A Numerical-Experimental Method For Dama

This article appeared in a journal published by Elsevier.
The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elseviers archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright
Author's personal copy
Computers and Structures 89 (2011) 17541770
Contents lists available at ScienceDirect
Computers and Structures

journal homepage: www.elsevier.com/locate/compstruc
A numerical-experimental method for damage location based on rotation elds

spatial differentiation
H.M. Reis Lopes a, J.V. Arajo dos Santos b,, C.M. Mota Soares b, R.J. Miranda Guedes c, M.A. Pires Vaz c
a
ESTIG, Instituto Politcnico de Bragana, Campus de Sta. Apolnia, Apartado 134, 5301-857 Bragana, Portugal
IDMEC/IST, Instituto Superior Tcnico, Av. Rovisco Pais, 1049-001 Lisboa, Portugal
c
DEMEGI, Faculdade de Engenharia do Porto, Rua Dr. Roberto Frias, 4200-465 Porto, Portugal
b
a r t i c l e
i n f o
Article history:
Received 27 November 2008
Accepted 5 September 2010
Available online 16 April 2011
Keywords:
Multiple damage location
Laminated plate
Curvature eld
Shear interferometry
Gaussian function derivative
Filtering techniques
a b s t r a c t
This paper presents a structural damage location method that decreases the number of spatial differentiations needed to compute modal curvature elds. The method is numerically and experimentally
applied to isotropic and laminated rectangular plates, respectively. A speckle shear interferometer is used
to measure the rotation elds of the laminated plate, while the isotropic plate is analysed by nite elements. It was found that the Gaussian differentiation is the most suited technique to compute the curvature elds. It is also demonstrated the superior performance of the curvature method using measured
rotation elds instead of measured displacement elds.
2010 Civil-Comp Ltd and Elsevier Ltd. All rights reserved.
1. Introduction
The interest in predictive structural monitoring, aiming at damage early detection, has been motivated by, amongst other factors,
the risk of human lives loses, due to unpredictable structural failures, such as airplane crashes and bridges or buildings collapses.
Damage detection and localisation aims at preventing the structures failure by repairing or replacing the damaged component.
The need of global methods for damage detection has been essentially motivated by aeronautical and aerospace applications. In
these applications, the less critical components have been progressively replaced by components manufactured with composite
materials. Although their high specic stiffness and strength, compared to those of metals, composite materials are more sensitive to
certain type of damages and present different kinds of defects or
damage mechanisms. The main damage types in laminated composites are interlaminar debonding, micro-cracks and micro-buckling, besides inclusions. These internal damages usually result from
manufacturing processes or in service loadings. Delamination or
interlaminar debonding is undetected by visual means and, therefore, one of the most critical type of damage, being also the most
common in aeronautical components.
There are several types of methods which use dynamic characteristics, namely those based on modal frequencies, displacements,
Corresponding author. Tel.: +44 958 8846.
E-mail address: viriato@ist.utl.pt (J.V. Arajo dos Santos).
rotations and curvatures [114]. Among these, the global methods

based on curvature changes can be given some prominence. Pandey et al. [3] introduced the mode shapes curvature method. In order to localise damage in beams, the proposed method uses the
difference in the curvature elds associated with mode shapes in
the undamaged and damaged states. The curvatures are extracted
by applying second order central nite differences to the mode
shapes. Contrary to the mode shapes, one can prove that the higher
changes in mode shapes curvatures are coincident with the damage region and their magnitudes are proportional to its severity.
An extension of the curvature differences method was proposed
by Sampaio et al. [7], who also used second order central nite differences to compute the frequency response functions curvatures.
Recently, Guan and Karbhari [12] stated that the computation of
the curvatures by numerical differentiation is the main cause for
the poor performance of the curvature method, in particular under
sparse and noisy measurements. In order to improve the numerical
differentiation results they proposed the use of a fourth order central difference method, instead of applying the second order central difference. The damage index method formulation presented
by Stubbs et al. [4] is also based on mode shapes curvatures information. This damage index relates the undamaged and damaged
curvature elds in each segment of the structure. Its deviation from
the indices normal distribution average is used as an indicator for
the identication of the probable damaged region. Also in the context of curvature methods, Ratcliffe [5] developed a procedure that
does not require undamaged structure information. The curvature
0045-7949/$ - see front matter 2010 Civil-Comp Ltd and Elsevier Ltd. All rights reserved.
doi:10.1016/j.compstruc.2011.03.014
H.M. Reis Lopes et al. / Computers and Structures 89 (2011) 17541770
1755
Fig. 1. Computation of curvature eld from measured: (a) displacement eld and (b) rotation eld.
Fig. 2. Rectangular plate nite element mesh and simulated damaged areas.
computation is carried out by applying a Laplacian operator to the

