International Journal of Solids and Structures 41 (2004) 6661–6676

www.elsevier.com/locate/ijsolstr

Vibration-based construction and extraction

of structural damage feature index
a,*
Y.J. Yan , H.N. Hao b, L.H. Yam c

a
Department of Engineering Mechanics, Institute of Vibration Engineering, Northwestern Polytechnical University,
127 You Yi Road, Xi’an 710072, China
b
Science School, Xi’an Shi You University, Xi’an 710065, China
c
Department of Mechanical Engineering, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong SAR, China
Received 27 May 2004
Available online 3 July 2004

Abstract
The finite element dynamic model of a honeycomb sandwich plate is presented using different mesh division for the
surface plates and the sandwich plate to accurately express the crack damage status of the plate. The experimental
measurements of plate natural frequency and dynamic responses are carried out for dynamic model verification. The
feasibility of detecting small crack damage according to structural natural frequency and dynamic responses is eval-
uated. The results show that the energy spectrum of the decomposed wavelet signals of dynamic responses has a higher
sensitivity to small crack damage, and more high order modes should be included in the dynamic model for structural
damage detection.

2004 Elsevier Ltd. All rights reserved.

Keywords: Crack detection; Natural frequency; Dynamic responses; Wavelet transform; Sandwich plate

1. Introduction

For a long time, great attention has been paid to researches on structural damage detection using
structural vibration characteristics. Even for a complex engineering structure in practice, its natural
frequencies and dynamic responses at few measured spots can be easily acquired, this fact gives a
potential feasibility for the realization of online damage detection and health monitoring of various
in-service structures (Farrar et al., 2001). The structural natural frequencies were the earliest used
parameters for structural damage detection. Collins et al. (1992) computed the frequency spectra and
studied the effects of crack location on longitudinal vibrations of a cantilevered bar with a transverse
crack. Nandwana and Maiti (1997) added a rotational spring to a slender beam for crack simulation in

*
Corresponding author. Tel.: +86-29-88492895; fax: +86-29-88492216.
E-mail address: [email protected] (Y.J. Yan).

0020-7683/$ - see front matter
2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijsolstr.2004.05.069
6662 Y.J. Yan et al. / International Journal of Solids and Structures 41 (2004) 6661–6676

order to check the feasibility of detecting crack location based on the measurement of natural frequencies,
and pointed out that when the internal crack depth is more than 20% of the section depth, the detec-
tion performance was satisfactory. Ramamurti and Neogy (1998) considered the feasibility of using
natural frequency as a criterion for damage detection, and concluded that natural frequency does not
appear to be an appropriate criterion for integrity analysis in a simplified model. Salawu (1997) reviewed
plentiful literatures about structural damage detection using the natural frequency and discussed
the possible limit factors for successful application of vibration monitoring to damage detection and
structural assessment. Many researches on structural damage detection using the online measured
structural vibration responses have also been carried out (Zou et al., 2000). Hou et al. (2000) used
the characteristics of the wavelet transformation of simulated vibration response signals generated from
a simple structural model subjected to a harmonic excitation. They showed a great promise of the
wavelet approach for damage detection and structural health monitoring. Zhang et al. (1999) adopted
vibration measurements to detect structural damage using Transmittance Function Monitoring, and the
parameters used for damage detection were computed from different types of measured structural
responses.
Although the two above-mentioned methods are simple and easy to execute, there are still many
problems in the realization of damage detection for practical engineering structures. Because a small
quantity of structural information is adopted, it is difficult to detect a practical complex structural damage
status, such as determination of damage location, damage category and extent as well as some small
structural damage. Therefore, some researchers adopted more structural information in structural damage
detection, for example, mode shapes (Kosmatka and Ricles, 1999; Hu et al., 2001), modal strain energy
(Doebling et al., 1997; Chen et al., 2000), etc. However, these methods require a large amount of measured
data or numerical simulation using an accurate structural dynamic model. This is not advantageous for
online damage detection of an in-service structure.
Identifying damage status from global dynamic behaviors of a structure is essentially an inverse problem
of structural dynamic analysis. The issues to be solved include identification of structural damage category,
extent and location. Because practical engineering problems are generally very complex, structural damage
detection using vibration method is always conducted in two steps, one is to determine if damage has
occurred, the other is to identify the category, extent and location of structural damage. Many researchers
have adopted some simple structural model, such as a beam, to study damage detection, but for complex
engineering structures some conclusions may not be the same.
Honeycomb sandwich composite plates have been widely applied to aeronautical structures as well as
building, automobile and train structures because it possesses many advantages, such as lighter weight,
higher stiffness, heat insulation and preservation, and anti-radialization (the structure can resist the radi-
alization from electromagnetic wave or infrared ray when structural material is mixed by some material
that can absorb electromagnetic wave or infrared ray). This kind of structural materials are made of very
thin aluminum alloy, FRP (fiberglass-reinforced plastics), PVC and CFRBP (ceramic fiber round braided
rope), etc. One of its most excellent properties is the lightweight, and its weight is only 10–15% of that of a
solid structure with the same material. However, the ability of resisting impact of a honeycomb sandwich
plate is very poor, so crack or unglued damage occurs frequently. This will seriously affect the function of
the structural components, such as the propeller of a helicopter, aerofoil and sealed-cabin. Obviously, the
study of in-service damage detection for honeycomb sandwich structures possesses significant application
values.
This paper aims at evaluating feasibility of structural crack damage detection using vibration parame-
ters. A more accurate finite element dynamic model of a honeycomb sandwich plate, which is closer to some
practical engineering structures, is established. Its natural vibration traits and response characteristics are
checked by experimental measurements. Various possible influences of crack damage status on structural
natural frequency and responses are discussed.
 Y.J. Yan et al. / International Journal of Solids and Structures 41 (2004) 6661–6676 6663

