Weifan 2010

Structural Health Monitoring
http://shm.sagepub.com/
Vibration-based Damage Identification Methods: A Review and Comparative Study

Wei Fan and Pizhong Qiao
Structural Health Monitoring 2011 10: 83 originally published online 20 April 2010
DOI: 10.1177/1475921710365419
The online version of this article can be found at:

http://shm.sagepub.com/content/10/1/83
Published by:
http://www.sagepublications.com
Additional services and information for Structural Health Monitoring can be found at:
Email Alerts: http://shm.sagepub.com/cgi/alerts
Subscriptions: http://shm.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations: http://shm.sagepub.com/content/10/1/83.refs.html
>> Version of Record - Feb 2, 2011
OnlineFirst Version of Record - Apr 20, 2010
What is This?
Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Vibration-based Damage Identification
Methods: A Review and Comparative Study
Wei Fan and Pizhong Qiao*
Department of Civil and Environmental Engineering and Composite

Materials and Engineering Center, Washington State University,
Pullman, WA, 99164-2910, USA
A comprehensive review on modal parameter-based damage identification methods

for beam- or plate-type structures is presented, and the damage identification algorithms
in terms of signal processing are particularly emphasized. Based on the vibration features,
the damage identification methods are classified into four major categories: natural
frequency-based methods, mode shape-based methods, curvature mode shape-based
methods, and methods using both mode shapes and frequencies, and their merits
and drawbacks are discussed. It is observed that most mode shape-based and curvature
mode shape-based methods only focus on damage localization. In order to precisely locate
the damage, the mode shape-based methods have to rely on optimization algorithms
or signal processing techniques; while the curvature mode shape-based methods are
in general a very effective type of damage localization algorithms. As an implementation,
a comparative study of five extensively-used damage detection algorithms for beam-type
structures is conducted to evaluate and demonstrate the validity and effectiveness of
the signal processing algorithms. This brief review aims to help the readers in identifying
starting points for research in vibration-based damage identification and structural
health monitoring and guides researchers and practitioners in better implementing
available damage identification algorithms and signal processing methods for beam- or
plate-type structures.
Keywords frequency mode shapes modal curvatures vibration damage identifi-

cation beams plates signal processing.
1 Introduction nondestructive damage identification technique is

crucial to maintain safety and integrity of
Structural damage identification has gained structures.
increasing attention from the scientific and Most nondestructive damage identification
engineering communities because the unpredicted methods can be categorized as either local or
structural failure may cause catastrophic, eco- global damage identification techniques [1]. Local
nomic, and human life loss. A reliable and effective damage identification techniques, such as
*Author to whom correspondence should be addressed. ß The Author(s), 2011. Reprints and permissions:
E-mail: [email protected] http://www.sagepub.co.uk/journalsPermissions.nav
Figures 2, 3 and 5–8 appear in color online: Vol 10(1): 83–29
http://shm.sagepub.com [1475-9217 (201101) 10:1;83–29 10.1177/1475921710365419]
83

84 Structural Health Monitoring 10(1)
ultrasonic methods and X-ray methods, require has been achieved in this highlighted area. A
that the vicinity of damage is known a priori and broad range of techniques, algorithms, and meth-
readily accessible for testing, which cannot be ods are developed to solve various problems
guaranteed for most cases in civil or aerospace encountered in different structures, from basic
engineering. Hence, the vibration-based damage structural components (e.g., beams and plates) to
identification method as a global damage identifi- complex structural systems (e.g., bridges and build-
cation technique is developed to overcome these ings). Doebling et al. [1] presented an extensive
difficulties. review of vibration-based damage detection meth-
The fundamental idea for vibration-based ods up to 1996. Sohn et al. [2] then presented an
damage identification is that the damage-induced updated version of this review on the literature up
changes in the physical properties (mass, damping, to 2001. In both the articles, the features extracted
and stiffness) will cause detectable changes in for identification were considered to classify the
modal properties (natural frequencies, modal damage identification methods. Following closely
damping, and mode shapes). For instance, reduc- this classification, Carden and Fanning [3] pres-
tions in stiffness result from the onset of cracks. ented a literature survey with particular emphasis
Therefore, it is intuitive that damage can be iden- on the papers and articles published from 1996 to
tified by analyzing the changes in vibration fea- 2003.
tures of the structure. Damage identification methods can also be
Although in vibration test, the excitation and classified as ‘model-based method’ or ‘response-
response are always measured and recorded in the based method’. The model-based method assumes
form of time history, it is usually difficult to exam- that a detailed numerical model of the structure is
ine the time domain data for damage identification. available for damage identification; while the
A more popular method is to examine the modal response-based method depends only on experi-
domain data through modal analysis technique, in mental response data from structures.
which the time domain data is transformed into the The objective of the present study is to provide
frequency domain, and then the modal domain data a summary review on damage identification
can be further extracted from the frequency domain algorithms for beam-type or plate-type structures.
data. During the past three decades, great effort has There are two main reasons for the focus on simple
been made in the researches within all three structures: (1) most structures or their major com-
domains (i.e., time, frequency, and modal domains). ponents in civil and mechanical engineering can be
It seems that this effort will continue since no single simplified as a beam or plate, and (2) the problem
existing method can solve all the damage identifica- of identifying a specific damage in a beam/plate
tion problems from various types of damages and provides an important benchmark for effectiveness
structures. However, the modal domain methods and accuracy of identification techniques.
attract more attention and play a dominant role in Furthermore, only the damage identification algo-
the state-of-the-art of structural damage identifica- rithms, as the key step of damage identification,
tion. The modal domain methods evolve along with will be reviewed in this article. Other related fac-
the rapid development of experimental modal ana- tors, such as environmental effects and excitation/
lysis technique, and they gain their popularity sensing methods, will not be discussed. Based on
because the modal properties (i.e., natural frequen- finite element simulation of a beam-type structure,
cies, modal damping, modal shapes, etc.) have their a comparative study of several damage identifica-
physical meanings and are thus easier to be inter- tion algorithms is provided to illustrate the validity
preted or interrogated than those abstract mathe- and effectiveness of the algorithms in different sce-
matical features extracted from the time or narios (e.g., single vs. multiple damages, noise and
frequency domain. sensor spacing effects, etc.). This brief review is
During the last three decades, extensive intended to help readers in identifying starting
research has been conducted in vibration-based points for research in vibration-based damage
damage identification, and significant progress identification and structural health monitoring

Fan & Qiao Vibration-based Damage Identification Methods 85
and guide researchers and practitioners in choosing approach. Frequency sensitivity turns to be propor-
and implementing the available damage identifica- tional to the potential energy stored at the cracked
tion algorithms and signal processing methods in cross section of the undamaged beam. Moreover,
damage detection of simple structures. the ratio of the frequency changes of two different
modes turns to be a function of damage location
2 Literature Review only. Both Liang’s and Morassi’s methods were
based on Euler–Bernoulli beam theory and modeled
This review is organized by the classification crack as a massless, infinitesimal rotational spring.
using the features extracted for damage identifica- All of the above-mentioned three explicit expres-
tion, and these damage identification methods are sions are valid only for small defects. Kasper et al.
categorized as follows: [7] derived the explicit expressions of wavenumber
shift and frequency shift for a cracked symmetric
. Natural frequency-based methods; uniform beam. These expressions apply to beams
. Mode shape-based methods; with both shallow and deeper cracks. But the expli-
. Curvature/strain mode shape-based methods; cit expressions are based on high frequency approx-
. Other methods based on modal parameters. imation, and therefore, they are generally
inaccurate for the fundamental mode and for a
crack located in a boundary-near field.
2.1 Natural Frequency-based Methods
2.1.2 The Inverse Problem The inverse problem
Natural frequency-based methods use the nat- of this issue is determination of damage location
ural frequency change as the basic feature for and size of a given structure based on natural fre-
damage identification. The choice of the natural quency measurement. The study on such an inverse
frequency change is attractive because the natural problem dates back to 1978, when Adams et al. [8]
frequencies can be conveniently measured from just presented a method for detection of damage in a
a few accessible points on the structure and are usu- 1D component utilizing the natural frequencies of
ally less contaminated by experimental noise. longitudinal vibrations. In 1997, Salawu [9] pres-
ented an extensive review of publications before
2.1.1 The Forward Problem The forward prob- 1997 dealing with the detection of structural
lem of this issue is to determine the natural fre- damage through frequency changes. In the conclu-
quency changes of a given structure based on sion of this review paper, Salawu suggested that
damage location and severity, and it serves as a the- natural frequency changes alone may not be suffi-
oretical foundation for natural frequency-based cient for a unique identification of the location of
methods. Gudmundson [4] used an energy-based structural damage because cracks associated with
perturbation approach and derived an explicit similar crack lengths but at two different locations
expression for the resonance frequencies of a wide may cause the same amount of frequency change.
range of damaged structure. This method can Messina et al. [10] proposed a correlation coef-
account for a loss of mass in addition to a loss of ficient termed the multiple damage location assur-
stiffness. Liang et al. [5] addressed the issue of deter- ance criterion (MDLAC) by introducing two
mining frequency sensitivity for simply supported methods of estimating the size of defects in a struc-
or cantilevered beam with one crack and developed ture. The method is based on the sensitivity of the
analytical relationships between the first-order frequency of each mode to damage in each loca-
changes in the eigenfrequencies and the location tion. ‘MDLAC’ is defined as a statistical correla-
and severity of the damage. This method requires tion between the analytical predictions of the
symbolic computation of the characteristic equa- frequency changes f and the measured frequency
tion. Morassi [6] showed that the frequency sensi- changes f. The analytical frequency change f can
tivity of a cracked beam-type structure can be be written as a function of the damage extent
explicitly evaluated by using a general perturbation vector D. The required damage state is obtained

by searching for the damage extent vector D the approximate method given by Liang et al. [11]
which maximizes the MDLAC value: for a single segment beam to multi-segment beams.
Murigendrappa et al. [17,18] later applied Patil and
MDLACðfDgÞ
Maiti’s approach to single/multiple crack detection
T 2
f f ðfDgÞ ð1Þ in pipes filled with fluid.
¼ T T : Morassi [19] presented a single crack identifi-
f f f ðfDgÞ f ðfDgÞ cation in a vibrating rod based on the knowledge
of the damage-induced shifts in a pair of natural
Two algorithms (i.e., first and second order frequencies. The analysis is based on an explicit
methods) were developed to estimate the absolute expression of the frequency sensitivity to damage
damage extent. Both the numerical and experimen- and enables nonuniform bars under general
tal test results were presented to show that the boundary conditions to be considered. Some of
MDLAC approach offers the practical attraction
the results are also valid for cracked beams in
of only requiring measurements of the changes in a
bending. Morassi and Rollo [20] later extended
few of natural frequencies of the structure between
the method to the identification of two cracks of
the undamaged and damaged states and provides
equal severity in a simply supported beam under
good predictions of both the location and absolute
flexural vibrations. However, the inverse problem
size of damage at one or more sites.
is ill-posed, just as many other frequency-based
Liang et al. [11] developed a method based on
methods; namely, even by leaving symmetrical
three bending natural frequencies for the detection
positions aside, cracks with different severity in
of crack location and quantification of damage
two sets of different locations can produce identi-
magnitude in a uniform beam under simply sup-
cal changes in the first three natural frequencies.
ported or cantilever boundary conditions. The
Kim and Stubbs [21] proposed a single damage
method involves representing crack as a rotational
indicator (SDI) method to locate and quantify a
spring and obtaining plots of its stiffness with crack
crack in beam-type structures by using changes in
location for any three natural modes through the
a few natural frequencies. A crack location model
characteristic equation. The point of intersection
and a crack size model were formulated by relating
of the three curves gives the crack location and stiff-
fractional changes in modal energy to changes in
ness. The crack size is then computed using the stan-
dard relation between stiffness and crack size based natural frequencies due to damage. In the crack
on fracture mechanics. This method had been location model, the measured fractional change in
extended to stepped beams by Nandwana and the i-th eigenvalue Zi and the theoretical (FEM-
Maiti [12] and to segmented beams by Chaudhari based) modal sensitivity of the i-th modal stiffness
and Maiti [13] using the Frobenius method to solve with respect to the j-th element Fij is defined,
Euler-Bernoulli type differential equations. respectively. The theoretical modal curvature is
Chinchalkar [14] used a finite element-based numer- obtained from a third order interpolation function
ical approach to mimic the semi-analytical of theoretical displacement mode shape, and they
approach using the Frobenius method. This are expressed as
approach does not require quadruple precision Zi ¼ !2i =!2i , ð2Þ
computation and is relatively easy to apply to dif- R xjþ1 00 2
ferent boundary conditions. Lele and Maiti [15] x f g dx
extended Nandwana and Maiti’s method to short Fij ¼ Kij =Ki ¼ RjL : ð3Þ
f 00 g2 dx
beam, taking into account the effects of shear defor- 0
mation and rotational inertia through Timoshenko Then, an error index eij is introduced to repre-
beam theory. Patil and Maiti [16] proposed a fre- sent the localization error for the i-th mode and the
quency shift-based method for detection of multiple j-th location
open cracks in an Euler–Bernoulli beam with vary-
Zi Fij
ing boundary conditions. This method is based on eij ¼ PNM PNM ð4Þ
the transfer matrix method and extends the scope of k¼1 Zk k¼1 Fkj

