Mechanical Systems and Signal Processing 25 (2011) 630643

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing

journal homepage: www.elsevier.com/locate/jnlabr/ymssp

Sensitivity of fundamental mode shape and static deection for

damage identication in cantilever beams
Maosen Cao a,c, Lin Ye a,b,n, Limin Zhou a, Zhongqing Su a, Runbo Bai c
a
Department of Mechanical Engineering, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong, Peoples Republic of China
b
Laboratory of Smart Materials and Structures (LSMS), Centre for Advanced Materials Technology (CAMT), School of Aerospace, Mechanical and Mechatronic
Engineering, The University of Sydney, NSW2006, Australia
c
Department of Engineering Mechanics, College of Mechanics and Materials, Hohai University, Nanjing 210098, Peoples Republic of China

a r t i c l e i n f o abstract

Article history: Fundamental mode shape and static deection are typical features frequently used for
Received 12 November 2009 identication of damage in beams. Regarding these features, an interesting question,
Received in revised form still pending, is which one is most sensitive for use in damage identication. The present
17 June 2010
study addresses the key sensitivity of these features for damage identication in
Accepted 28 June 2010
Available online 3 July 2010
cantilever beams, wherein these features are extremely similar in congurations. The
intrinsic relation between the fundamental mode shape and static deection is
Keywords: discussed, and in particular, an explicit generic sensitivity rule describing the sensitivity
Fundamental mode shape of these features to damage in cantilever beams is proposed. The efciency of this rule
Static deection
in identifying damage is investigated using EulerBernoulli cantilever beams with a
Crack modeling
crack. The validity of the approach is supported by three-dimensional elastic nite
Damage identication
Cantilever beams element simulation, incorporating the potential scatter in actual measurements. The
results show that the generic sensitivity rule essentially provides a theoretical basis for
optimal use of these features for damage identication in cantilever beams.
& 2010 Elsevier Ltd. All rights reserved.

1. Introduction

Damage identication in beam-like structures has been a basic research topic in structural health monitoring for the
late decades [1]. Various algorithms have been derived for damage diagnosis in a cantilever beam [2,3]. In studies of
damage identication in cantilever beams, fundamental mode shape is the most typical dynamic property [3,49]
employed for damage localization and quantication. The extensive use of fundamental mode shape in damage
identication is, to a large extent, attributed to its good sensitivity, reliability and relative convenience in experimental
acquisition using a standard modal testing method. Although the higher mode shapes may theoretically be more sensitive
to small damage, difculties in acquisition considerably decrease their practicability in damage diagnosis. Parallel to
fundamental mode shape, static deections including deection under tip-concentrated loading and deection under
uniformly distributed loading are alternative simple properties [1014] for damage identication in cantilever beams.
Hereafter, for conciseness in statement, fundamental mode shape, deection under tip-concentrated loading and deection
under uniformly distributed loading are symbolized as FMS, dtcl and dudl, respectively.

n
Corresponding author at: Laboratory of Smart Materials and Structures (LSMS), Centre for Advanced Materials Technology (CAMT), School of
Aerospace, Mechanical and Mechatronic Engineering, The University of Sydney, NSW2006, Australia. Tel.: +61 02 93514798; fax: + 61 02 93517060.
E-mail address: l.ye@usyd.edu.au (L. Ye).

0888-3270/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ymssp.2010.06.011
 M. Cao et al. / Mechanical Systems and Signal Processing 25 (2011) 630643 631

For damage identication in cantilever beams based on FMS, dtcl or dudl, a key question behind the popular algorithms is
which of these characteristic properties is most sensitive for use in damage identication in a cantilever beam. This issue
presents a challenge for the optimal uses of these properties for damage identication.
In this study, sensitivity analyses of FMS, dtcl and dudl for crack identication in cantilever beams are investigated, using
analytical models in conjunction with a three-dimensional nite element method. A generic sensitivity rule characterizing
these features in cantilever beams is developed. The outcomes are benecial for optimal selection of these features for
crack identication in cantilever beams.

