Signal-Based Damage Detection Methods – Algorithms and Applications

International Journal of Engineering Research and Development
e-ISSN: 2278-067X, p-ISSN: 2278-800X, www.ijerd.com
Volume 11, Issue 05 (May 2015), PP.44-57
44
Signal-Based Damage Detection Methods – Algorithms
and Applications
Long Qiao1
, Asad Esmaeily2
1
Department of Engineering Technology, Missouri Western State University, Saint Joseph, MO 64507, USA
2
Department of Civil Engineering, Kansas State University, Manhattan, KS 66506, USA
Abstract:-Civil structures are susceptible to damages over their service lives due to aging, environmental
loading, fatigue and excessive response. Such deterioration significantly affects the performance and safety of
structure. Therefore, it is necessary to monitor the structural performance, detect and assess damages at the
earliest possible stage in order to reduce the life-cycle cost of structure and improve its reliability.Recently,
signal-based methods have been widely used for structural health monitoring and damage detection. These
methods examine changes in the features derived directly from the measured time histories or their
corresponding spectra through proper signal processing methods and algorithms to detect damage.Based on
different signal processing algorithms for feature extraction, these methods are classified into time-domain
methods, frequency-domain methods, and time-frequency (or time-scale)-domain methods.This paper provided
an overview of these methods based on two aspects: (1) feature extraction algorithms, and (2) successful
applications. Signal-based methods are particularly more effective for structures with complicated nonlinear
behaviour and the incomplete, incoherent, and noise-contaminated measurements of structural response.
Keywords:-Damage detection, signal-based methods, features extraction, pattern recognition, algorithms and
applications.
I. INTRODUCTION
Deterioration of structures due to aging, cumulative crack growth or excessive response decreases their
stiffness and integrity, and therefore significantly affects the performance and safety of structures during their
service life. Structural Health Monitoring (SHM) and damage detection denotes the ability to monitor the
performance of structure, detect and assess any damage at the earliest stage in order to reduce the life-cycle cost
of structure and improve its reliability and safety.Fig. 1 shows a brief classification of different damage
detection categories, methods and basic algorithms.
Recent advances in computer, sensors and other electronic technologies make Non-destructive Damage
Detection (NDD) techniques far more convenient and cost effective than destructive detection techniques which
usually evaluate the safety of a structure by testing samples removed from the structure. NDD techniques can
be classified into two categories: (1) local methods; and (2) global methods.
Current highly effective localized NDD methods include acoustic or ultrasonic methods, magnetic field
methods, radiograph, microwave/ground penetrating radar, fiber optics, eddy-current methods and thermal field
methods. These methods are visual or localized experimental methods that detect damage on or near the surface
of the structure by measuring light, sound, electromagnetic field intensity, displacements, or temperature.Some
of these methods are particularly effective for a specific application. For example, eddy current is very effective
for crack detection at welded joint. But these methods have several limitations when testing large and complex
structures. First, the depth of wave penetration is limited. Second, the vicinity of the damage should be known
and the portion of the structure being inspected should readily be accessible. However, there is no easy way to
determine the global health condition of a structure. Chang and Liu [1] provided detailed information about
“local” methods.
Static-based and vibration-based NDD methods provide the opportunity to detect and assess damage on
a global basis. Static-based methods rely on the strain or displacement measurements from a structure under
known static loads and the finite-element model updating to determine changes in deflection, stiffness, and load-
carrying capacity of the structure. These methods are widely used for bridge health monitoring and evaluation.
Examples of such work are Barr et al. [2] and Cardinale and Orlando [3]. The drawbacks of static-based NDD
methods are: (1) they require a large amount of measured data; (2) they require the finite-element model
updating using accurate material properties; (3) they require static-load tests which will interrupt the structure
service. These drawbacks will make static-based NDD methods more difficult for online damage detection of
an in-service structure. Vibration-based NDD methods rely on the change of vibration characteristics and
signals as indication of damage due to the reason that the damage changes the physical properties of a structure,
which in turn will cause changes to the vibration characteristics and signals of the structure.Over the last two

