You are on page 1of 16

STRUCTURAL CONTROL AND HEALTH MONITORING

Struct. Control Health Monit. 2006; 13:472–487


Published online 15 December 2005 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/stc.118

Damage detection in Euler–Bernoulli beams via spatial


wavelet analysis

Pol D. Spanos1,*,y, Giuseppe Failla2,z,k, Adolfo Santini2,}, and Massimiliano


Pappatico2,},yy
1
George R. Brown School of Engineering, L. B. Ryon Chair in Engineering, Rice University, P.O. Box 1892,
Houston, TX 77251, U.S.A.
2
Dipartimento di Meccanica e Materiali, Università ‘Mediterranea’ di Reggio Calabria,
Località Graziella Feo di Vito, 89060 Reggio Calabria, Italy

SUMMARY
A spatial wavelet transform (WT) is used for damage detection in Euler–Bernoulli beams subject to static
loads. It is shown that by applying the WT on the difference between the displacement responses of the
damaged and the undamaged beams for various loading conditions, boundary effects are eliminated and
damage-related local maxima are clearly identified in the WT modulus map. Estimates of damage locations
and amplitudes are then obtained by two separate optimization procedures, in which each damaged section
is modelled by an equivalent ‘reduced-stiffness’ spring. The effectiveness of the method is assessed using
digitally-simulated data obtained via cubic interpolation on a finite number of nodal displacements. In this
context, results pertaining to multi-crack beams prove quite accurate even for small damage amplitudes.
Copyright # 2005 John Wiley & Sons, Ltd.

KEY WORDS: damage detection; static displacement response; wavelet analysis

1. INTRODUCTION

Originally developed for seismic records processing [1, 2], wavelet analysis has now become an
important tool in a plethora of engineering applications, including time–frequency analysis of
random processes, random field simulation, system identification, damage detection, and

*Correspondence to: Pol D. Spanos, George R. Brown School of Engineering, L. B. Ryon Chair in Engineering,
Rice University, P.O. Box 1892, Houston, TX 77251, U.S.A.
y
E-mail: spanos@rice.edu
z
E-mail: gfailla@ing.unirc.it
}
E-mail: adolfo.santini@unirc.it
}
E-mail: massimiliano.pappatico@unirc.it
k
Researcher.
 Full Professor.
yy
Ph.D. Student.

Contract/grant sponsor: NSF


Contract/grant sponsor: M.I.U.R.

Received 30 May 2005


Copyright # 2005 John Wiley & Sons, Ltd. Revised 6 September 2005
DAMAGE DETECTION IN EULER–BERNOULLI BEAMS 473

material characterization [3]. Indeed, local features and transient phenomena which cannot be
detected by conventional methods such as the traditional Fourier analysis may be captured by
the WT.
In damage detection applications, the WT may be used in conjunction with non-destructive
tests to estimate the damage state, its evolution in time, and its propagation in space. Early
studies, primarily devoted to fault diagnosis in machineries, have shown that transient vibrations
due to developing damage can be disclosed by the local maxima of the WT mean-square map [4].
Also, visual inspection of the WT modulus and phase has been used to localize a fault [5]. Based
on these preliminary results, wavelet analysis has been extended to damage detection in structural
components. Various methods have been formulated in this context, depending on the wavelet-
transformed function. For multi-crack beams flexural waves induced by an impact hammer have
been considered [6, 7]; and for delaminated composites, vibration response [8], acoustic emission
waves generated by low-velocity impact loads [9], or Lamb waves [10].
For damage detection in beams, methods where the WT is applied on spatial signals, either
static displacement responses or vibration modes, are most commonly employed. Securing
signals with a sufficient resolution is certainly a crucial issue when implementing spatial wavelet
analysis. For this reason methods based on vibration modes, which can be readily acquired via
standard modal analysis, have become preferable over the last decade. First Liew and Wang
[11], who have studied a simply-supported beam with a single non-propagating crack, have
shown that a damaged section corresponds to local maxima in the WT of the vibration modes.
However, only numerical modes have been considered, and no estimate of the damage
amplitude has been provided in Reference [11]. An attempt in this regard has been made by
Okafor and Dutta [12] who have used a least-square regression analysis to correlate the damage
amplitude in cantilever beams with the WT of experimental modes recorded by a laser
vibrometer. Next Hong et al. [13], who have applied the WT to the experimental modes of a
free–free beam, have provided further insight into the correlation existing between damage-
related singularities and local maxima of the WT modulus. Specifically, based on the concept of
Lipschitz singularity [14], they have shown that WT modulus local maxima, associated with a
damaged section, exhibit asymptotically-decaying amplitudes across the WT scales. A proper
wavelet family for this task must have an adequate number of vanishing moments n. That is, the
inequality n5a must hold, where a is the damage-related Lipschitz singularity index. Within this
framework, Hong et al. have also devised a method to estimate the damage amplitude. The
results obtained by Hong et al. have been confirmed by Douka et al. [15], who have applied
the WT to the first experimental mode of a single-crack cantilever beam. Later, Haase and
Widjajakusuma [16] have proposed a fast algorithm to determine the WT maxima lines.
Another interesting method involving the vibration modes of a multi-crack beam has been
devised by Chang and Chen [17] who have used the WT local maxima to determine the position
of the cracks; and experimental natural frequencies to predict the crack depths via the
characteristic equation. Chang and Chen have also pointed out that the WT may exhibit local
maxima at the beam ends, even if no damaged section is located therein. These boundary effects,
relevant for any function defined on a finite interval, represent a non-negligible limitation of the
approach, since no reliable diagnosis can be formulated on the damage state at the beam ends.
In a later work, Gentile and Messina [18] have proposed to eliminate boundary local maxima by
applying the WT on windowed vibration modes, which decay to zero at the beam ends. The
result sought in Reference [18], however, is to build a WT where boundary local maxima do not
appear, and the only local maxima retained are those related to damaged sections in the central