where NM denotes the number of measured vibra- in [22], all of the frequencies-based methods
tion modes. To account for all available modes, a are model-based. Most methods for damage
SDI is defined to indicate the damage location identification in beam-type structures rely on
" #1=2 Euler–Bernoulli beam theory (except the shear
X
NM
deformable model in [15]) and modeling crack
2
SDIj ¼ eij ð5Þ
i¼1
as a rotational spring. It is well known that
Euler–Bernoulli beam theory over-predicts natural
While in the crack size model, the damage frequencies in short beams and high frequency
inflicted aj at predefined locations can be predicted bending modes. It is also known that modeling
using the sensitivity equation. The crack depth can of crack as a rotational spring based on frac-
be computed from aj and the crack size model based ture mechanics will lose its credibility in high
on fracture mechanics. The feasibility and practical- frequency modes or deep crack cases. Therefore,
ity of the crack detection scheme were evaluated by these methods are only applicable to a slender
applying the approach to the 16 test beams. beam-type structure with small cracks. Moreover,
Zhong et al. [22] recently proposed a new the range of vibration modes which are suited for
approach based on auxiliary mass spatial probing damage detection is narrowed to the first few
using the spectral centre correction method modes due to the basic assumptions of these
(SCCM), to provide a simple solution for damage methods.
detection by just using the output-only time history Another critical limitation is that the frequency
of beam-like structures. A SCCM corrected highly changes caused by damage are usually very small
accurate natural frequency versus auxiliary mass and may be buried in the changes caused by envi-
location curve is plotted along with the curves of its ronmental and operational conditions. For this
derivatives (up to third order) to detect the crack. reason, most successful damage identification
However, only the FE verification was provided to methods using frequency are verified only in con-
illustrate the method. Since it is not so easy to get a trolled laboratory scale rather than testing of real
high resolution natural frequency versus auxiliary structures in their operating environment. Given
mass location curve in experiment as in numerical that it is acknowledged that in situ measured fre-
simulation, the applicability and practicality of the quency variation due to ambient vibration and
method in in situ testing or even laboratory testing environmental effects can be as high as 5–10%,
are still in question. generally, it would be necessary for a natural fre-
There are also several frequency change-based quency to change by about 5% for damage to be
methods using soft optimization process for detected with confidence [9]. To account for this
damage detection. Maity and Tripathy [23] used effect, a statistical damage detection model using
genetic algorithm (GA) to detect and assess the pattern recognition technique is needed to distin-
structural damage from changes in natural fre- guish damage-induced changes from
quencies. Pawar and Ganguli [24] applied the environment-induced changes. Ni et al. [25]
genetic fuzzy system to detect the matrix crack in addressed the modeling of temperature effects on
thin-walled composite structures based on changes modal frequencies based on a long-term structural
in natural frequencies. health monitoring system of Ting Kau Bridge. The
support vector machine (SVM) technique is
2.1.3 Limitations of Frequency-based Methods applied to formulate regression models which
Although the use of natural frequency changes for quantify the effect of temperature on modal fre-
crack detection has been extensively studied in quencies. It shows that the SVM models exhibit
the past three decades, the aforementioned tradi- good capabilities for mapping between the temper-
tional damage detection methods based on fre- ature and modal frequencies. Kim et al. [26] pres-
quency change still have several common ented a vibration-based damage monitoring
limitations. scheme to give warning of the occurrence, location,
One fundamental limitation lies in the model- and severity of damage under temperature-induced
ing of structure and damage. Except the approach uncertainty conditions. A damage warning model

is selected to statistically identify the occurrence of 2.2 Mode Shape-based Methods

damage by recognizing the patterns of damage-
Compared to using natural frequencies, the
driven changes in natural frequencies of the struc-
advantage of using mode shapes and their deriva-
ture and by distinguishing temperature-induced
tives as a basic feature for damage detection is
off-limits.
quite obvious. First, mode shapes contain local
Another limitation is that the damage identifi-
cation problem is often ill-posed even without noise information, which makes them more sensitive to
pollution, which leads to nonuniqueness of the solu- local damages and enables them to be used directly
tions of damage location and severity. First, it is in multiple damage detection. Second, the mode
obvious that damage with same severity in symmet- shapes are less sensitive to environmental effects,
ric locations of a symmetric structure will cause such as temperature, than natural frequencies [29].
identical frequency changes. Furthermore, damage The disadvantages are also apparent. First, mea-
with different severity in different locations can also surement of the mode shapes requires a series of
produce identical changes in a few measured natural sensors; second, the measured mode shapes are
frequencies. This problem is even more severe for more prone to noise contamination than natural
detection of multiple cracks. To avoid nonunique- frequencies.
ness, several damage detection methods require that In the past three decades, many damage iden-
the number of measurable modal frequencies tification methods have been developed based on
should be close to the number of system parameters direct or indirect use of measured mode shapes.
required to characterize the damage. However, the These methods are roughly categorized into two
number of accessible vibration modes which is types in this article. The traditional ‘mode shape
suited for damage detection is very limited (typically change methods’ try to establish a relationship
the first few modes), usually much less than the between damage location/severity and modal
number of system parameters required. To over- shape change through a finite element model or
come this difficulty, Dilena and Morassi [27] sug- experimental test. Hence, they depend on modal
gested that an appropriate use of frequencies and shape data from both intact and damaged struc-
anti-resonances may avoid the nonuniqueness of tures. With the development of modern signal pro-
the damage location problem which occurs when cessing technique and its application in damage
only frequency measurements data is used in identification, a series of relatively new damage
single crack identification. Jiang et al. [28] incorpo- identification methods have been developed. The
rated a tunable piezoelectric transducer circuitry modern ‘signal processing methods’ can be applied
into the structure to enrich the modal frequency to either mode shape change or mode shape data
measurements, meanwhile implementing a from damaged structures alone. These methods
high-order identification algorithm to sufficiently take mode shape change or mode shape data as a
utilize the enriched information. It is shown that spatial-domain signal and use signal processing
the modal frequencies can be greatly enriched by technique to locate damage by detecting the local
inductance tuning, which, together with the discontinuity of mode shape curve caused by
high-order identification algorithm, leads to a damage.
fundamentally-improved performance on the iden-
tification of single and multiple damages with the
usage of only lower-order frequency measurements. 2.2.1 Traditional Mode Shape Change
In summary, the frequency change-based Methods The traditional mode shape change
damage identification method can be successfully method uses the mode shape change from intact
applied to simple structures with small cracks (typ- and damaged structures as a basic feature for
ically, a slender beam-type structure with an arti- damage identification. The baseline data from
ficially induced crack) in a controlled laboratory intact structure can be obtained from either an
condition. However, due to its inherent drawbacks, experimental test on the intact structure or an
its applications for real complex structures or mul- accurate numerical model of the intact structure.
tiple/severe damage detection are limited. When experimental data on the intact structure is

not available, a finite element model is usually the damage location and size using Fourier coeffi-
adopted to generate the baseline data. cients as input. Numerical studies showed that
Shi et al. [30] extended the damage localization damage detection using Fourier coefficients and
method based on multiple damage location assur- neural networks has the capability to detect the
ance criterion (MDLAC) [10] by using incomplete location and damage size accurately. However,
mode shape instead of modal frequency. The the use of this method is limited to beams with
two-step damage detection procedure is to prelimi- clamped-clamped boundary condition. Abdo and
narily localize the damage sites by using incom- Hori [34] suggested that the rotation (i.e., the first
plete measured mode shapes and then to detect derivative of displacement) of mode shape is a sen-
the damage site and its extent again by using mea- sitive indicator of damage. Based on a finite ele-
sured natural frequencies. No expansion of the ment analysis of a damaged cantilevered plate and
incomplete measured mode shapes or reduction a damaged simply-supported plate, the rotation of
of finite element model is required to match the mode shape is shown to have better performance of
finite-element model, and the measured informa- multiple damage localization than the displace-
tion can be used directly to localize damage sites. ment mode shape itself.
The method was demonstrated in a simulated 2D In general, the traditional mode shape change
planar truss model. Comparison showed that the methods are not very sensitive to damage, and they
proposed method is more accurate and robust in are commonly subjected to several limitations.
damage localization with or without noise effect For example, they are only sensitive to damage in
than the original MDLAC method. In this certain area of the structures (e.g., mid-span of a
method, the use of mode shape is only for preli- clamped-clamped beam), and without further
minary damage localization, and the accurate signal processing or pattern recognition technique,
localization and quantification of damage still they can only be adopted for preliminary damage
rely on measured frequency changes. localization rather than accurate localization or
Lee et al. [31] presented a neural network- quantification of damage. Furthermore, their
based technique for element-level damage assess- applications in in situ structures are also very
ments of structures using the modal shape limited.
differences or ratios of intact and damaged struc-
tures. The effectiveness and applicability of the 2.2.2 Mode Shape Analysis using Modern Signal
proposed method using the mode shape differences Processing Methods In most in-situ cases, exper-
or ratios were demonstrated by two numerical iment on intact structures is not feasible. For
example analyses on a simple beam and a the model-based method, the credibility of a
multi-girder bridge. Hu and Afzal [32] proposed numerical model must be established through care-
a statistical algorithm for damage detection in ful model verification and validation [35]. The
timber beam structures using difference of the lacks of experimental data from intact structures
mode shapes before and after damage. The differ- also increase the difficulty in establishing an accu-
ent severities of damage, damage locations, and rate numerical model. Furthermore, numerical
damage counts were simulated by removing mass analysis requires considerable computational cost,
from intact beams to verify the algorithm. The especially when integrated in an iterative optimiza-
results showed that the algorithm is reliable for tion process. This limitation makes model-based
the detection of local damage under different seve- methods not suitable for real-time structure
rities, locations, and counts. Pawar et al. [33] inves- damage detection.
tigated the effect of damage on beams with In order to solve these problems, developing a
clamped boundary conditions using Fourier analy- response-based method that depends on experi-
sis of mode shapes in the spatial domain. The mental data only from damaged structures has
damaged mode shapes were expanded using a spa- recently become a focused research topic in
tial Fourier series, and a damage index (DI) in the damage identification. These methods do not
form of a vector of Fourier coefficients was formu- require a theoretical or numerical model. Their
lated. A neural network (NN) was trained to detect basic assumption is that the mode shape data