Signal-Based Damage Detection Methods – Algorithms and Applications
45
decades, extensive research has been conducted on Vibration-based detection approach, leading to various
experimental techniques, methodologies, and signal processing algorithms. Doebling et al. [4] and Sohn et al. [5]
presented comprehensive literature reviews of vibration based damage detection and health monitoring methods
for structural and mechanical systems. These methods can be classified into either modal-based or signal-based
categories.
Fig.1:SHM and damage detection categories
Modal-based methods use changes in measured modal parameters (resonant frequencies, modal
damping, mode shapes, etc.) or their derivatives as a sign of change in physical-dynamic properties of the
structure (stiffness, mass and damping). The basic premise behind the methods is that a change in stiffness leads
to a change in natural frequencies and mode shapes. Modal-based methods have been applied successfully to
identify the dynamic properties of linearized and time-invariant equivalent structural systems. The methods
include mode shape curvature method, the change in flexibility method, the change in stiffness method, modal
strain energy, etc. Examples of such work are Kosmatka and Ricles [6], Ren and Roeck [7], Shi et al. [8] and
Kim et al. [9]. Recently, wavelet-based and Hilbert-based approaches have been developed as enhanced
techniques for parametric identification of non-linear and time-variant systems. Examples of such work are
Staszewski [10], Kijewski and Kareem [11], Yang et al. [12], Huang et al. [13], Hou et al. [14], Chen et al. [15]
and Yan and Miyamoto [16]. Although modal-based methods are generally applicable for the purpose of
damage detection and structural health monitoring, they still have many problems and challenges: (1) damage is
a local phenomenon and may not significantly influence modal parameters, particularly for large structures; (2)
variation in the mass of the structure or environmental noise may also introduce uncertainties in the measured
modal parameters; (3) the number of sensors, the types of sensors, and the coordinates of sensors may have a
crucial effect on the accuracy of the damage detection procedure.
Signal-based methods examine changes in the features derived directly from the measured time
histories or their corresponding spectra through proper signal processing methods and algorithms to detect
damage. Based on different signal processing algorithms for feature extraction, these methods are classified into
time-domain methods, frequency-domain methods, and time-frequency (or time-scale)-domain methods. Time-
domain methods use linear and nonlinear functions of time histories to extract the signal features. Examples of
this category are Auto-Regressive (AR) model, Auto-Regressive Moving Average (ARMA) model, Auto-
Regressive with eXogenous input (ARX) model and Extended Kalman Filter (EKF). Frequency-domain
methods use Fourier analysis and cepstrum (the inverse Fourier transform of the logarithm of the Fourier spectra
magnitude squared) analysis to extract features in a given time window. Examples of this category are
Frequency Response Functions (FRFs), frequency spectra, cross power spectra, power spectra, and power

46
spectral density.Time-frequency domain methods employ Wigner-Ville distribution and wavelet analysis to
extract the signal features. Examples of this category are spectrogram, continuous wavelet transform coefficients,
wavelet packet energies and wavelet entropy.Detailed descriptions of these feature extraction algorithms and
successful applications for damage detection were discussed in this paper.
II. SIGNAL-BASED DAMAGE DETECTION
Recently, signal-based damage detection methods have received many attentions. These methods
involve two main processes: (1) feature extraction and selection, and (2) pattern recognition. Feature extraction
and selection is the process of identifying and selecting damage-sensitive features derived from the measured
dynamic response, to quantify the damage state of the structure. A variety of algorithms are employed to
improve the feature extraction and selection procedure. Based on different signal processing algorithms for
feature extraction, these methods are classified into time-domain methods, frequency-domain methods, and
time-frequency methods.
A. Time-Domain Methods
Time-domain methods use linear and nonlinear functions of time histories to extract features. Sohn et
al. [17] used an auto-regressive (AR) model to fit the measured time history on a structure. Damage diagnoses
using X-bar control chart were performed using AR coefficients as damage-sensitive features. In
the𝐴𝑅(𝑛)model, the current point in a time series is modeled as a linear combination of the previous 𝑛 points.
     
1
n
j x
j
x t x t j e t

    (1)
wherex(t) is the time history at time t; ϕj is the unknown AR coefficient; and ex(t) is the random error with
zero mean and constant variance. The value of ϕj is estimated by fitting the AR model to the time history data.
The AR coefficients of the model fit to subsequent new data were monitored relative to the baseline AR
coefficients. The X-bar control chart was used to provide a framework for monitoring the changes in the mean
values of the AR coefficients and identifying samples that were inconsistent with the past data sets. A
statistically significant number of AR coefficients outside the control limits indicated that the system was
transited from a healthy state to a damaged state. Principal component analysis and linear and quadratic
projections were applied to transform the time series from multiple measurement points into a single time series
in an effort to reduce the dimensionality of the data and enhance the discrimination between features from
undamaged and damaged structures. For demonstration, the authors applied the AR model combined with X-bar
control chart to determine the existence of damage on a concrete bridge column as the column was progressively
damaged. The AR coefficients on the X-bar control chart as detailed in the method indicated the damage
existence.
Sohn and Farrar [18] proposed a two-stage time history prediction model, combining auto-regressive
(AR) model and an autoregressive with exogenous inputs (ARX) model. The residual error, which was the
difference between the actual acceleration measurement for the new signal and the prediction obtained from the
AR-ARX model from the reference signal, was defined as the damage-sensitive feature. The increase in residual
errors was monitored to detect system anomalies. In this method, the ARX model is expressed as
       1 1
a b
i j x x
i j
x t x t i e t j t
 
