy of thickness
glue and it was utilised for elastic wave actuation. Excitation signal in 0.08 mm was modelled. Excitation signal was applied to the piezo-
the form of sinusoid modulated by Hanning window of carrier fre- electric transducer in the form of sine modulated by Hanning window.
quency 150 kHz was utilised. Full wavefield measurements of elastic The signal contained 5 cycles and carrier frequency 150 kHz exactly
waves were conducted in a non-contact way based on scanning laser like in experimental research.
Doppler vibrometer (SLDV). Measurements were taken at a surface Mesh consisted of 3D spectral elements with 108 nodes each
opposite to the surface where elastic waves were excited by a piezo- (6 × 6 × 3 nodes) and 3 degrees of freedom in each node: ux , uy and uz .
electric transducer. Only one scanning head of laser vibrometer was In order to be consistent with SLDV measurements, particle velocities

319
T. Wandowski, et al. Composite Structures 215 (2019) 317–330

Fig. 4. Cross-section through the modelled region around dent caused by an

impact. The assumed dent depth is denoted by d and it is proportionally de-
creasing with the ply number so that the bottom surface of the composite plate
remains flat. The dented area with reduced stiffness is grey whereas intact area
is white.

been used for many years as a foundation of damage identification al-

gorithms. These algorithms often rely on Lamb wave effects like wave
reflection, scattering, refraction or mode conversion. Extracted features
are a time of flight, amplitude change, velocity, etc. Thin plate-like
structures made out of GFRP have found many applications and are of
great interest from the point of view of NDT as well as SHM techniques.
Modelling of Lamb wave phenomenon in GFRP laminates of various
stacking sequences is a challenging task. It requires high computational
power. Hence, in modelling techniques balance between accuracy and
computation time has to be found. One of the main complexities in-
volved in the propagation of guided waves is the attenuation effect.
However, due to Lamb wave modelling complexities it is often ne-
glected. The most common viscous damping description is the Rayleigh
Fig. 3. GFRP specimen with impact damage (energies: 5 J, 10 J and 15 J).
damping model, where a linear combination of mass and stiffness ma-
trices are used. This Rayleigh damping is defined as:
ϑx , ϑy and ϑz were stored for each frame of propagating elastic waves.
C = αM + βK (1)
Delaminations which reflects Teflon inserts were modelled by node
separation between appropriate spectral elements. The delamination at where α and β are the mass M and stiffness K proportionality coeffi-
a cross-section along the thickness has parabolic shape and the maximal cients, respectively. The mass proportional damping simulates the ef-
opening gap is set up to 0.1 mm. Contact effect at delamination region fect of particles moving through a viscous media which are driven by
was not taken into account allowing penetration of nodes. the absolute velocities. The stiffness proportional damping is propor-
Delamination which reflects impact damage was modelled in a tional to the strain rate which can be associated with the material
different way. The shape of the dent on the top surface was modelled as damping.
the paraboloid of diameter 8 mm for 5 J impact, 10 mm for 10 J impact The model consists of an irregular mesh of 3D spectral elements
and 12 mm for 15 J impact which roughly resembles diameters of de- based on 3D elasticity theory of solids. Each spectral element has 36 in-
fects on experimental specimens estimated by visual inspection (the plane nodes and 3 nodes across the thickness. The nodes are located at
colour on damaged area of GFRP has changed – see Fig. 3). The depth of Gauss-Lobatto-Legendre points. The spectral solid element has 3 de-
paraboloid dent as it is denoted in Fig. 4 was assumed as 0.01 mm, grees of freedom: displacements in x, y and z direction, respectively.
0.03 mm and 0.05 mm for the case of 5 J, 10 J and 15 J impact, re- The equation of motion has the classic form:
spectively. Moreover, the Young modulus of epoxy was reduced in all
¨ + Cu̇ + Ku = F
Mu (2)
elements across the thickness of the plate in the dented area (see Fig. 4).
The reduction was 25%, 50% and 75% for the case of 5 J, 10 J and 15 J where M is the mass matrix, C is the damping matrix, K is the stiffness
impact, respectively. Young modulus of fibres was kept intact. The matrix, F is the force matrix, u is the displacement matrix and the dot
material properties used in simulations are summarized in Table 1. It is accent denotes differentiation in respect to time. The Eq. (1) is solved
assumed that the volume fraction of reinforcing fibres is 45%. Ply by the central difference method which leads to:
material properties are calculated by using the rule of mixtures as for
1 1 2 1
unidirectional fibres and averaged (see further sections for more de- ⎛ M+C ⎞ ut + Δt = Ft − Kut + Mut − 2 Mut − Δt
⎝ Δt 2 2Δt ⎠ Δt 2 Δt
tails). The mechanical properties of the lamina with reinforcing fibres at
1
0° for the intact plate as well as for considered reductions of Young’s + Cut − Δt
2Δt (3)
modulus of epoxy are given in Table 1.
where Δt denotes the time increment. Due to the orthogonality of shape
functions of spectral elements mass, the matrix is diagonal. It is as-
4. Model enhancement sumed that the damping matrix is proportional to the mass matrix,
C = α M, so that the damping matrix is also diagonal. Then the dis-
Guided waves propagating through thin plate-like structures are placement field at the next time step ut + Δt can be efficiently calculated.
known as Lamb waves. Anomalies of propagating Lamb waves have It should be noted that this is contrary to the approach presented by