2. Dynamic finite element model of a sandwich plate with crack damage

When a crack exists in a honeycomb sandwich plate (as shown in Fig. 1), it can be described using five
parameters: depth d, length l, directional angle a and location coordinates xc and yc . Assume that only very
narrow crack is considered, the crack width can be approximately taken as zero. A crack damage status can
be expressed as
g ¼ gðxc ; yc ; l; d; aÞ ð1Þ

Crack damage in a sandwich plate will lead to a structural stiffness reduction of the local area as well as
the whole structure. Therefore, structures with different location and extent of crack will exhibit different
dynamic features. In the finite element model established for this study, the crack parameters including
location, length and directional angle are expressed using the nodal coordinates of two adjacent eight nodes
quadrangular elements I and II as shown in Fig. 2(a). According to different size and shape of the two
elements, various cracks can be simulated.

Fig. 1. Crack damage in a honeycomb sandwich plate.

Fig. 2. Element division of a honeycomb sandwich plate with crack damage.

6664 Y.J. Yan et al. / International Journal of Solids and Structures 41 (2004) 6661–6676

Assume that the crack depth reaches the inside of sandwich ply of the plate. In the present study, a kind
of sandwich ply made of stiffened plate is studied. In order to deal with the crack depth, the top and bottom
plates, and the stiffened plate are independently divided into quadrangular elements as shown in Fig. 2(b),
and the different vertical dimension of the two adjacent quadrangular elements III and IV can represent the
variation of the crack depth.
Assume that the top and bottom surfaces, and all the stiffened plates are isotropic thin plates and
subjected to a small elastic bending distortion. Three types of mid-plane coordinates: ot xt yt zt ; ob xb yb zb and
oi xi yi zi , are used respectively for these three parts. Displacements in x-, y- and z-directions can be expressed
respectively as
9
ut ¼ zt htx ; vt ¼ zt hty ; wt ¼ wt ðxt ; yt Þ; =
ub ¼ zb hbx ; vb ¼ zb hby ; wb ¼ wb ðxb ; yb Þ; ð2Þ
;
ui ¼ yi hix ; vi ¼ vi ðxi ; zi Þ; wi ¼ yi hiz :
At the common nodes of the top and stiffened plates or the bottom and stiffened plates, the continuous
conditions of the displacement give wt ¼ wi , ut ¼ ui , and vt ¼ vi or wb ¼ wi , ub ¼ ui and vb ¼ vi . Strains in
these three parts can be expressed as
  T
ohtx ohty ohtx ohty
fegt ¼  zt ;  zt ;  zt þ ; ð3Þ
oxt oxt oxt oxt
  T
ohbx ohby ohbx ohby
fegb ¼  zb ;  zb ;  zb þ ; ð4Þ
oxb oxb oxb oxb
  
ohiz ohix ohiz ohix
fegi ¼  yi ;  yi ;  yi þ : ð5Þ
ozi oxi ozi oxi
An arbitrary quadrangular element in a Cartesian coordinate system can be transformed into a rect-
angular element using a set of the following transform
   
x n
¼f : ð6Þ
y g

Assume that the nodal coordinate of an arbitrary quadrangular element is ðxj ; yj Þ; j ¼ 1; 2; . . . ; 8, and the
element shape function is Nj ðn; gÞ; j ¼ 1; 2; . . . ; 8, then Eq. (6) can be expressed as
X
8 X
8
x¼ Nj ðn; gÞxj ; y¼ Nj ðn; gÞyj : ð7Þ
j¼1 j¼1

The displacements in Eq. (2) can be written as

X
8 X
8 X
8
wt ¼ Nj wtj ; htx ¼ Nj htxj ; hty ¼ Nj htyj ; ð8Þ
j¼1 j¼1 j¼1

X
8 X
8 X
8
wb ¼ Nj wbj ; hbx ¼ Nj hbxj ; hby ¼ Nj hbyj ; ð9Þ
j¼1 j¼1 j¼1

X
8 X
8 X
8
hiz ¼ Nj hizj ; hix ¼ Nj hixj ; vi ¼ Nj vij ; ð10Þ
j¼1 j¼1 j¼1
 Y.J. Yan et al. / International Journal of Solids and Structures 41 (2004) 6661–6676 6665

where the element shape function Nj ðn; gÞ is as follows

Nj ðn; gÞ ¼ ð1 þ nnj Þð1 þ ggj Þðnnj þ ggj  1Þn2j g2j =4 þ ð1  n2 Þð1 þ ggj Þð1  n2j Þg2j =2
þ ð1  g2 Þð1 þ nnj Þð1  g2j Þn2j =2; j ¼ 1; 2; . . . ; 8: ð11Þ
The displacement vector at the jth node for bending deformation of a thin plate can be written as
T T T
fdjt g ¼ ½wtj ; htxj ; htyj ; fdjb g ¼ ½wbj ; hbxj ; hbyj ; fdji g ¼ ½hizj ; hixj vij : ð12Þ
For the top plate, using the regular procedures of FEM (Huebner, 2001), one can obtain the equation of
motion of an element as follows
½M e €e g þ ½K e e
¼ fF e g24 1 ;
24 24 fd 24 1 24 24 fd g24 1 ð13Þ
where ½M e , ½K e and fF e g are element mass matrixes, stiffness matrix and nodal force vector, respectively.
fd€e g and fde g are nodal acceleration and displacement vectors, respectively. Hereinto,
Z 1 Z 1 Z 1 Z 1
T
e
½M 24 24 ¼ G PGjJj dn dg; e
½K 24 24 ¼ BT DBjJj dn dg; ð14Þ
1 1 1 1