Copyright # 2005 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2006; 13:472–487
474 P. D. SPANOS ET AL.

part of the beam. Obviously, in this manner the WT cannot detect any damaged section located
near the beam ends.
Note that scarce contributions may be found in the literature on damage detection via wavelet
analysis of beam static displacements. Specifically, Wang and Deng [19] and later Quek et al. [20]
have shown that damage-related local maxima occur in the WT of the numerical displacement of
beams with either vertical or embedded cracks, featuring different orientation and width. An
estimate of the minimum number of measurement points to implement wavelet analysis has been
also attempted, for a damage of a specified amplitude [19]. Nevertheless, in these references
several critical aspects have not been discussed, such as the applicability of the method on
experimental data, remedies to boundary effects, and estimation of the damage amplitude.
Further, note that most studies have involved single-crack beams. Multi-crack beams have
been considered only in Reference [17], but the same amplitude has been considered for all
damaged sections.
Clearly then, beyond the considerable research effort devoted in the last decade on damage
detection via spatial wavelets, important issues remain to be resolved.
This short paper focuses on spatial wavelet analysis applied on data pertaining to the static
displacement of Euler–Bernoulli cracked beams. A WT modulus map is built where boundary
effects are eliminated and damage-related local maxima are clearly identifiable. Estimates of
damage parameters are then obtained via two separate optimization problems, based on a
displacement solution for cracked beams recently proposed in Reference [21]. Numerical tests
are conducted on the (simulated) pseudo-experimental response of beams featuring up to three
cracks. It is found out that the WT modulus map discloses all damaged sections as long as the
ratio between the damage amplitudes does not exceed a certain threshold. A complementary
approach is adopted for multi-crack beams where damage amplitudes may differ significantly.

2. WAVELET TRANSFORM

The concept of WT was first introduced in References [1, 2]. It decomposes a function f(x)
belonging to the set of finite-energy one-dimensional functions, f ðxÞ 2 L2 ðRÞ; into a two-
parameter family of elementary functions, each derived from a basic or mother wavelet, cðxÞ:
The first parameter, a, corresponds to a dilation or compression of the mother wavelet, and is
generally referred to as the scale. The second parameter, b, defines a shift of the mother
wavelet along the x-domain. In mathematical terms
Z 1
ÿ1=2
 
Wf ða; bÞ ¼ a f ðxÞc ðx ÿ bÞ=a dx ð1Þ
ÿ1

where a2R+, b2R and the bar denotes complex conjugation. In the literature, Equation (1) is
generally referred to as continuous wavelet transform.
Obviously, the localization properties of the WT depend on the value of the parameter a. As a
approaches zero, the dilated wavelet aÿ1/2c(x/a) is highly concentrated at x ¼ 0 and the WT
displays the small-scale features of the function f(x), at various locations b. However, as a
approaches +1, the WT yields an increasingly coarser spatial resolution, displaying the large-
scale features of the function f(x).

Copyright # 2005 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2006; 13:472–487
DAMAGE DETECTION IN EULER–BERNOULLI BEAMS 475

Discrete versions of Equation (1), where both a and b vary on a finite grid, have also been
developed in conjunction with fast decomposition and reconstruction algorithms. More details
and relevant references may be found in Reference [3].

3. PROPOSED IDENTIFICATION METHOD

In this section a damage detection method is presented in conjunction with an Euler–Bernoulli


cracked beam.

3.1. WT modulus map and boundary effects for beam deflection


Denote by L the length of an Euler–Bernoulli cracked beam, h the height and I the moment of
inertia of its undamaged cross-section, E the material’s Young’s modulus. Further, be xi’s the
locations of ND non-propagating vertical cracks, and di’s the corresponding crack depths.
Denote by q(x) an arbitrary applied load. Following Reference [21], where a vertical crack is
modelled via a reduced-stiffness equivalent spring, the displacement response, measured in cm,
is represented as
ND 
q½4Š ðxÞ X q½2Š ðxi Þ

2 3 ai
sðxÞ ¼ c1 þ c2 x þ c3 x þ c4 x þ þ 2c3 þ 6c4 xi þ ðx ÿ xi ÞUðx ÿ xi Þ
EI i¼1
EI 1 ÿ ai A
ð2Þ
In Equation (2), U(xÿxi) is the Heaviside function; q[2](x) and q[4](x) are a second- and fourth-
order primitive of the load q(x); A ¼ 2:013 cmÿ1 ; cj’s are four constants depending on the
boundary conditions; and ai’s are given in terms of the crack depths di’s by the formula
2Zi ÿ Z2i
ai ¼ ; Zi ¼ di =h for i ¼ 1; . . . ; ND ð3Þ
ð0:9=h ÿ AÞZ2i ÿ ð0:9=h ÿ AÞ2Zi þ 0:9=h
Based on Equation (3), the stiffness of the ith crack-equivalent spring is given by
kfi ¼ ð1 ÿ ai AÞEI=ai [21].
Next apply the WT, Equation (1), on the displacement response, Equation (2). That is,
compute the integral
Z L
ÿ1=2
 
Ws ða; bÞ ¼ a sðxÞc ðx ÿ bÞ=a dx ð4Þ
0

As reported in References [13–15], the damage-related singularities in Equation (2) determine


local maxima in the Ws ða; bÞ modulus map, whose amplitude decays asymptotically across scales.
Consider then the two-crack clamped–clamped (CC) beam shown in Figures 1(a)–2(a), where
L ¼ 100 cm; E ¼ 210 GPa; P ¼ 2 kN: The cross section is a square with h ¼ 5 cm: Be the
damage parameters {x1 ¼ L=20; Z1 ¼ 0:1} and {x2 ¼ L=4; Z2 ¼ 0:5}. To compute Ws ða; bÞ;
the Mexican Hat wavelet
2
cðxÞ ¼ pffiffiffi ðx2 ÿ 1Þ expðÿx2 =2Þ ð5Þ
3p1=4