from a healthy structure contains only dimension (GFD)’ method by introducing a scale
low-frequency signal in spatial domain compared factor S in the FD algorithm
to the damage-induced high-frequency signal
log10 ðnÞ
change. Several signal processing technique-based GFD ¼ ,
damage detection algorithms, such as fractal log10 ðds =Ls Þ þ log10 ðnÞ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
dimension (FD) method and wavelet transform 2 2
ds ¼ max yiþj yi þS2 xiþj xi ,
method, have been investigated. It should be 1 5 jM
M qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X 2 2
noted that the ‘signal processing method’ can be
applied to both the mode shape change and Ls ¼ yiþj yiþj1 þS2 xiþj xiþj1
j¼1
damaged mode shape data alone. When only the
damaged mode shape data are used, these methods ð8Þ
cannot be used for damage quantification due to Instead of directly applying the algorithm to
its inherent limitation. the fundamental mode shape, the GFD is applied
Fractal Dimension Method: Hadjileontiadis to the ‘uniform load surface’(ULS) [39] to detect
et al. [36] and Hadjileontiadis and Douka [37] the damage in the paper. Three different types of
proposed a response-based damage detection damage in laminated composite beams have been
algorithm for beams and plates using FD. successfully detected by the GFD [38,40]. It should
For beam-type structures, the FD of a curve is esti- be pointed out that the GFD bears no conven-
mated by tional physical meaning as compared to the FD,
and it only serves as an indicator of damage.
log10 ðnÞ A scale factor S has to be carefully chosen in
FD ¼ , ð6Þ
log10 ðd=LÞ þ log10 ðnÞ order to detect damage successfully.
Recently, Qiao and Cao [41] proposed a novel
where n is the number of steps in the curve; d is the waveform fractal dimension-based damage identi-
diameter estimated as the distance between the first fication algorithm. An approximate waveform
point of the sequence P1 and the i-th point Pi of the capacity dimension (AWCD) was formulated
sequence that provides the farthest distance; and L first, from which an AWCD-based modal irregu-
is the total length of the curve or the sum of dis- larity algorithm (AWCD-MAA) was systemati-
tances between successive points. cally established. Then, the basic characteristics
of AWCD-MAA on irregularity detection of
d ¼ max distðP1 , Pi Þ mode shapes, e.g., crack localization, crack quan-
X
n1 ð7Þ tification, noise immunity, etc., were investigated
L¼ distðPi , Piþ1 Þ: based on an analytical crack model of cantilever
i
beams using linear elastic fracture mechanics. In
This method calculates the localized FD of the particular, from the perspective of isomorphism,
fundamental mode shape directly. The damage fea- a mathematical solution on the use of applying
tures are established by employing a sliding waveform FD to higher mode shapes for crack
window of length M across the mode shape and identification was originally proposed, from
estimating the FD at each position for the regional which the inherent deficiency of waveform FD to
mode shape inside the window. Damage location identify crack when implemented to higher mode
and size are determined by a peak on the FD curve shapes is overcome. The applicability and effective-
indicating the local irregularity of the fundamental ness of the AWCD-MAA was validated by an
mode shape introduced by the damage. experimental program on damage identification
If the higher mode shapes were considered, this of a cracked composite cantilever beam using
method might give misleading information as directly measured strain mode shape from smart
demonstrated in their study. To overcome this piezoelectric sensors.
shortcoming, Wang and Qiao [38] proposed a Wavelet Transform Method: Recently, wavelet
modified FD method termed ‘generalized fractal analysis has shown its inherent merits in damage

detection over traditional methods due to its ability cause an interruption of the singularity in the
to closely examine the signal with multiple scales to signal towards the finest scales. The wavelets are
provide various levels of detail and approxima- shown to be able to denoise the mode shape data
tions. The use of wavelet transform to identify and preserve the peaks indicating damage location
damage from mode shape has been one of the by adopting a tradeoff between finest and larger
most popular techniques. These methods treat scales. It is also shown that weighting the mode
mode shape data as a signal in spatial domain, shape through a window can effectively reduce
and they use spatial wavelet transform technique the boundary effect caused by signal discontinuity.
to detect the signal irregularity caused by damage. For application of classical Gaussian wavelets, a
Liew and Wang [42] first used spatial wavelet relatively high density of sensors is required due
coefficients for crack detection based on the to their natural difficulty to process a low
numerical solution for the deflection of a beam number sampling points.
under oscillating excitation. Wang and Deng [43] Zhong and Oyadiji [49] proposed a crack
considered Haar wavelet transforms to analyze detection algorithm in symmetric beam-like struc-
spatially distributed signals (displacement and tures based on stationary wavelet transform (SWT)
strain measurements) for damage detection of a of mode shape data. Two sets of mode shape data,
cracked beam and plate. Quek et al. [44] examined which constitute two new signal series, are, respec-
the sensitivity of wavelet technique in the detection tively, obtained from the left half and recon-
of cracks in beam structures. Although their works structed right half of modal displacement data of
focused on deflection under static or impact load- a damaged simply supported beam. The difference
ing other than mode shape, they demonstrated the of the detail coefficients of the two new signal
potential of using wavelet transform on mode series was used for damage detection. The
shape for damage detection. method was verified using modal shape data gen-
Hong et al. [45] showed that the continuous erated by a finite element analysis of 36 damage
wavelet transform (CWT) of mode shape using a cases of a simply supported beam with an artificial
Mexican hat wavelet is effective to estimate the random noise of 5% SNR. The effects of crack
Lipschitz exponent for damage detection of a size, depth and location as well as the effects of
damaged beam. The magnitude of the Lipschitz sampling interval were examined. The results
exponent can be used as a useful indicator of the show that all the cases can provide evidence of
damage extent. It was also proved in their work that crack existence at the correct location of the
the number of the vanishing moments of wavelet beam and that the proposed method can be recom-
should be at least 2 for crack detection in beams. mended for identification of small cracks as small
Douka et al. [46,47] applied 1D symmetrical 4 wave- as 4% crack ratio in real applications with mea-
let transform on mode shape for crack identification surement noise present. However, there are two
in beam and plate structures. The position of the main disadvantages of this method. First, the use
crack is determined by the sudden change in the of this method based on SWT requires fairly accu-
wavelet coefficients. An intensity factor is also rate estimates of the mode shapes. Second, the
defined to estimate the depth of the crack from method cannot tell the crack location from its
the coefficients of the wavelet transform. mirror image location due to its inherent limita-
Gentile and Messina [48] presented a compre- tion. Therefore, in applying the method, both the
hensive investigation on the application of CWT crack location predicted and its mirror image loca-
on mode shape for damage detection of a cracked tion should be checked for the presence of a crack.
beam. It is shown that when the scale s is small, Chang and Chen [50] presented a spatial Gabor
CWT with m vanishing moment can be a good wavelet-based technique for damage detection of a
approximation of the m-th derivatives of the spa- multiple cracked beam. Given natural frequencies
tial domain signal except for the discrepancies at and crack positions, the depths of the cracks are
the boundaries. The Gaussian derivative wavelet then solved by an optimization process based on
family (Gaus1 4) is suggested due to its particular traditional characteristic equation. Analysis and
advantage in damage detection because it will not comparison showed that it can detect the cracks

positions and depths and also has high sensitivity to The proposed methodology is evaluated using the
the crack depth, and the accuracy of this method is mode shape data from the numerical finite element
good. The limitation of this method is very common analysis and experimental testing of a cantilever
in wavelet transform methods, that is, there are beam with a through-width crack. The methodol-
peaks near the boundaries in the wavelet plot ogy presented provides a robust and viable tech-
caused by discontinuity and the crack cannot be nique to identify minor damage in a relatively
detected when the crack is near the boundaries. lower signal-to-noise ratio environment.
Chang and Chen [51] also applied this technique Most recently, the authors presented a 2D
to detect damage in a rectangular plate. CWT-based damage detection algorithm using
Poudel et al. [52] proposed a damage detection ‘Dergauss2d’ wavelet for plate-type structures
method for beam-type structures based on complex [55]. The 2D CWT considered is based on the for-
Gaussian wavelet transform of mode shape differ- mulation by Antoine et al. [56]. A concept of iso-
ence. The uses of three different wavelet transforms surface of 2D wavelet coefficients is proposed, and
(i.e., Gaussian wavelet, complex Gaussian wavelet, it is generated to indicate the location and approx-
Morlet wavelet) are compared. The determination imate shape or area of the damage. The proposed
of maximum and minimum scale to avoid edge algorithm is a response-based damage detection
effect and noise effect is also discussed. The labo- technique which only requires the mode shapes of
ratory experiment on a simply supported beam the damaged plates. This algorithm is applied to
using high-speed digital video camera was con- the numerical vibration mode shapes of a cantile-
ducted to verify the method. Results showed that ver plate with different types of damage to illus-
wavelet transformation of the mode shape differ- trate its effectiveness and viability. A comparative
ence function provides an inspection-friendly dis- study with other two 2D damage detection algo-
play of local changes due to structural damage. rithms, i.e., 2D gapped smoothing method (GSM)
The maximum modulus and sign change of phase and 2D strain energy method (SEM), is performed,
angle in the wavelet coefficients indicate the and it demonstrates that the proposed 2D
damage locations. CWT-based algorithm is superior in noise immu-
Rucka and Wilde [53] applied respective nity and robust with limited sensor data. The algo-
Gaussian wavelet and reverse biorthogonal wavelet rithm is further implemented in an experimental
transform on the fundamental mode shapes of modal test to detect impact damage in an FRP
beam and plate structures for damage detection. composite plate using smart piezoelectric actuators
Recently, Cao and Qiao [54] proposed a novel and sensors, demonstrating its applicability to the
wavelet transform technique (so called ‘integrated experimental mode shapes. The 2D CWT-based
wavelet transform’), which takes synergistic advan- algorithm presented in [55] is among a few limited
tage of the SWT and the CWT, to improve the studies in the literature to explore the application
robustness of irregularity analysis of mode shapes of 2D wavelets in damage detection, and as demon-
in damage detection. Two progressive wavelet ana- strated, it can be used as a viable and effective
lysis steps are considered, in which the SWT-based technique for damage identification of plate- or
multiresolution analysis (MRA) is first employed shell-type structures.
to refine the retrieved mode shapes, followed by
CWT-based multiscale analysis (MSA) to magnify
2.3 Mode Shape Curvature/Strain Mode
the effect of slight irregularity. The SWT-MRA is
Shape-based Methods
utilized to separate the multi-component modal
signal, eliminate random noise and regular inter- It has been shown by many researchers that the
ferences, and thus extract purer damage informa- displacement mode shape itself is not very sensitive
tion; while the CWT-MSA is employed to to small damage, even with high density mode shape
smoothen, differentiate or suppress polynomials measurement [57–59]. As an effort to enhance the
of mode shapes to magnify the effect of irregular- sensitivity of mode shape data to the damage, the
ity. The choice of the optimal mother wavelet in mode shape curvature (MSC) is investigated as
damage detection is also elaborately discussed. a promising feature for damage identification.

Once again, the modal curvature-based methods shapes of all modes, is introduced. The technique
are categorized into two types in this article: is further applied to a real structure, namely bridge
(1) the traditional ‘modal curvature change meth- Z24, to show its effectiveness in multiple damage
ods’, and (2) the modern ‘signal processing meth- location. Just-Agosto et al. [67] used the MSC data
ods’. The curvature mode shapes also closely and a properly trained NN scheme to implement a
involve in the modal strain energy-based method successful damage detection for composite sand-
[30,60–63] (see Section 2.3.3). wich beam.
Deriving the MSC from curve fitting or finite
difference approximation may introduce consider-
2.3.1 Traditional Modal Curvature Change
able error. According to Euler–Bernoulli beam
Method Pandey et al. [64] suggested for the first
theory, the strain at top surface of a beam under
time that the MSC, that is, the second derivatives
bending is proportional to the curvature of the
of mode shape, are highly sensitive to damage and
beam. Therefore, the curvature mode shape of a
can be used to localize it. The curvature mode
slender beam can be directly measured by measur-
shapes are derived using a central difference
ing the strain mode shape.
approximation as follows
Swamidas and Chen [68] performed a finite
i ¼ ðwiþ1 þ wi1 2wi Þ=h2 ð9Þ element-based modal analysis on a cantilever plate
with a small crack. It was found that the surface
where w is the modal displacement at i-th point, crack in the structure will affect most of the modal
h is the sensor spacing. parameters, such as the natural frequencies of the
Result showed that the difference of curvature structure, amplitudes of the response and mode
mode shapes from intact and damaged structure shapes. Some of the most sensitive parameters are
can be a good indicator of damage location. It is the difference of the strain mode shapes and the
also pointed out [64,65] that for the higher modes, local strain frequency response functions. By mon-
the difference in modal curvature shows several itoring the changes in the local strain frequency
peaks not only at the damage location but also at response functions and the difference between the
other positions, which may lead to a false indication strain mode shapes, the location and severity of the
of damage. Hence, in order to reduce the possibility crack that occurs in the structure can be determined.
of a false alarm, only first few low curvature mode Laboratory experiment by Zhou et al. [69] shows
shapes can be used for damage identification. that the direct measurement of curvature using
Salawu and Williams [58] compared the perfor- appropriately placed strain gauges was found to
mance of both curvature and displacement mode improve the localization resolution, a fact attribu-
shapes for locating damage and confirmed MSC as ted to the elimination of several data processing
a sensitive indicator of damage. However, the sen- steps required when the acceleration data are
sitivity and effectiveness of MSC method were also used. Li et al. [70] presented a crack damage detec-
questioned by several researchers with experimen- tion using a combination of natural frequencies and
tal evidence showing that the modal curvature by strain mode shapes as input in artificial neural net-
itself cannot locate small damages [66]. Abdel works (ANN) for location and severity prediction
Wahab and De Roeck [65] investigated the accu- of crack damage in beam-like structures. In the
racy of using the central difference approximation experiment, several steel beams with six distributed
to compute the MSC based on finite element ana- surface-bonded strain gauges and an accelerometer
lysis. The authors suggested that a fine mesh is mounted at the tip were used to obtain modal para-
required to derive the modal curvature correctly meters such as resonant frequencies and strain
for the higher modes and that the first mode will mode shapes.
provide the most reliable curvature in practical Recently, Qiao and his coworkers (i.e., Qiao
application due to the limited number of sensors et al., Hamey et al., Lestari and Qiao, Lestari
needed. Then, a damage indicator called ‘curvature et al. [40,71–74]) conducted a series of studies on
damage factor’, which is the average absolute dif- using strain mode shapes measured from
ference in intact and damaged curvature mode surface-bonded piezoelectric sensors to identify