where the Jacobi determinant jJj, elastic matrix D and mass density matrix P are as follows:
2 3 2 3
ox
; oy 3 1 l 0 h 0 0
Eh 4 l 1 0 5; 4 0 1 h3
jJj ¼ onox on
oy ; D¼ P ¼ q 12
0 5; ð15Þ
og og
12ð1  l2 Þ 0 0 1l 0 0 1 3
h
2 12

where E, l and q are the elastic modulus, Poisson’s ratio and mass density of the structural material,
respectively, and h is the plate thickness. In Eq. (14), the strain matrix B and velocity matrix G are as
follows:
)
B ¼ ½ðB1 Þ3 3 ; ðB2 Þ3 3 ; . . . ; ðB8 Þ3 3 ;
ð16Þ
G ¼ ½ðG1 Þ3 3 ; ðG2 Þ3 3 ; . . . ; ðG8 Þ3 3 ;
where
2 oN 3 2 3
 oxtj 0 0 Nj 0 0
6 7
Gj ¼ 4 0 0 5;
oN
Bj ¼ 4 0  oytj 0 5; Nj j ¼ 1; 2; . . . ; 8: ð17Þ
oN
 oxtj
oN
 oytj 0 0 0 Nj

Similar procedures can be used to establish the element equations of motion of the bottom and stiffened
plates. Assembling all element equations of motion of the three kinds of plates, one can obtain the fol-
lowing equation of motion of the whole honeycomb sandwich plate without considering damping:
€ þ KD ¼ F;
MD ð18Þ
where M, K and F are the global mass matrix, stiffness matrix of the structure and the external force vector,
€ and D are the global nodal acceleration and displacement vectors, respectively. Assume that
respectively. D
the structure has proportional damping as follows:
C ¼ aM þ bK; ð19Þ
where C is the global damping matrix, and a and b are the proportional damping coefficients. To execute
modal transform to Eq. (19) using the normalized modal matrix U, one can get

UT CU ¼ UT ðaM þ bKÞU ) 2fi xi ¼ a þ bx2i ði ¼ 1; 2; . . . ; nÞ;

6666 Y.J. Yan et al. / International Journal of Solids and Structures 41 (2004) 6661–6676

where xi and fi are the ith natural frequency and modal damping ratio, respectively. For arbitrary i 6¼ j, the
a and b can be solved using following equation:
) 
a þ bx2i ¼ 21i xi ; a ¼ 21i xi  2x2i ð1j xj  1i xi Þ=ðx2j  x2i Þ;
)
a þ bx2j ¼ 21j xj ; b ¼ 2ð1j xj  1i xi Þ=ðx2j  x2i Þ:

Though modal damping ratio fi and fj may have some difference for different modes, they mainly depend
on structural material property. Generally, steel fi  0:005, concrete fi  0:08, felt or cork fi  0:06, nat-
ural rubber fi  0:05, etc. Supposing that 1i  1j ¼ 1, one can get the approximate formula as follows
)
a  21xi ½1  xi =ðxi þ xj Þ ;
ð20Þ
b  21=ðxj þ xi Þ:
Combining Eqs. (18) and (19), the structural equations of motion can be expressed as
€ þ CD_ þ KD ¼ F:
MD ð21Þ

3. Vibration-based structural damage feature index

When some crack occurs in a honeycomb sandwich plate, the plate natural frequency, modal shapes,
frequency response functions and dynamic response properties, etc. will vary with the location and extent of
the cracks because of the variations of local structural stiffness. In order to emphatically evaluate the
influence of crack location and extent on the above-mentioned dynamic characteristics, the situation with
only one crack is considered in this paper.
For a honey comb sandwich plate with crack damage status g ¼ gða; c; l; d; aÞ, if the structural damping
is neglected and the external load equals zero, Eq. (21) can be written as
€ þ KðgÞD ¼ 0:
MD ð22Þ