Copyright # 2005 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2006; 13:472–487
476 P. D. SPANOS ET AL.

x 10-3
4 1.5
P
L/ 3 1.25
L/ 4 3.5
1

3 0.75
L/ 20 EI 0.5
2.5
L 0.25

a, scale
2 0

1.5
P
1

0.5

0 10 20 30 40 50 60 70 80 90 100
(a) (b) b, position

Figure 1. Two-crack CC beam subject to a load P at x ¼ L=3: (a) geometry and bending moment diagram;
and (b) WT, Equation (4), modulus contour map of the displacement response, Equation (2).

x 10-3
4 1.5
P
L/2 1.25
L/4 3.5
1

3 0.75
L / 20 EI 0.5
2.5
L 0.25
a, scale

2 0

1.5
P
1

0.5

0 10 20 30 40 50 60 70 80 90 100
(a) (b) b, position

Figure 2. Two-crack CC beam subject to a load P at x ¼ L=2: (a) geometry and bending moment diagram;
and (b) WT, Equation (4), modulus contour map of the displacement response, Equation (2).

is selected. The same wavelet family has been used in Reference [13]. Figure 1(b) shows that, for
P applied at x ¼ L=3; the Ws ða; bÞ modulus map exhibits local maxima at x1 ¼ L=20 and
x2 ¼ L=4; with amplitude decaying across the scales. It may be also observed that:
(i) Local maxima with amplitude decaying across scales appear also near the beam end
x ¼ L; despite the fact that no damaged section is present [17, 18].
(ii) Damage-related local maxima may not be visible depending on the position of the applied load.
Recognize in Figure 2(b) that, for P applied at x ¼ L=2; no local maxima appear at x2 ¼ L=4
in the Ws ða; bÞ modulus map, as a result of the low bending moment value at x2 ¼ L=4:

Both issues (i)–(ii) are next addressed to formulate a reliable spatial wavelet identification method.

Copyright # 2005 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2006; 13:472–487
DAMAGE DETECTION IN EULER–BERNOULLI BEAMS 477

First, consider the WT of the undamaged beam displacement response u(x) to a load q(x).
That is,
Z L
Wu ða; bÞ ¼ aÿ1=2 uðxÞc½ðx ÿ bÞ=aŠ dx ð6Þ
0

where
q½4Š ðxÞ
uðxÞ ¼ c01 þ c02 x þ c03 x2 þ c04 x3 þ ð7Þ
EI
It is seen in Figure 3(a) that for P ¼ 2 kN applied at x ¼ L=3 boundary local maxima,
analogous to those in Figure 1(b), appear in the Wu ða; bÞ modulus map. This fact suggests that
boundary effects can be eliminated if the WT is applied on the difference between the
displacement responses of the damaged and undamaged beam, subject to the same load. Thus,
the WT Z L
Wd ða; bÞ ¼ Ws ða; bÞ ÿ Wu ða; bÞ ¼ aÿ1=2 ½sðxÞ ÿ uðxފc½ðx ÿ bÞ=aŠ dx ð8Þ
0

is introduced. Combining Equations (2) and (7) one finds that


ND 
q½2Š ðxi Þ

00 00 00 2 00 3
X ai
sðxÞ ÿ uðxÞ ¼ c1 þ c2 x þ c3 x þ c4 x þ 2c3 þ 6c4 xi þ ðx ÿ xi ÞUðx ÿ xi Þ ð9Þ
i¼1
EI 1 ÿ ai A
for c00k ¼ ck ÿ c0k : Figure 3(b) shows that boundary effects are eliminated in the Wd ða; bÞ modulus
map, and the only local maxima with amplitude decaying across scales are those associated with
the two damaged sections at x1 ¼ L=20 and x2 ¼ L=4:
In a most general case, to ensure that all damaged sections are clearly displayed regardless of
the selected load position, a certain number NC of loading conditions may be considered to
build the WT modulus map. To this purpose, the WT modulus
NC
X
Wtot ða; bÞ ¼ jWdðkÞ ða; bÞj ð10Þ
k¼1

x 10-3 x 10-4
4 1.5 4 7.5

1.25
3.5 3.5
1 5

3 0.75 3
0.5 2.5
2.5 2.5
0.25
a, scale
a, scale

2 0 2 0

1.5 1.5

1 1

0.5 0.5

0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
(a) b, position (b) b, position

Figure 3. Two-crack CC beam of Figure 1(a): (a) WT, Equation (6), modulus contour map of the
undamaged beam displacement response, Equation (7); and (b) WT, Equation (8), modulus contour map.

Copyright # 2005 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2006; 13:472–487
478 P. D. SPANOS ET AL.

can be introduced, where the symbol WdðkÞ ða; bÞ denotes Wd ða; bÞ; Equation (8), for the kth
loading condition. That is,
Z L
WdðkÞ ða; bÞ ¼ WsðkÞ ða; bÞ ÿ WuðkÞ ða; bÞ ¼ aÿ1=2 ½sðkÞ ðxÞ ÿ uðkÞ ðxފc½ðx ÿ bÞ=aŠ dx ð11Þ
0
(k) (k)
The symbols s (x) and u (x) in Equation (11) refer to the displacement responses (2) and (7),
for a given load q(k)(x). For the two-crack beam in Figures 1(a)–2(a), consider then the NC ¼ 5
loading conditions described in Figure 4(a), where a single load P ¼ 2 kN is applied at different
positions equally-spaced along the beam. Note in Figure 4(b) that all damage-related local
maxima are clearly captured by Equation (10).
Consider next a beam for which the end displacements and/or rotations are not restrained. In
this case boundary local maxima reflect the discontinuity of the displacement function and/or its
first derivative at the beam ends. The approach devised for a CC beam can be then generalized
by enforcing the continuity of the displacement function and/or its first derivative at the beam
ends. For instance, this means that for a cantilever and a simply-supported (SS) beam the
extended displacement functions

cantilever beam (free end at x ¼ L)