320
T. Wandowski, et al. Composite Structures 215 (2019) 317–330

Table 1
Material properties of composite constituents and composite lamina calculated by the rule of mixtures.
Property Epoxy Glass fibres

Young’s modulus [GPa] 3.49 85

Poisson ratio 0.35 0.23
Density [kg/m3] 1250 2250

Lamina for intact plate

E11 E22 E33 υ12 υ13 υ23 G12 G13 G23 ρ

40.1 GPa 8.3 GPa 8.3 GPa 0.296 0.296 0.391 3.1 GPa 3.1 GPa 2.98 GPa 1700 kg/m3

Lamina for 25% Young’s modulus reduction of the epoxy

E11 E22 E33 υ12 υ13 υ23 G12 G13 G23 ρ

39.7 GPa 6.3 GPa 6.3 GPa 0.296 0.296 0.393 2.37 GPa 2.37 GPa 2.28 GPa 1700 kg/m3

Lamina for 50% Young’s modulus reduction of the epoxy

E11 E22 E33 υ12 υ13 υ23 G12 G13 G23 ρ

39.2 GPa 4.3 GPa 4.3 GPa 0.296 0.296 0.395 1.6 GPa 1.6 GPa 1.55 GPa 1700 kg/m3

Lamina for 75% Young’s modulus reduction of the epoxy

E11 E22 E33 υ12 υ13 υ23 G12 G13 G23 ρ

38.7 GPa 2.2 GPa 2.2 GPa 0.296 0.296 0.397 0.8 GPa 0.8 GPa 0.8 GPa 1700 kg/m3

Gresil et al. [22] in which comparison between the experimental and Let’s consider a simple average of elastic constants:
the finite element method received signals were generated by stiffness − −
proportional damping (α = 0, C = β K ). It turns out that attenuation Dij = [(Qij )0 + (Qij )90]/2 (6)
can be modelled quite accurately both ways but only for a single mode
However, it was observed that for such an approach the group ve-
of propagation. Gresil et al. [22] shown that only for certain frequency,
locity of S0 mode is slightly overestimated in the numerical model in
coefficients α and β can be selected so that both A0 and S0 modes have
comparison to experimental results. Due to the fact that the S0 mode
appropriate attenuation characteristics. Otherwise, only mono-mode
can be relatively well approximated by in-plane displacement compo-
Lamb waves can be simulated properly. The simplest extension of the
nents, the group velocity can be reduced by the appropriate decrease of
proposed approach is the replacement of scalar proportionality coeffi-
in-plane elastic constants:
cient, α , by a row vector of the form [46]:
− −
1
α = [α xy 2
, α xy N −2
, αz3, ⋯, α xy N −1
, α xy , αzN ] Qij = ϰQij i, j = 1, 2 (7)
(4)
where ϰ ≤ 1 is the correction parameter.
where α xy is in-plane damping coefficient and α z is out-of-plane damping
coefficient and N is the number of degrees of freedom. Such an ap-
proach can be justified by the fact that A0 mode behaviour can be 6. Numerical results
approximated quite well by out-of-plane displacements whereas S0
mode can be approximated by in-plane displacements. In this paper for A large number of simulations has been performed but only the
1
simplicity, it was assumed that α xy = α xy 2
= α xy N −2
= ⋯=α xy N −1
= α xy and most interesting results are selected and characterized especially con-
sidering mode conversion effect. In order to be consistent with ex-
α z = αz3 = ⋯=αzN .
periment, full wavefield data is interpolated on a grid of uniform
333 × 333 points in space covering the whole plate. Signals with 512
5. Homogenization
time instances are stored and analysed. Numerical and experimental
signals are compared at 3 points of coordinates in mm: A(0,50), B
It was assumed that the composite laminate is orthotropic. It allows
(101,0), C(101,−101) in respect to the origin at plate centre. If it is not
simple approach for calculation of elastic constants of unidirectional
stated differently, presented signals and full wavefields have a form of
fibre reinforced composite by using the rule of mixture. The crucial
transverse particle velocities ϑz .
parameter is the volume fraction of reinforcing fibres (glass) in relation
to the matrix (epoxy resin). Then the matrix of elastic constants is
6.1. Influence of damping parameters
transformed depending on the angle of reinforcing fibres in each ply
separately. In the case of homogenization the material properties are
First of all, the numerical model was calibrated in order to best
averaged over the thickness of the composite laminate according to the
match experimental data. Damping parameters α xy and α z influence on
following formula:
signals of propagating elastic waves were investigated. A grid search
N − approach was utilized for finding α xy and α z parameters so that the
Dij = ∑ (Qij )k (hk /h) difference between numerical and experimental signals is minimal.
k=1 (5)
−
However, only a few cases were analysed due to long computation time
where (Qij )k are elastic constants in layer k , hk is the thickness of layer (the number of degrees of freedom for each simulation was about 8
k , h is the total thickness of composite laminate and N is the number of millions) so that the final values are not optimal. Selected results are
layers. It is intuitive that plain weave composite can be modelled by a presented in Fig. 5.
combination of procedures for unidirectional ply at 0° and 90° angles. Nevertheless, it can be noted that damping plays a very important