Solving Eq. (22), one can get its eigenvalue xi ðgÞ and mode vector Ui ðgÞ, which vary with crack damage
status g, and i ¼ 1; 2; . . . ; n, where n is the number of structural mode considered in this study.
Vibration responses at few spots of an in-service structure can be easily measured using the technology of
piezoelectric smart structures (Gobin et al., 2000). However, the raw response signal in time domain cannot
be used directly to identify structural damage quantitatively. Some representative indexes have to be se-
lected and constructed. Wavelet transform of response signals is one of the available methods. Wavelet
analysis of time-varying signal is a kind of localization analysis method in time and frequency domains, and
the time and frequency windows can both be changed. This signal processing method has higher frequency
and time resolution (Chui, 1997).
A continuous wavelet transform of a function f ðtÞ 2 L2 ðRÞ is defined as
Z  
1  tb
Wf ða; bÞ ¼ pffiffiffiffiffiffi f ðtÞW dt; ð23Þ
jaj R a
where b is the translation parameter, a is the scale parameter, f ðtÞ is the function (signal) to be transformed,
W ðtÞ is the transforming function (mother wavelet), Wf is the calculated wavelet coefficients, which can be
used to recompose the original function f ðtÞ. The re-composition equation can be expressed as
Z þ1 Z þ1  
1 1 tb
f ðtÞ ¼ W f ða; bÞW da db; ð24Þ
CW 1 1 a2 a
 Y.J. Yan et al. / International Journal of Solids and Structures 41 (2004) 6661–6676 6667
R þ1 j b
W ðxÞj2
where CW ¼ 0 b
dx < 1, and WðxÞ is the Fourier transform of Mother Wavelet WðtÞ. The various
jxj
forms of mother wavelet WðtÞ have now been developed.
In practical application for wavelet transforms, especially in order to realize numerical simulation in
computer, the continuous wavelet must be changed into discrete form. One of the most usually used
discrete wavelet is Dyadic Wavelet (binary wavelet), i.e.
Wj;k ðtÞ ¼ 2j=2 Wð2j t  kÞ; j; k 2 Z: ð25Þ
Therefore, the discrete wavelet transforms and re-composing (invert wavelet transform) of a function f ðtÞ
can be written as
Z
W2j f ðkÞ ¼ 2j f ðtÞW ð2j t  kÞ dt; ð26Þ
R

XZ
f ðtÞ ¼ W2j f ðkÞW2j ð2j t  kÞ dk; ð27Þ
j2Z

where W2j f ðkÞ denotes one variable, i.e. the wavelet transform of f ðtÞ.
The wavelet package analysis (WPA) is the most useful method of wavelet transform. It can adaptively
choose the corresponding frequency bandwidth according to the characteristics of the signal to be analyzed,
so that the resolution in frequency and time domains can both be enhanced.
The WPA algorithm is as follows.
Let gjn ðtÞ 2 Ujn , then gjn ðtÞ can be expressed as
X j;n
gjn ðtÞ ¼ dl un ð2j t  lÞ; ð28Þ
l

where gjn ðtÞis an arbitrary function in sub-space Ujn . un ð2j t  lÞ is the orthogonal wavelet packet, and dlj;n is
the decomposed wavelet packet coefficient. The WPA decomposed coefficient is calculated by equation as
follows
P 
dlj;2n ¼ k ak2l dkjþ1;n ;
P ð29Þ
dlj;2nþ1 ¼ k bk2l dkjþ1;n ;
where ak2l is the low-pass digital filter and bk2l is the high-pass digital filter.
The WPA re-composing is to calculate fdljþ1;n g using fdlj;2n g and fdlj;2nþ1 g, and equation as follows
X
dljþ1;n ¼ ½hl2k dkj;2n þ gl2k dkj;2nþ1 ; ð30Þ
k

where hl2k is the low-pass digital filter and hl2k is the high-pass digital filter.
The theory of wavelet analysis is very profuse, and the interested researcher can refer to references
(Mallat, 1999; Strang and Nguyen, 1996).
Energy of dynamic response of cracked structures compared with that of the intact structure in some
special frequency bands will exhibit some remarkable difference. This is because the structural damage will
suppress or enhance some components of response signal in special frequency bands, i.e., the structural
damage can cause energy increase of some response signal components or energy decrease of other response
signal components. Therefore, the energy of structural vibration response signals with different frequency
components contains ample information of structural damage, and the energy variation of one or several
frequency components of the signals can indicate a special status of structural damage.
In order to extract structural damage information from structural response signal, the signal is first
decomposed into multiple sub-signals in various frequency bands using WPA. Let S00 ðtÞ denote the original
signal of structural response, it can be expressed as
6668 Y.J. Yan et al. / International Journal of Solids and Structures 41 (2004) 6661–6676

2k1
X
S00 ðtÞ ¼ Skj ðtÞ; ð31Þ
j¼1

where Skj ðtÞ is the sub-signal with orthogonal frequency band and k indicates the layer number of the tree
structure of wavelet decomposition.
The energy of these sub-signals can be expressed as
Z
2
Ukj ¼ jSkj ðtÞj dt: ð32Þ

A non-dimensional index vector can be composed using Ukj0 and Ukj ðj ¼ 0; 1; 2; . . . ; 2k1 Þ, i.e.
Ukj
Vd ¼ fK1 ; K2 ; . . . ; K2k1 g; Kj ¼ 1  ; j ¼ 1; 2; . . . ; 2k1 ; ð33Þ
Ukj0
where Ukj0 and Ukj are the sub-signal energy of the intact and crack damaged plates, respectively. Kj
indicates the magnitude variation of the jth order sub-signal energy, it is a measurement of the enhance-
ment or attenuation of the jth order sub-signal energy.
We also find that when different mother wavelet is adopted for decomposition of vibration response of
structure with damage, the obtained ‘‘the energy index’’ is quite different. In this study, we have attempted
to use several mother wavelets, and found the Daubechies wavelet (db5) has the better effect for indication
of structural damage. In numerical simulation, the wavelet analysis toolbox in the MATLAB software is
used, so that programme design can be greatly simplified.

4. Numerical simulation and experiment

The structural damage status, such as damaged locations, extents and categories, of a practical engi-
neering structure is related to a large amount of information and data. It is not reasonable to acquire such
information only using experiment or numerical simulation. Superfluous experimental work is time
consuming and not economical, and numerical simulation without experimental verification is unbeliev-
able. A more scientific way is to use a more accurate structural dynamic model checked by experiment
for numerical simulation to acquire a large amount of information and data related to structural damage
status.
The specimen of the numerical model is a honeycomb sandwich cracked plate as shown in Fig. 3, and it
has length L, width B and thickness h of 295, 98 and 8 mm, respectively. The plate is composed of PVC
materials, and its top surface has a thin aluminium raincoat. The plate weight is only 12.67% of a solid
structure with the same dimension and material. The PVC material parameters are E ¼ 3:5 GPa, l ¼ 0:34
and q ¼ 1:36 kg/m3 .