2. Formulation

The free vibration of a cantilever beam with given initial displacement and velocity can be found by superposing
contributions from each mode as follows [15]:
X
1
yx,t Yn xAn cos on t Bn sin on t 1
n1

where on and Yn (in Appendix A) are the nth order natural frequency and mode shape, respectively, and coefcients An and
Bn are in the following form:
RL RL
y0 xYn xdx v0 xYn xdx
An 0 R L , Bn 0 R L 2
2 2
0 Yn xdx 0 Yn xdx

with y0 and v0 being initial displacement and velocity, respectively. As the discussion should be independent of the initial
velocity in this study, v0 = 0 with Bn = 0 for all n [15] are adopted in the discussion.
Essentially, 9An9 or 9Bn9 quantitatively signies the contribution of the nth mode shape Yn to displacement y. An index,
RAn, can be established as follows:
9An 9
RAn 3
P
1
9Am 9
m1

which reects the contribution ratio of the nth mode shape Yn to the displacement y.
The specic dtcl and dudl in Fig. 1, Yc and Yd, with the subscripts c and d identifying concentrated and distributed
loadings, respectively, expressed in Appendix A, are taken as initial displacements, respectively. According to Eq. (2),
substitutions of y0 =Yc and y0 = Yd into An individually, one obtains coefcients Acn and Adn , respectively, as follows:
4PL3 Hc1 gn
Acn  4-a
3CEI Hc2 gn

qL4 Hd1 gn
Adn  4-b
2CEI Hd2 gn
p
where E is the elastic modulus, I the second moment of inertia, C the arbitrary constant in Yn, and gn L 4 o2n rA=EI with r, A
and L being the density of material, the cross-sectional area and the length of the beam, respectively. Hc1 gn , Hc2 gn , Hd1 gn
and Hd2 gn are, respectively, detailed as

Hc1 gn sin gn sinh gn g3n cosh gn g3n cos gn 3sin gn 3 cos gn sinh gn ,
Hc2 gn g3n 3 sinh2gn cos2 gn 6 sinh gn cos gn gn cos 2gn 3 cosh gn 2 sin gn cosh gn sin 2gn gn cosh 2gn 4 sin gn sinh gn ,

Hd1 gn fg4n 8cos gn g4n 8cos gn 8cosh gn 8gsin gn sinh gn ,

Hd2 gn g4n 3 sinh2gn cos2 gn 6 sinh gn cos gn gn cos 2gn 3cosh gn 2 sin gn cosh gn sin 2gn gn cosh 2gn 4sin gn sinh gn :

Fig. 1. Static deections of a cantilever beam. (a) Tip-concentrated loading and (b) uniformly distributed loading.
632 M. Cao et al. / Mechanical Systems and Signal Processing 25 (2011) 630643

Table 1
Contributions of rst ve mode shapes to vibration displacement.

Mode order Type of initial displacement

Tip-concentrated loading Uniformly distributed loading

Acn RAcn % Adn RAdn %

First  0.9707 97.08  1.0134 98.42

Second 0.0247 2.472  0.0143 1.389
Third  0.0032 0.315  0.0011 0.104
Fourth 0.0008 0.082  0.0002 0.019
Fifth  0.0003 0.030  0.0001 0.005

Note: Acn andAdn are calculated using the normalized static deections in Appendix A.

Fig. 2. Comparison between static deection and fundamental mode shape. (a) FMS versus dtcl and (b) FMS versus dudl.

Thus the relative contribution ratios, RAcn and RAdn , can be dened as

9Acn 9 9fc gn 9
RAcn 1 5-a
P
1 P
9Am 9c 9fc gm 9
m1 m1

9Adn 9 9fd gn 9
RAdn 5-b
P
1 P
1
9Acm 9 9fd gm 9
m1 m1

where fc Hc1 =Hc2 and fd Hd1 =Hd2 .

RAcn or RAdn is a constant as a result of the constant gn in Hc1 , Hc2 , Hd1 and Hd2 , and these constants are irrelevant to material
properties, geometrical dimensions and magnitude of external loads. Herein, the rst 20 mode shapes are calculated, from
which the RAcn and RAdn for the rst ve mode shapes, with respect to the two loading cases in Fig. 1, are dened according to
Eq. (5), listed in Table 1. Clearly, the relative contribution ratios, RAc1 and RAd , of FMS is nearly two orders of magnitude
1
higher than those of the second mode and the others, being dominant in the dynamic displacement. As a consequence, the
static deections, dtcl and dudl, in Fig. 1 closely resemble the FMS, as depicted in Fig. 2. This implies that for damage
identication in cantilever beams, the FMS acquired by dynamic testing extremely approximates dtcl and dudl, and vice
versa.
 M. Cao et al. / Mechanical Systems and Signal Processing 25 (2011) 630643 633