321
T. Wandowski, et al. Composite Structures 215 (2019) 317–330

the experimental case. In order to better match these parameters,

complex optimization of elastic constants should be conducted which is
out of the scope of this paper.
In the frame taken for time instant 0.05 ms in numerical results
interaction of S0 mode with Teflon inserts is visible. This interaction
results in the conversion of symmetric mode S0 to antisymmetric mode
A0′. This could be also clearly seen in the frame for the experimental
result taken at time instant 0.078 ms.
Close-ups of wavefield near delaminations at various depths at time
instant t = 0.078 [ms] for numerical and experimental results were
presented in Fig. 7. As results of S0 mode conversion, new mode A0′
with a different wavelength is generated at the locations of delamina-
tion/Teflon insert. In the frames in Fig. 6 for the numerical and ex-
perimental result for time instant 0.1 ms interaction of A0 mode with
delamination and reflections of S0 mode from specimen edges are
visible. It can be observed that the interaction of S0 mode with dela-
minations is more visible in the case of numerical results than in the
case of experimental results. The effect of S0/A0′ mode conversions for
both cases is similar. Good agreement of numerical and experimental
results can be noticed. Based on mode conversion effect and interaction
of A0 mode with delamination the shape of the delamination can be
identified (see also Fig. 12).

6.3. Interaction of Lamb waves with delamination of various shapes

In this case comparison of full wavefield results obtained numeri-

cally and experimentally for the case of the specimen with delamina-
tions of various shapes (Fig. 2) is presented in Fig. 8. The frequency of
excitation in the investigated case was equal to 150 kHz. Here also the
propagation of fundamental modes S0 and A0 is visible. Moreover, S0/
A0′ mode conversion on delaminations with different shapes can be
seen. It can be observed that the interaction of S0 mode with delami-
nations is more visible in the case of numerical results than in the case
of experimental results.
The effect of S0/A0′ mode conversions for both cases is similar.
Fig. 5. Selected results of damping parameters influence on signals of propa-
Based on mode conversion effect the shape of the delamination can be
gating elastic waves acquired at three points A, B and C, respectively.
identified. This is especially visible in Fig. 9 where close-ups of wave-
field near delaminations are presented (see also Fig. 13). Moreover, in
role in modelling aspects and has a huge influence on the amplitude and the frame from experimental results for time instant 0.1 ms presented in
resulting shapes of propagating wave packets (see Fig. 5). In this figure Fig. 8, S0/A0′ mode conversion is also visible at the edges. There was
numerical and experimental results were presented for the specimen no such effect in numerical results. This was caused by the fact of
depicted in Fig. 1 for delaminations at various depths. Excitation fre- perfect symmetry of geometry of the numerical model and dis-
quency was equal to 150 kHz. It can be seen that the parameter α z continuities near the edges of real panel (non-uniform thickness near
strongly affects the amplitude of the second wave packet which re- the edges).
sembles A0 mode. The green signal represents the overdamped case
whereas the red signal slightly better fit experimental signal (black). 6.4. Interaction of Lamb waves with impact damage
It also has been found that the damping parameters can be tweak
only for one selected mode. In the analysed case the amplitude of the In the third case comparison of full wavefield results for frequency
first wave packet (S0 mode) was fitted very well but the amplitude of 150 kHz which were obtained numerically and experimentally for the
the A0 mode was fitted poorly. More information about this approach case of the specimen with impact damage (Fig. 3) is presented in
could be found in [46]. In this research the final parameters which were Fig. 10. It can be observed that the interaction of Lamb waves with
used for further simulations were: α xy = 4e2 , α z = 1e 4 . impact damage is much less pronounced than interaction with Teflon
inserts.
6.2. Interaction of Lamb waves with delamination of various depths The effect of S0/A0′ mode conversion is clearly visible only in ex-
perimental wavefield for impacts with energies 10 J and 15 J. It sug-
Comparison of full wavefield results obtained numerically and ex- gests that the assumed reduction of Young’s modulus of epoxy was too
perimentally for the case of the specimen with delaminations at various low. Both in numerical and experimental data interaction of A0 mode
depths (Fig. 1) is presented in Fig. 6. The frequency of excitation was with the strongest impact can be distinguished (Fig. 10, t = 0.109
equal to 150 kHz. [ms]). Interactions of Lamb waves with impact damage are weak for
In the case of numerical and experimental results propagation of impacts of lower energies. This could be also noticed in Fig. 11 where
two fundamental Lamb wave modes, A0 and S0 are seen. Both modes close-ups of wavefield near impact damage for numerical and experi-
can be distinguished based on different wavelengths and velocities of mental results were presented.
propagation. S0 mode is faster and its wavelength is larger than for A0 Close-ups of delamination regions showing S0/A0′ mode conversion
mode. It should be noted that certain discrepancies between numerical effect in numerical results are shown in Fig. 12 for circular delamina-
and experimental signals occur in terms of velocities of A0 and S0 tions at various depths and in Fig. 13 for the case of delaminations with
mode. Also, both fundamental modes appear to be more dispersive in various shapes. These plots were created for time instant t = 0.078 ms.