various damage in composite beam or plate struc- functions. The DI at each point is determined as
tures. Hamey et al. [71] evaluated several damage the squared magnitude of the difference between
detection algorithms in carbon/epoxy composite the two cubic polynomials and the experimental
beams with several possible damage configurations curvatures. It is shown in experiment that using
using directly measured curvature mode shape broadband data can offer a better damage detec-
(strain mode shape). The results showed that all tion sensitivity than using resonant data due to
the methods of the curvature modes measured by higher data accuracy. Yoon et al. [77] expand the
the piezoelectric strain sensors (PVDF) could be GSM by introducing a ’globally optimized smooth
used as a potential tool in damage detection techni- shape’ from the analytic mode shape function.
ques, and the damage index method (DIM) [63] Yoon et al. [78] generalized the 1D GSM to 2D
detected and isolated the damage better than other plate-like structural applications. Local features
ones studied. Lestari and Qiao [72] adopted a sim- in vibration curvature shapes are extracted using
ilar approach of using the directly measured curva- a 2D localized surface fit. A structural irregularity
ture mode shapes from PVDF sensor to detect index is defined to represent the structural stiffness
debonding and core crushing in composite honey- variability for each test point on the structure. A
comb sandwich beams, and both the location and statistical-based outlier detection method is applied
relatively size of damage were evaluated. to the indices to detect damaged areas. The proce-
dure can either use mode shape data or broadband
operating shape data. The results confirm
2.3.2 Modern Signal Processing Methods using
Ratcliffe’s conclusion that using broadband data
Modal Curvature Ratcliffe [75] developed a
can offer a better damage detection sensitivity
response-based GSM to localize the damage by
than using resonant data.
fitting a localized cubic polynomial curve to the
Amaravadi et al. [79] proposed an orthogonal
MSC and calculating a difference between the
wavelet transform technique that operates on cur-
cubic and Laplacian as
vature mode shape for enhancing the sensitivity

i ¼ a0 þ a1 x þ a2 x2 þ a3 x3 i , ð10Þ and accuracy in damage location. First, the curva-
ture mode shape is calculated by central difference
where a0, a1, a2, a3 are the coefficients determined approximation from the displacement mode shapes
by fitting MSC i2 , i1 , iþ1 , iþ2 of damaged experimentally obtained from SLV. Then, a thresh-
structure. The MSC can be obtained by a finite old wavelet map is constructed for the curvature
difference approximation using Equation (9) or mode shape to detect the damage. The experimen-
by directly measuring the strain mode shape. This tal results are reasonably accurate.
method was verified by a laboratory experiment on Kim et al. [80] proposed a curvature mode
a slotted steel beam under free–free boundary con- shape-based damage identification method for
dition. The GSM is similar to Pandey et al.’s cur- beam-like structures using wavelet transform.
vature mode shape difference method in both using Using a small damage assumption and the Haar
mode shape curvature, but it does not require the wavelet transformation, a set of linear algebraic
baseline mode shape. Ratcliffe also suggested that equations is given by damage mechanics. With
the mode shape data from the fundamental mode the aid of singular value decomposition, the singu-
are most suited to this technique. Ratcliffe and larities in the damage mechanism were discarded.
Bagaria [76] successfully applied the GSM to Finally, the desired DI was reconstructed using the
locate a delamination in a composite beam. A DI pseudo-inverse solution. The performance of the
defined as the square of the difference between the proposed method was compared with two existing
curvature and the gapped cubic was used to deter- NDE methods (the MSC method by Pandey et al.
mine the location and size of the damage. Ratcliffe [64] and the DIM by Stubbs et al. [62,63]) for an
[66] later extended the GSM to broadband operat- axially loaded beam without any special knowledge
ing shape curvature. Separate gapped cubic poly- about mass density and an applied axial force. The
nomials are calculated for each of the real and effect of random noise on the performance was
imaginary parts of the complex curvature examined. The proposed method was verified by

a finite element model of a clamped-pinned i-th mode, respectively; and m is the number of
pre-stressed concrete beam and by field test data measured bending modes. The curvature mode
on the I-40 Bridge over the RioGrande. The pro- shape can be obtained from a third order interpo-
posed method resolves some existing deficiencies of lation function of displacement mode shape.
the MSC method and the DI method, including the Then, assuming that the damage indices at
mode selection problem, the singularity problem, different sub-regions is a normally distributed
the axial force consideration, and the estimation of random variable, a normalized damage index
the absolute severity of damage. Despite its strong (NDI) Z at sub-region j can be obtained using
features, the proposed method still has at least
j
the following two weaknesses. First, a dense mea- Zj ¼ , ð12Þ
surement of grid is needed for good accuracy.
Second, the accurate extraction of the mode where and r represent the mean and standard
shapes is considered to be a prerequisite. deviation of the damage indices j, respectively.
Usually, a damage detection criterion can be set
2.3.3 Modal Strain Energy-based Methods as Zj (NDI) larger than 2.
Another category of widely used damage identifi- Cornwell et al. [81] generalized the DI method
cation methods is modal strain energy method. for 1D beam-type structure into plate-type struc-
These methods use fractional modal strain energy tures. Shi and Law and Shi et al.[61,82] presented
change (MSEC) for damage detection. For a damage localization method for beam, truss or
beam-type or plate-type structures, the modal frame type structures based on the modal strain
strain energy can be directly related to (and usually energy change. The MSEC at the element level is
derived from) mode shape curvatures. Hence, the suggested as an indicator for damage localization.
modal strain energy-based method can also be con- Law et al. [83] applied this SEM to detect the
sidered as a special case of mode shape damage location in a structure with incomplete
curvature-based method in the context of and noisy measured modal data. The method con-
beam-type or plate-type structures. sists of three stages: expansion of the measured
Stubbs and Kim and Stubbs et al. [62,63] devel- mode shapes, localization of the damage domain
oped a DIM based on the change in modal strain using the elemental energy quotient difference,
energy. This method assumes that if the damage is and damage quantification based on sensitivity of
primarily located at a single sub-region, then the the modal frequency. The complete procedure of
fractional strain energy will remain relatively con- analysis is evaluated by several damage case studies
stant in sub-regions. For beam-type structures, the of a finite element model of a European Space
bending stiffness EI is assumed to be essentially Agency structure and verified by a laboratory
constant over the length of the beam for both the experiment on a two-story steel plane frame struc-
undamaged and damaged modes. The DI at ture. Results indicate that this method is effective
sub-region j can be obtained by and practical in detecting and quantifying single or
Pm multiple damages in the structure. Au et al. [84]
i¼1 Fij þ 1 followed a similar approach as proposed by Law
j ¼ Pm et al. [83] in damage detection using incomplete
i¼1 Fij þ 1
Pm hR 2 R L 2 R L 2
i and noisy modal test data; but they adopted a
i¼1 j i dx þ 0 i dx 0 ð i Þ dx micro-genetic algorithm in damage quantification
¼ P h R RL R i, stage instead of the traditional iterative algorithm
m 2 2 L 2
i¼1 j ði Þ dx þ 0 ði Þ dx 0 i dx based on minimum rank update.
ð11Þ
2.4 Other Methods based on Modal
where Fij and Fij are the fractional strain energy of
Parameters
undamaged and damaged beam for the i-th mode
at sub-region j; i and i are the curvature mode Topole and Stubbs [85] presented for the first
shapes of undamaged and damaged beam for the time a damage detection method to locate and size

structural damage from measured natural frequen- on the singular value decomposition (SVD) is
cies and mode shapes of the damaged structure most effective. The method has been verified by a
without a priori knowledge of the modal character- simple beam and a continuous beam numerical
istics of a baseline structure. However, its applica- model with numbers of simulated damage scenar-
tion is limited by several requirements which are ios. The method is further verified by a laboratory
difficult to meet in real experiment, such as the experiment of a reinforced concrete beam.
knowledge of baseline mass and stiffness, measure- Wong et al. [91] developed a general order per-
ment of mode shapes from all degrees of freedom turbation method involving multiple perturbation
(DOFs) and many modes. Since then, in order to parameters for eigenvalue problems with changes
find effective and accurate damage identification in the stiffness parameters. The perturbation
methods, extensive research effort has also been method is then used iteratively with an optimiza-
put into damage identification methods utilizing tion method to identify the stiffness parameters of
both the mode shapes and frequencies or modal structures. The generalized inverse method is used
parameters derived from mode shapes and fre- efficiently with the first order perturbations, and
quencies, such as modal flexibility and ULS. the gradient and quasi-Newton methods are used
with the higher order perturbations.
2.4.1 Modal Flexibility-based Methods Pandey Rahai et al. [92] presents a finite element-based
and Biswas [86] suggested using changes in modal approach for damage detection in structures utiliz-
flexibility for damage detection. Zhang and Aktan ing incomplete measured mode shapes and natural
[39] studied the modal flexibility and its derivative frequencies. Mode shapes of a structure are char-
ULS. It is easy to note that the ULS is essentially a acterized as a function of structural stiffness para-
weighted average of mode shapes. Consequently, it meters. More equations were obtained using
is less sensitive to noise compared to mode shapes. elemental damage equation which requires com-
Wu and Law [87,88] applied the ULS curvature to plete mode shapes. This drawback is resolved by
plate structures for damage localization and quan- presenting the mode shape equations and dividing
tification. It is found that the ULS curvature is sen- the structural DOFs to measured and unmeasured
sitive to the presence of local damages, even with parts. The nonlinear optimization problem is then
truncated, incomplete, and noisy measurements. solved by the sequential quadratic programming
Wang and Qiao [38] modified the GSM proposed (SQP) algorithm. Monte Carlo simulation is
by Ratcliff [75] to localize delamination in a com- applied to study the sensitivity of this method to
posite beam. A simplified gapped-smoothing (SGS) noise in measured modal displacements.
method using a fourth order polynomial is consid- Since the ill-posed nature of this type of optimi-
ered to fit the ULS curve of a damaged beam, and zation problems, soft optimization algorithms, such
the difference between the polynomial and ULS as genetic algorithm, simulation annealing or NN
curves is squared to obtain the damage index. algorithm, are considered useful tools for solving
these problems. Ruotolo and Surace [93] utilized
2.4.2 Optimization Algorithm-based Methods genetic algorithm to solve the optimization prob-
There are also several methods that formulate the lem. The objective function is formulated by intro-
damage identification method as an optimization ducing terms related to global damage and the
problem and use multiple modal parameters as dynamic behavior of the structure, i.e., natural fre-
objective function. quencies, mode shapes and modal curvature.
Ren and De Roeck [89,90] proposed a damage The damage assessment technique has been applied
identification technique from the finite element to both the simulated and experimental data related
model using frequencies and mode shape change. to cantilevered steel beams, each one with a differ-
The element damage equations have been estab- ent damage scenario. It is demonstrated that this
lished through the eigenvalue equations that char- method can detect the presence of damage and esti-
acterize the dynamic behavior. Several solution mate both the crack positions and sizes with satis-
techniques are discussed and compared. factory precision. The problems related to the
The results show that the SVD-R method based tuning of the genetic search and to the virgin state