mode shapes in order to locate damage experimentally. The damage position is identied from discontinuities in the curvature
eld. Data extracted from nite elements allows the damage identication of local thickness reduction higher than 10%. Subsequently and in order to show graphical discontinuities of damage
lower than 10%, a modication on the Laplacian was presented.
This method was named gapped smoothing damage detection
method. A smoothed curvature prole, representative of the
undamaged structure, is subtracted to the curvature prole. The
performance of this method was experimentally tested in a steel
beam with localised damage, characterised by a half thickness
transversal cut. The damage was successfully located using the rst
two mode shapes. However, the results low accuracy lead these
researchers to suggest the use of strain gauges as an alternative
technique to measure the curvatures. The gapped smoothing damage detection method was applied to a composite beam in order to
locate a delamination [6]. The extension of this method to two
dimensions and the subsequent application to plates was performed to localise delaminations [10]. A nite model update algorithm based on experimental modal data was proposed for damage
detection [13]. A quantitative assessment of damage detection
using different modal selection criteria is presented. The nearreal-time composite structures health monitoring based on wavelet signal processing of lamb wave propagation was presented by
Sohn et al. [14]. A set of piezoelectric patches works as an active
actuators-sensors pair to generate a specic wavelet and to measure the propagation of lamb wave signals. In this study, the proposed damage detection strategy was proved to correctly detect
the simulated delamination. A more complete description of these
and other damage identication methods can be found in a recently published survey [15].
The presence of material properties discontinuities results in
discontinuities or perturbations in the modal curvature eld. This
Fig. 3. Damaged plate fourth mode: (a) x and (b) y rotation elds obtained using
ANSYS.
1756
0.8
0.6
y [m]
0.05
0.4
0.2
0.1
0
-0.2
0.15
-0.4
0.198
0.069
0.138
x [m]
0.207
0.2765
-0.6
(a)
0
0.8
0.6
0.4
0.05
y [m]
0.2
0
0.1
-0.2
-0.4
-0.6
0.15
-0.8
-1
0.198
0.069
0.138
x [m]
0.207
0.2765
(b)
Fig. 4. Differences between the rst mode: (a) x and (b) y rotation elds magnitude.
fact was investigated by Arajo dos Santos et al. [11], who

measured mode shape displacements of a laminated composite
plate using double pulsed TV holography. The numerical differentiation method proposed by Lopes et al. [16] was used to compute
the second order spatial derivatives of the mode shapes displacement elds. In order to minimise the propagation and amplication of experimental high frequency noise, this method uses both
differentiation and ltering techniques. It allowed the localisation
of a delamination in the plate centre due to low velocity impact.
Small perturbations or discontinuities in the curvature eld, associated with the existence of delaminations, are in general mixed up
with high frequency noise. Furthermore, these signal components,
essential for damage localisation, are also attenuated or removed
by ltering techniques, which are usually applied in the computation of high order derivatives. In order to partially overcome these
problems, a new numerical-experimental method is proposed in
this paper. This method is based on the computation of the curvature eld from the experimental rotation eld, and not, as in [11],
from the experimental displacement eld. In the present work, the
rotation elds are directly measured using a speckle shear interferometric technique with a double pulse laser. Therefore, the number of spatial derivative computations is reduced to one, since
the curvature eld is obtained by differentiating only once the

rotation eld. The present technique yields an improvement on
the sensitivity of the curvature difference method in structural
damage localisation.
2. Methodology
Curvature elds can only be obtained by numerical spatial differentiation of measured displacement or rotation elds. These
experimental elds are contaminated with high frequency noise
which is propagated and amplied by the differentiation process
[17]. This difculty can be partially overcome by using low-pass lters. Nevertheless, the signal higher order spatial terms, essential
for the correct representation of derivatives, in particular their discontinuities can also be eliminated in this process [18]. Therefore, a
robust methodology is needed to assure that this information is
not eliminated by applying lters. The solution lies on a progressive ltering in the course of the successive spatial derivatives
computation. This approach aims at preserving the signal higher
order components and simultaneously avoids the propagation of
experimental high frequency noise.
1757
35
30
0.05
25
y [m]
20
0.1
15
10
0.15
5
0
0.198
0.069
0.138
x [m]
0.207
0.2765
-5
(a)
0
30
20
0.05
y [m]
10
0.1
0
-10
0.15
-20
0.198
-30
0
0.069
0.138
x [m]
0.207
0.2765
(b)
0
30
20
10
0.05
y [m]
0
-10
0.1
-20
-30
0.15
-40
-50
0.198
0.069
0.138
x [m]
0.207
0.2765
(c)
Fig. 5. Differences between the rst mode: (a) xx, (b) xy and (c) yy curvature elds magnitude.
-60
1758
0
1
0.05
y [m]
0.5
0.1
-0.5
0.15
-1
0.198
0
0.069
0.138
x [m]
0.207
0.2765
(a)
0
1
y [m]
0.05
0.5
0.1
-0.5
0.15
-1
0.198
-1.5
0
0.069
0.138
x [m]
0.207
0.2765
(b)
Fig. 6. Differences between the rst mode: (a) x and (b) y rotation elds magnitude (rotation elds with 1% signal amplitude of pseudorandom noise).
2.1. Numerical differentiation and ltering techniques
where Hn(x) are the Hermite polynomials:
Central nite differences techniques are commonly used for the