Fringe transverse crack Inner transverse crack

Fringe lognitudinal
h

L
B

Fig. 3. Specimen of a honeycomb sandwich plate with crack damage.

 Y.J. Yan et al. / International Journal of Solids and Structures 41 (2004) 6661–6676 6669

Wn FREQ RESP H1 X1, Y1 MAG MAIN Y: 12.3

Y: 25.ok 80dB X: 0Hz
X: 0Hz + 400 Hz Lin
#AVM: 1
11

10

2
0 50 100 150 200 250 300 350 Hz
Set up W
Measurement: MULTICH. SPECTRUM AVERAGING 800 lines
Trigger: SIG X1 Slope: + Level: +0.02*max input
Delay: Lin 5
FREQ. SPAN: 400Hz ∆ f: 500mHz T: 2s ∆ t : 977µs

WEIGHT. Y1: RECTANGULAR

CONDIT. X1: Input Signal No 1 Slot 1;1 (A;1)
100mV DIRECT DC GROUND 10.0mV/unit

Fig. 4. Experimental frequency–response curve of the specimen.

The natural frequencies of an undamaged specimen of honeycomb sandwich plate are experimentally
measured to verify the reliability of the theoretical formula and programs. An experimental frequency
response curve is given in Fig. 4. The lowest 10 natural frequencies acquired by experiment and simulation
and the percentage errors between the results obtained using these two methods are listed in Table 1, which
shows that the errors are below 5%. This is an acceptable numerical precision in engineering problems. The
lowest 10 elastic mode shapes acquired by simulation are shown in Fig. 5.
Based on the dynamic model verified by the experiment, the natural frequencies of the honeycomb
sandwich plate for various crack damage status are numerically computed. When the crack width is ex-
tremely small and negligible, crack length will be the most important factor to affect structural dynamic

Table 1
Natural frequencies of the intact honeycomb sandwich plate obtained by experiment and numerical simulation
Order Numerical Experiment Errors
1 29.711 Hz 28.5 Hz 4.2%
2 50.900 Hz 52.5 Hz 3.0%
3 80.826 Hz 82 Hz 1.4%
4 106.99 Hz 108 Hz 1.0%
5 139.70 Hz 134 Hz 4.2%
6 162.70 Hz 158.5 Hz 2.6%
7 192.16 Hz 184 Hz 4.4%
8 212.43 Hz 208 Hz 2.1%
9 250.42 Hz 261.5 Hz 4.2%
10 277.07 Hz 287.5 Hz 3.6%
6670 Y.J. Yan et al. / International Journal of Solids and Structures 41 (2004) 6661–6676

Undamaged Plate Mode1 29.7109Hz Undamaged Plate Mode2 50.9003Hz

1 1

0.5 0.5

0 0
z

z
-0.5 -0.5

-1 -1
1 1
1 1
0.5 0.8 0.8
0.6 0.5 0.6
0.4 0.4
y 0.2 0.2
0 0 y 0 0
x x

Undamaged Plate Mode3 80.8255Hz Undamaged Plate Mode4 106.9928Hz

1 1

0.5 0.5

0 0

z
z

-0.5 -0.5

-1 -1
1 1
1 1
0.8 0.5 0.8
0.5 0.6 0.6
0.4 0.4
0.2 y 0.2 x
y 0 0 0
0 x
Undamaged Plate Mode5 139.6993Hz Undamaged Plate Mode6 162.7042Hz

1 1

0.5 0.5

0 0
z
z

-0.5 -0.5

-1 -1
1 1
1 1
0.8 0.5 0.8
0.5 0.6 0.6
0.4 0.4
y 0.2 y 0.2
0 0 0 0
x x

Undamaged Plate Mode7 192.1592Hz Undamaged Plate Mode8 212.432Hz

1 1

0.5 0.5

0 0
z

-0.5 -0.5

-1 -1
1 1
1 1
0.5 0.8 0.8
0.6 0.5 0.6
0.4 0.4
y 0.2 y 0.2
0 0 0 0
x x
Undamaged Plate Mode9 250.419Hz
Undamaged Plate Mode10 277.0733Hz

1
1
0.5
0.5
0
z

0
z

-0.5
-0.5
-1
1 -1
1 1
0.5 0.8 1
0.6 0.5 0.8
0.4 0.6
y 0.2 0.4
0 0 y 0.2
x 0 0 x

Fig. 5. Elastic mode shapes of a plate with free boundary conditions.