3. Generic sensitivity

Curvature has often been used as a feature for effectively quantifying damage in beam-type structures [16]. The
curvatures derived from dtcl, Yc, dudl, Yd and FMS, Y1 (in Appendix A), are
d2 Yc P
jYc xL 6-a
dx2 EI

d2 Yd q
jYd  Lx2 6-b
dx2 2EI

d2 Y1 2 
jY1 C l1 cos l1 x cosh l1 x 0:7341sin l1 x sinh l1 x 6-c
dx2
In Eq. (6), P, q and C are the parameters that are responsible for a global magnitude of individual curvatures.
Normalization of curvature magnitude is necessary for reecting the sensitivity of the three deformation features to
damage in a quantitatively comparable manner. The following normalized forms, YcN , YdN and Y1N , with beam tip
deformation being unity, Fig. 3(a), are used in the subsequent discussion:
x2 3Lx
YcN 7-a
2L3

x2 6L2 4Lx x2
YdN 7-b
3L4

1 
Y1N cos l1 xcosh l1 x0:7341sin l1 xsinh l1 x 7-c
2
The curvatures resulting from Eq. (7) for dtcl, dudl and FMS are expressed as follows and as shown in Fig. 3(b)
3Lx
jc 8-a
L3

4Lx2
jd 8-b
L4

1  
j1 l21 cos l1 x cosh l1 x 0:7341sin l1 x sinh l1 x 8-c
2
where the asterisk marks the derivation based on the normalized deformation.
The points of intersection marked by K1, K2 and K3 in the curvature curves in Fig. 3(b) are obtained through solving
different combination of two simultaneous equations in Eq. (8):
xK1 0:2058L, xK2 0:2500L, xK3 0:2866L 9

Fig. 3. Normalized FMS, dtcl, and dudl (a) and their curvatures (b).
634 M. Cao et al. / Mechanical Systems and Signal Processing 25 (2011) 630643

Thus, the curvature of the beam is divided into four segments: 0 o x o xK1 , xK1 o x oxK2 , xK2 o x o xK3 and xK3 ox o L. The
sensitivity of the three deformation features with respect to damage is dependent on the segment. For example, for an
arbitrary value of x within xK3 ox o L, we have
jd x o j1 x o jc x 10
Owning to the constitutive equation of the EulerBernoulli beam, M =EIj, the following relation is obtained via
multiplying each term in Eq. (10) by EI
Md x o M1 x o Mc x 11
Assuming that a damage event occurs at location x=xt and it alters the local exural stiffness from EI(xt) to EIxt
1KEIxt with 0 o K o 1, variations in curvature at the site of damage from the initial intact state are formulated as
 
1 1 K
Djd xt Md xt  Md xt 12-a
EIxt EIxt 1KEIxt

 
1 1 K
Dj1 xt M1 xt  M1 xt 12-b
EIxt EIxt 1KEIxt

 
1 1 K
Djc xt Mc xt  Mc xt 12-c
EIxt EIxt 1KEIxt

where L/(1 L)EI(xt) is a constant.

Based on Eqs. (11) and (12), variations in curvature at the site of damage from the initial intact state are obtained as the
following rst inequality:
Djd xt o Dj1 xt o Djc xt , xK3 o xt o L
Djc xt o Dj1 xt o Djd xt , 0 o xt oxK1
13
Djc xt o Djd xt o Dj1 xt , xK1 o xt o xK2
Djd xt o Djc xt o Dj1 xt , xK2 o xt o xK3

Then, the relations of the damage-induced variation in curvature to the other three segments can be elaborated using
the rest of Eq. (13).
In Eq. (13), the magnitude of Dj quantitatively characterizes the sensitivity of the curvature to damage in a cantilever
beam with respect to each deformation feature. Thus, a generic sensitivity rule can be established as follows: for damage
location xt 2 0 xK1 , the degree of sensitivity to the crack for three deformation features is ranked with dudl rst, FMS
second and dtcl third; for xt 2 xK1 xK2 , FMS rst, dudl second and dtcl third; for xt 2 xK2 xK3 , the ranking is in the order FMS,
dtcl and dudl; and for xt 2 xK3 L, dtcl has the greatest sensitivity, followed by FMS and dudl. Moreover, in the narrow interval
xK1 0:2058L, xK3 0:2866L, the sensitivity of the different features becomes less distinguishable.