differentiation of sparse numerical data. When such techniques are
applied to experimental measurements an amplication of high
frequency noise occurs. Therefore, a more robust differentiation
technique should be used in these cases. For full-eld measurements the high order signal and high frequency noise can be partially split [17]. This allows the application of more advanced
differentiation techniques. The differentiation and smoothing techniques are combined to obtain the spatial derivatives. In this work,
the Gaussian, Fourier and spline differentiation techniques were
applied.
The rst differentiation and smoothing technique used is based
on the one-dimensional Gaussian function G(x) dened by [19]:
H0 x 1;
2
1
x
Gx p exp
:
2
2p
The spatial differentiation can be obtained from this Gaussian function, since the one-dimension n-th order Gaussian derivative is
[19]:

n

n
d Gx
1
x
p
p
Gx;
H
n
n
dx
2p
2p
H1 x 2x;
Hn1 x 2xHn x 2n 1Hn1 x:

The application of such a differentiation scheme to a measured eld
makes it necessary that the Gaussian function derivatives be dened in a discrete domain. In a discrete domain the x represents
the kernel points of the one-dimensional Gaussian derivative. The
use of discrete Gaussian derivatives has the advantage of spatial
variables separation, allowing a straightforward computation of
two-dimensional eld derivatives.
Spatial derivatives can also be obtained using a spline smoothing technique, which is based on third order coefcients of Bsplines functions, computed by tting these functions to a set of
points of a measured eld. For a uniformly spaced set of onedimensional points xi, a cubic B-spline function is written as [20]:
Sx
m
X
Bj;3 xi aj
for i 1; . . . ; m;
j1
where the coefcient Bj,3 are obtained by the de Boors recurrence

relation [20], m stands for the number of points and aj are non-zero
1759
0
15
0.05
y [m]
10
0.1
5
0.15
0.198
0
0.069
0.138
x [m]
0.207
0.2765
(a)
0
5
0.05
y [m]
-5
0.1
-10
-15
0.15
-20
0.198
0
0.069
0.138
x [m]
0.207
0.2765
(b)
0
2
0
0.05
-2
y [m]
-4
-6
0.1
-8
-10
0.15
-12
-14
0.198
-16
0
0.069
0.138
x [m]
0.207
0.2765
(c)
Fig. 7. Differences between the rst mode: (a) xx, (b) xy and (c) yy curvature elds magnitude (differentiation of rotation elds with 1% signal amplitude of pseudorandom
noise using Gaussian technique).
1760
30
20
0.05
10
y [m]
0
0.1
-10
-20
0.15
-30
-40
0.198
0
0.069
0.138
x [m]
0.207
0.2765
(a)
0
100
0.05
y [m]
50
0.1
-50
0.15
-10
0.198
0
0.069
0.138
x [m]
0.207
0.2765
(b)
0
20
15
0.05
10
y [m]
5
0
0.1
-5
-10
0.15
-15
-20
-25
0.198
0
0.069
0.138
x [m]
0.207
0.2765
(c)
noise using FFT technique).
1761
0
15
10
0.05
y [m]
5
0.1
0
-5
0.15
-10
0.198
0
0.069
0.138
x [m]
0.207
0.2765
(a)
0
15
10
0.05
y [m]
5
0
0.1
-5
-10
0.15
-15
0.198
0
0.069
0.138
x [m]
0.207
0.2765
(b)
0
6
4
2
0.05
y [m]
0
-2
0.1
-4
-6
0.15
-8
-10
-12
0.198
0
0.069
0.138
x [m]
0.207
0.2765
(c)
noise using spline technique).
1762
Table 1
Localisations and energies of impacts.
Impact 1
Impact 2
Table 2
Natural frequencies for the three states of the plate.
Position (x, y) (mm)
Impact energy (J)
(74, 51)
(215, 160)
13.5
26.2
coefcients of the function. These coefcients are computed by

minimising the least square estimator:
m
X
jSxi f xi j2 ;
j1
where f(xi) is the value of the measured eld at point xi. The nth order derivative, up to the fourth order, of the cubic B-spline can now
be directly computed according to the equation:
Frequency
number
Undamaged
Impact 1
Frequency
(Hz)
Frequency
(Hz)
Difference
(%)
Frequency
(Hz)
Difference
(%)
1
2
3
4
5
6
7
8
9
10
11
12
13
97.7
158.6
251.6
267.3
329.1
440.2
492.7
532.5
730.9
776.1
787.8
858.9
970.6
97.8
158.6
251.6
267.2
328.9
440.2
492.5
532.8
730.9
776.1
788.0
858.8
970.5
0.16
0.00
0.00
0.04
0.06
0.00
0.04
0.06
0.00
0.00
0.03
0.01
0.01
98.1
158.3
250.3
266.7
328.4
438.8
492.0
526.8
728.3
771.1
780.8
858.8
965.5
0.33
0.19
0.52
0.19
0.15
0.32
0.10
1.13
0.36
0.64
0.91
0.00
0.52
Impact 1 plus 2
m
d Sx X
Bj;3 xi ajn
n
dx
j1
n 1; 2; 3; 4:
with ajn 0 for j n > m;