 Y.J. Yan et al. / International Journal of Solids and Structures 41 (2004) 6661–6676 6671

characteristics. Besides, crack directions also have influence on an anisotropic plate, and the crack location
in a plate also needs to be studied. First, the plate natural frequencies for damage status of different crack
length with longitudinal and transverse cracks (parallel and perpendicular to the stiffened plates, see Fig. 3)
are calculated, and they are listed in Table 2.
According to the analysis of data in Table 2, some conclusions can be drawn as follows:

(1) If a crack length is less than 5% of plate length or width, the natural frequencies nearly have no change
except for some particular mode, such as the 7th frequency in damage status of longitudinal inner
crack, i.e., crack away from the plate edges.
(2) If a crack length is less than 10% of plate length or width, some changes may occur in certain orders of
natural frequencies, such as the 6th and 7th frequencies in damage status of longitudinal and transverse
inner crack.
(3) If a crack length is longer than 20% of plate length or width, there will be remarkable changes in multi-
ple natural frequencies. However, a practical structure with such large crack may have already failed.
(4) Results show that changes in the natural frequency of a cracked plate do not always appear in the low-
est modal frequencies.
(5) For an anisotropic sandwich plate, the sensitivity of natural frequency to transverse crack is lower than
that to longitudinal crack.

In order to evaluate the influence of crack location on plate natural frequency, the natural frequencies
of the honeycomb sandwich plate with given crack length, crack direction and different locations in x- or
y-directions are calculated, and the results are shown in Fig. 6(a) and (b). The results show that the natural
frequencies rarely change with crack locations in the plate. This is because the given crack length is only
10% of plate length or width, and such a crack length is not large enough to be detected by structural
natural frequencies.

Table 2
Natural frequencies of the cracked honeycomb sandwich plate (Hz)
Natural frequency 1 2 3 4 5 6 7 8 9 10
order no.
Crack Crack
category length
Intact 0 29.711 50.900 80.826 106.99 139.70 162.70 192.16 212.43 250.42 277.07
Longitudinal 5% L 29.710 50.880 80.755 106.94 139.66 161.77 178.51 209.71 249.60 275.62
inner cracka 10% L 29.712 50.841 80.800 106.24 139.71 158.37 174.31 210.44 250.76 274.57
20% L 29.707 50.801 80.762 102.83 137.61 149.42 151.15 209.33 239.93 268.04
30% L 29.865 50.784 80.867 90.507 111.40 139.31 140.62 196.72 220.48 267.51

Longitudinal 5% L 29.710 50.883 80.827 106.97 139.66 162.12 190.47 211.19 249.76 277.72
edge crackb 10% L 29.713 50.599 80.809 100.37 100.34 146.06 146.57 201.17 237.80 264.23
20% L 29.716 49.682 72.518 80.868 93.41 145.67 146.55 202.32 239.64 263.81
30% L 29.658 30.890 52.306 80.863 91.29 144.48 137.09 155.19 223.34 254.45
Transverse 5% B 29.639 50.723 80.410 106.61 139.39 161.88 192.02 211.99 249.62 276.59
edge crackb 10% B 29.373 50.000 79.123 104.99 138.02 159.93 191.88 209.17 245.24 275.32
20% B 28.925 48.966 77.088 102.40 135.32 157.81 191.19 203.36 240.34 293.03
30% B 25.703 30.526 50.499 75.054 90.73 136.88 147.79 192.67 225.89 250.68
a
Inner crack – crack from plate edge.
b
Edge crack – crack starting from plate edge.
6672 Y.J. Yan et al. / International Journal of Solids and Structures 41 (2004) 6661–6676

As a summary, one can conclude that structural natural frequency is not suitable to detect crack damage
less than 10% of the plate dimension, even up to 20% of the total size of a plate-like structure. Besides, it is
very difficult to determine the location and severity of crack damage using natural frequencies.
In order to acquire the dynamic responses of a PVC honeycomb sandwich plate, two piezo-patches with
a size of 25 · 15 · 0.28 mm are bonded on the surface of the plate. One of them acts as an actuator and the
other acts as a sensor. The experimental set-up for acquisition of the dynamic responses of the plate with
different crack lengths is shown in Fig. 7. A square wave signal with 150 mV magnitude and 30 Hz
frequency generated by the signal generator TGA 1230 is fed into the TRek Model 700 Piezo-driver. The
30 V voltage signal from the output of the piezo-driver is exerted on the piezo-patch actuator. Dynamic
responses are measured using the piezo-patch sensor, and this signal is first fed into the B&K 2525 mea-
suring amplifier, which can amplify the signal and filter out the noise using the 3–3 kHz band-pass function.
Then, the output signal from the measuring amplifier is taken as the input of a computer with AD card for
data sampling and storage.
In the experimental study, the plates are put on a soft sponge so that the free-free boundary conditions
are simulated. Three honeycomb sandwich plates with different crack lengths of 3, 9 and 15 mm in lon-
gitudinal direction are studied, these crack lengths equal to 1%, 3%, and 5% of the plate length. Each
dynamic response signal of these plates is decomposed into the 5th layer ðk ¼ 5Þ of wavelet transform, and

Hz Hz
Natural Frequency

Natural Frequency

40.00 60.00
30.00
Mode 1

Mode 2

40.00
20.00
20.00
10.00
0.00 0.00
0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L 0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L
Crack Location along x direction Crack Location along x direction

Hz Hz
Natural Frequency

150.00
Natural Frequency

100.00
80.00
Mode 4
Mode 3

100.00
60.00
40.00 50.00
20.00
0.00 0.00
0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L 0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L
Crack Location along x direction Crack Location along x direction

Hz Hz
Natural Frequency

200.00
Natural Frequency

150.00
150.00
Mode 6
Mode 5

100.00
100.00
50.00 50.00
0.00 0.00
0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L 0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L
Crack Location along x direction Crack Location along x direction

Hz Hz
Natural Frequency

Natural Frequency

200.00 250.00
150.00 200.00
Mode 7

Mode 8

150.00
100.00
100.00
50.00 50.00
0.00 0.00
0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L 0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L
Crack Location along x direction Crack Location along x direction

Hz Hz
Natural Frequency

Natural Frequency

300.00 300.00
Mode 10
Mode 9

200.00 200.00

100.00 100.00

0.00 0.00
0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L 0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L
(a)

Fig. 6. Natural frequencies of a plate with different crack status.