calibration of the model are also discussed. Hao and damage indicator, called mode shape area index,
Xia [94] applied a genetic algorithm with real based on changes of mode shapes, was developed
number encoding to minimize the objective func- and found as a sensitive damage detection
tion, in which three criteria are considered: the fre- approach. Damage detection or localization via
quency changes, the mode shape changes, and a changes of the flexibility matrix performed better
combination of the two. A laboratory tested canti- than natural frequencies or mode shapes alone.
lever beam and a frame structure were used to verify Zhou et al. [69] conducted a laboratory experi-
the proposed technique. The algorithm did not ment and finite element analysis study to evaluate
require an accurate analytical model and gave five different vibration-based damage detection
better damage detection results for the beam than methods in detection and localization of small-scale
the conventional optimization method. damage on the deck slab of a simply supported
bridge. The five methods include three curvature-
2.5 Literature on Comparative Study based (i.e., MSC method, SEM and flexibility cur-
vature method) and two other methods (i.e., mode
Farrar and Jaurequi [95,96] conducted a com- shape method and flexibility method). The mode
prehensive experimental and numerical study of shape curvatures were obtained from a natural
different methods of damage detection (i.e., cubic spline interpolation of mode shape. The com-
damage index, curvature mode shape, change in parative study showed that the three curvature-
flexibility, change in ULS curvature, and change based methods tend to predict damage location at
in stiffness) on the interstate 40 bridge. the measurement point due to its piecewise linear
Ndambi et al. [97] presented a comparative distribution caused by cubic spline interpolation.
study of damage detection methods based on lab- They produced virtually identical results, displaying
oratory test of two cracked RC beams. The a maximum error in damage localization of approx-
damage detection methods based on eigenfrequen- imately half of the sensor spacing. The other two
cies, MAC, COMAC, flexibility matrices and methods performed better when the finite element
strain energy are evaluated. The results show that data were used, but were similar to the curvature
(1) The eigenfrequency evolutions can follow the methods when experimental data were used. The
damage severity but are not influenced by the crack performance of all five methods investigated
damage locations; (2) The MAC factors are, in declined when damage was located in a near-
contrast, less sensitive to crack damage compared support region. An increase in the number of mea-
with eigenfrequencies; (3) With the COMAC factor surement points led to a proportional improvement
evolution, it is possible to detect and localize in localization resolution for the three curvature-
damage in the tested RC beams but difficult to based methods, but increasing the number of
follow severity and spreading; (4) The change in modes considered from one to three did not signifi-
flexibility matrices allows also detection of the cantly improve the performance of the techniques.
crack damage in RC beams, but the damage local- Qiao et al. [40] evaluated three response-based
ization is difficult; and (5) Damage indices method damage detection algorithms (i.e., simplified GSM,
based on the strain energy appears to be more pre- GFD, and SEM) for composite laminated plates
cise than the others in damage localization, but the using the data acquired from two measurement sys-
difficulty remains when the damage is spread out tems (i.e., a surface-bonded piezoelectric (PVDF)
over a certain length of the RC beam. sensor system and a scanning laser vibrometer
Huth et al. [59] compared several (SLV) system). From the damage detection out-
response-based identification techniques based on comes, it is observed that the non-contact SLV
test data on a progressively damaged prestressed system proves to be more convenient and effective,
concrete bridge. Although the bridge was severely and it is capable of scanning high-density mode
cracked, natural frequencies as well as mode shape data over the entire plate specimens; while
shapes display only minor changes. However, the the PVDF system, in which the curvature mode
relative changes of mode shapes are larger than shapes are directly acquired, exhibits good sensitiv-
those observed for natural frequencies. A novel ity to damage, and it is capable of performing

onboard and in situ health monitoring. The damage analysis to generate the natural frequencies and
detection algorithms, like GSM, GFD and SEM mode shapes of the healthy and damaged beam.
based on the utilization of three consecutive mode For simplicity, the structure is assumed to be a
curvatures (modes 3–5) and resulting ULS curva- cantilevered beam of 0.01 m width 0.01 m
ture, successfully identify the presence, location, depth 1 m length. The material is assumed to
and relative size of delamination in the composite be isotropic with Young’s modulus E ¼ 1 GPa
plates. and Poisson’s ratio m ¼ 0.25. The beam is evenly
divided into 50 3-node quadratic beam elements
3 Comparative Study based on Finite B32. The mode shape data are extracted from the
Element Model 101 nodes along the beam length. At the damage
location, the Young’s modulus of the element(s) is
In this section, a comparative study based on reduced to simulate the damage-induced stiffness
finite element model was conducted to evaluate loss. The different damage severities are simulated
five extensively used damage detection algorithms by different stiffness reductions in the damaged
for beam-type structures. The five damage detec- element, i.e., when damage severity is a, the
tion algorithms considered are (1) frequency-based Young’s modulus of the damaged element E0 is
SDI method [21], (2) mode shape-based GFD set to (1 a)E.
method [37], (3) curvature-based MSC method
[64], (4) GSM [75], and (5) strain energy-based
3.1 Single Damage Detection
DIM [63]. Although some of these methods are
eligible for both damage localization and identifi- First, a single damage with three different seve-
cation, only its damage localization ability will be rities is considered. The damage is assumed to be at
investigated here. For the SDI, MSC and DIM the 20th element which locates 0.38–0.40 m away
methods, the modal parameters from the first five from the clamped end (x ¼ 0). The simulated nat-
modes are adopted. For the GFD and GSM meth- ural frequency change ratio and damage severity
ods, only the fundamental mode shape data are relationship in the first five modes is given in
used as suggested in the literature [37,75]. The Table 2. The nature frequency change ratio sug-
characteristics of these five methods are summar- gests that mode 5 should be the most sensitive
ized in Table 1. mode to this specific damage.
The commercial FE analysis package Since the damage index/indicator from
ABAQUS is used to perform an eigenvalue each algorithm has its unique physical meaning,
Table 1 Characteristics of five comparative damage detection algorithms
Basic
Algorithms Type Modal parameters requirement assumption DI
SDI Model-based Measured natural frequencies in Single Defined at element/
healthy and damaged beam, or damage sub-region
theoretical mode shape in healthy
beam model
GFD Response-based, Measured fundamental mode shape in – Defined at node/
damaged state damaged beam sensor
response only
MSC Response-based Measured mode shape curvatures in – Defined at node/
healthy and damaged beams sensor
GSM Response-based, Measured fundamental MSC in – Defined at node/
damaged state damaged beam sensor
response only
DIM Model-based Measured mode shape curvatures in Single Defined at element/
damaged beam, or theoretical mode damage sub-region
shape in healthy beam model

it is meaningless to compare them directly. method is not so sensitive to small damage and
For comparison, assuming that the DI at different may lead to a false negative indication of damage
location is a normally distributed random variable, in some cases of damage severity. The MSC
an NDI can be obtained using Equation (12). A method always gives the largest peak values, and
threshold value NDI ¼ 2 is usually used to detect its peak values are insensitive to damage severity.
damage, that is, an NDI larger than the threshold The SDI method gives peak values decreasing with
value indicates the location of potential damage. the damage severity, which indicates that this
It should also be noted that for the GFD method may gradually lose its effectiveness when
method [38], the DI is subjected to the choice of damage severity increases. It is also not surprising
scale parameter S (see Equation (8)). In this study, to see the GSM gives oscillating peak values since
investigation shows that an S value larger than 100 the NDI from the GSM should not be expected to
should be adopted in order to detect the damage correlate with damage severities when only
and that the NDI varies little when S is chosen damaged state data are used.
between 103 and 106. Hence, a constant scale
parameter S ¼ 1000 is chosen for all cases when 3.2 Multiple Damage Detection
the GFD algorithm is implemented.
A comparison of NDI for all the five algo- Second, two multiple damage cases with two
rithms in the case of a single damage with severity separate damage elements are considered. For the
a ¼ 0.1 are shown in Figure 1. For the SDI and first case, the multiple damages with the same
DIM methods with element-wise DIs other than severity a ¼ 0.1 are assumed to be at the 3rd and
node-wise indices, their indices are placed at the 20th elements which locate 0.04–0.06 m and 0.38–
midpoint of the corresponding element. It can be 0.40 m, respectively, away from the cantilever
seen that all five algorithms can detect and localize clamped end (x ¼ 0). For the second case, the mul-
the damage correctly. Further results show that tiple damages are assumed to be at the 20th and
these algorithms can successfully localize the 30th elements, which locate 0.38–0.40 m and 0.58–
damage in all three damage severity cases. It is 0.60 m, respectively, away from the cantilever
interesting to note that there is a small bump in clamped end (x ¼ 0).
the SDI curve around x ¼ 0.6 m, since due to the The comparison of NDI in two multiple
frequency-based nature of SDI method, this bump damage cases are shown in Figure 3. It shows
will become an equally sharp peak as the one at that MSC, GSM and DIM can localize the multi-
x ¼ 0.4 m if the boundary condition is symmetric. ple damages successfully. It is interesting to note
The peak values of NDI in three different cases of that the DIM method can correctly locate the mul-
damage severities are collected in Figure 2. tiple damages even though it is originally derived
It shows that all the methods give satisfactory from the single damage assumption. In the first
damage indices, except the GFD. The GFD attempt of using the GSM to detect the multiple
method gives relatively low NDI and drops damages, it shows that the GSM cannot localize
below zero when a ¼ 0.3, which indicates that this the multiple damages if the fundamental mode is
Table 2 Nature frequency change ratio in several damage severities
Natural frequency change ratio ¼ (fh fd)/fh
Damage Bending stiffness

severity, a ratio, (EI)d/(EI)h Mode 1 Mode 2 Mode 3 Mode 4 Mode 5
0.1 0.9 0.0010 0.0014 0.0011 0.0004 0.0022
0.2 0.8 0.0023 0.0032 0.0026 0.0007 0.0049
0.3 0.7 0.0038 0.0054 0.0043 0.0012 0.0083
0.5 0.5 0.0089 0.0125 0.0098 0.0027 0.0183
0.7 0.3 0.0204 0.0279 0.0214 0.0059 0.0385
0.9 0.1 0.0726 0.0888 0.0617 0.0175 0.0980

8
6
4
DIM
2
0
–2
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
8
6
(1st mode)
4
GSM
2
0
–2
–4
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
8
6
MSC
4
2
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
3
(1st mode)
2
GFD
1
0
–1
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
8
6
SDI
4
2
0
–2
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Location x (m)
Figure 1 NDI along beam length for five algorithms in the case of single damage (damage location x ¼ 0.38 0.40 m,
damage severity a ¼ 0.1).
10 adopted as suggested in the literature [75].