6
Finally, a two-dimensional eld spatial derivative and smoothing

technique was also applied. It relies on the direct fast Fourier
transform (FFT) and its inverse (FFT1). The computations are performed in the wave number domain, where the high frequency
components, usually associated with experimental noise, are suppressed and the remaining components multiplied by a wave number vector. The m and n order spatial derivative of the eld f(x, y) in
the directions x and y, respectively, are dened by [21]:

onm f x; y
v ;
FFT1 jun jv m FFTf x; y hu;
oxn oym
Fig. 10. (a) Experimental set-up to perform impact and (b) location of the two impacts.
Fig. 11. General view of experimental set-up for the measurements of the mobility frequency response functions (FRF).
1763
Fig. 12. Experimental equipment for rotation elds measurements: (a) shear interferometric system, (b) suspended plate and loudspeaker, (c) LUMONICS double pulse Ruby
laser.
Fig. 13. Post-processing of mode shape phase maps measurements: (a) phase map obtained with interferometric system, (b) phase map after applying the phase lter
technique, (c) mode shape rotation eld.
y is the ltering function, where u and v are the wave

where hx;
p
number vector in the direction x and y, respectively and j 1.
The ltering technique used in this work is the average technique, which is usually applied for the removal of high frequency
spatial distribution noise, e.g. speckle noise, to improve the quality
of images [22]. This lter performs spatial ltering by applying the
arithmetic mean to the level values in a square or rectangular window surrounding a point, and can be computed by using an image
convolution technique, according to the equation:
f x; y f x; y hm; n;
where the symbol is the convolution operator, f(x, y) is a measured full-eld, h(m, n) the window lter, such that m and n are
the x and y axes window sizes, respectively. This technique proves
to be more computational efcient than others, such as the Gaussian, median, Wiener lter, cubic splines, FFT, windowed FFT, short
time FFT or wavelet transform techniques [23].
Instead of applying a unique lter to the measured eld, it is
presently proposed a progressive ltering of the information in
the course of the process of the successive spatial derivatives computation. A single lter is normally used for the elimination of high
frequency noise. In this process, the high order signal components,
important for the localisation of damages, are also eliminated. On
the other hand, the high frequency noise is propagated and amplied by the numerical differentiation, becoming preponderant to
the signal components. The proposed methodology tries to deal
simultaneously with these two opposite problems by attenuating
both high frequency signal and noise, without removing the most
important high order signal components. For experimental measurements with high spatial sample resolution, an effective implementation of this method can be performed by relaxing the lter
and applying progressive low-pass ltering technique trough the
numerical differentiation process. Fig. 1(a) shows a scheme of the
xx curvature eld kxx(x, y) computation from the displacement eld
w(x, y) as performed in [11]. In the present work, and since the
measured eld are the rotation eld hx(x, y) or hy(x, y), the curvature eld computation is highly simplied, as can be seen in
Fig. 1(b). In this case, only one spatial differentiation and two ltering processes are needed to obtain the xx curvature eld. The yy
and xy curvatures elds, kyy(x, y) and kxy(x, y), can be computed
in a similar way.
The ltering and differentiation techniques parameters are
found by the minimisation of the root mean square error calculate
between the post-processed experimental eld and the numerical
simulation eld. The optimal parameters strongly dependent of the
case study and can only are found by an iterative and heuristic process. It was seen that, when using experimental data, these parameters need few adjustments and with few iterations one achieves
the optimal solution [23].
2.2. Damage localisation method

Each mode shape l of an undamaged plate can be characterised
by two rotation elds, hx(l)(x, y), and hy(l)(x, y), and three curvature
elds, kxx(l)(x, y), kxy(l)(x, y), and kyy(l)(x, y). If there is some kind of
damage, the plate will present a different dynamic behaviour.
1764
Fig. 14. Undamaged plate modes x rotation elds.
Hence, for a damaged plate the two rotation elds and the three
~ x; y; k
~ x; y, and
curvature elds will be ~
hxl x; y; ~
hyl x; y; k
xxl
xyl
~ x; y. Based on this, the differences between undamaged and
k
yyl
damaged rotation elds and undamaged and damaged curvature
elds can be dened as damage location indicators. For instance,
if one takes the lth mode shape, the difference between the magnitude of the undamaged and the damaged x rotation in the point
with coordinates (xi, yi) is given by ~
hxl xi yi hxi xi ; yi . The difference between the magnitude of the undamaged and the damaged y rotations, xx curvatures, xy curvatures, and yy curvatures
in the same point and for the same mode are obtained in a similar
way.
3. Applications
The damage localisation method was applied to: (i) a rectangular isotropic plate numerically studied and (ii) a rectangular laminated plate subjected to an experimental analysis. In the rst
application, the undamaged and damaged curvature elds of the
isotropic plate were computed by differentiating numerical rotation elds. For the analysis of the laminated plate, its undamaged
curvature elds were computed relying on numerical rotation