 Y.J. Yan et al. / International Journal of Solids and Structures 41 (2004) 6661–6676 6673

Hz Hz

Natural Frequency

Natural Frequency
40.00 60.00
30.00
Mode 1

Mode 2
40.00
20.00
20.00
10.00
0.00 0.00
0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L 0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L
Crack Location along y direction Crack Location along y direction

Hz Hz
Natural Frequency

100.00

Natural Frequency
150.00
80.00
Mode 3

Mode 4
60.00 100.00
40.00
50.00
20.00
0.00 0.00
0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L 0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L
Crack Location along y direction Crack Location along y direction

Hz Hz
Natural Frequency

150.00

Natural Frequency
200.00
150.00
Mode 5

Mode 6
100.00
100.00
50.00
50.00
0.00 0.00
0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L 0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L
Crack Location along y direction Crack Location along y direction
Hz Hz
Natural Frequency

250.00 250.00
Natural Frequency

200.00 200.00
Mode 8
Mode 7

150.00 150.00
100.00 100.00
50.00 50.00
0.00 0.00
0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L 0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L
Crack Location along y direction Crack Location along y direction
Hz Hz
Natural Frequency

300.00
Natural Frequency

300.00
Mode 10
Mode 9

200.00 200.00

100.00 100.00

0.00 0.00
0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L 0% 10% 20% 30% 40% 50% 60% 70% 80% 90%L
(b)

Fig. 6 (continued)

Fig. 7. Schematic diagram of experimental set-up for crack damage detection.

6674 Y.J. Yan et al. / International Journal of Solids and Structures 41 (2004) 6661–6676

Table 3
Crack damage index Vd (%) of a honeycomb sandwich plate
Crack 3 mm (1% L) 9 mm (3% L) 15 mm (5% L)
W.N E S Error (%) E S Error (%) E S Error (%)
0 0.001 0.0009 8.48 0.069 0.0742 7.53 0.131 0.1403 7.07
1 )0.306 )0.329 7.55 2.129 2.3219 9.06 4.269 4.6567 9.08
2 )0.357 )0.380 6.58 0.465 0.5019 7.93 1.761 1.8941 7.56
3 1.687 1.9166 13.6 )12.59 )13.22 5.00 )28.48 )29.73 4.36
4 0.316 0.3496 10.6 4.619 5.0826 10.0 )1.060 )1.130 6.68
5 0.001 0.0009 9.19 )15.06 )15.72 4.37 )32.04 )33.18 3.55
6 1.985 2.2752 14.6 )9.385 )9.914 5.64 )24.98 )26.39 5.64
7 )0.541 )0.588 8.61 0.856 0.9273 8.32 3.930 4.2732 8.73
8 0.863 0.9657 11.8 2.899 3.1819 9.75 4.799 5.2526 9.45
9 1.240 1.4015 13.0 )17.16 )17.78 3.57 )16.68 )17.70 6.13
10 1.767 2.0161 14.0 )8.499 )9.022 6.16 )26.65 )28.00 5.09
11 0.001 0.0011 9.92 1.403 1.5259 8.75 3.245 3.5154 8.33
12 4.275 4.9169 15.0 )2.849 )3.052 7.15 7.815 8.6011 10.0
13 )1.509 )1.674 11.1 )4.966 )5.298 6.69 7.725 8.4745 9.70
14 0.472 0.5253 11.2 2.504 2.7392 9.39 2.140 2.3112 8.00
15 1.212 1.3633 12.4 )18.59 )19.07 2.57 )45.71 )46.89 2.59
W.N: wavelet number, E: experimental, S: numerical simulation.

16 wavelet sub-signals are obtained. The relative energy variations in these cases acquired using Eq. (33) are
listed in Table 3.
According to the data in Table 3, some hints of crack damage detection can be found as follows: (1)
When crack length is longer than 3% of the plate length, the largest element value of damage index vector
will be larger than 10%, and errors between experimental and numerical results do not exceed 10%. (2) The
larger the element value of the damage index vector, the smaller the errors between the experimental and
numerical results. Especially, the errors of those larger element values (larger than 10%) are usually less
than 5%. (3) When the crack length is shorter than 1% of the plate length, the element values of damage
index vector are smaller, and the errors between the experimental and numerical results are generally larger
than 10%. This indicates that these data are not reliable. The occurrence of so large errors for small crack
may be due to random errors in experiment and modal truncation in numerical simulation, and the latter
may be the main reason. This will be discussed in detail later.
Thus, one can conclude that if the largest element value of the damage index vector reaches 10%, some
crack damage in a honeycomb sandwich plate can be detected using the method proposed in this paper.
Describing complex damage status of a practical engineering structure requires a large amount of structural
dynamic response data, which can be acquired by numerical simulation using an experimentally checked
dynamic model.
In order to evaluate the ability of detecting crack extent using the method proposed in this paper, the
damage index vector for crack length from 1% to 10% of the plate length are obtained using numerical
simulation. The first three largest element values of each damage index vector are shown in Fig. 8. The
results show that the largest element values of the damage index vector have an increasing tendency with
the length of a crack, but the relation between them is not linear; if the largest element value reaching 10% is
taken as an indicator of discovering structural crack, the smallest detectable crack extent using the damage
index vector is about 2% of the structural length. Apparently, this method is more sensitive than using
structural natural frequencies.
A special phenomenon discovered in this study is that structural damage information is often contained
in some high order modes. Hence, the requirement for the established structural dynamic model for damage
 Y.J. Yan et al. / International Journal of Solids and Structures 41 (2004) 6661–6676 6675