322
T. Wandowski, et al. Composite Structures 215 (2019) 317–330

Numerical Experimental

t=0.05 [ms]

t=0.078 [ms]

t=0.1 [ms]

Fig. 6. Full wavefield of transverse (out-of-plane) particle velocities at selected time instances: numerical and experimental results for delaminations at various
depths; frequency 150 kHz.

Fig. 7. Close-ups of wavefield near delaminations at various depths at time instant t = 0.078 [ms]: a) numerical results, b) experimental results; frequency 150 kHz.

All plots for four cases on top and bottom surfaces are normalized in propagation of A0′ mode outside the delamination. Part of the wave is
respect to the largest amplitude. These plots were created using the transmitted through the edges of delamination and propagates in the
different colour map in order to better illustrate the effects of S0/A0′ material of specimen outside delamination. It is evident that S0/A0′
mode conversion. Two effects of S0/A0′ mode conversion can be dis- mode conversion can be successfully used for precise estimation of the
tinguished. The first effect is related to entrapment of A0′ mode in the shape of delamination. Moreover, the S0/A0 mode conversion effect is
delaminated/Teflon insert region. Mode A0′ is reflecting internally on more visible if the number of composite layers n between acquired
the edges of delamination. In the case of experimental results, this is wavefield surface and Teflon insert is lower.
caused by differences in acoustic impedances of composite plate and
delamination (Teflon layer and air). The second effect is related to the

323
T. Wandowski, et al. Composite Structures 215 (2019) 317–330

Numerical Experimental

t=0.05 [ms]

t=0.078 [ms]

t=0.1 [ms]
Fig. 8. Full wavefield of transverse (out-of-plane) particle velocities at selected time instances: numerical and experimental results for the various delamination
shapes; frequency 150 kHz.

6.5. Signal analysis GFRP laminate was used. Nevertheless, some manufacturing im-
perfections and measurement inconsistencies always appear. As it was
In this subsection comparison of signals at points A, B and C ob- mentioned in the real GFRP specimens some non-uniform thickness
tained by the time domain SEM to signals measured by SLDV are pre- near the edges was noticed what was the source of additional S0/A0′
sented in Figs. 14–16. Points A and B lay at angle 90° and 0° in respect mode conversions on the edges of the specimen. All signals are nor-
to wave source, respectively. Point C lay at angle −45° in respect to malized in respect to an excitation signal at the edge of PZT.
wave source. It should be noted that for each damage scenario the same The relatively good agreement can be observed between numerical

Fig. 9. Close-ups of wavefield near delaminations with various shapes at time instant t = 0.078 [ms]: a) numerical results, b) experimental results; frequency
150 kHz.