elds and the damaged curvature elds were computed by differentiating experimental rotation elds.
3.1. Numerical application
In order to validate the damage localisation method in the presence of two different size damages and chose the best differentiation technique, a thin rectangular isotropic plate was numerically
studied using the nite element method. The plate was discretised
and a modal analysis was performed in ANSYS using 60 80
SHELL93 nite elements [24]. In order to analyse the interaction
of two different sized damages in the rst twelve natural frequencies, their position and dimension were selected according to the
anti-nodal regions distribution of the twelve mode shapes translation elds (Fig. 2). The two damages are numerically simulated by
reducing in 50% the stiffness of some of the elements.
A visual inspection of mode shapes translation elds does not
reveal the damages presence. However, some of the mode shapes
rotation elds show local changes that can be correlated to the
damages, in particular the larger one. It was also found that the
success of damage location depends on the rotation eld consid-
1765
Fig. 15. Damaged plate modes x rotation elds (impacts 1 plus 2).
ered. As an example, the fourth mode y rotation eld hy(4)(x, y) has

visible perturbations in the damaged areas, while the corresponding x rotation eld hx(4)(x, y) has almost no irregularities, as can be
seen in Fig. 3.
3.1.1. Damage localisation with clean rotation elds

The rotation elds that came directly from ANSYS were subtracted in order to compute their differences. There are large
spread areas with high differences in the rotation elds magnitude,
as shown in Fig. 4. These areas contain the damaged elements.
Fig. 5 shows the differences between the rst mode curvature
elds magnitude. Their high values are now restricted to smaller
areas, which contain the damaged elements. This is more noticeable in the differences between the xx curvature eld magnitude
(Fig. 5(a)). Since in this damage analysis the numerical rotation
elds present no noise, the curvature elds were computed by differentiating the rotation elds using nite differences. By comparing Figs. 4 and 5 one sees that the differences in curvature elds are
one order of magnitude larger than the differences in the rotation
elds, which shows that curvature differences are a better indicator of damage.
3.1.2. Damage localisation with noisy rotation elds

To simulate the noise that is always present in experimental
data, the damage method was also applied to the rotation elds
previously contaminated with 1% signal amplitude of pseudorandom noise. It can be seen in Fig. 6 that the differences in rotation
elds fail to locate the damages. However, the curvature differences perform better, although only the larger damage is located.
The best damage localisation is obtained when the Gaussian technique is used to compute the curvature elds (Fig. 7). The worst
damage localisations are obtained with the FFT technique
(Fig. 8), because of the border effect. This technique, contrary to
the Gaussian one, is not able to deal with the edge discontinuities,
even if the eld is replicated beyond the plate edges. Its poor performance is due to the difculty in obtaining continuous derivatives of non-periodic elds, such as the present rotation elds. It
is also possible to locate the larger damage using the spline technique, which does not require a mirrored eld (Fig. 9). The Gaussian technique shows a better performance, because it deals with
high frequencies at a local level, i.e. at pixel level. For global analysis techniques, such as the FFT and the spline techniques, the high
frequencies signals are superimposed to high frequency noise,
being removed in the smoothing process.
1766
plate. The two impact locations are indicated in Fig. 10(b) and no
damage can be visually detected.
3.2. Experimental application

A rectangular carbon bre reinforced epoxy plate was experimentally analysed. The plate is 0.2765 m long, 0.198 m wide,
1.825 103 m thick, and the layers elastic properties, specied
by the manufacture, are: E1 = 126.103 GPa, E2 = 15.0 GPa,
E3 = 15.746 GPa, m12 = 0.217, m23 = m13 = 0.208, G12 = 5.678 GPa,
G23 = 5.547 GPa, G13 = 3.603 GPa. The layers stacking sequence is
[0/90/+45/45/0/90]s and specic mass is q = 1690 kg/m3.
3.2.1. Damage characteristics
The carbon bre reinforced epoxy plate was subjected to delaminations produced by two low velocity impacts, using a steel spherical impactor of 0.988 kg. The impact energy was computed based
on the height of the sphere fall. The localisations and energies of
the two impacts are presented in Table 1. The plate was xed
against a support structure along the two shorter edges
(Fig. 10(a)). The sphere was guided trough an acrylic tube with
small cuts, in order to reduce air resistance. The plate surface
was painted with a thin coat of removable white paint so that
the impact zone can be registered and the supporting structure is
slightly tilted relatively to the tube axis in order to project the
sphere, after the impact and, hence, avoid a second impact in the
3.2.2. Natural frequencies and rotation elds