However, further investigation shows that the
higher modes actually work better than the funda-
8
mental modes in this case. Therefore, the NDI
Normalized damage indices
curve of the GSM shown in Figure 3 is evaluated

6
from the fifth mode shape curvature. However,
using the fifth mode expands the irregular bump
SDI
4 area, which leads to inaccuracy in damage localiza-
GFD
MSC
tion. For different modes, there is a trade-off
2 GSM between the sensitivity and the damage localization
DIM accuracy in the GSM method. The GFD method,
0 on the other hand, can localize damages in the first
case, but it can only locate one of the two damages
0.1 0.2 0.3 in the second case. It can be seen from the NDI
Damage severity a curve that the sensitivity of the GFD to damage
Figure 2 Peak values of NDI for five algorithms with three highly depends on the location of damage. The
different single-damage severities (a ¼ 0.1, 0.2, and 0.3). bump in the NDI curve is more obvious in the

6
4
DIM
2
0
–2
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
(5th mode) 4
2
GSM
0
–2
–4
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
6
4
MSC
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
3
2 Damage at element 3/20
(1st mode)
GFD
1 Damage at element 20/30
0
–1
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
6
4
SDI
2
0
–2
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Location x (m)
Figure 3 NDI along beam length for five algorithms in two multiple-damage cases (damage case 1: located at
3rd (0.04–0.06 m) and 20th (0.38–0.40 m) elements, and damage case 2: 20th (0.38–0.40 m) and 30th (0.58–0.60 m)
elements).
3rd element than in the 20th element; while the frequency-based SDI method is that in a cantile-
bump in 30th element can barely be noticed. vered beam the natural frequency change is more
Hence, it is more effective for the GFD to detect sensitive to damage near the cantilevered end than
the damage near the cantilever end than that near that near the free end if their damage severities are
the free end. Although the frequency-based SDI similar. Due to its single damage assumption and
method is derived form a single damage assump- frequency-based nature, the SDI method tends to
tion and designed mainly for single damage detec- predict the damage location at the primary damage
tion, surprisingly it can still approximately locate which dominates the damage-induced frequency
both damages in the second case with both the change. To support this conclusion, it can also be
peaks shift toward each other a little. But it is noticed in the second case that the peak value at
obviously not a reliable algorithm for multiple the 20th element is higher than that of the 30th
damage detection, e.g., it cannot detect the element. In summary, MSC and GSM work well
damage at the 20th element in the first case. for localizing multiple damages if the mode is care-
The reason of the failure in the first case for the fully chosen. DIM also works in well separated

multiple damage case, even though using it for cantilevered end (x ¼ 0). The damage severity of
multiple damage detection violates its single each damaged element is assumed to be a ¼ 0.1.
damage assumption; while GFD and SDI are not A comparison of NDI for the five algorithms in
suitable for multiple damage localization. the large-area damage case is shown in Figure 4.
It shows that both the MSC and DIM are capable
3.3 Large-area Damage Detection of localizing the large-area damage successfully;
GFD method is not very sensitive to the large-area
After the single damage and multiple damage damage although the slight irregularity in its NDI
cases, a large-area damage case is considered. curve around the damaged region can still be
Large-area damage can be caused by impact noticed. The GSM and SDI methods show only
damage or delamination in composite laminates. one major peak in their curves which indicates the
In this study, the damage is assumed to be at the approximate location of damage, but they fail to
region from the 20th element to 23rd element identify the exact location and the length of the
which locates 0.38–0.46 m away from the damage.
8
6
4
2
DIM
0
–2
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
8
6
(5th mode)
4
GSM
2
0
–2
–4
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
8
6
4
MSC
2
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
4
(1st mode)
2
GFD
–2
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
8
6
4
SDI
2
0
–2
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Location x (m)
Figure 4 NDI along beam length for five algorithms in a large-area damage case (damage located from the 20th to
23rd elements (0.38–0.46 m)).

3.4 Evaluation of Measurement Noise Effect The contaminated mode shapes are repre-
sented as
In experimental modal analysis, measurement
noises are inevitable. To evaluate the robustness of ’0ij ¼ ’ij þ rij ’rms,j ð13Þ
damage detection algorithms under measurement
noise, the simulated mode shape data are contami- where ’0ij and ’ij are the displacement mode shape
nated with certain level of artificial random noise or strain mode shape components of the j-th mode
to generate ‘measured’ mode shapes. The original at the i-th DOF with noise and without noise,
simulated mode shapes are extracted from a respectively; r is the normally distributed random
single-damage case with a damage severity a ¼ 0.5 variables with a mean equal to zero and a variance
at the 20th element. Only the mode shape-based equal to 1; q is the random noise level; and ’rms,j is
and curvature-based methods are studied here the root-mean-square of the j-th displacement
for comparison. The frequency-based SDI mode shape or strain mode shape.
method is excluded because it is difficult to find a When the mode shape curvatures are derived
common ground to compare the level of noise from from the displacement mode shape, the comparison
frequency and mode shape measurement. The of damage detection resulted from the different
mode shape curvatures are obtained from two algorithms are shown in Figure 5. It should
methods: derived from the displacement mode be noted that for the GSM and GFD methods, it
shape data or directly measured from the strain is difficult to detect damage using the fundamental
mode shape. mode in this case. Since the most sensitive mode to
8
6
4
DIM
2
0
–2
–4
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
8
6
(5th mode)
GSM
4
2
0
–2
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
8
6
MSC
4
2
0
–2
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
r =0.0001
8
r =0.001
(5th mode)
6
r =0.01
GFD
4
2
0
–2
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Location x (m)
Figure 5 Measurement noise effect on damage detection algorithms (Case 1: curvature derived from the displacement
mode shape; single damage at 0.38–0.40 m and a ¼ 0.5).

this damage location is the 5th mode as suggested by the directly measured curvature mode shape is
the frequency change ratio in Table 2, the 5th mode used, the GSM method can correctly localize
shape instead of the first mode shape is used. When damage even under level of noise q ¼ 0.1; the
q ¼ 0.0001, all four methods can localize the damage MSC method can correctly localize damage under
correctly. When q ¼ 0.001, only the GFD method level of noise q ¼ 0.2; while the DIM method can
is capable of localizing the damage; while both even localize damage under level of noise q ¼ 0.4.
the GSM and DIM methods give the false indica- All the three methods perform much better than
tion of damage location; the MSC method indicates the mode shape-based GFD method when the
the correct damage location but also gives a processing-induced error is eliminated. Therefore,
false indication with the small peak at x ¼ 0.18 m. for the curvature-based methods, if at the similar
When q ¼ 0.01, all the methods fail to correctly noise level, the directly measured curvature mode
localize damage. Although the GFD shows a peak shape has a great advantage over the derived cur-
at the damage location, it also gives a false positive vature mode shape from laboratory or in situ
indication of damage at x ¼ 0.62 m. This result experiments.
shows that the GFD method is more robust
than the MSC method; while the MSC method
is better than the other two remaining methods 3.5 Evaluation of Sensor Spacing Effect
(i.e., GSM and DIM) under measurement noise. In experimental modal analysis, the mode
The main reason is that the mode shape curvatures shapes usually can only be measured at a relatively
used in the MSC method are derived from the dis- small number of locations, especially for in situ
placement mode shape by the central difference
approximation (see Equation (9)), while the 8 r =0.01
damaged state curvatures used in the DIM are 6 r =0.1
derived by the cubic spline interpolation. For the r =0.2
4
DIM
GSM method, the damaged state curvatures are r =0.4

2
derived by the central difference approximation,
0
while the baseline curvature derivation involves
cubic polynomial interpolation. This derivation –2
process, no matter by the central difference approx- 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
8
imation or the cubic interpolation, will greatly
6
magnify the measurement error. This processing
(5th mode)
4
error weakens the robustness of the curvature-
GSM
based algorithms under measurement noise. The 2
results also show that the cubic interpolation will 0
introduce larger processing error than the finite dif- –2

ference approximation, which makes the GSM and 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
DIM methods very sensitive to measurement noise.
8
To manifest this conclusion, a similar damage
6
detection case is studied, in which the mode shape
MSC
curvatures are directly obtained from the simulated 4

strain mode shape and then contaminated by arti- 2
ficial measurement noise to generate the ‘mea- 0
sured’ strain mode shape. Again, the original
–2
simulated mode shape is extracted from a 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
single-damage case with a damage severity a ¼ 0.5 Location x (m)
at the 20th element (0.38–0.40 m from the cantile- Figure 6 Measurement noise effect on damage detec-
ver clamped end of x ¼ 0). The comparison of tion algorithms (Case 2: curvature directly measured from
curvature-based damage detection algorithms are the strain mode shape; single damage at 0.38–0.40 m,
shown in Figure 6. The result shows that when a ¼ 0.5).

experiments. The sparse distribution of sensors will When the derived curvatures are used, the
often pose difficulties for damage detection algo- results are illustrated in Figure 7. The GFD
rithms to detect and localize the damage. Even cannot detect the damage when the sensor spacing
though novel measurement systems, such as SLV, reaches s ¼ 0.05. When s ¼ 0.10, the DIM method
can be employed to obtain high-density mode gives a false indication of damage at the cantile-
shape data in laboratory-scale experiments, the vered end of beam, and it tends to give periodical
robustness of damage detection algorithms under false indication along the beam; the MSC and GSM
limited measured data points is still of interest for can barely detect the damage, and the GSM also
their practicality for in situ experiment. tends to give ambiguous prediction at x ¼ 0.6 m.
To simulate the data acquisition process, the Another feature can also be noticed as pointed out
‘measured’ mode shape data are extracted from by Zhou et al. [69], that the GSM and MSC tends to
the FE-generated mode shapes at uniformly spaced predict damage location at the sensor location due
locations. The effect of sensor spacing on the algo- to the piecewise nature of derived curvatures and
rithms is studied. The mode shape curvatures are can only predict the damage location with the accu-
again obtained from two methods: derived from racy of half of the sensor spacing.
the displacement mode shape data or directly mea- When the directly measured curvatures are
sured from the strain mode shape. Three cases with used, the results are illustrated in Figure 8. The
sensor spacing of s ¼ 0.02, 0.05, and 0.10 m are results show that all three curvature-based meth-
studied. ods (i.e., the GSM, MSC, and DIM) can correctly
s=0.02 m
8
6 s=0.05 m
4 s=0.1 m
DIM
2
0
–2
–4
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
4
(5th mode)
2
GSM
–2
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
4
MSC
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
4
(5th mode)
GFD
–2
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Location x (m)
Figure 7 Sensor spacing effect on damage detection algorithms (Case 1: curvature derived from the displacement
mode shape; single damage at 0.38–0.40 m, a ¼ 0.5).

localize the damage without any false indication. and the accuracy of damage location prediction
For the GSM method, all the five modes can be will decrease accordingly.
used independently to localize the damage.
Therefore, for the curvature-based methods, the 3.6 Summary on Comparative Study
directly measured curvatures has its advantage
Based on the above case illustrations and dis-
over the derived curvatures in the case of large
cussions, the capabilities of the aforementioned
sensor spacing. With the increase of sensor spac-
five damage detection algorithms in term of their
ing, the predicted damage zone tends to spread,
single and multiple damage identification,
large-area damage identification, degree of noise
8 immunity, and tolerance for sensor spacing are
6 summarized in Table 3.
4
4 Concluding Remarks
DIM
2
0
This article provides a summary review on
–2 modal parameters-based damage identification
80.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 algorithms for beam-type or plate-type structures.
6 These methods are categorized as natural
4 frequency-based methods, mode shape-based
2
methods, curvature mode shape-based methods
GSM
and methods using both mode shape and

0
frequencies.
–2
The frequency change-based damage identifi-
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 cation method can be successfully applied to local-
8 ize and quantify damage in simple structures with
s =0.02
small damage (typically, a slender beam-type struc-
6 s =0.05
s = 0.1
ture with an artificially induced crack) in a con-
MSC
4 trolled laboratory condition. However, due to its

2 inherent limitations, in general it is not reliable for
0 damage detection in real complex structures or
multiple/severe damage cases. Most mode
–2
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 shape-based and curvature-based methods only
Location x (m) focus on damage localization. The direct use of
mode shape change can only roughly localize the
Figure 8 Sensor spacing effect on damage detection
algorithms (Case 2: curvature directly measured from damage. In order to precisely localize the damage,
the strain mode shape; single damage at 0.38–0.40 m, the mode shape-based methods have to rely on
a ¼ 0.5). optimization algorithms or signal processing
Table 3 Capabilities of five comparative damage detection algorithms
Single Multiple Large-area Sensor

damage damage damage Noise spacing
Algorithm detection detection detection immunity tolerance
SDI Yes No No N/A N/A
GFD Yes No No Excellent Fair
MSC Yes Yes Yes Good (Good) Good (Good)
GSM Yes Yes No Fair (Good) Good (Good)
DIM Yes Yes Yes Fair (Excellent) Fair (Excellent)
Note: The results from directly measured curvature mode shape are shown in parentheses.