324
T. Wandowski, et al. Composite Structures 215 (2019) 317–330

Numerical Experimental

t=0.05 [ms]

t=0.078 [ms]

t=0.109 [ms]

Fig. 10. Full wavefield of transverse (out-of-plane) particle velocities at selected time instances: numerical and experimental results for the impact damage; fre-
quency 150 kHz.

and experimental signals. 6.6. Cross-section analysis

It also can be noted that the amplitude of first incoming wave packet
matches quite well the amplitude of the experimental signal at each In this subsection numerical results of detailed analysis of wavefield
analysed point resembling attenuation of S0 mode with propagating of the particle velocities in cross-section of specimens are presented.
distance. However, other wave packets in the signal corresponding to Wavefield of the particle velocities component ϑx at selected time in-
propagating A0 mode and reflections have larger amplitudes in the stances for the cross-section of the GFRP specimen no. 1 (Fig. 1) at a
numerical model than it is in case of experimental signals. It seems that cross-section through delaminations T2 and T4 is presented in Fig. 17.
the damping parameters can be further tuned but it requires some There is the propagation of S0 mode only visible in the time frame at
iterative procedure and it is not feasible due to long computation time. t = 0.0250 ms. In the time frame at t = 0.05 propagation of S0 and A0
Moreover, it is very difficult to model accurately weave fabric re- mode is visible. S0 mode is faster and its wavelength is longer than for
inforced composites such as GFRP laminates investigated here. A0 mode. Moreover, mode S0 start to interact with delaminations T2
and T4. In the time frame at time instant t = 0.078 ms. S0 mode
reached the edge of the specimen. There is also S0/A0′ mode

Fig. 11. Close-ups of wavefield near impact damage at time instant t = 0.078 [ms]: a) numerical results, b) experimental results; frequency 150 kHz.

325
T. Wandowski, et al. Composite Structures 215 (2019) 317–330

Top view

T1 (circle), n=10 T2 (circle), n=8 T3 (circle), n=6 T4 (circle), n=4

Bottom view

T1 (circle), n=2 T2 (circle), n=4 T3 (circle), n=6 T4 (circle), n=8

Fig. 12. Close-ups of full wavefield on top and the bottom surface of the plate around delaminated regions for the case of delaminations at various depths; n is the
number of composite layers between acquired wavefield surface and Teflon insert; frequency 150 kHz.
Top view

T1 (circle), n=10 T2 (square), n=10 T3 (rectangle), T4 (ellipse), n=10

n=10
Bottom view

T1 (circle), n=2 T2 (square), n=2 T3 (rectangle), n=2 T4 (ellipse), n=2

Fig. 13. Close-ups of full wavefield on top and the bottom surface of the plate around delaminated regions for the case of delaminations of various shapes; n is the
number of composite layers between acquired wavefield surface and Teflon insert; frequency 150 kHz.

conversion at regions with Teflon inserts. New mode A0′ can be noticed magnified time frame at t = 0.078 ms with marked modes was pre-
in the region of the specimen above and below delaminations. More- sented in Fig. 18. Propagation of the S0 mode, effect of mode conver-
over, propagation of A0 mode that has not reached the region of de- sion S0/A0′ at delamination (new A0′ mode is generated in the region
laminations is visible. The time frame for t = 0.1 ms shows the inter- of delamination) and propagation of A0 mode (which have not reached
action of A0 mode with delamination. In the next time frame for the delamination) can be distinguished. In the region of delamination,
t = 0.114 ms propagation of S0 mode reflected from the edge of the the wave is splitting into to waves propagating above and below de-
specimen is visible. In the time frame at t = 0.2 ms A0 mode is inter- lamination. This could be noticed in a region marked as S0/A0′ in
acting with the edge of the specimen. In the time frame for t = 0.3 ms Fig. 18. The same effect was observed for the A0 mode in Fig. 17 (for
propagation of S0 edge reflected mode is visible. Moreover, the effects t = 0.1 ms) whereas interaction of propagating S0 mode with delami-
of S0/A0′ mode conversion at the regions of delaminations are still nation does not produce any visible effects at the cross-section.
visible. Delamination could be a source of layers separation and due to this
For the better illustration of mode propagation and conversion, wave splits into wave propagating in two separated parts of the plate.