The experimental modal response tests performed, considering
the plate free in space, consisted in the measurement of its natural
frequencies and respective natural modes rotation elds before
damage, after impact 1, and impacts 1 plus 2. The plate was suspended using elastic rubber bands, creating nearly free boundary
conditions (Fig. 11). The natural frequencies were obtained by
measuring the mobility frequency response functions (FRF). The
transient excitation was applied using a small impact hammer
PCB model 084A17 and the velocity response was measured without contact, using a laser vibrometer Polytec model OFV-3001. A
reective tape was xed in the plate surface to increase the laser
beam reectivity and, therefore, improve the signal quality. Both
signals, i.e. the excitation and response signals, were amplied
and conditioned in the Bruel & Kjaer 2525 amplier and analysed
in frequency with the dynamic signal analyser Oros-OR35 system.
For each one of the plate states, i.e. undamaged case, impact 1,
and impact 1 plus 2, the mobility function H1 type was measured
in the upper right corner of the plate in the same surface as the
one where the mode shapes are measured. These functions charac-
Fig. 16. Undamaged plate modes xx curvature elds.
teristics lead to the conclusion that, in general, damage produces a

decrease in the natural frequencies values. This decrease is more
pronounced for higher order natural frequencies.
Table 2 shows the natural frequencies for the three states studied and their respective differences. One can see that after impact 1
the plate natural frequencies present a negligible variation relatively to those in the undamaged state, being of the same order
as the frequency resolution (156.25 mHz). However, the damage
case induced by impact 1 plus 2 produce, in general, a noticeable
reduction in the plates natural frequencies, being the maximum
value of 1.13% registered in the eighth natural frequency. For this
particular state, there is an increase in the rst natural frequency.
Similar results have been reported in the literature [9,11].
Speckle shear interferometry is a full-eld technique that allows
non-contact measurements of displacements spatial gradients with
a resolution of a few tenth of micrometers [25]. Since it measures
the displacement spatial gradient in a predetermined direction, this
technique is almost insensitive to rigid body motions. It belongs to
the optical techniques class that also include other full-eld techniques, such as holographic interferometry, projections Moir, digital holography, and point-wise techniques, such as Doppler
vibrometry [26]. The shear interferometric technique was rst
developed by two independent research groups [27,28].
1767
A MachZehnder shear interferometer (Fig. 12(a)) with a spatial

carrier in the primary fringes [29,30] was built to measure the natural modes rotation elds of the plate in free-free condition
(Fig. 12(b)). The measurement system also uses a LUMONICS double pulse Ruby laser to generate a pair of impulses with 690 nm
wave length, separated by 400 ls (Fig. 12(c)). The holograms obtained from this double exposure were recorded using a
1608 1208 pixels CCD camera JAI model CV-M2 and then are
processed to obtain the wrapped phase maps, which represents
the spatial variation of rotation elds. In order to obtain the full
rotation elds, the high frequency phase noise and phase discontinuities need to be removed by dedicated image processing techniques [23]. The phase lter technique proposed by Kemao et al.
[31] and phase unwrapping technique proposed by Volkov et al.
[32] were applied to obtain the mode shapes rotation elds.
Fig. 13 presents the results of the ltering and unwrapping processes of the phase map of the plate tenth mode shape.
The plate was suspended by elastic rubber bands and excited
using a loudspeaker placed behind it (Fig. 12(b)). The plate frontal
surface was painted with a thin coat of removable white paint to
assure a uniform irradiation intensity of the measured surface.
A Tektronix model AFG320 signal generator along with the
loudspeaker was used to excite the plate at its natural frequencies.
Fig. 17. Damaged plate modes xx curvature elds (Impacts 1 plus 2).
1768
For each natural frequency, the signal amplitude was adjusted to

frame the vibration amplitude within the systems measurement
range. This same signal was used to synchronize the laser trigger
and the plate harmonic motion.
3.2.3. Damage localisation
The modes of the undamaged plate rotation elds were numerically obtained using a model of 254 404 ANSYS SHELL93 nite
elements [24], while the modes of the damaged plate rotation
elds were measured using the pulsed laser and the shear interferometer, with a spatial phase modulation [29]. The image that describes each rotation eld has a resolution of 996 1295 pixels.
In Figs. 14 and 15 are presented the x axis rotation elds of the
rst twelve modes of the undamaged plate and the corresponding
ones after impacts 1 and 2, respectively. A global comparative analysis between the several rotation elds pairs show that, with the
exception of the fourth mode, they present similar congurations.
As seen in the numerical application, small perturbations or discontinuities in the rotation elds, due to the presence of damage,
are more pronounced in the curvature elds. The xx curvature
elds are obtained by applying the Gaussian differentiation technique described in Section 2.1. These curvature elds of the
undamaged plate and the corresponding ones after impacts 1
and 2 can be seen in Figs. 16 and 17, respectively.
The visual cross-examination of the curvature elds distribution reveals perturbations in the congurations relative to the
undamaged structure, therefore exposing the presence of a