Fig. 4. Cracked cantilever beam models with a single-edge crack. (a) A cantilever beam with a single-edge crack, (b) a two-segment model with a
massless rotational spring, (c) tip-concentrated loading and (d) uniformly distributed loading.
 M. Cao et al. / Mechanical Systems and Signal Processing 25 (2011) 630643 635

4. A cracked beam model

A single-edge Mode I crack in a cantilever beam, as depicted in Fig. 4, is commonly evaluated as a massless rotational
spring [1719] in analytical analysis of the beam. The equivalent stiffness of the rotational spring, KT, obtained from the crack
strain energy function and Castiglianos theorem, can be expressed in the following form, as deduced in Appendix B:
EIh
KT Ra 14
6p 0 af 2 a=hda

where b and h are the width and height of the beam, a is the depth of the crack, and f(a/h) is the geometric function [20] given
as follows:
r
a 2h pa 0:923 0:1991sin pa=2h4
f tan
h pa 2h cospa=2h
The presence of the crack is represented by an additional rotation, yn(Lc)= M(Lc)/KT with M(Lc)= EId2YI/(Lc)/dx2 at the
crack location (x =Lc), resulting in a discontinuity in the slope of the beam. The transition of slope is consequently expressed
as [21]
dYI Lc EI d2 YI Lc dYII Lc
15
dx KT dx2 dx
where YI and YII are the amplitudes of the exural deformation of the beam segments I and II, respectively.

4.1. Dynamic response

Based on the EulerBernoulli beam theory, the governing equations of exural vibration of beam segments I and II,
Fig. 4(b), are expressed as
d4 YI 4
ln YI 0; 0 r x rLc 16-a
dx4

d4 YII 4
ln YII 0; Lc r x rL 16-b
dx4
p
where ln 4 on 2 rA=EI
The general solutions to Eq. (16) can be written as
Yin x Bi,1 cos ln x Bi,2 sin ln x Bi,3 cosh ln x Bi,4 sinh ln x i I,II 17

where BI,j and BII,j with j =1,2,3,4 are constants to be determined from the boundary and continuity conditions.
The boundary conditions at the clamped and free ends are specied as
dYI 0 d2 YII L d3 YII L
YI 0 0, 0, 0, 0 18
dx dx2 dx3
Eq. (15) gives the compatibility in slope at the crack location; moreover, the continuity conditions of the displacement,
bending moment and shear force are stated as
d2 YI Lc d2 YII Lc d3 YI Lc d3 YII Lc
YI Lc YII Lc , , 19
dx2 dx2 dx3 dx3
An 8  8 matrix of coefcients for the eight constants (Bi,j, i= I, II and j = 1,2,3,4) can be constructed from the general
solution in conjunction with the boundary and continuity conditions (Eqs. (15), (18) and (19)). The associated set of
simultaneous equations with respect to the eight constants will have nontrivial solutions only if the determinant of the
coefcients becomes zero. Expansion of the determinant leads to the frequency equation, from which the natural
frequencies can be numerically determined. The natural frequencies prompt the solution of the constants in Eq. (17).

4.2. Static models

dtcl in Fig. 4(c) for the cracked cantilever beam is obtained from a particular bending equation EId4Yc/dx4 =0 considering
the boundary conditions
00
Yc;I 0 0, Yc;I
u 0 0, Yc;II L 0, u00 L P
EIYc;II 20

with the prime denoting differentiation, and compatibility conditions including Eq. (15) with the shear effect neglected,
and

Yc;I Lc Yc;II Lc 21
636 M. Cao et al. / Mechanical Systems and Signal Processing 25 (2011) 630643

Fig. 5. Numerical model of a cantilever beam with a single-edge crack at Lc = 0.1L. (a) Finite element mesh, (b) zoomed-in crack.

The resultant solution is in the form:

Yc;I x x3L=6EI; 0 r x r Lc
22
Yc;II x P6EILLc Lc x K T x 2 x3L=6EIKT ; Lc r x r L

where KT is given in Eq. (15).

dudl in Fig. 4(d) for the cracked cantilever beam is obtained from EId4Yd/dx4 = q considering zero shearing force at x =L
and other boundary and compatibility conditions similar to those for the tip-concentrated loading case. The resultant
solution has the form:
qx2 6L2 4Lx x2
Yd;I x  , 0 r x r Lc 23-a
24EI

q12EILLc 2 Lc xKT x2 6L2 4Lx x2 

Yd;II x , Lc rx rL 23-b
24EIKT

4.3. Numerical simulation

A three-dimensional elastic nite element simulation is performed to further validate the generic sensitivity rule. The
beams are modeled using 20-node 3D structural solid elements (SOLID 95) in the commercial software ANSYSs. In
particular, the single-sided crack is modeled via pairs of coincident nodes with coordinates identical to those of adjoining
elements. An illustration of the nite element mesh of a cracked beam is shown in Fig. 5. dtcl with P= 100 N, dudl with
q= 100 kN/m2 and FMS are evaluated, followed by a crack identication procedure implemented on the acquired dynamic
and static deformation responses.