         (2)
where a and b are the order of the ARX model; αi and βj are the coefficients of the AR and the exogenous input,
respectively; εx(t) is the residual error after fitting the ARX(a, b) model to the ex(t) and x(t) pair in the one-
stage ahead AR model. If the ARX model obtained from the reference signal block pair x(t) and ex(t) were not
be a good representation of the newly obtained block pair y(t) and ey(t), there would be a significant change in
the residual error, εy(t), compared to εx(t). The standard deviation ratio of the residual errors, σ(εy) σ εx ,
would reach its maximum value at the sensors instrumented near the actual damage locations. The applicability
of this approach was demonstrated by the authors using acceleration time histories obtained from an eight
degree-of-freedom mass-spring system. Sohn et al. [19] developed a unique combination of the AR-ARX model,
auto-associative neural network, and statistical pattern recognition techniques for damage classification
explicitly taking the environmental and operational variations of the system in the consideration. In this method,
AR-ARX model is developed to extract damage sensitive features, which are the αi andβj coefficients of the
ARX model. An auto-associative neural network is trained to characterize the dependency of the extracted
features on the variations caused by environmental and operation conditions. A damage classifier is constructed
using a sequential probability ratio test to automatically determine the damage condition of the system. The
authors demonstrated the proposed approach using a numerical example of a computer hard disk and an
experimental study of an eight degree-of-freedom spring-mass system.

47
Bodeux and Golinval [20] applied the autoregressive moving average vector (ARMAV) model and
statistical tools such as confidence interval and the normal distribution of random variable for damage detection.
In the state space, the ARMAV model is expressed as
     1x n Ax n W n   (3)
Where x[n] is the observed vibration vector at the nth discrete time point; A is the matrix containing the
different coefficients of the autoregressive (AR) part; W[n] is a matrix containing the moving average (MA)
terms. The natural eigenfrequenciesfr and damping ratios ζr can be extracted from the eigenvalues τr of the AR
matrix A as
 
2
ln r
rf
t


 
(4)
  
 
r
r
Real ln
ln
r

 

(5)
whereΔt is the discrete time interval. The authors used the changes in the frequencies estimated by the ARMAV
model to detect the damage on the Steel-Quake structure at the Joint Research Center in Ispra, Italy. The
frequencies were assumed to be independently distributed variables and a negative change in frequencies
indicated damage caused by structure change. As damage indicator, the probability of negative change Pδfi
in
frequency fi is given by
0
2 2
0
1i
i i
f
i i
f f
P
 
  
   
(6)
Where σi
2
and σi0
2
are the variances of the frequencies fi and fi0 corresponding to the damaged and undamaged
states. Φis the unit normal distribution function. The structure was assumed damaged if the probability was
close to one. The proposed method was limited to only detecting the damage existence.
Nair et al. [21] applied an Auto-Regressive Moving Average (ARMA) model for damage identification
and localization. A damage-sensitive feature, DSF, was defined as a function of the first three auto regressive
(AR) components. The mean values of the DSF obtained from the damaged and undamaged signals were
significantly different. In this method, the vibration signals obtained from sensors are modeled as ARMA time
series as
     1 1
p q
ij k ij k ij ij
k k
x x t k t k t
 
         (7)
Where xij(t) is the normalized acceleration signal; φk and θk are the k-th AR (Auto-Regressive) and MA
(Moving Average) coefficients, respectively; p and q are the model orders of the AR and MA processes,
respectively; and εij(t) is the residual term. DSF is defined as
1
2 2 3
1 2 3
DSF


    
(8)
where α1 , α2 and α3 are the first three AR coefficients. A hypothesis test involving the t-test was used to
determine the existence of damages on the structure. Two indices, LI1and LI2, were introduced based on the AR
coefficient space to localize damages. At the sensor locations where damage was introduced, LI1 and LI2had
comparatively large values.The authors tested the proposed methodologies on the analytical and experimental
results of the ASCE benchmark structure. The results of the damage detection indicated that DSF was able to
detect the existence of all damage patterns in the ASCE Benchmark simulation experiment. The results of the
damage localization indicated that LI1 and LI2 were all able to localize minor damages but LI1 was more robust
than LI2.
Liu et al. [22] presented a damage sensitive feature index for damage detection based on Auto-
Regressive Moving Average (ARMA) time series analysis. The acceleration signal was modeled as ARMA
models, and a principal component matrix derived from the AR coefficients of these models was utilized to
establish the Mahalanobis distance criterion function. The Mahalanobis-distances of m-dimensional vector xi
from the principal component matrix of damaged structure to the ones of undamaged structure were defined as
the damage sensitive feature (DSF) index. It is expressed as
   