( ðkÞ
s ðxÞ; 04x4L
ðkÞ
se ðxÞ ¼ ð12aÞ
sðkÞ ðLÞ þ dsðkÞ =dxjx¼L  ðx ÿ LÞ; x5L

P
L/ 6
L/ 4

L/ 20 EI

L x 10-3
4 1.5

1.25
P 3.5
L/ 3 1
L/4
3 0.75

0.5
L/ 20 EI 2.5
0.25
a, scale

L
2 0

...............................
1.5

P
5L/ 6 1
L/ 4
0.5

L/20 EI
0 10 20 30 40 50 60 70 80 90 100
L b, position
(a) (b)
Figure 4. Two-crack CC beam subject to a load P at NC ¼ 5 equally-spaced positions: (a) geometry; and
(b) WT, Equation (10), contour map.

Copyright # 2005 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2006; 13:472–487
DAMAGE DETECTION IN EULER–BERNOULLI BEAMS 479

uðkÞ ðxÞ;
(
04x4L
uðkÞ
e ðxÞ ¼ ð12bÞ
uðkÞ ðLÞ þ duðkÞ =dxjx¼L  ðx ÿ LÞ; x5L

simply-supported beam
8 ðkÞ
>
> ds =dxjx¼0  x; x40
<
seðkÞ ðxÞ ¼ sðkÞ ðxÞ; 04x4L ð12cÞ
>
>
: ðkÞ
ds =dxjx¼L  ðx ÿ LÞ; x5L

8 ðkÞ
>
> du =dxjx¼0  x; x40
<
ðkÞ ðkÞ
ue ðxÞ ¼ u ðxÞ; 04x4L ð12dÞ
>
>
: ðkÞ
du =dxjx¼L  ðx ÿ LÞ; x5L

can be introduced to compute the WT in Equation (10).

In this context, a two-crack SS beam of square cross-section, with L ¼ 100 cm; h¼ 5 cm;
E ¼ 210 GPa; is considered. Be the damage parameters {x1 ¼ L=20; Z1 ¼ 0:1} and {x2 ¼ L=2;
Z2 ¼ 0:1}. Figure 5(a) shows Equation (10) for s(k)(x) and u(k)(x) given by Equations (2) and (7),
when NC ¼ 5 equally-spaced loads P ¼ 2 kN are applied individually on the beam. Amplitude-
decaying local maxima appear at x ¼ 0 and x ¼ L; due to the discontinuity of the first derivative
of both s(k)(x) and u(k)(x). Clearly, no damaged section may be detected at x1 ¼ L=20: Figure
5(b) shows Equation (10) when applied on the displacement responses (12c) and (12d), defined
on the dummy interval [ÿL/3, 4/3L]. Recognize that boundary effects are eliminated, and only
damage-related amplitude-decreasing local maxima remain.

x 10-3 x10-3
4 1.5 4 1.5

1.25 1.25
3.5 3.5
1 1

3 0.75 3 0.75

0.5 0.5
2.5 2.5
0.25 0.25
a, scale
a, scale

2 0 2 0

1.5 1.5

1 1

0.5 0.5

0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
(a) b, position (b) b, position

Figure 5. Equation (10) contour map for a two-crack SS beam subject to a load P at NC ¼ 5
equally-spaced positions: (a) for s(k)(x), u(k)(x) defined on [0, L]; and (b) for seðkÞ ðxÞ; uðkÞ
e ðxÞ defined by
Equations (12c) and (12d) on the dummy interval [ÿL/3, 4/3L].

Copyright # 2005 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2006; 13:472–487
480 P. D. SPANOS ET AL.

3.2. Damage location estimation


Figures 4(b) and 5(b) demonstrate that Equation (10) yields the exact location of the damaged
sections when the theoretical displacement responses s(k)(x) are used. However, when dealing
with experimental data damage-related local maxima in Equation (10) may shift across scales,
within a small but finite range. In this case, an estimate of the damage locations becomes
important.
First recognize that, at each damaged section xi, the approximation
Z L
ÿ1=2
Wd ða; xi Þ  Ki  a ðx ÿ xi ÞUðx ÿ xi Þc½ðx ÿ xi Þ=aŠ dx a ! 0 ð13Þ
0

holds, in which Ki is a constant depending on the damage state of the beam, Ki ¼ ½2c3 þ
6c4 xi þ q½2Š ðxi Þ=EIŠai =ð1 ÿ ai AÞ: Equation (13) may be readily derived by observing that, as the
scale a ! 0 in Equation (8), the contribution of the singular part ðx ÿ xi ÞUðx ÿ xi Þ dominates
over the contribution of the remaining terms involved in Equation (9).
Next, for each damaged section xi introduce the error function
ei ða; bÞ ¼ jjWd ða; bÞj ÿ jWh ða; bÞjj b 2 ½xi ÿ Di ; xi þ Di Š ð14Þ
where Di is small, and Wh(a,b) is given by
Z L
ÿ1=2
Wh ða; bÞ ¼ a ðx ÿ bÞUðx ÿ bÞc½ðx ÿ bÞ=aŠ dx ð15Þ
0