5. Results and discussion

A cracked cantilever steel beam is considered with dimensions: length (L)=0.30 m, width (b)=0.02 m, and depth (h)=0.02 m,
and Youngs modulus (E), Poissons ratio (v), and mass density (r) are taken as 206 GPa, 0.3, and 7850 kg m  3, respectively.
Four crack scenarios of Lc =0.1, 0.25, 0.5 and 0.75L from the clamped end with a/h=0.15 for analytical models and a/h=0.3 for
numerical simulation are considered. The adoption of larger a/h for numerical simulation means to offer a clearer graphic
presentation of the sensitivity of FMS, dtcl and dudl to a crack under considerations of modeling and simulation error.

5.1. Sensitivity

For the scenario with the crack located at 0.5L, Fig. 6(a) presents the resultant three deformation features, with the
slopes in Fig. 6(b) and the curvatures in Fig. 6(c), arising from the rst-order difference and the second-order central
difference, respectively. The effect of the crack on the slope is barely visible, with only a tiny increase in the local slope at
the crack location, which cannot be used as an effective damage signature, especially taking into account the presence of
noise and systematic errors. However, a sharp singular peak in the curvature makes the crack clearly identiable. The
height of bars in Fig. 6(c) denotes the magnitude of the singular peak for curvature. Clearly, the magnitudes of the singular
peak for curvature are ranked with dtcl rst, FMS second and dudl third. Since the magnitude of the singular peak indicates
the degree of sensitivity to the crack, this ranking implies that when the crack is at x= 0.5L, dtcl is most sensitive to the
 M. Cao et al. / Mechanical Systems and Signal Processing 25 (2011) 630643 637

Fig. 6. Curvatures of FMS, dtcl and dudl with a crack at Lc =0.5L. (a) Deformations for Lc = 0.5L, (b) slopes for Lc = 0.5L and (c) curvatures for Lc = 0.5L.

presence of the crack, followed by FMS and dudl. Fig. 7 depicts the curvatures for the scenarios when the crack is located at
0.1, 0.25 and 0.75L, respectively. For the crack located at 0.1L (Fig. 7(a)), the magnitude of the singular peak for dudl is
greater than that of FMS and dtcl, respectively. This implies that dudl is most sensitive to the presence of the crack, followed
by FMS and dtcl. Similarly, the ranking of sensitivity for the crack located at 0.25L can be observed in Fig. 7(b), in the order
of FMS, dudl and dtcl. In addition, the ranking of sensitivity for the crack located at 0.75L is the same as that for the crack
located at 0.5L, because the two crack scenarios occur in the same segment of the beam.
For the numerical simulation, one-dimensional LoG (Laplacian of Gaussian) [22], rather than the second-order central
difference, is employed to acquire the curvature from deformation responses with greater accuracy. This operator is
expressed as

x2 s2 2
=2s2
cx p ex 24
2 2ps5
where s is the standard variance of normal distribution. In essence, the one-dimensional LoG, as shown in Fig. 8, is really a
Gaussian wavelet of two-order vanishing moments. The features of convolution manipulation, smoothing, and exibility in
choosing s to match the signal under investigated entails it a more versatile differential operator than the second-order
central difference. In the present analysis, a particular LoG with s = 1.8 (Fig. 8) is optimally chosen as the operator. The
distortion near the end of beam induced by the convolution is treated by a cubic spline extrapolation, as stated in [3]. For a
crack located at 0.1L from the clamped end, the curvatures from the normalized deformation responses with normalization
manipulation similar to that adopted in Eq. (7), are shown in Fig. 9(a). It can be observed that the magnitudes of the
singular peaks in curvature are ranked in the order of dudl, FMS, and dtcl. This is the same result as that obtained from the
theoretical analysis, as shown in Fig. 7(a). The curvatures for the other scenarios with the crack located at Lc = 0.25, 0.5 and
638 M. Cao et al. / Mechanical Systems and Signal Processing 25 (2011) 630643

Fig. 7. Comparison of curvatures of FMS, dtcl and dudl for a crack at Lc = 0.1 (a), 0.25 (b), 0.75L (c), respectively.