delamination in the region where the impact 2 took place. This
result conrms that the curvature elds are more sensitive to
damage than the rotation elds. With the technique now proposed, where the rotation elds are experimentally measured,
the numerical differentiation to obtain the curvature elds is performed only once and not twice, decreasing the computational
cost. Furthermore, in each differentiation is applied a low-pass lter that is responsible for eliminating the experimental noise and
the higher order terms that characterise the perturbations in the
curvatures associated with the damage. Hence, by decreasing the
number of differentiations the perturbations due to damage are
preserved.
In order to corroborate the localisation results obtained by the
analysis of the damage plate curvature elds, the differences between them and the undamaged plate curvature elds were also
computed. It should be noted that the maximum amplitude, the
spatial resolution and the phase of the undamaged and damaged
curvature elds had to be previously adjusted. The curvature elds
magnitude differences of the undamaged plate and the damaged
plate, after impacts 1 plus 2, are shown in Fig. 18. Its analysis reveals, in several natural modes, the presence of an area that contains the location of impact 2. There is a lack of perturbations in
the curvature elds differences near the low energy impact area,
i.e. impact 1. One should note that the small impact area is represented mainly by high signal components. This means that the per-
Fig. 18. Differences between the xx curvature elds magnitude (impacts 1 plus 2).
1769
y [m]
0.05
0.1
0.15
0.2
0.075
0.14
x [m]
0.205
0.2765
Fig. 19. Plate image and delamination area located using xx curvature elds differences of the eighth mode.
turbations due to small damages are mixed with high frequency