180
170
160
150

Element values of vector Vd

140
130
120
110
(%) 100
90
80
70
60
50
40
30
20
10
0
0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10%L
Crack Length
The first largest value
The second largest value
The third largest value

Fig. 8. The first three largest element values of damage index vector for different crack lengths.

detection is different from that for active structural vibration control, for which only several low order
modes are usually considered. Structural damage detection requires more precise structural dynamic model,
i.e. more modes should be included in the dynamic model. Otherwise, some small structural damage may be
overlooked. Because only 20 modes are adopted in the dynamic model in this study, and the frequencies of
these modes are very close, this may be a reason for the discrepancy between experimental and numerical
results for small crack.

5. Conclusions

In this study, the ability of detecting crack damage in a honeycomb sandwich plate by using two types of
structural vibration parameters (natural frequency and dynamic response) have been evaluated, and the
feasibility of detecting small crack using method proposed in this study is also evaluated. We find that using
structural natural frequency may not be suitable for detecting crack damage less than 10%, even up to 20%
of the total size of a plate-like structure. Besides, it is very difficult to determine the location and category of
crack damage with such a dimension. However, energy spectrum of wavelet transform signals of structural
dynamic response has higher sensitivity to crack damage, it can exhibit structural damage status for a crack
length close to 2% of the dimension of a plate-like structure. We also found that structural damage
information is often contained in some high order modes of a structure, and more vibration modes should
be included in structural dynamic model for detection of small damage.

Acknowledgements

The authors would like to thank for the support by Natural Science Foundation of China under the
grant 50375123 and the Research Grants Council of Hong Kong Special Administrative Region of China
under the project no. PolyU 5313/03E and the Research Committee of the Hong Kong Polytechnic Uni-
versity (project no. GT 621).
6676 Y.J. Yan et al. / International Journal of Solids and Structures 41 (2004) 6661–6676

References

Chen, S.E., Venkatappa, S., Moody, J., Petro, S., GangaRao, H., 2000. A novel damage detection technique using scanning laser
vibrometry and a strain energy distribution method. Materials Evaluation 58 (12), 1389–1394.
Chui, C.K., 1997. Wavelets: A Mathematical Tool for Signal Processing. Society for Industrial and Applied Mathematics,
Philadelphia.
Collins, K.R., Plaut, R.H., Wauer, J., 1992. Free and forced longitudinal vibrations of a cantilevered bar with a crack. Journal of
Vibration and Acoustics––Transactions of the ASME 114 (2), 171–177.
Doebling, S.W., Hemez, F.M., Peterson, L.D., Farhat, C., 1997. Improved damage location accuracy using strain energy-based mode
selection criteria. AIAA Journal 35 (4), 693–699.
Farrar, C.R., Doebling, S.W., Nix, D.A., 2001. Vibration-based structural damage identification. Philosophical Transactions of the
Royal Society of London Series A––Mathematical Physical And Engineering Sciences 359 (1778), 131–149.
Gobin, P.F., Jayet, Y., Baboux, J.C., et al., 2000. New trends in non-destructive evaluation in relation to the smart materials concept.
International Journal of Systems Science 31 (11), 1351–1359.
Hou, Z., Noori, M., St Amand, R., 2000. Wavelet-based approach for structural damage detection. Journal of Engineering
Mechanics––ASCE 126 (7), 677–683.
Hu, N., Wang, X., Fukunaga, H., Yao, Z.H., Zhang, H.X., Wu, Z.S., 2001. Damage assessment of structures using modal test data.
International Journal of Solids and Structures 38 (18), 3111–3126.
Huebner, K.H., 2001. The Finite Element Method for Engineers. Wiley, New York.
Kosmatka, J.B., Ricles, J.M., 1999. Damage detection in structures by modal vibration characterization. Journal of Structural
Engineering––ASCE 125 (12), 1384–1392.
Mallat, S.G., 1999. A Wavelet Tour of Signal Processing, 2nd ed. Academic Press, London, UK.
Nandwana, B.P., Maiti, S.K., 1997. Modeling of vibration of beam in presence of inclined edge or internal crack for its possible
detection based on frequency measurements. Engineering Fracture Mechanics 58 (3), 193–205.
Ramamurti, V., Neogy, S., 1998. Effect of cracks on the natural frequency of cantilevered plates––a Rayleigh–Ritz solution. Mechanics
of Structures and Machines 26 (2), 131–143.
Salawu, O.S., 1997. Detection of structural damage through changes in frequency: a review. Engineering Structures 19 (9), 718–723.
Strang, G., Nguyen, T., 1996. Wavelets and Filter Banks. Cambridge Press, Wellesley, MA.
Zhang, H., Schulz, M.J., Ferguson, F., Pai, P.F., 1999. Structural health monitoring using transmittance functions. Mechanical
Systems and Signal Processing 13 (5), 765–787.
Zou, Y., Tong, L., Steven, G.P., 2000. Vibration-based model-dependent damage (delamination) identification and health monitoring
for composite structures––a review. Journal of Sound and Vibration 230 (2), 357–378.