Fig. 8. One-dimensional Laplacian of Gaussian (LoG).

 M. Cao et al. / Mechanical Systems and Signal Processing 25 (2011) 630643 639

Fig. 9. Comparison of curvatures of FMS, dtcl and dudl for a crack at Lc = 0.1 (a), 0.25 (b), 0.5 and 0.75L (c), respectively.

0.75L, respectively, are shown in Fig. 9(b)(d). The rankings of the magnitudes of the singular peaks in curvature for the
crack located at 0.5 and 0.75L, are rstly dtcl, secondly FMS and thirdly dudl, which is consistent with the rankings obtained
from the theoretical analysis (Figs. 6(c) and 7(c), respectively). For the crack located at 0.25L, the discrepancy in magnitude
of the singular peak of curvature between the three deformation features is relatively small and the sensitivity of three
deformation features to the crack is almost identical. Thus the results yielded by both theoretical analyses and numerical
simulation fall into the scope of the generic sensitivity rule, validating the generic sensitivity rule well.

5.2. Effects of scatter in measurement

In the actual measurement of FMS, dtcl, and dudl, there is always some scatter in the data because of random noise and
systematic error. To assess the inuence of potential noise, the actual measurements are mimicked by adding Gaussian
white noise to these deformation responses obtained from the nite element simulation. The noise level is dened by
the signal-to-noise ratio (SNR), SNR =20 log10(As/An), with As and An denoting the root-mean-square (RMS) magnitude of the
simulated deformation response and added noise, respectively. The Monte Carlo method [23] is employed to simulate the
deformation responses taking into account noise. In the analysis, the LoG operator (Eq. (24)) is applied to acquire
curvatures from the mimicked responses. A correlation coefcient between the local portions of curvatures with and
640 M. Cao et al. / Mechanical Systems and Signal Processing 25 (2011) 630643

Fig. 10. Curvatures from dtcl for signifying effective (a) and ineffective (b) identication of the crack located at Lc = 0.1L.

Fig. 11. Curvatures from FMS for signifying effective (a) and ineffective (b) identication of the crack located at Lc = 0.1L.

Fig. 12. Curvatures from dudl for signifying effective (a) and ineffective (b) identication of the crack located at Lc =0.1L.

without noise is adopted as a measure for assessing the impact of the noise. The correlation coefcient adopted is
expressed as
Pn
i 1 Ui UVi V
r q
Pn q
Pn 25
2 2
i 1 Ui U i 1 Vi V

where U and V are the local curvatures with a window size of 0.1L (covering 30 nodal points of the FE model) around the
crack location for the intact and noisy responses, respectively. A value of correlation coefcient below 0.5 is used to
 M. Cao et al. / Mechanical Systems and Signal Processing 25 (2011) 630643 641

Table 2
Effects of random noise on three deformation features for crack identication.

SNR [R0.1L(%), RL(n)] dtcl FMS dudl

Effective (%) Ineffective (%) Effective (%) Ineffective (%) Effective (%) Ineffective (%)

100 [2.385, 0.351] 97.96 2.04 98.28 1.72 98.03 1.97

90 [8.858, 1.302] 93.10 6.90 93.61 6.39 94.23 5.77
85 [13.355, 1.960] 88.25 11.75 88.58 11.42 88.72 11.28
80 [30.867, 4.531] 79.20 20.80 80.79 19.21 80.99 19.01
75 [35.637, 5.234] 64.25 35.75 66.39 33.61 66.3 33.7
70 [101.596,14.910] 41.09 58.91 45.91 54.09 45.74 54.26
S 77.31 22.69 78.93 21.07 79.00 21.00