techniques. The curvature mode shape-based directly measured curvature has great advan-
methods are in general a very effective type of tage over derived curvatures in the case of
damage localization algorithms. A curvature large sensor spacing. If at the same noise
mode shape-based algorithm, either using direct level, the directly measured curvatures are pre-
change in curvature or applying signal processing ferred than the derived curvatures for damage
techniques on curvature, usually can successfully detection in all studied cases.
localize the damage. (5) In general, the GFD method is not a very sen-
Furthermore, a comparative study based on sitive damage detection method, and it
the finite element model is conducted to evaluate requires high density mode shape measure-
five relatively common damage detection algo- ment. The MSC and DIM methods are rela-
rithms for beam-type structures. The five studied tively robust in high measurement noise and
damage detection algorithms are frequency-based large sensor spacing condition. Although the
SDI method, mode shape-based GFD Method, GSM is not as robust as the MSC and DIM
Curvature-based MSC method, GSM, and Strain methods, it is still a good choice for damage
energy-based DIM. The following conclusions are detection algorithm when only the damaged
drawn from the comparisons among the five state structures are available as in most in
damage detection algorithms in terms of single situ experiment.
versus multiple damages, damage area, methods
of acquiring mode shape curvatures, noise, and In summary, the comprehensive review in
sensor spacing: vibration-based damage identification methods for
simple beam- or plate-type structures provide read-
(1) In noise-free and small sensor spacing condi- ers a vibration feature-based categorization of the
tion, all five methods are capable of localizing methods as well as pros and cons of different modal
single damage; while only the MSC, GSM and parameters-based damage detection algorithms;
DIM methods can localize multiple damages, while the comparative study of five common
and only the MSC and DIM can localize the damage detection algorithms (signal processing
large-area damage. methods) demonstrates the advantages and validity
(2) The fundamental mode may not always be the of some of the algorithms and guides researchers
most effective mode for damage detection in and practitioners when implementing these damage
the GFD and GSM methods. A higher mode algorithms in damage identification process.
may be more sensitive in some cases. The nat- There are still many questions to be addressed
ural frequency change ratio can be a good indi- in this area. In particular, the following research
cator of sensitivity of certain mode to the topics of interest need immediate attention:
damage.
(3) When the mode shape curvatures are derived (1) Development of robust multiple damage iden-
from the displacement mode shape, the mode tification methods under environmental noise.
shape-based GFD method is more robust Among all available techniques, the pattern
under the measurement noise than the recognition technique has shown its great
curvature-based methods due to their potential in building a robust statistical
processing-induced errors. However, when cur- model to distinguish multiple damages in dif-
vatures are directly measured from the strain ferent locations and/or different types under
mode shape, the curvature mode shape-based environmental noise.
methods are much more robust and promising (2) Development of quantification techniques for
than the mode shape-based GFD method. damage magnitude. The localization of the
(4) When the mode shape curvatures are derived damage is primarily studied in the literature;
from the displacement mode shape, the MSC while the quantification of the damage, as the
and GSM methods are more robust than follow-up step in damage identification, is still
others in the large sensor spacing case. For in a relatively immature stage, worth immedi-
the curvature mode shape-based methods, the ate attention and investigation;

(3) Development of viable damage identification 7. Kasper, D.G., Swanson, D.C. and Reichard, K.M.
methods for 2D plates. There are only limited (2008). Higher-frequency wavenumber shift and fre-
studies [37,53,55,78,81,98,99] available for quency shift in a cracked, vibrating beam. Journal of
plates, with most of them being generalized Sound and Vibration, 312, 1–18.
8. Adams, R.D., Cawley, P., Pye, C.J. and Stone, B.J.
from the 1D algorithms. It should be men-
(1978). A vibration technique for non-destructively
tioned that the methods using novel signal pro- assessing the integrity of structures. Journal of
cessing techniques, such as 2D wavelet Mechanical Engineering Science, 20, 93–100.
transform as demonstrated by the authors 9. Salawu, O.S. (1997). Detection of structural damage
[55], show their potential to be a powerful through changes in frequency: A review. Engineering
tool in damage identification of plates and Structures, 19, 718–723.
thus demand further research. 10. Messina, A., Williams, E.J. and Contursi, T. (1998).
Structural damage detection by a sensitivity and
statistical-based method. Journal of Sound and
Vibration, 216, 791–808.
Acknowledgements 11. Liang, R.Y., Choy, F.K. and Hu, J.L. (1991).
Detection of cracks in beam structures using measure-
This study is partially supported by the Alaska University ments of natural frequencies. Journal of the Franklin
Transportation Center (AUTC)/US Department of Institute-Engineering and Applied Mathematics, 328,
Transportation (Contract/Grant No.: DTRT06-G-0011) 505–518.
on Smart FRP Composite Sandwich Bridge Decks in 12. Nandwana, B.P. and Maiti, S.K. (1997). Detection of
Cold Regions and the NASA STTR (Proposal No.: the location and size of a crack in stepped cantilever
08-1 T2.01-9951) on Structural Health Monitoring with beams based on measurements of natural frequencies.
Fiber Bragg Grating and Piezo Arrays. Journal of Sound and Vibration, 203, 435–446.
13. Chaudhari, T.D. and Maiti, S.K. (2000). A study of
vibration of geometrically segmented beams with and
References without crack. International Journal of Solids and
Structures, 37, 761–779.
1. Doebling, S.W., Farrar, C.R., Prime, M.B. and Shevitz, 14. Chinchalkar, S. (2001). Determination of crack location
D.W. (1996). Damage Identification and Health in beams using natural frequencies. Journal of Sound
Monitoring of Structural and Mechanical Systems from and Vibration, 247, 417–429.
Changes in their Vibration Characteristics: A Literature 15. Lele, S.P. and Maiti, S.K. (2002). Modelling of trans-
Review. Los Alamos National Laboratory report, verse vibration of short beams for crack detection and
(LA-13070-MS). measurement of crack extension. Journal of Sound and
2. Sohn, H., Farrar, C., Hunter, N. and Worden, K. (2003). Vibration, 257, 559–583.
A Review of Structural Health Monitoring Literature: 16. Patil, D.P. and Maiti, S.K. (2003). Detection of multi-
1996–2001. Los Alamos National Laboratory report, ple cracks using frequency measurements. Engineering
(LA-13976-MS). Fracture Mechanics, 70, 1553–1572.
3. Carden, E.P. and Fanning, P. (2004). Vibration based 17. Murigendrappa, S.M., Maiti, S.K. and Srirangarajan,
condition monitoring: A review. Structural Health H.R. (2004). Experimental and theoretical study on
Monitoring, 3, 355–377. crack detection in pipes filled with fluid. Journal of
4. Gudmundson, P. (1982). Eigenfrequency changes of Sound and Vibration, 270, 1013–1032.
structures due to cracks, notches or other geometrical 18. Murigendrappa, S.M., Maiti, S.K. and Srirangarajan,
changes. Journal of the Mechanics and Physics of H.R. (2004). Frequency-based experimental and
Solids, 30, 339–353. theoretical identification of multiple cracks in straight
5. Liang, R.Y., Hu, J.L. and Choy, F. (1992). pipes filled with fluid. NDT & E International, 37,
Theoretical-study of crack-induced eigenfrequency 431–438.
changes on beam structures. Journal of Engineering 19. Morassi, A. (2001). Identification of a crack in a rod
Mechanics-ASCE, 118, 384–396. based on changes in a pair of natural frequencies.
6. Morassi, A. (1993). Crack-induced changes in eigenpara- Journal of Sound and Vibration, 242, 577–596.
meters of beam structures. Journal of Engineering 20. Morassi, A. and Rollo, M. (2001). Identification of two
Mechanics-ASCE, 119, 1798–1803. cracks in a simply supported beam from minimal

frequency measurements. Journal of Vibration and 34. Abdo, M.A.B. and Hori, M. (2002). A numerical study
Control, 7, 729–739. of structural damage detection using changes in the
21. Kim, J.T. and Stubbs, N. (2003). Crack detection in rotation of mode shapes. Journal of Sound and
beam-type structures using frequency data. Journal of Vibration, 251, 227–239.
Sound and Vibration, 259, 145–160. 35. Hemez, F.M. (2004). Uncertainty quantification and
22. Zhong, S., Yadiji, S.O. and Ding, K. (2008). the verification and validation of computational
Response-only method for damage detection of models. In: Damage Prognosis for Aerospace, Civil and
beam-like structures using high accuracy frequencies Mechanical Systems, Chichester, England, Hoboken,
with auxiliary mass spatial probing. Journal of Sound NJ: Wiley.
and Vibration, 311, 1075–1099. 36. Hadjileontiadis, L.J., Douka, E. and Trochidis, A.
23. Maity, D. and Tripathy, R.R. (2005). Damage assess- (2005). Fractal dimension analysis for crack identifica-
ment of structures from changes in natural frequencies tion in beam structures. Mechanical Systems and Signal
using genetic algorithm. Structural Engineering and Processing, 19, 659–674.
Mechanics, 19, 21–42. 37. Hadjileontiadis, L.J. and Douka, E. (2007). Crack
24. Pawar, P.M. and Ganguli, R. (2005). Matrix crack detection in plates using fractal dimension.
detection in thin-walled composite beam using genetic Engineering Structures, 29, 1612–1625.
fuzzy system. Journal of Intelligent Material Systems 38. Wang, J. and Qiao, P.Z. (2007). Improved damage
and Structures, 16, 395–409. detection for beam-type structures using a uniform
25. Ni, Y.Q., Hua, X.G., Fan, K.Q. and Ko, J.M. (2005). load surface. Structural Health Monitoring-an Interna-
Correlating modal properties with temperature using tional Journal, 6, 99–110.
long-term monitoring data and support vector machine 39. Zhang, Z. and Aktan, A.E. (1998). Application
technique. Engineering Structures, 27, 1762–1773. of modal flexibility and its derivatives in structural
26. Kim, J.T., Park, J.H. and Lee, B.J. (2007). identification. Research in Nondestructive Evaluation,
Vibration-based damage monitoring in model 10, 43–61.
plate-girder bridges under uncertain temperature condi- 40. Qiao, P.Z., Lu, K., Lestari, W. and Wang, J. (2007).
tions. Engineering Structures, 29, 1354–1365. Curvature mode shape-based damage detection in com-
27. Dilena, M. and Morassi, A. (2004). The use of antire- posite laminated plates. Composite Structures, 80,
sonances for crack detection in beams. Journal of Sound 409–428.
and Vibration, 276, 195–214. 41. Qiao, P. and Cao, M. (2008). Waveform fractal dimen-
28. Jiang, L.J., Tang, J. and Wang, K.W. (2006). An sion for mode shape-based damage identification of
enhanced frequency-shift-based damage identification beam-type structures. International Journal of Solids
method using tunable piezoelectric transducer circuitry. and Structures, 45, 5946–5961.
Smart Materials & Structures, 15, 799–808. 42. Liew, K.M. and Wang, Q. (1998). Application of
29. Farrar, C.R. and James, G.H. (1997). System identifi- wavelet theory for crack identification in structures.
cation from ambient vibration measurements on a Journal of Engineering Mechanics-ASCE, 124, 152–157.
bridge. Journal of Sound and Vibration, 205, 1–18. 43. Wang, Q. and Deng, X.M. (1999). Damage detection
30. Shi, Z.Y., Law, S.S. and Zhang, L.M. (2000). Damage with spatial wavelets. International Journal of Solids and
localization by directly using incomplete mode shapes. Structures, 36, 3443–3468.
Journal of Engineering Mechanics-ASCE, 126, 656–660. 44. Quek, S.T., Wang, Q., Zhang, L. and Ang, K.K. (2001).
31. Lee, J.J., Lee, J.W., Yi, J.H., Yun, C.B. and Jung, H.Y. Sensitivity analysis of crack detection in beams by
(2005). Neural networks-based damage detection for bri- wavelet technique. International Journal of Mechanical
dges considering errors in baseline finite element models. Sciences, 43, 2899–2910.
Journal of Sound and Vibration, 280, 555–578. 45. Hong, J.-C., Kim, Y.Y., Lee, H.C. and Lee, Y.W.
32. Hu, C.S. and Afzal, M.T. (2006). A statistical algorithm (2002). Damage detection using the Lipschitz exponent
for comparing mode shapes of vibration testing before estimated by the wavelet transform: applications to
and after damage in timbers. Journal of Wood Science, vibration modes of a beam. International Journal of
52, 348–352. Solids and Structures, 39, 1803–1816.
33. Pawar, P.M., Reddy, K.V. and Ganguli, R. (2007). 46. Douka, E., Loutridis, S. and Trochidis, A. (2003).
Damage detection in beams using spatial Fourier ana- Crack identification in beams using wavelet analysis.
lysis and neural networks. Journal of Intelligent International Journal of Solids and Structures, 40,
Material Systems and Structures, 18, 347–359. 3557–3569.