By comparing Equation (15) with (13), it is clear that Equation (14) attains its minimum value
when b ¼ xi : Note also that, unlike Equation (13), Equation (15) does not involve the constant
Ki. The latter, in fact, applies for all positions b and can be thus eliminated when computing the
error function (14).
Denote by s# ðkÞ ðxÞ the experimental displacement response to the kth loading condition. Also,
denote by
NC
X
W# tot ða; bÞ ¼ jW ðkÞ
# ða; bÞj ð16Þ
d
k¼1
and Z L
W ðkÞ ðkÞ ðkÞ
# ða; bÞ ¼ W s# ða; bÞ ÿ Wu ða; bÞ ¼ a
ÿ1=2
½#s ðkÞ
ðxÞ ÿ uðkÞ ðxފc½ðx ÿ bÞ=aŠ dx ð17Þ
d
0
the counterparts of Equations (10) and (11). Based on Equation (14), an estimate x* i for each
damage location can be sought via the minimization problem
NC X
NAP
jjW ðkÞ
X
min Pðx* i Þ ¼ # ðaj ; x
* i Þj ÿ jWh ðaj ; x* i Þjj
d
ð18Þ
x* i
k¼1 j¼1
where NAP is a certain number of scales aj introduced to smoothen the noise effects. Note that
the computational effort involved in Equation (18) is quite small, since the damage locations x* i
can be sought only within the vicinity of the local maxima of Equation (16). Also note that no
a priori knowledge on the damage amplitudes is required, as no damage amplitude-related term
is involved in Equation (15).

3.3. Damage amplitude estimation


Upon determining the damage locations by Equation (18), the damage amplitudes may be
estimated by seeking the optimal set of parameters Z* i ’s that minimize the difference between the

Copyright # 2005 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2006; 13:472–487
DAMAGE DETECTION IN EULER–BERNOULLI BEAMS 481

experimental data and the theoretical displacement responses s(k)(x), given as Equation (2). For
this, the following minimization problem:
NC X
NAO X
NB
jjWs#ðkÞ ðaj ; bm Þj ÿ jWsðkÞ ðaj ; bm Þjj
X
min Oð*Z1 ; . . . ; Z* ND Þ ¼ ð19Þ
Z* 1 ;...;*ZND
k¼1 j¼1 m¼1

is formulated, where NAO is a number of scales aj , and NB is a number of reference points


selected along the total length of the beam. The damage locations x* i ; estimated via the
minimization problem defined by Equation (18), will be taken as parameters xi’s in the
theoretical displacement responses s(k)(x) involved in Equation (19).
Note that the range of the NAP/NAO scales in Equations (18) and (19) can be selected to
minimize the noise effects. For this purpose high scales are always preferable, as they are
associated with wavelet functions of large spatial support as compared to the length of the
beam. The maximum scale involved in the functional Pðx* i Þ; however, should always be such
that the corresponding wavelet does not extend up to the next damaged section. Otherwise, the
approximation by Equation (13) will not be reliable.
It is worth noting that the proposed identification method has significant advantages as
compared to alternative methods available in the literature [21–23]. Clearly the exact number of
damaged sections is estimated by Equation (16). Also, the two separate minimization problems
posed by Equations (18) and (19) do not require a significant computational effort, even for
multi-crack beams. Methods such as those in References [21, 22], however, imply an a priori
knowledge of the number of damaged sections, which is rather unlikely to happen in practice.
Further, they demand an almost prohibitive computational effort for multi-crack beams, since
damage locations and amplitudes are estimated in conjunction, via a single optimization
problem [21–23].

4. NUMERICAL RESULTS ON PSEUDO-EXPERIMENTAL DATA

4.1. Noise and discrete-sampling effects


The effectiveness of the proposed method is next assessed using numerical data. The latter are
produced by a cubic interpolation on a finite number of M pseudo-measured displacements,
given as random deviates of the corresponding theoretical ones [21, 22]. That is,
L
s#ðkÞ ðxm Þ ¼ sðkÞ ðxm Þð1 þ rm
ðkÞ
Þ; xm ¼ m for m ¼ 1; 2; . . . ; M ð20Þ
Mþ1
where r(k)
m ’s are independent random variables uniformly-distributed over an interval [ÿr; r]. To
detect the damage-related local maxima in Equation (16), intuition suggests that both the
sampling rate and the noise magnitude are critical.
Consider then a three-crack CC beam of square cross-section, with L ¼ 100 cm; h ¼ 5 cm;
E ¼ 210 GPa: Be the damage parameters {x1 ¼ L=20; Z1 ¼ 0:1}, {x2 ¼ L=4; Z2 ¼ 0:8}, {x3 ¼ 3L=5;
Z3 ¼ 0:3}. Again, compute Equation (16) for NC ¼ 5 equally-spaced loads P ¼ 2 kN applied
individually on the beam.
Figures 6(a) and 7(a) describe Equation (16) for {M ¼ 10; rðkÞ
m ¼ 0} and {M ¼ 300; r ¼ 0:05}
in Equation (20), respectively. It is seen that additional amplitude-decaying local maxima may
appear at undamaged sections of the beam, as a result of poor measurement sampling

Copyright # 2005 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2006; 13:472–487
482 P. D. SPANOS ET AL.

x 10-3 x 10-3
4 2 4 3.5
3
3.5 1.5 3.5 2.5
2
3 1 3
1.5

0.5 1
2.5 2.5
0.5
a, scale

a, scale
2 0 2 0

1.5 1.5

1 1

0.5 0.5

0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
(a) b, position (b) b, position

Figure 6. Equation (16) contour map for a three-crack CC beam subject to a load P at NC ¼ 5
equally-spaced positions: (a) for fM ¼ 10; rðkÞ ðkÞ
m ¼ 0:0g in Equation (20); and (b) for fM550; rm ¼ 0:0g
in Equation (20).

x 10-3 x 10-3
4 3.5 4 3.5

3 3
3.5 2.5 3.5 2.5

2 2
3 3
1.5 1.5

1 1
2.5 2.5
0.5 0.5
a, scale

a, scale

2 0 2 0

1.5 1.5

1 1

0.5 0.5

0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
(a) b, position (b) b, position

Figure 7. Equation (16) contour map for a three-crack CC beam subject to a load P at NC ¼ 5 equally-
spaced positions, for fM ¼ 300; r ¼ 0:05g in Equation (20): (a) 1 test for each loading condition; and
(b) 30 tests for each loading condition.