1
1 2T
DSFD x x    
 
(9)
Where μ and Σ are mathematics expectation and covariance matrices of the m-dimensional vector from the
principal component matrix of undamaged structure, respectively. A hypothesis test involving the t-test method
was further applied to make a damage alarming decision by determining the statistical significance in the
difference of mean values of DDSF obtained from the damaged and undamaged cases. These methodologies were
tested on a numerical three-span-girder beam model containing some subtle damages. The results show that the

48
defined index is sensitive to these subtle structure damages, and the proposed algorithm can be applied to the
on-line damage alarming in structural health monitoring.
B. Frequency-Domain Methods
Frequency-domain methods analyze any stationary event localized in time domain. They use Fourier
analysis, cepstrum (the inverse Fourier transform of the logarithm of the Fourier spectra magnitude squared)
analysis, spectral analysis, frequency response technique, etc to extract features in a given time window. Tang et
al. [23] quantitatively diagnosed gear-wear through cepstrum analysis of gear noise signals. The amplitude value
of the peak in cepstrum represented gear mesh-harmonics in spectrum. The trend of the change of gear-wear
degree was about the same as that of the change of the value of a peak in cepstrum. The value was independent
of intensity of gear noise signal and was used as an indicator for quantitatively diagnosing gear-wear. Based on
analyzing the results of experiments with gearboxes, the thresholds of the gear wear by cepstrum diagnosis was
determined to distinguish normal, moderate and serious wears. The theoretical analysis agreed with the
experimental results very well.
Kamarthi and Pittner [24] presented sensor data representation schemes for flank wear estimation in
turning processes. The sensor data representation algorithm based on fast Fourier transform (FFT) transformed a
time series vector X of the sensor signal from turning experiments into the spectral vector x, and then formed
the vector xf with the set i1, i2, … , id . The features xr , the d-dimensional sensor data representation of X, was
computed through the relation
1 2/
fr wx S x
 (10)
The features were used by recurrent neural network architecture to continually compute the flank wear estimates.
Lee and Kim [25] used the frequency analysis to detect and localize damage. A signal anomaly index
(SAI) which quantified the change of frequency response was developed as damage feature. The SAI is defined
as a Euclidean norm of the difference between two frequency response function (FRFs) of basis and compared
state as
   
 
1
1
1
2 2
2
B C
B
FRF FRF
FRF
n
i
n
i
f B C
f f i i
f B
f f i
H f H f
SAI
H f




  
  
 
 
(11)
where, H( ) and FRF represent the frequency response function in continuous form and discrete form
respectively, superscript B and C stand for the state of Basic and Compared. The symbols, f1 and fn are the
lowest and highest frequency of the considering frequency range, respectively. Changes in the shape of the FRF
due to the reason of structural damage caused the increase of SAI value. The presence of damage was identified
from the SAI value. All SAI values calculated from different sensors and different frequency ranges formed a
SAI matrix which showed variation patterns of the FRF in both the space and the frequency domain. The SAI
matrix was used as input for the neural network to identify the location of damage. The authors conducted a
series of experimental tests and numerical simulation on an experimental model bridge to demonstrate the
feasibility of the proposed algorithm. The results of this example application show that the SAI based pattern
recognition approach has the great potential for structural health monitoring on a real bridge.
Fasel et al. [26] used a frequency domain auto-regressive model with exogenous inputs (ARX) to
detect joint damage in steel moment-resisting frame structures. Damage sensitive features were extracted from
the ARX model in the consideration of non-linear system input/output relationships. A frequency domain ARX
model was used to predict the response at a particular frequency based on the input at that frequency, as well as
responses at surrounding frequencies. The responses at the surrounding frequencies were included as inputs to
the model to account for sub-harmonics and super-harmonics introduced to the system through non-linear
feedback. To accounts for non-linearity in the system, first-order ARX model in the frequency domain is built as
             1 11 1Y k B k U k A k Y k A k Y k     2 3 1fk , ,...,N  (12)
where Nf is the highest frequency value examined, Y(k) is the response at the k-th frequency, U(k) is the input
at the k-th frequency, and Y(k − 1) and Y(k + 1) are the responses at the (k-1)th and (k+1)th frequencies,
respectively.A1(k)andA−1(k) are the frequency domain auto-regressive coefficients, and B(k) is the exogenous
coefficient. The frequency response of one accelerometer was treated as an input and the other accelerometer
response was treated as an output. The auto-regressive coefficients in this frequency domain model were used as
features. These features were then analyzed using extreme value statistics (EVS) to differentiate between
damage and undamaged conditions. The suitability of the ARX model, combined with EVS, to non-linear
damage detection was demonstrated on a three-story building model.
Qiao et al. [27, 28] noted that in the Fast Fourier Transform (FFT) spectrums of acceleration signals of
the structure under different damage scenarios, the peak magnitude changes were more sensitive than the peak
frequency shifts. The authors selected the FFT magnitude vectors in frequency domain as the sensitive features
which also preserved the information of frequency shifting, forming a one-dimension pattern, presenting a