identify the threshold when the noise masks the presence of the crack. For the crack at 0.1L and the SNR =90, the inuence
of uncertainty on the identication of the crack is illustrated in Figs. 1012, in which two sets of 200 proles of curvature
along the beam axis are presented for each deformation feature. In the gures, similar proles of curvature with respect to
three deformation features exist for either the effective or ineffective crack identication situation, and the discrepancy is
almost indiscernible. To comprehensively investigate the impact of noise level, 10,000 noisy deformation responses for
each deformation feature at one noise level are further constructed using the Monte Carlo method in order to approach a
stable statistic character for the impact of noise. The noise levels of SNR =100, 90, 85, 80, 75 and 70 are considered. In
Table 2 the percentages of effective and ineffective crack identication for each deformation feature at each noise level are
evaluated and listed, where R0.1L and RL denote, at a noise level, the ratio of the mean magnitude of noise to the magnitude
of deformation responses at 0.1L and L of the beam, respectively.
For SNR =80, approximately 80% successful identication of the crack was achieved for each deformation feature, which
may be regarded as an acceptable boundary, below which the efciency of crack identication deteriorates markedly and
has little reliability. At such a level of signal-to-noise ratio, the mean noise magnitude approximately equals 3.1% of the
deformation response magnitude at the 0.1L of the beam, which provides a prior precision criterion for the optimal
selection of deformation features from a perspective of experimental measurement. For conventional linear variable
differential transformer (LVDT)-based measurement methods, dtcl is potentially the most feasible to meet this precision
requirement, due to convenience in measurement, whereas use of FMS is probably more difcult because of the relatively
complex procedure involved in the modal analysis. Advanced optical or electromagnetic spectrum-based measurement
apparatus, such as a laser vibrometer, may provide the opportunity to meet this precision requirement in actual
measurement of all deformation features.
Clearly, all the deformation features are able to identify the crack for the SNRZ80, but their capabilities decrease
signicantly when SNRr75. At the level of SNR =70, these features are almost inadequate to act as signatures for crack
identication. In Table 2, the average effective and ineffective ratios for each deformation feature imply that the impact of
scatter is most signicant for dtcl, followed by FMS and ducl. In other words, in the case of a cantilever beam with a crack,
ducl has greatest immunity to noise, followed by FMS and dtcl. This result coincides with that for the sensitivity of the three
deformation features to the presence of the crack. The nding that higher sensitivity corresponds to greater immunity to
noise also applies to the other crack cases with the crack at 0.25, 0.5, and 0.75L, respectively.

6. Conclusions

The most typical dynamic and static deformation features used in various algorithms for the detection of damage in
cantilever beams are fundamental mode shape, deection under tip-concentrated loading and deection under uniformly
distributed loading. A fundamental question behind these popular algorithms is which is most sensitive for use in damage
identication. With the EulerBernoulli beam theory and nite element simulation, the sensitivity of these features to
damage in cantilever beams is comprehensively investigated in this study, and a generic sensitivity rule characterizing the
sensitivity characteristics of the three deformation features to an edge crack in a cantilever beam is developed, addressing
the potential scatter of actual measurement of these features. This generic sensitivity rule holds promise for the rational
and optimal use of fundamental mode shape, deection under tip-concentrated loading and deection under uniformly
distributed loading for damage detection in cantilever beams.

Acknowledgements

L. Ye, L. Zhou and Z. Su thank the Hong Kong Polytechnic University for a special research grant on Advanced Composite
and Fundamental Structure. L. Zhou also thanks the Research Grant Council of the Hong Kong SAR for the research Grant of
PolyU5333/07E. M.S. Cao is grateful for National Natural Science Foundation of China (Grant no. 50978084).
642 M. Cao et al. / Mechanical Systems and Signal Processing 25 (2011) 630643

Appendix A

Px2 x3L
Yc x A1
6EI

qx2 6L2 4Lx x2

Yd x  A2
24EI

Yn x Ccos ln xcosh ln xwn sin ln xsinh ln x A3

where C is an arbitrary constant and, wn cos ln L cosh ln L=sin ln L sinh ln L with lnL= gn being the solutions of
cosgn coshgn 1 for an EulerBernoulli cantilever beam.
Eq. (A3) gives the fundamental model shape when n = 1:
Y1 x Ccos l1 xcosh l1 xw1 sin l1 xsinh l1 x A4
with w1 = 0.7341 arising from g1 = 1.87510406871.

Appendix B
The stress intensity factor of mode I for a single-edge crack due to bending moment is expressed as
p a
K KI s paf A5
h
2
where s =6M/bh with M being the bending moment at the cross-section of crack, a is the crack depth, and f(a/h) is the
geometric function.
If the crack can be represented by a spring, the additional work due to the bending moment equals to the crack strain
energy:
Z a
M2
DU Js bda A6
2KT 0

where KT is the equivalent stiffness of the rotational spring, and Js KI2 =E is the strain energy density function in the plane-
stress condition [20]. Substituting Eqs. (A5) into Eq. (A6), KT can be formulated as
M2 M2 M2 M2
KT Ra Ra 2 R a s2 paf 2 a=h R  2 2 paf 2 a=h