(see Figure 6(a), at the vicinity of x ¼ 10; 40, 90 cm), and noise magnitude (see Figure 7(a), at
the vicinity of x ¼ 45 cm). Further, in Figure 6(a) it is apparent that local maxima
corresponding to small levels of damage may not be visible, as at x1 ¼ L=20; and at severe
levels of damage they may spread over a wider area, as at x2 ¼ L=4: As far as poor sampling is
concerned, its effects may be eliminated by increasing the number of base points M. Figure 6(b)
shows in fact that only damage-related amplitude-decaying local maxima remain in Equation
ðkÞ
(16), when {M550, rm ¼ 0} is set in Equation (20). It may be then concluded that the cubic
interpolation yields a displacement response almost identical to the theoretical one, as long as a
sufficient number of base points M is taken in Equation (20). As far as noise magnitude is
concerned, its effects may be ameliorated by performing a certain number of tests for each

Copyright # 2005 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2006; 13:472–487
DAMAGE DETECTION IN EULER–BERNOULLI BEAMS 483

loading condition. In this context, Figure 7(b) shows that only damage-related amplitude-
decaying local maxima are retained in Equation (16), when the displacement responses s# ðkÞ ðxÞ
are averaged over 30 tests for each loading condition.
In References [21, 22], the authors have computed the minimum number of loading tests to
estimate the damage parameters with a certain accuracy. Herein, the minimum number of tests
may be evaluated to smoothen out additional amplitude-decaying local maxima in Equation
(16), for given values of r and M in Equation (20). For this purpose, M is taken within the range
20–300. A great number of data, in fact, may be readily obtained using modern laser technology,
by which a resolution up to 512  512 is achieved in a few mm2 [24]. Laser scanning is also
extremely fast and numerous tests may be conveniently performed.
Various CC beams featuring up to three cracks have been considered, where the ratio between
the largest and smallest damage amplitude is Zmax /Zmin 57:0: Specifically, the results reported in
Tables I and II refer to a beam of square cross section with L ¼ 100 cm; h ¼ 5 cm; E ¼ 210 GPa;
and NC ¼ 5 equally-spaced loads P ¼ 2 kN applied individually on the beam. Data in each row
Z of Tables I and II indicate the minimum number of tests to detect all the cracks, when
Z ¼ Zmin : As expected, the number of tests increases with r; and decreases with M. Also small

Table I. Minimum number of tests for each loading condition to detect damage-related local
maxima in Equation (16), for r ¼ 0:01 in Equation (20).
M
Z (%) 20 30 40 50 100 150 200 250 300
10 } 40 20 20 20 20 10 5 5
20 10 5 5 5 5 1 1 1 1
30 5 5 5 1 1 1 1 1 1
40 1 1 1 1 1 1 1 1 1
50 1 1 1 1 1 1 1 1 1
60 1 1 1 1 1 1 1 1 1
70 1 1 1 1 1 1 1 1 1
80 1 1 1 1 1 1 1 1 1
90 1 1 1 1 1 1 1 1 1

Table II. Minimum number of tests for each loading condition to detect damage-related local
maxima in Equation (16), for r ¼ 0:05 in Equation (20).
M
Z (%) 20 30 40 50 100 150 200 250 300
10 } } } } } } } } 30
20 } } } } 40 40 20 20 10
30 30 30 20 20 20 20 10 10 5
40 20 20 20 10 10 5 5 5 5
50 10 5 5 5 5 5 5 5 1
60 10 5 5 5 5 1 1 1 1
70 5 5 5 1 1 1 1 1 1
80 5 1 1 1 1 1 1 1 1
90 1 1 1 1 1 1 1 1 1

Copyright # 2005 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2006; 13:472–487
484 P. D. SPANOS ET AL.

x 10-3
4 1 76.055

3.5 0.75
76.0545

3 0.5

76.054
2.5 0.25
a, scale

Π
2 0 76.0535

1.5
76.053

1
76.0525
0.5

76.052
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90
(a) b, position (b) b

Figure 8. Three-crack CC beam subject to a load P at NC ¼ 5 equally-spaced positions, for


fM ¼ 300; r ¼ 0:05g in Equation (20), 30 tests for each loading condition: (a) Equation (16) contour map;
and (b) functional Pðx* i Þ in Equation (18), for rðkÞ
m ¼ 0:0 (solid line) and for r ¼ 0:05 in Equation (20).

damage amplitudes require, in general, a higher number of tests and a higher number of
interpolation points M. Empty cells indicate that the minimum number of tests required is 550.

4.1.1. Identification of a three-crack beam: case (a). Consider next a three-crack CC beam of
square cross section, with L ¼ 100 cm; h ¼ 5 cm; E ¼ 210 GPa: Be the damage parameters
{x1 ¼ L=20; Z1 ¼ 0:1}, {x2 ¼ L=2; Z2 ¼ 0:2} and {x3 ¼ 7L=10; Z3 ¼ 0:3}. Apply Equation (16)
for NC ¼ 5 equally-spaced loads P¼ 2 kN applied individually on the beam. For {M ¼ 300;
r ¼ 0:05} in Equation (20), it is assumed that 30 tests are performed for each loading condition.
Figure 8(a) shows that all damage-related local maxima are clearly captured by Equation (16).
Since discrete sampling and noise determine a certain shift of the local maxima across the scales,
the minimization problem posed by Equation (18) is used to estimate the damage locations.
Figure 8(b) shows the corresponding functional Pðx* i Þ for NA ¼ 5 equally-spaced scales within
the range [1, 3]. The estimated damage locations are x* 1 ¼ 5:0; x* 2 ¼ 50:8; x* 3 ¼ 69:6 cm: The
maximum percentage error is equal to 1.6% at x2. The minimization problem posed by
Equation (19) is then solved, for NAO ¼ 5 equally-spaced scales within the range [20, 30] and
NB ¼ 10 locations along the beam length. Specifically, b1 ¼ L=20 and bm ¼ m  L=10 for
m ¼ 1; . . . ; 9; have been selected. The derived estimates obtained for the damage amplitudes
are Z* 1 ¼ 0:1; Z* 1 ¼ 0:2; Z* 1 ¼ 0:3; with no error as compared to the exact values.