49
unique damage condition. In order to separate the feature changes caused by operational and environmental
variations of the system from the structure changes of interest, each magnitude vector in a pattern was
normalized with respect to the square root of the sum of square of the corresponding pattern. The dynamic
response patterns of a damaged structure were compared with a wide range of numerically generated damaged
cases stored in a pattern database to detect damage severity and location. To demonstrate the feasibility of the
proposed method, numerical and experimental studies were conducted on a simple three-story steel building.
The results showed that the FFT patterns were successfully used as sensitive features for damage detection.
C. Time-Frequency (or Scale)-Domain Methods
In contrast to the frequency-domain methods, the time-frequency (or scale) methods can be used to
analyze any non-stationary event localized in time domain. Staszewski et al. [29] applied the Wigner-Ville
distribution (WVD) to detect local tooth faults in spur gears. The authors showed that the visual observation of
the WVD contour plots could be used for fault detection. Dark zones and curved bands in the contour plots were
the main features of an impulse produced by the fault in the spur gear. The changes in features of the
distribution were used to monitor the progression of a fault. For the sake of automatic fault detection, the
authors chose the two-dimensional contour plots of the WVD as patterns, and the amplitude values of the
contour plots as features. Pattern recognition procedures based on the statistical and neural approaches were
used for classification of different fault conditions.
Biemans et al. [30] employed the orthogonal wavelet analysis of the strain data measured from
piezoceramic sensors to detect crack growth in the middle of a rectangular aluminum plate. The strain data
measured from the plate under the Gaussian white noise excitation was decomposed into orthogonal wavelet
levels. The logarithm of the variance of the orthogonal wavelet coefficients was calculated for all wavelet levels.
The mean vector μ, of the logarithms for the undamaged plate formed the template for the similarity analysis. A
Euclidean distance between the template μ and the logarithms x, for the damaged plate was used as a damage
index. The damage index is denoted as
   2 T
x,d x x    (13)
The mean and standard deviation of the damage index representing the undamaged condition of the
plate were used to establish an alarm level. The damage could be considered existence in the plate if the damage
index was above the alarm level. The experimental results on the aluminum plate show that such damage index
can be used to successfully detect as small as 6-7mm crack and to monitor the crack growth.
Hou et al. [31] presented the great potential of wavelet analysis for singularity extraction in the signals.
Characteristics of four types of representative vibration signals were examined by continuous and discrete
wavelet transforms. The singularity in these signals were extracted and best illustrated in the plot of wavelet
coefficient in the time-scale plane. The fringe pattern in the continuous wavelet coefficient contour plot
indicated the existence of a singularity in the local time and the spike in the discrete wavelet coefficient plot also
indicated the existence of a singularity in the local time. The sensitivity of wavelet results to a singularity was
effectively used to detect possible structural damage using measured acceleration response data. To demonstrate
the feasibility of the proposed method, the authors used both numerical simulation data from a simple structural
model with multiple paralleled breakable springs and actual acceleration data recorded on the roof of a building
during an earthquake event. The detection results showed that occurrence of damage could be detected by spikes
in the detailed of the wavelet decomposition of the response data, and the locations of these spikes could
accurately indicate the moments when the damage occurred. The similar work can also be found on Hera and
Hou [32], Ovanesova and Suarez [33], Melhem and Kim [34] and Qiao et al. [35].
Kim and Kim [36] used the ratio of the incident wave toward and the reflected wave from the damage
as index to assess the damage size. The ratio was estimated by the continuous wavelet transform of the
measured signal and the ridge analysis. In the time-frequency plane of the continuous wavelet transform, the
ridge was traced to compare the magnitude of the incident wave and the magnitude of the reflected wave from
the damage. It was found that the ratio of these magnitudes along the two ridges was the same as the ratio of the
magnitude of the incident wave to the magnitude of the reflected wave. Due to the fact that the magnitude and
frequency-dependent pattern of the ratio varied with damage size, it was able to correlate the ratio and the
damage size except when the damage size was very small. The authors conducted the wave experiments in a
cylindrical ferromagnetic beam. Magnetostrictive sensors were used to measure the bending waves in the beam
cross section. The continuous Gabor wavelet transform was employed to estimate the crack size in the beam.
Yen and Lin [37] investigated the feasibility of applying the Wavelet Packet Transform (WPT) to
detect and classify the mechanical vibration signals. They introduced a wavelet packet component energy index
and demonstrated that the wavelet packet component energy had more potential for use in signal classification
as compared to the wavelet packet component coefficients alone. The component energy is defined as