2 0 Js bda 2 0 KI =Ebda a 6M=bh
2 0 E bda 2 0 E bda
Ebh4 EIh
Ra Ra A7
72p 0 af 2 a=hda 6p 0 af 2 a=hda

References

[1] Y.J. Yan, L. Cheng, Z.Y. Wu, L.H. Yam, Development in vibration-based structural damage detection technique, Mechanical Systems and Signal
Processing 21 (2007) 21982211.
[2] C.R. Farrar, D.A. Jauregui, Comparative study of damage identication algorithms applied to a bridge: II, Smart Materials and Structures 7 (1998)
720731.
[3] M. Rucka, K. Wilde, Application of continuous wavelet transform in vibration based damage detection method for beams and plates, Journal of Sound
and Vibration 297 (35) (2006) 536550.
[4] V. Pakrashi, A. OConnor, B. Basu, A study on the effects of damage models and wavelet bases for damage identication and calibration in beams,
Computer-Aided Civil and Infrastructure Engineering 22 (8) (2008) 555569.
[5] C.P. Ratcliffe, Damage detection using a modied Laplacian operator on mode shape data, Journal of Sound and Vibration 204 (3) (1997) 505517.
[6] J. -C. Hong, Y.Y. Kim, H.C. Lee, Damage detection using the Lipschitz exponent estimated by the wavelet transform: applications to vibration modes of
a beam, International Journal of Solids and Structures 39 (2001) 18031816.
[7] Z.R. Lu, S.S. Law, Dynamic condition assessment of a cracked beam with the composite element model, Mechanical Systems and Signal Processing 23
(2009) 415431.
[8] G.J. Prakash, S. Swarnamani, Damage identication in composite beam using continuous wavelet transform applied to mode shape and strain energy
data, Advances in Vibration Engineering 7 (2) (2008) 127141.
[9] M.S. Cao, P.Z. Qiao, Novel Laplacian scheme and multiresolution modal curvatures for structural damage identication, Mechanical Systems and
Signal Processing 23 (2009) 12231242.
[10] P.D. Spanos, G. Failla, A. Santini, M. Pappatico, Damage detection in EulerBernoulli beams via spatial wavelet analysis, Structural Control and Health
Monitoring 13 (1) (2005) 472487.
[11] M. Rucka, K. Wilde, Crack identication using wavelets on experimental static deection proles, Engineering Structures 28 (2006) 279288.
[12] Salvatore Caddemia, A. Morassi, Crack detection in elastic beams by static measurements, International Journal of Solids and Structures 44 (16)
(2007) 53015315.
[13] K.D. Hjelmstad, S. Shin, Damage detection and assessment of structures from static response, Journal of Engineering Mechanics 123 (6) (2007)
568576.
[14] G. Buda, S. Caddemi, Identication of concentrated damages in EulerBernoulli beams under static loads, Journal of Engineering Mechanics 133 (8)
(2007) 942956.
[15] S. Timoshenko, D.H. Young Jr., W. Weaver, in: Vibration Problems in Engineering, 4th ed., Wiley, New York, 1974.
[16] A.K. Pandy, M. Biswas, M.M. Samman, Damage detection from changes in curvature mode shape, Journal of Sound and Vibration 145 (2) (1991) 321332.
[17] K.V. Singh, Transcendental inverse eigenvalue problems in damage parameter estimation, Mechanical Systems and Signal Processing 23 (2009)
18701883.
 M. Cao et al. / Mechanical Systems and Signal Processing 25 (2011) 630643 643

[18] G. Gounaris, A. Dimarogonas, A nite element of a cracked prismatic beam for structural analysis, Computers & Structures 28 (3) (1988) 309313.
[19] Lee. Jinhee, Identication of multiple cracks in a beam using vibration amplitudes, Journal of Sound and Vibration 326 (2009) 205212.
[20] H. Tada, P. Paris, G. Irwin., in: The Stress Analysis of Cracks Handbook, 2nd Ed., Paris Production Inc., St. Louis, MI, USA, 1985.
[21] W.M. Ostachowicz, M. Krawczuk, Analysis of the effect of cracks on the natural frequencies of a cantilever beam, Journal of Sound and Vibration 150
(2) (1991) 191201.
[22] S. Mallat, in: A Wavelet Tour of Signal Processing, Academic Press, USA, 1998.
[23] Y. Xia, H. Hao, Statistical damage identication of structures with frequency changes, Journal of Sound and Vibration 263 (4) (2003) 853870.