5. AN ALTERNATIVE METHOD FOR MULTI-CRACK BEAMS

The preceding conclusions based on the data reported in Tables I and II cannot be generalized
to multi-crack beams where Zmax =Zmin 57:0: This is due to the fact that in this case local maxima
related to the smallest damages may not be detectable in the WT modulus map. In fact, the WT
modulus map is flat everywhere, while sharp local maxima appear where the damage is severe.
To determine whether one or more damaged sections are hidden in the flat regions of the WT
modulus map, the identification method can be modified by formulating the generalized

Copyright # 2005 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2006; 13:472–487
DAMAGE DETECTION IN EULER–BERNOULLI BEAMS 485

4 0.02 36.48

3.5 0.015 36.475

3 0.01
36.47

2.5 0.005
36.465
a, scale

Π
2 0
36.46
1.5
36.455
1
36.45
0.5

36.445
0 10 20 30 40 50 60 70 80 90 100 0 5 10 15 20 25 30
(a) b, position (b) b

Figure 9. Three-crack CC beam subject to a load P at NC ¼ 5 equally-spaced positions, for


fM ¼ 300; r ¼ 0:1g in Equation (20), 10 tests for each loading condition: (a) Equation (16) contour
ðkÞ
map; and (b) functional Pðx* i Þ in Equation (18), for rm ¼ 0:0 (solid line) and for r ¼ 0:1 in Equation (20).

minimization problem
NC X
NAO X
NB
jjWs#ðkÞ ðaj ; bm Þj ÿ jWsðkÞ ðaj ; bm Þjj
X
min % x* ND1þ1 ; . . . ; x* ND ; Z* 1 ; . . . ; Z* ND Þ ¼

x* ND1þ1 ;...;x* ND Z* 1 ;...;*ZND
k¼1 j¼1 m¼1
ð21Þ
where x* ND1þ1 ; . . . ; x* ND are ND2 ¼ ND ÿ ND1 potential damage locations, in addition to the ND1
visible in the contour map of Equation (16), and estimated via the minimization problem posed
by Equation (18). Since the actual damage state is unknown, the number ND2 shall be set
arbitrarily.
Finally, note that the minimization problem posed by Equation (21) may be also used to
identify the nature, real ones or artefacts, of damage-like amplitude-decaying local maxima such
as in Figure 7(a). In this manner, only a limited number of tests may be performed to derive an
averaged displacement function to be used in Equation (16).

5.1. Identification of a three-crack beam: case (b)


Consider a three-crack CC beam of square cross section with L¼ 100 cm; h¼ 5 cm; E ¼
210 GPa: Be the damage parameters {x1 ¼ L=20; Z1 ¼ 0:8}, {x2 ¼ 3L=20; Z2 ¼ 0:8}, and
{x3 ¼ L=2; Z3 ¼ 0:1}. Compute Equation (16) for NC ¼ 5 equally-spaced loads P ¼ 2 kN
applied individually on the beam. For {M ¼ 300; r ¼ 0:1} in Equation (20), 10 tests are carried
out for each loading condition. Figure 9(a) shows sharp amplitude-decaying local maxima at x1
and x2 in Equation (16), while local maxima at x3 ¼ L=2 are not visible. However, it is also seen
that Equation (16) features a quite flat region, despite the rather high level of noise, and the
limited number of tests performed. To determine whether additional damaged sections are
hidden in Equation (16), the minimization approach posed by Equation (21) is used next.
The damage locations x1 and x2 are first estimated by solving the minimization problem posed
by Equation (18). Figure 9(b) shows the functional Pðx* i Þ; when NA ¼ 5 equally-spaced scales
are used within the range [1, 2]. The estimates obtained are x* 1 ¼ 5:0 cm and x* 2 ¼ 15:02 cm; with
a maximum percentage error equal to 0.1%.

Copyright # 2005 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2006; 13:472–487
486 P. D. SPANOS ET AL.

Based on x* 1 and x* 2 ; the generalized functional in Equation (21) is constructed. Specifically


ND ¼ 4 is assumed; that is, two additional couples of parameters fx* 3 ; a* 3 g and fx* 4 ; a* 4 g are
sought. Further, NAO ¼ 5 equally-spaced scales are selected within the range [20–30] and
NB ¼ 10 locations along the beam length. As in the previous case, b1 ¼ L=20; b2 ¼ 3L=20 and
bm ¼ m  L=10 for m ¼ 2; . . . ; 9; are taken. The estimates obtained for the damage amplitudes
corresponding to x* 1 and x* 2 are Z* 1 ¼ 0:8 and Z* 2 ¼ 0:8; respectively. The estimates for the
additional parameters are fx* 3 ¼ 49:4 cm; Z* 3 ¼ 0:1g; fx* 4 ¼ 70:0 cm; Z* 4 ¼ 0:0g: Thus, the exact
number of damaged sections is captured. The damage amplitudes are correctly evaluated, while
the damage location x3 is estimated with a percentage error as low as 1.2%.