326
T. Wandowski, et al. Composite Structures 215 (2019) 317–330

Fig. 16. Comparison of numerical and experimental signals at points A, B and C

Fig. 14. Comparison of numerical and experimental signals at points A, B and C for the case of impact damage and excitation frequency 150 kHz.
for the case of delaminations at various depths and excitation frequency
150 kHz. The thickness of two parts differs from the thickness of the specimen
without delamination. It is well known that for Lamb waves velocity of
propagation depends on the product of excitation frequency and plate
thickness. Change of the thickness is directly related to change of ve-
locity assuming unchanged excitation frequency. On the other hand
change of velocity is related to change of wavelengths based on the
general formula:

cp
λ=
f (8)

where: λ – the wavelength of considered wave mode, c p – the phase

velocity of considered mode, f – the frequency. Therefore wavelength of
new mode A0′ resulting from S0 mode conversion and propagating in
the delamination region could differ from wavelength to A0 mode
propagating in the region of the plate without delamination.
In Fig. 19 wavefield in the form of the particle velocity component
ϑz (transverse, out-of-plane velocity) for specimen no. 1 (Fig. 1) at a
cross-section through delaminations T2 and T4 is presented. The same
conclusions can be drawn for analysis of this wavefield like for wave-
field for velocities component ϑx (Fig. 17). It can be noticed that the S0
mode is converting to A0′ mode at delamination for frames at time
instants t = 0.05 ms and t = 0.078 ms. It could be noticed that wave is
splitting into to waves propagating above and below delamination.
Moreover, waves above and below delamination have different wave-
lengths what could be noticed in Fig. 20.
It is interesting to note the distribution of the particle velocity
component ϑx (Fig. 21) and ϑz (Fig. 22) across the thickness of the plate
for the location of delaminations T2 and T4 for a time frame in which
S0/A0′ mode conversion is observed. Depending on the case, it is
symmetric or antisymmetric at the undamaged cross-section in respect
Fig. 15. Comparison of numerical and experimental signals at points A, B and C to the neutral plane. Though, for the cross-section through delamina-
for the case of delaminations of various shapes and excitation frequency tion, the distribution of the particle velocity has a characteristic zig-zag
150 kHz.
character which can be useful for estimation of the depth of delami-
nation.

327
T. Wandowski, et al. Composite Structures 215 (2019) 317–330

Fig. 17. Wavefield of the particle velocity compo-

nent ϑx , at selected time instances for the cross-sec-
tion of the sample no. 1 (cross-section through de-
laminations T2 and T4); frequency 150 kHz.

Fig. 18. Wavefield of the particle velocity component ϑx , at time instant t = 0.078 [ms] for the cross-section of the sample no. 1 with marked S0, A0 modes and S0/
A0′ mode conversion; frequency 150 kHz.

Fig. 19. Wavefield of the particle velocity compo-

t=0.0250 [ms] nent ϑz at selected time instances for the cross-sec-
tion of the sample no. 1 (cross-section through de-
laminations T2 and T4); frequency 150 kHz.
t=0.05 [ms]

t=0.078 [ms]

t=0.1 [ms]

t=0.114 [ms]

t=0.2 [ms]

t=0.3 [ms]

Fig. 20. Wavefield of the particle velocity component ϑz , at time instant t = 0.078 [ms] for the cross-section of the sample no. 1 with marked S0, A0 modes and S0/
A0′ mode conversion; frequency 150 kHz.

7. Conclusions It is easier to detect delamination which is located closer to the surface

of the plate. A good agreement between numerical and experimental
Based on presented results it could be concluded that full wavefield results was achieved but still, some small differences can be noticed.
displacements or velocities acquired at the surface of composite lami- The source of this is non-uniform geometry (thickness near the edges)
nate can be effectively utilized for delamination detection, localization of specimens. Moreover, it is very difficult to model accurately weave
and estimation of the size and its shape. Moreover, acquisition of full fabric reinforced composites such as GFRP laminates investigated here.
wavefield particle velocities at both surface of the plate enables to es- It is also difficult to calibrate damping parameters to achieve a good
timate if the delamination is located above or below the neutral plane. match between numerical and experimental signals. In particular,

328
T. Wandowski, et al. Composite Structures 215 (2019) 317–330

Fig. 21. Distribution of the particle velocity component ϑx across the thickness of the plate for the cross-section of the plate; propagation of S0, A0 modes and S0/A0′
mode conversion visible.

Fig. 22. Distribution of the particle velocity component ϑz across the thickness of the plate for the cross-section of the plate; propagation of S0, A0 modes and S0/A0′
mode conversion visible.