noise, being both eliminated in the differentiation and ltering
processes.
Fig. 19 shows a plate image and the delamination area located using the xx curvature elds differences of the eighth natural mode shape, which is also the mode that presents the
higher difference between the undamaged and the damaged
plate natural frequencies (Table 2). The delamination area is dened by the contour that inscribes the two maximum juxtaposed amplitudes, presenting a disposition near the region
where impact 2 occurred.
4. Conclusion
This paper presents a numerical-experimental technique that
decreases the number of spatial differentiations needed to
numerically compute modal curvature elds, by means of measuring modal rotation elds. This process leads to a signicant
improvement in structural damage localisation. This improvement is due to the preservation of higher order terms that characterise the perturbations in the curvatures associated with the
damage. If two differentiations instead of one are performed,
these perturbations, if not sufciently high, are lost in the
smoothing and ltering processes. The use of different spatial differentiation techniques to compute the curvature elds, such as
the Gaussian function derivative, the FFT based differentiation
and the spline differentiation, was discussed. It was found that
the Gaussian differentiation is the one which present better results, since it deals with high frequencies at local level. This technique also deals well with the discontinuities at the edges. A
comparison between modal rotation and modal curvature based
methods for damage location was also presented. The numerical
and experimental applications of both methods to an isotropic
and a laminated plate show that the differences in curvatures
are more efcient than the differences in rotations. The proposed
methodology could be very useful in the development of a structural health monitoring tool for the inspection of laminate structures in industrial environment. Another major improvement in
the characterisation of damages will be the development experimental/numerical tool for quantication of their severity, which
will be performed by integrating the experimental mode, shapes
rotation elds into a nite elements model.
Acknowledgements
The authors greatly appreciate the nancial support of FCT/
POCTI (2010)/FEDER through Projects POCTI/EME/56616/2004,
PPCDT/EME/56316/2004, POCI/EME /63236/2004, the EU through
FP6-STREP Project contract No. 013517-NMP3-CT-2005-0135717.
References
[1] Adams RD, Walton D, Flitcroft JE, Short D. Vibration testing as a nondestructive test tool for composite materials. Composite Reliability, vol. 580,
ASTM STP; 1975; p. 15975.
[2] Cawley P, Adams RD. A vibration technique for non-destructive testing of bre
composite structures. J Compos Mater 1979;12:16175.
[3] Pandey AK, Biswas M, Samman MM. Damage detection from changes in
curvature mode shapes. J Sound Vib 1991;145(2):32132.
[4] Stubbs N, Kim JT, Farrar CR. Field verication of a nondestructive damage
localization and severity estimation algorithm. In: Proceedings 13th
international modal analysis conference, Nashville, USA; 1995.
[5] Ratcliffe CP. Damage detection using a modied Laplacian operator on mode
shape data. J Sound Vib 1997;204(3):50517.
[6] Ratcliffe CP, Bagaria WJ. Vibration technique for locating delamination in a
composite beam. Am Inst Aeronaut Astronaut J 1998;36(6):10747.
[7] Sampaio RPC, Maia NMM, Silva JMM. Damage detection using the frequencyresponse-function curvature method. J Sound Vib 1999;226(5):102942.
[8] Abdo MAB, Hori M. A numerical study of structural damage detection using
changes in the rotation of mode shape. J Sound Vib 2002;251(2):22739.
[9] Hamey CS, Lestari W, Qiao PZ, Song GB. Experimental damage identication of
carbon/epoxy composite beams using curvature mode shapes. Struct Health
Monit 2004;3(4):33353.
[10] Yoon MK, Heider D, Gillespie Jr JW, Ratcliffe CP, Crane RM. Local damage
detection using the two-dimensional gapped smoothing method. J Sound Vib
2005;279(12):11939.
[11] Arajo dos Santos JV, Lopes HMR, Vaz M, Mota Soares CM, Mota Soares CA, de
Freitas MJM. Damage localization in laminated composite plates using mode
shapes
measured
by
pulsed
TV
holography.
Compos
Struct
2006;76(3):27281.
[12] Guan H, Karbhari VM. Improved damage detection method based on element
modal strain damage index using sparse measurement. J Sound Vib
2008;309(35):46594.
[13] Doebling SW, Hemez FM, Barlow MS, Peterson LD, Farhat C. Selection of
experimental modal data sets for damage detection via model update. In
Proceedings of 34th AIAA/ASME/ASCE/AHS/ASC structures, structural
dynamics, and materials conference, AIAA-93-1481-CP; 1993. p. 150617.
[14] Sohn Hoon, Park Gyuhae, Wait Jeannette R, Limback Nathan P, Farrar Charles
R. Wavelet-based active sensing for delamination detection in composite
structures. Smart Mater Struct 2004;13:15360.
[15] Arajo dos Santos JV, Maia NMM, Mota Soares CM, Mota Soares CA. Structural
damage identication: a survey. In: Topping BHV, Papadrakakis M, editors.
Trends in computational structures technology. Stirlingshire, Scotland: SaxeCoburg Publications; 2008.
[16] Lopes HMR, Guedes RM, Vaz MA. An improved mixed numerical-experimental
method for stress eld calculation. Opt Laser Technol 2007;39(5):106673.
1770
[17] Blom J, Romeny BMtH, Bel A, Koenderink JJ. Spatial derivatives and the
propagation of noise in gaussian scale space. J Vis Commun Image Represent
1993;4(1):113.
[18] Lopes HMR. Estudo do uxo de energia vibratria em vigas e placas. M.Sc.
Thesis. Faculdade de Engenharia da Universidade do Porto; 2002 [In
Portuguese].
[19] Boomgaard Rvd, Smeulders A. The morphological structure of images: the
differential equations of morphological scale-space. IEEE Trans Pattern Anal
Mach Intell 1994;16(11):110113.
[20] Boor Cd. A practical guide to splines. Springer-Verlag; 1978.
[21] Arfken GB. Mathematical methods for physicists. 5th ed. Academic Press;
2000.
[22] Gonzalez RC, Wintz PA. Digital image processing. 2nd ed. Reading,
MA: Addison-Wesley; 1987.
[23] Lopes HMR. Desenvolvimento de tcnicas interferomtricas, contnuas e
pulsadas, aplicadas anlise do dano em estruturas compsitas. Ph.D. Thesis.
Faculdade de Engenharia da Universidade do Porto; 2007 [In Portuguese].
[24] ANSYS, Theory reference (Release 10.0); 2005.
[25] Kreis T. Handbook of holographic interferometry: optical and digital
methods. Weinheim: Wiley-VCH; 2005.
[26] Moreau A, Borza D, Nistea I. Full-eld vibration measurement by time-average

speckle interferometry and by Doppler vibrometry a comparison. Strain
2008;44:38697.
[27] Leendertz JA, Butters JN. An image-shearing speckle pattern interferometer
for measuring the bending moments. J Phys E Sci Instrum
1973;6(11):110710.
[28] Hung MYY, Taylor CE. Speckle shearing interferometric camera a tool for
measurement of derivatives of surface displacements. Proc SPIE
1974:16975.
[29] Pedrini G, Zou Y-L, Tiziani HJ. Quantitative evaluation of digital shearing
interferogram using the spatial carrier method. Pure Appl Opt
1996;5(3):31321.
[30] Santos F, Vaz M, Monteiro J. A new set-up for pulsed digital shearography
applied to defect detection in composite structures. Opt Laser Eng
2004;42(2):13140.
[31] Kemao Q, Soon Seah Hock, Assundi Anand. A simple phase unwrapping
approach based on ltering by windowed Fourier transform. Opt Laser Technol
2005;37:45862.
[32] Volkov VV, Zhu YM. Deterministic phase unwrapping in the presence of noise.
Opt Lett 2003;28(22):21568.

A Numerical-Experimental Method For Dama

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

A Numerical-Experimental Method For Dama

Încărcat de

Drepturi de autor:

Formate disponibile

This article appeared in a journal published by Elsevier.

Author's personal copy

Computers and Structures 89 (2011) 17541770

Contents lists available at ScienceDirect

Computers and Structures