47. Douka, E., Loutridis, S. and Trochidis, A. (2004). 61. Shi, Z.Y. and Law, S.S. (1998). Structural damage
Crack identification in plates using wavelet analysis. localization from modal strain energy change. Journal
Journal of Sound and Vibration, 270, 279–295. of Sound and Vibration, 218, 825–844.
48. Gentile, A. and Messina, A. (2003). On the continuous 62. Stubbs, N. and Kim, J.T. (1996). Damage localization
wavelet transforms applied to discrete vibrational data in structures without baseline modal parameters. AIAA
for detecting open cracks in damaged beams. Interna- Journal, 34, 1644–1649.
tional Journal of Solids and Structures, 40, 295–315. 63. Stubbs, N., Kim, J.T. and Farrar, C.R. (1995).
49. Zhong, S.C. and Oyadiji, S.O. (2007). Crack detection Field verification of a nondestructive damage localiza-
in simply supported beams without baseline modal tion and severity estimation algorithm. Proceedings of
parameters by stationary wavelet transform. Mechani- 13th International Modal Analysis Conference, 1,
cal Systems and Signal Processing, 21, 1853–1884. 210–218.
50. Chang, C.C. and Chen, L.W. (2005). Detection of the 64. Pandey, A.K., Biswas, M. and Samman, M.M. (1991).
location and size of cracks in the multiple cracked beam Damage detection from changes in curvature
by spatial wavelet based approach. Mechanical Systems mode shapes. Journal of Sound and Vibration, 145,
and Signal Processing, 19, 139–155. 321–332.
51. Chang, C.C. and Chen, L.W. (2004). Damage detection 65. Abdel Wahab, M.M. and De Roeck, G. (1999).
of a rectangular plate by spatial wavelet based Damage detection in bridges using modal curvatures:
approach. Applied Acoustics, 65, 819–832. Application to a real damage scenario. Journal of
52. Poudel, U.P., Fu, G.K. and Ye, H. (2007). Wavelet Sound and Vibration, 226, 217–235.
transformation of mode shape difference function for 66. Ratcliffe, C.P. (2000). A frequency and curvature based
structural damage location identification. Earthquake experimental method for locating damage in structures.
Engineering & Structural Dynamics, 36, 1089–1107. Journal of Vibration and Acoustics-Transactions of the
53. Rucka, M. and Wilde, K. (2006). Application of con- Asme, 122, 324–329.
tinuous wavelet transform in vibration based damage 67. Just-Agosto, F., Shafiq, B. and Serrano, D. (2007).
detection method for beams and plates. Journal of Development of a damage detection scheme applicable
Sound and Vibration, 297, 536–550. to sandwich composites. Journal of Sandwich Structures
54. Cao, M. and Qiao, P. (2008). Integrated wavelet trans- & Materials, 9, 343–363.
form and its application to vibration mode shapes for 68. Swamidas, A.S.J. and Chen, Y. (1995). Monitoring
the damage detection of beam-type structures. Smart crack-growth through change of modal parameters.
Materials and Structures, 17, 055014. Journal of Sound and Vibration, 186, 325–343.
55. Fan, W. and Qiao, P.Z. (2009). A 2-D continuous 69. Zhou, Z.G., Wegner, L.D. and Sparling, B.F. (2007).
wavelet transform of mode shape data for damage Vibration-based detection of small-scale damage on a
detection of plate structures. International Journal of bridge deck. Journal of Structural Engineering-ASCE,
Solids and Structures, 46, 4379–4395. 133, 1257–1267.
56. Antoine, J.P., Murenzi, R., Vandergheynst, P. and Ali, 70. Li, H.J., Li, H., He, C., Ji, J., Wang, H. and Hao, C.
S.T. (2004). Two-Dimensional Wavelets and Their (2005). Crack damage detection in beam-like structures
Relatives, Cambridge: Cambridge University Press. using RBF neural networks with experimental valida-
57. Khan, A.Z., Stanbridge, A.B. and Ewins, D.J. (1999). tion. International Journal of Innovative Computing
Detecting damage in vibrating structures with a scan- Information and Control, 1, 625–634.
ning LDV. Optics and Lasers in Engineering, 32, 71. Hamey, C.S., Lestari, W., Qiao, P. and Song, G. (2004).
583–592. Experimental damage identification of carbon/epoxy
58. Salawu, O.S. and Williams, C. (1994). Damage location composite beams using curvature mode shapes.
using vibration mode shapes. International Modal Structural Health Monitoring, 3, 333–353.
Analysis Conference, 933–939. 72. Lestari, W. and Qiao, P.Z. (2005). Damage detection of
59. Huth, O., Feltrin, G., Maeck, J., Kilic, N. and fiber-reinforced polymer honeycomb sandwich beams.
Motavalli, M. (2005). Damage identification using Composite Structures, 67, 365–373.
modal data: Experiences on a prestressed concrete 73. Qiao, P.H., Lestari, W., Shah, M.C. and Wang, J.
bridge. Journal of Structural Engineering-ASCE, 131, (2007). Dynamics-based damage detection of composite
1898–1910. laminated beams using contact and noncontact mea-
60. Kim, J.T., Ryu, Y.-S., Choi, H.-M. and Stubbs, N. surement systems. Journal of Composite Materials, 41,
(2003). Damage identification in beam-type structures: 1217–1252.
frequency-based method vs mode-shape-based method. 74. Lestari, W., Qiao, P.Z. and Hanagud, S. (2007).
Engineering Structures, 25, 57–67. Curvature mode shape-based damage assessment of

carbon/epoxy composite beams. Journal of Intelligent 87. Wu, D. and Law, S.S. (2004). Damage localization in
Material Systems and Structures, 18, 189–208. plate structures from uniform load surface curvature.
75. Ratcliffe, C.P. (1997). Damage detection using a mod- Journal of Sound and Vibration, 276, 227–244.
ified laplacian operator on mode shape data. Journal of 88. Wu, D. and Law, S.S. (2005). Sensitivity of uniform
Sound and Vibration, 204, 505–517. load surface curvature for damage identification in
76. Ratcliffe, C.P. and Bagaria, W.J. (1998). Vibration plate structures. Journal of Vibration and
technique for locating delamination in a composite Acoustics-Transactions of the ASME, 127, 84–92.
beam. AIAA Journal, 36, 1074–1077. 89. Ren, W.X. and De Roeck, G. (2002). Structural
77. Yoon, M.K., Heider, D., Gillespie, J.W., Ratcliffe, C.P. damage identification using modal data. I: Simulation
and Crane, R.M. (2001). Local damage detection using verification. Journal of Structural Engineering-ASCE,
a global fitting method on mode shape data. 128, 87–95.
Proceedings of SPIE, the International Society for 90. Ren, W.X. and De Roeck, G. (2002). Structural damage
Optical Engineering, 4359, 231–237. identification using modal data. II: Test verification.
78. Yoon, M.K., Heider, D., Gillespie Jr, J.W., Ratcliffe, Journal of Structural Engineering-ASCE, 128, 96–104.
C.P. and Crane, R.M. (2005). Local damage detection 91. Wong, C.N., Zhu, W.D. and Xu, G.Y. (2004). On an
using the two-dimensional gapped smoothing method. iterative general-order perturbation method for multi-
Journal of Sound and Vibration, 279, 119–139. ple structural damage detection. Journal of Sound and
79. Amaravadi, V., Rao, V., Koval, L.R. and Derriso, Vibration, 273, 363–386.
M.M. (2001). Structural health monitoring using wave- 92. Rahai, A., Bakhtiari-Nejad, F. and Esfandiari, A.
let transforms. Smart Structures and Integrated (2007). Damage assessment of structure using incom-
Systems. Proceedings of SPIE, 4327, 258–269. plete measured mode shapes. Structural Control &
80. Kim, B.H., Park, T. and Voyiadjis, G.Z. (2006). Health Monitoring, 14, 808–829.
Damage estimation on beam-like structures using the 93. Ruotolo, R. and Surace, C. (1997). Damage assessment
multi-resolution analysis. International Journal of of multiple cracked beams: Numerical results and
experimental validation. Journal of Sound and
Solids and Structures, 43, 4238–4257.
81. Cornwell, P., Doebling, S.W. and Farrar, C.R. (1999). Vibration, 206, 567–588.
94. Hao, H. and Xia, Y. (2002). Vibration-based dam-
Application of the strain energy damage detection
age detection of structures by genetic algorithm.
method to platelike structures. Journal of Sound and
Journal of Computing in Civil Engineering, 16, 222–229.
Vibration, 224, 359–374.
95. Farrar, C.R. and Jauregui, D.A. (1998). Comparative
82. Shi, Z.Y., Law, S.S. and Zhang, L.M. (2000). Structural
study of damage identification algorithms applied to a
damage detection from modal strain energy change.
bridge: I. Experiment. Smart Materials & Structures, 7,
Journal of Engineering Mechanics-ASCE, 126, 1216–1223.
704–719.
83. Law, S.S., Shi, Z.Y. and Zhang, L.M. (1998). Structural
96. Farrar, C.R. and Jauregui, D.A. (1998). Comparative
damage detection from incomplete and noisy modal test
study of damage identification algorithms applied to a
data. Journal of Engineering Mechanics-Asce, 124,
bridge: II. Numerical study. Smart Materials &
1280–1288.
Structures, 7, 720–731.
84. Au, F.T.K., Wang, J.J. and Liu, G.D. (2003).
97. Ndambi, J.M., Vantomme, J. and Harri, K. (2002).
Structural damage detection based on a micro-genetic
Damage assessment in reinforced concrete beams
algorithm using incomplete and noisy modal test data.
using eigenfrequencies and mode shape derivatives.
Journal of Sound and Vibration, 259, 1081–1094.
Engineering Structures, 24, 501–515.
85. Topole, K.G. and Stubbs, N. (1995). Nondestructive
98. Loutridis, S., Douka, E., Hadjileontiadis, L.J. and
Damage Evaluation in Complex Structures from a Trochidis, A. (2005). A two-dimensional wavelet trans-
Minimum of Modal Parameters. Modal Analysis-the form for detection of cracks in plates. Engineering
International Journal of Analytical and Experimental Structures, 27, 1327–1338.
Modal Analysis, 10, 95–103. 99. Kim, B.H., Kim, H. and Park, T. (2006). Nondestruc-
86. Pandey, A.K. and Biswas, M. (1994). Damage tive damage evaluation of plates using the multi-
Detection in Structures Using Changes in Flexibility. resolution analysis of two-dimensional Haar wavelet.
Journal of Sound and Vibration, 169, 3–17. Journal of Sound and Vibration, 292, 82–104.

Weifan 2010

Uploaded by

Copyright:

Available Formats

Weifan 2010

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Weifan 2010

Uploaded by

Copyright:

Available Formats

Structural Health Monitoring

Vibration-based Damage Identification Methods: A Review and Comparative Study

The online version of this article can be found at:

Email Alerts: http://shm.sagepub.com/cgi/alerts

>> Version of Record - Feb 2, 2011

OnlineFirst Version of Record - Apr 20, 2010

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Wei Fan and Pizhong Qiao*

Department of Civil and Environmental Engineering and Composite

A comprehensive review on modal parameter-based damage identification methods

Keywords frequency mode shapes modal curvatures vibration damage identifi-

1 Introduction nondestructive damage identification technique is

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

is selected to statistically identify the occurrence of 2.2 Mode Shape-based Methods

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Table 1 Characteristics of five comparative damage detection algorithms

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Table 2 Nature frequency change ratio in several damage severities

Natural frequency change ratio ¼ (fh fd)/fh

Damage Bending stiffness

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

10 adopted as suggested in the literature [75].

curve of the GSM shown in Figure 3 is evaluated

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

1 Damage at element 20/30

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

GSM method, the damaged state curvatures are r =0.4

based algorithms under measurement noise. The 2

results also show that the cubic interpolation will 0

introduce larger processing error than the finite dif- –2

curvatures are directly obtained from the simulated 4

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

and methods using both mode shape and

4 trolled laboratory condition. However, due to its

Table 3 Capabilities of five comparative damage detection algorithms

Single Multiple Large-area Sensor

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

Downloaded from shm.sagepub.com at University of Sussex Library on March 7, 2013

You might also like