damping can be effectively modelled only for one selected propagating [3] Naidjate M, Helifa B, Feliachi M, Lefkaier IK, Heuer H, Schulze M. A smart Eddy
mode. However, presented approach is very useful for study the effects current sensor dedicated to the nondestructive evaluation of carbon fibers re-
inforced polymers. Sensors (Basel) 2017;17(9).
of S0/A0′ mode conversion in fibre reinforced composite plates with [4] Xu X, Ji H, Qiu J, Takagi T. Detection of delamination in laminated CFRP compo-
different types of discontinuities. sites using eddy current testing: Simulation and experimental study. Int J Appl
Electromagnet Mech 2018;57(2):1–16.
[5] Garcea SC, Wang Y, Withers PJ. X-ray computed tomography of polymer compo-
Acknowledgements sites. Compos Sci Technol 2018;156:305–19.
[6] Khan MH, Elamin M, Li B, Tan KT. X-ray micro-computed tomography analysis of
Authors would like to gratefully acknowledge financial support impact damage morphology in composite sandwich structures due to cold tem-
perature arctic condition. J Compos Mater 2018;52(25).
given by Polish National Science Centre under grant agreement no. [7] Chimenti DE. Review of air-coupled ultrasonic materials characterization.
UMO-2014/13/D/ST8/03167 in the frame of SONATA project entitled: Ultrasonics 2014;54(7):1804–16.
“Investigation of elastic wave mode conversion phenomenon in thin- [8] Neuenschwander J, Furrer R, Roemmeler A. Application of air-coupled ultrasonics
for the characterization of polymer and polymer-matrix composite samples. Polym
walled structures with discontinuities”.
Test 2016;56:379–86.
[9] Qiu L, Yuan S, Zhang X, Wang Y. A time reversal focusing based impact imaging
Appendix A. Supplementary data method and its evaluation on complex composite structures. Smart Mater Struct
2011;20(10):105014.
[10] Qiu L, Liu B, Yuan S, Su Z. Impact imaging of aircraft composite structure based on
Supplementary data to this article can be found online at https:// a model-independent spatial-wavenumber filter. Ultrasonics 2016;64:10–24.
doi.org/10.1016/j.compstruct.2019.02.076. [11] Michaels TE, Michaels JE, Ruzzene M. Frequency-wavenumber domain analysis of
guided wavefields. Ultrasonics 2011;51:452–66.
[12] Zhang K, Zhou Z. Quantitative characterization of disbonds in multilayered bonded
References composites using laser ultrasonic guided waves. NDT and E Int 2018;97:42–50.
[13] Pieczonka L, Ambrozinski L, Staszewski WJ. Damage detection in composite panels
[1] Klepka A, Pieczonka L, Staszewski WJ, Aymerich F. Impact damage detection in based on mode-converted lamb waves sensed using 3D laser scanning vibrometer.
laminated composites by non-linear vibro-acoustic wave modulations. Compos B Opt Lasers Eng 2017;99:80–7.
Eng 2014;65:99–108. [14] Leckey CAC, Wheeler KR, Hafiychuk VN, Hafiychuk H, Timuçin DA. Simulation of
[2] Rahammer M, Kreutzbruck M. Fourier-transform vibrothermography with fre- guided-wave ultrasound propagation in composite laminates: benchmark compar-
quency sweep excitation utilizing local defect resonances. NDT and E Int isons of numerical codes and experiment. Ultrasonics 2018;84:187–200.
2017;86:83–8. [15] Kudela P, Radzienski M, Ostachowicz W. Impact induced damage assessment by
means of Lamb wave image processing. Mech Syst Sig Process 2018;102:23–36.