50
 
2i i
j jE f t dt


 (14)
wherefj
i
(t) is the ith component after j levels of decomposition. Sun and Chang [38] applied the wavelet packet
component energy index to assess structural damage. The vibration signals of a structure were decomposed into
wavelet packet components. The component energies were calculated and the ones which were both significant
in value and sensitive to the change in rigidity were selected as damage indices and then used as inputs into
neural network models for damage assessment. The authors performed numerical simulations on a three-span
continuous bridge under impact excitation. Various levels of damage assessment including identifying the
occurrence, location, and severity of the damage were studied. The results show that the WPT-based component
energies are sensitive to structural damage and can be used for various levels of damage assessment.
Sun and Chang [39] also derived two damage indicators from the WPT component energies. The
acceleration signals of a structure excited by a pulse load were decomposed into wavelet packet components.
The energies of these wavelet packet components were calculated and sorted by their magnitudes. The dominant
component energies which were highly sensitive to structural damage were defined as the wave packet signature
(WPS). Two damage indicators, SAD (sum of absolute difference) and SSD (square sum of difference), were
then formulated to quantify the changes of these WPSs. SAD and SSD are defined as
1
^
m
i i
j j
i
SAD E E

  (15)
2
1
^
m
i i
j j
i
SSD E E


 
  
 
(16)
whereEj
i
(i = 1,2, … , m) are termed as the baseline WPS that are used as a reference; and Ej
i
(i = 1,2, … , m) are
WPS obtained from any subsequent measurement. These two indicators basically quantified the deviations of
the WPS from the baseline reference. To monitor the change of these damage indicators, the X-bar control
charts were constructed and one-sided confidence limits were set as thresholds for damage alarming. For
demonstration, the authors conducted an experimental study on the health monitoring of a steel cantilever I
beam. Four damage cases, involving line cuts of different severities in the flanges at one cross section, were
introduced. Results show that the health condition of the beam can be accurately monitored by the proposed
method; the two damage indicators are sensitive to structural damage and yet insensitive to measured noise.
Yam et al. [40] constructed a non-dimensional damage feature proxy vector for damage detection of
composite structures. The damage feature proxy vector was calculated based on energy variation of the wavelet
packet components of the structural vibration response before and after the occurrence of structural damage. The
damage feature proxy vector, Vd is defined as
1
1
1 2 2
0 0 0
1 2 2
1 1 1
L
L
T
dd d
L, L, L,
d
L, L, L,
UU U
V , ,...
U U U


  
    
  
(17)
where UL,j
0
and UL,j
d
are the energy of the jth order sub-signal of the intact and damaged structures, respectively;
L is the layer number of the tree structure of wavelet decomposition. Artificial neural network (ANN) was used
to establish the mapping relationship between the damage feature proxy and damage location and severity. The
method was applied to crack damage detection of a PVC sandwich plate. The results show that the damage
feature proxy exhibits high sensitivity to small damage.
Han et al. [41] proposed a damage detection index called wavelet packet energy rate index (WPERI)
for the damage detection. The rate of signal wavelet packet energy Δ(Ejj
) at j level is defined as
 
   
 
2
1
i ij
j j
j
i
j
f f
b a
f
i
f
a
E E
E
E


 
(18)
where Efj
i is the energy stored in the component signal fj
i
(t) after j levels of decomposition; (Efj
i )a is the
component signal energy Efj
i at j level without damage; and (Efj
i )b is the component signal energy Efj
i with some
damage. It was assumed that structural damage would affect the energies of wavelet packet components and
therefore altered this damage indicator. To establish threshold values for damage indexes, WPERIs, X-bar
control charts were constructed and one-sided confidence limits were set as thresholds for damage alarming.
The proposed method was applied to a simulated simply supported beam and to the steel beams with three
damage scenarios in the laboratory. Both simulated and experimental studies demonstrated that the WPT-based
energy rate index is a good candidate index that is sensitive to structural local damage.

51
Diao et al. [42] proposed a two–step structural damage detection approach based on wavelet packet
analysis and neural network. The wavelet packet component energy change γsi was selected as an input into
probabilistic neural network to determine the location of the damage. The γsi is defined as
d u
si si
si u
si
E E
E

  (19)
whereEsi
u
is the ith component energy at s level without damage, Esi
d
is the ith component energy at s level with
damage. The component energy was selected as input into back-propagation network to determine the damage
extent. The method was demonstrated by numerical simulation of a tree-dimensional four-layer steel frame.
Chen et al. [15] introduced an improved Hilbert-Huang Transform (HHT) to extract the structural
damage information from the response signals of a simulated composite wingbox. The signals was firstly
decomposed into sub-signals using Wavelet Packet Transform (WPT). These sub-signals were then
decomposed into multiple Intrinsic Mode Function (IMF) components by Empirical Mode Decomposition
(EMD). The IMF selection criterion was then applied to eliminate the unrelated IMF components. The retained
IMF components were transformed using HHT to obtain instantaneous energy of all sub-signals. By comparing
the instantaneous energy corresponding to IMFs of intact wingbox with those of damaged wingbox, it was
found that some instantaneous energy was changed obviously. Based on this fact, the authors constructed the
variation quantity of instantaneous energy ΔEt as feature index vector, which is defined as
0 1 100t
t
t
E
E %
E
 