6. CONCLUDING REMARKS

A wavelet-based method has been presented for damage identification in Euler–Bernoulli


beams, subject to static loads.
As an initial step, a simple approach has been proposed to eliminate boundary effects, and
to construct a WT modulus map where only damage-related amplitude-decaying local
maxima are retained. In this manner, the exact number of damaged sections can be determined.
Next, two separate optimization problems have been formulated. The first leads to an
estimate of the damaged locations. Based on the latter ones, the solution of the second
provides an estimate of the damage amplitudes. In this manner, non-negligible advantages as
compared to other methods in the literature are gained, by circumventing time-consuming
single optimization problems requiring an a priori knowledge of the number of damaged
sections.
The effectiveness of the proposed method has been investigated using digitally-simulated
noisy displacement responses, constructed via cubic interpolation on a finite grid. It has
been found that both poor sampling rate and noise may introduce additional damage-like
local maxima in the WT modulus map. These local maxima, however, may be smoothened out
by an appropriate number of measurement points, and repeated tests for each
loading condition. In this context, an extensive number of tests have been carried out on CC
beams featuring up to three cracks, with various geometry and damage characteristics. It has
been found that, as long as the ratio between maximum and minimum damage amplitude is
Zmax =Zmin 57:0; all damage-related local maxima are clearly visible in the WT modulus map, and
the identification method may be implemented via the two proposed separate optimization
problems. If Zmax/Zmin57.0, however, local maxima related to the smallest damage amplitudes
may not be visible in the WT modulus map. In this case, a generalized minimization problem
can be implemented, where additional damage parameters must be included. In both cases,
excellent estimates have been obtained for the damage parameters, even for crack ratios as small
as Z ¼ 0:1:
Obviously, additional work is warranted to test the proposed method on recorded
experimental data. Further, the reliability and possible modifications of the proposed wavelet
analysis when two or more cracks are close to each other is worth investigating. Finally, possible
extensions can be pursued to render the identification method applicable for cases involving
dynamic loading, and multi-dimensional continua.

Copyright # 2005 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2006; 13:472–487
DAMAGE DETECTION IN EULER–BERNOULLI BEAMS 487

ACKNOWLEDGEMENTS

The financial support of this effort from NSF, U.S.A., and from M.I.U.R. (P.R.I.N. 2003), Italy, is
gratefully acknowledged.

REFERENCES
1. Goupillaud P, Grossmann A, Morlet J. Cycle-octave and related transforms in seismic signal analysis.
Geoexploration 1984; 23:85–102.
2. Grossmann A, Morlet J. Decomposition of Hardy functions into square integrable wavelets of constant shape.
SIAM Journal on Mathematical Analysis 1984; 15:723–736.
3. Spanos PD, Failla G. Wavelets: theoretical concepts and vibrations related applications. Shock and Vibration Digest
2005, 37(5):359–375.
4. Wang WJ, McFadden PD. Application of orthogonal wavelets to early gear damage detection. Mechanical Systems
and Signal Processing 1995; 9:497–507.
5. Staszewski WJ, Tomlinson GR. Application of the wavelet transform to fault detection in a spur gear. Mechanical
Systems and Signal Processing 1994; 8:289–307.
6. Zhang LX, Li Z, Su XY. Crack detection in beams by wavelet analysis. Proceedings of SPIE 2001; 4537:229–232.
7. Tian J, Li Z, Su X. Crack detection in beams by wavelet analysis of transient flexural waves. Journal of Sound and
Vibration 2003; 261:715–727.
8. Yam LH, Yan YJ, Jiang JS. Vibration-based damage detection for composite structures using wavelet transform
and neural network identification. Composite Structures 2003; 60:403–412.
9. Sung D-U, Kim C-G, Hong C-S. Monitoring of impact damages in composite laminates using wavelet transform.
Composites B 2002; 33:35–43.
10. Paget CA, Grondel S, Levin K, Delebarre C. Damage assessment in composites by Lamb waves and wavelet
coefficients. Smart Materials and Structures 2003; 12:393–402.
11. Liew KM, Wang Q. Application of wavelet theory for crack identification in structures. Journal of Engineering
Mechanics 1998; 124:152–157.
12. Okafor AC, Dutta A. Structural damage detection in beams by wavelet transform. Smart Materials and Structures
2000; 9:906–917.
13. Hong JC, Kim YY, Lee HC, Lee YW. Damage detection using the Lipschitz exponent estimated by the wavelet
transform: applications to vibration modes of a beam. International Journal of Solids and Structures 2002; 39:
1803–1816.
14. Mallat S, Hwang WL. Singularity detection and processing with wavelets. IEEE Transactions on Information Theory
1992; 38:617–643.
15. Douka E, Loutridis S, Trochidis A. Crack identification in beams using wavelet analysis. International Journal of
Solids and Structures 2003; 40:3557–3569.
16. Haase M, Widjajakusuma J. Damage identification based on ridges and maxima lines of the wavelet transform.
International Journal of Engineering Science 2003; 41:1423–1443.
17. Chang C-C, Chen L-W. Detection of the location and size of cracks in the multiple cracked beam by spatial wavelet
based approach. Mechanical Systems and Signal Processing 2005; 19:139–155.
18. Gentile A, Messina A. On the continuous wavelet transforms applied to discrete vibrational data for detecting open
cracks in damaged beams. International Journal of Solids and Structures 2003; 40:295–315.
19. Wang Q, Deng X. Damage detection with spatial wavelets. International Journal of Solids and Structures 1999;
36:3443–3468.
20. Quek ST, Wang Q, Zhang L, Ang KK. Sensitivity analysis of crack detection in beams by wavelet technique.
International Journal of Mechanical Sciences 2001; 43:2899–2910.
21. Buda G, Caddemi S. Identification of concentrated damages in Euler–Bernoulli beams under static loads. Journal of
Engineering Mechanics 2005, submitted.
22. Di Paola M, Bilello C. An integral equation for damage identification of Euler–Bernoulli beams under static loads.
Journal of Engineering Mechanics 2004; 130:225–234.
23. Vestroni F, Capecchi D. Damage detection in beam structures based on frequency measurements. Journal of
Engineering Mechanics 2000; 126:761–768.
24. Pai PF, Young LG. Damage detection of beams using operational deflection shapes. International Journal of Solids
and Structures 2001; 38:3161–3192.

Copyright # 2005 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2006; 13:472–487

You might also like