329
T. Wandowski, et al. Composite Structures 215 (2019) 317–330

[16] Rahammer M, Solodov I, Bisle W, Scherling D, Kreutzbruck M. Thermosonic testing waveguides of arbitrary cross-section. J Sound Vib 2006;295(3–5):685–707.
with phase matched guided wave excitation. J Nondestr Eval 2016:35–47. [32] Ostachowicz W, Kudela P, Krawczuk M, Zak A. Guided waves in structures for SHM
[17] Di Ianni T, De Marchi L, Perelli A, Marzani A. Compressive sensing of full wave field – the time–domain spectral element method. John Wiley & Sons; 2012.
data for structural health monitoring applications. IEEE Trans Ultrason Ferroelectr [33] Kudela P. Parallel implementation of spectral element method for Lamb wave
Freq Control 2015;62(7):1373–83. propagation modeling. Int J Numer Meth Eng 2015;106(6):413–29.
[18] Sharif-Khodaei Z, Ghajari M, Aliabadi MH. Determination of impact location on [34] Patera AT. A spectral element method for fluid dynamics: laminar flow in a channel
composite stiffened panel. Smart Mater Struct 2012;21:105026. expansion. J Comput Phys 1984;54:468–88.
[19] Ciampa F, Meo M. Impact detection in anisotropic materials using a time reversal [35] Rekatsinas CS, Saravanos DA. A time domain spectral layerwise finite element for
approach. Struct Health Monitor 2012;11:43–9. wave structural health monitoring in composite strips with physically modeled
[20] Markus S, Tornow C, Dieckhoff S, Bousite M, Ecault R, Berthe L, Bockenheimer C. active piezoelectric actuators and sensors. J Intell Mater Syst Struct
Extended non-destructive testing of composite bonds. SAE Technical Papers; 2011. 2017;28(4):488–506.
DOI: 10.4271/2011-01-2514. [36] Murat BIS, Khalili P, Fromme P. Scattering of guided waves at delaminations in
[21] Kudela P, Ostachowicz W. 3D time-domain spectral elements for stress waves composite plates. J Acoust Soc Am 2016;139(6):3044–52.
modelling. J Phys Conf Ser 2009;181:89–104. [37] Wandowski T, Kudela P, Malinowski P, Ostachowicz W. Analysis of S0/A0 guided
[22] Howard R, Cegla F. Detectability of corrosion damage with circumferential guided wave mode conversion phenomenon. Proceedings of SPIE, Health Monitoring of
waves in reflection and transmission. NDT and E Int 2017;91:108–19. Structural and Biological Systems XII, 10600; 2018, Colorado, USA.
[23] Chakraborty A, Gopalakrishnan S. A spectrally formulated finite element for wave [38] Ghadami A, Behzad M, Mirdamadi HR. A mode conversion-based algorithm for
propagation analysis in layered composite media. Int J Solids Struct detecting rectangular notch parameters in plates using Lamb waves. Arch Appl
2004;41:5155–83. Mech 2015;85(6):793–804.
[24] Lonkar K, Chang F-K. Modeling of piezo-induced ultrasonic wave propagation in [39] Schaal C, Samajder H, Baid H, Mal A. Rayleigh to Lamb wave conversion at a de-
composite structures using layered solid spectral element. Struct Health Monitor lamination-like crack. J Sound Vib 2015;353:150–63.
2013;13(1):50–67. [40] Nakamura NN, Ogi H, Hirao M, Nakahata K. Mode conversion behavior of SH
[25] Shen Y, Cesnik CES. Hybrid local FEM/global LISA modeling of damped guided guided wave in a tapered plate. NDT and E Int 2012;45(1):156–61.
wave propagation in complex composite structures. Smart Mater Struct 2016;25(9). [41] Nakamura NN, Ogi H, Hirao M. EMAT pipe inspection technique using higher mode
095021 (20pp). torsional guided wave T(0,2). NDT and E Int 2017;87:78–84.
[26] Gresil M, Giurgiutiu V. Prediction of attenuated guided waves propagation in [42] Gan CS, Chia CC, Tan LY, Mazlan N, Harley JB. Statistical evaluation of damage size
carbon fiber composites using Rayleigh damping model. J Intell Mater Syst Struct based on amplitude mapping of damage-induced ultrasonic wavefield. IOP Conf
2015;26(16):2151–69. Series: Mater Sci Eng 2018;405:012006.
[27] Kudela P, Radzienski M, Ostachowicz W. Wave propagation modeling in composites [43] Wandowski T, Malinowski P, Ostachowicz W. Full wavefield analysis for damage
reinforced by randomly oriented fibers. J Sound Vib 2018;14:110–25. assessment in composite materials. Trans Nanjing Univ Aeronaut Astronomy
[28] Hennings B, Lammering R. Material modeling for the simulation of quasi-con- 2018;35(2):264–74.
tinuous mode conversion during Lamb wave propagation in CFRP-layers. Compos [44] Dong J, Kim B, Locquet A, et al. Nondestructive evaluation of forced delamination
Struct 2016;151:142–8. in glass fiber-reinforced composites by terahertz and ultrasonic waves. Compos B
[29] Shoja S, Berbyuk V, Boström A. Delamination detection in composite laminates 2015;79:667–75.
using low frequency guided waves: numerical simulations. Compos Struct [45] Dong J, Locquet A, Citrin DS. Enhanced terahertz imaging of small forced dela-
2018;203:826–34. mination in woven glass fibre-reinforced composites with wavelet de-noising. J
[30] Cho Y. Estimation of ultrasonic guided wave mode conversion in a plate with Infrared, Millimeter Terahertz Waves 2016;37:289–301.
thickness variation. IEEE Trans Ultrasonics, Ferroelectrics, Frequency [46] Wandowski T, Kudela P, Malinowski P, Ostachowicz W. Guided wave attenuation in
2000;47(3):591–603. composite materials. Proceedings of SPIE, SPIE Smart Structures NDE Conference,
[31] Bartoli I, Marzani A, di Scalea FL, et al. Modeling wave propagation in damped Vol. 10170, March 25–29, 2017, Portland, USA.

330