    
 
(20)
whereEt
0
and Et are instantaneous energy of intact and damaged structure respectively at time t. Reduction in
Young‟s modulus was used to characterize damage in wingbox. The detection results show that the feature
index vector distinctly reflects the wingbox damage status, and is more sensitive to small damage.
Ding et al. [43] developed a procedure for damage alarming of frame structures based on energy
variations of structural dynamic responses decomposed by wavelet packet transform. The damage alarming
index ERVD, extracted from the wavelet packet energy spectrum is expressed as
 
2
1
m
p
p
ERVD ERV ERV

  (21)
p up dpERV I I   1 2p , ,...,m (22)
2 1
1
2
i
i,p
p
i
i, j
j
E
I
E




 
 
 
 1 2p , ,...,m (23)
whereIup and Idp are the damage indication vector in the pth dominant frequency band of the intact and
damaged structures, respectively. Ei,j is the jth component energy at l level. The authors demonstrated the
practicability of the damage alarming method for the frame structures by using the ASCE structural benchmark
data. The results reveal that the WPT-based damage alarming index ERVD is sensitive to structural local
damage affected by the actual measurement noise; the index ERVD constructed under the lower decomposition
level and dominant frequency bands is efficient for the detection of the damage occurrence.
Ren and Sun [44] combined wavelet transform with Shannon entropy to detect structural damage from
measured vibration signals.Wavelet entropy, relative wavelet entropy and wavelet-time entropy were used as
features to detect and locate damage. The wavelet entropy is defined as
  0
lnWT WT j j
j
S S p p p

       
(24)
where{pj} is the wavelet energy vector, which represents energy distribution in a time-scale. It gives a suitable
tool for detecting and characterizing singular features of a signal in time-frequency domain. For the jth scale,
the wavelet energy ratio vector {pj} is defined as
j
j
tol
E
p
E
 (25)
The relative wavelet entropy (RWE) is defined as
  0
ln
j
WT j
j
j
p
S p / q p
q

 
   
  
(26)
which gives a measure of the degree of similarity between two probability distributions. The wavelet-time
entropy is defined as
 
     
0
ln
i i i
WT j j
j
S p p p

    
 
(27)

52
wherepj
(i)
is the time evolution of relative wavelet energy at a resolution level j in the time interval i
 
 
 
i
i j
j i
tol
E
p
E
 (28)
These features were investigated by numerically simulated harmonic signals and two laboratory test cases. It
was demonstrate that wavelet-time entropy is a sensitive damage feature in detecting the abnormality in
measured successive vibration signals; relative wavelet entropy is a good damage feature to detect damage
occurrence and damage location through the vibration signals measured from the intact and damaged structures;
and the relative wavelet entropy method is flexible in choosing the reference signal simultaneously measured
from any undamaged location of the target structure.
III. APPLICATIONS TO SPECIAL STRUCTURES
A. Damage Detection on Bridge
Omenzetter et al. [45] identified the unusual events in multi-channel bridge monitoring strain data
using wavelet transform and outlier analysis. The strain data was recorded during continuous, long-term
operation of a multi-sensor Structural Health Monitoring (SHM) system installed on a full-scale bridge. Outlier
detection in multivariate data was applied to find and localize abnormal, sudden events in the strain data and
wavelet transform was used to separate the abrupt strain changes from slowly varying ones. The method was
successfully tested using known events recorded during construction of the bridge and later effectively used for
detection of anomalous post-construction events.
Omenzetter and Brownjohn [46] proposed and examined the application of concepts of time series
analysis to the processing of data from a continuously operating SHM system installed in a major bridge
structure. The recorded static strain data was modeled using ARIMA models. The coefficients of the ARIMA
models were identified on-line using an extended Kalman filter. The method was first applied to strains
recorded during bridge construction, when structural changes corresponded to known significant events such as
cable tensioning. Then the method was used to analyze signals recorded during the post-construction period
when the bridge was in service. The results show that the method can provide information on structural
performance under normal environmental and operational conditions.
Ding and Li (47) proposed an online structural health monitoring method for long-term suspension
bridge using wavelet packet transform (WPT). The method was based on the wavelet packet energy spectrum
(WPES) variation of structural ambient vibration responses. As an example application, the WPES-based health
monitoring system was used on the Runyang Suspension Bridge to monitor the responses of the bridge in real-
time under various types of environmental conditions and mobile loads. As for the vibration monitoring of the
bridge, a total of 27 uni-axial servo type accelerometers were installed at the nine sections of the bridge deck. In
each deck section, one lateral accelerometer directly recorded the lateral response, and the vertical acceleration
of the deck section was obtained by averaging the accelerations measured by the two vertical accelerometers
located in the upriver and downriver cross section, respectively.The analysis showed that changes in
environmental temperature had a long-term trend influence on the WPES, and the effect of traffic loadings on
the WPES presented instantaneous changes.
Zhang [48] presented a statistical damage identification procedure for bridge health monitoring. The
damage features were extracted based on time series analysis combining auto-regressive and auto-regressive
with eXogenous input prediction models. The structural condition was evaluated in a statistical way based on
the damage possibilities that were derived from a quite large number of data samples to minimize the effect of
the variability in data acquisition process and in structural properties on the damage assessment.The proposed
damage identification procedure was applied to a numerical 3-span continuous girder bridge model under
random ground excitations. Reasonable damage severities for beam structures as well as realistic noise levels
were simulated.The results show that the damage identification procedure has great potential to detect structural
damage at early stage, in which the structural modal frequency changes are almost imperceptible.
B. Crack Detection on Beam and Plate
Wang and Deng [49] detected the crack on beam and plate structures based on wavelet analysis of
spatially distributed structural response measurements.Simulated deflection signals of a beam containing a
transverse crack and the displacement response of a plate with a through-thickness crack were used.Wavelet
transforms were performed on these signals to obtain the wavelet coefficients along the span of the structures.
The crack location was detected by observing a sudden change, such as a spike, in the distribution of the wavelet
coefficients. The magnitude of the spike in the wavelet analysis was the maximum when the measurement point
was next to the damage location.
Biemans et al. [50] applied the piezoceramic sensors to monitoring crack propagation. The specimens
used were two rectangular (400 × 150 × 2 mm) aluminum plate with cracks initiated by spark erosion in the

53
middle of the plates. Each plate was instrumented with 6 piezoceramics bonded in a symmetrical configuration
20 mm below and above the initiated crack. One of the piezoceramics was used as an actuator excited by a sine
sweep and Gaussian white noise signals to exploit broadband excitation. The plates were subjected to static and
dynamic tensile loading. The growing crack was monitored by two of the remaining piezoceramic sensors. The
response strain data was analyzed using a number of time, frequency, and wavelet domain statistical parameters.
The results show that low frequency broadband excitation offers a possible means of on-line detection of cracks
in metallic structures.
Yan et al. [51] detected the crack damage in a honeycomb sandwich plate by using two structural
vibration damage feature indexes: natural frequency and WPT energy index. The finite element dynamic model
of a honeycomb sandwich plate was presented using different mesh division for the surface plate and the
sandwich plate to accurately express the crack damage status (locations, length and direction) of the plate. In
order to acquire the experimental dynamic response of the plate, two piezo-patches with a size of 25×15×0.28
mm were bonded on the surface of the plate. One of them acted as an actuator and the other acted as a sensor.
The natural frequencies of the undamaged plate were experimentally measured to verify the numerical model.
Based on the dynamic model verified by the experiment, the damage feather indexes for various crack damage
status were numerically computed. Then the crack damage status was determined by comparing the damage
feature indexes obtained from the numerical and experimental results. The authors found that natural frequency
of structure might not be used to detect crack damage less than 10%, even up to 20% of the total size of a plate-
like structure; energy spectrum of wavelet transform signals of structural dynamic response was so sensitive to
crack damage that it could exhibit a crack length close to 2% of the dimension of a plate-like structure. They
also found that high order modes of a structure contain more structural damage information; in order to detect a
small damage, more vibration modes should be included in a structural dynamic model.
Chang and Chen [52] detected the locations and sizes of multi-cracks in a beam by spatial wavelet
analysis. The crack type was open crack and was represented as a rotational spring.The mode shapes of the
multi-cracked beam under free vibration were analyzed by wavelet transformation. The positions of the cracks
were observed as a sudden change in the plot of wavelet coefficients. The natural frequencies of the beam were
used to predict the depth of the cracks through the characteristic equation. The limitation of this method is that
there are two peaks near the boundaries in the wavelet plot and the crack can not be detected when the crack was
near the boundaries.
Poudel et al. [53, 54] employed high-resolution images for damage detection on a simply supported
prismatic steel beam. A high-speed digital video camera was used to recode the free vibration displacement of
the beam which was excited by imposing an initial displacement near the mid-span from the left support. The
camera had a Complimentary Metal Oxide Semiconductor (CMOS) sensor with 1280 × 1024 resolution and a
10-bit A/D converter. Its frame rate ranges was from 100 to 2000 frames/s. The displacement data with high
spatial resolution were then used to obtain the mode shapes and the mode shape difference function between the
reference and damage states of the structure.The spatial signal in terms of mode shape difference function was
decomposed by wavelet transformation to display the changes due to cracking damage. The appropriate range
of wavelet scale was determined by the spatial frequency bandwidths of the mode shape difference functions.
The maximum modulus and sign change of phase angle in the wavelet coefficients indicated the changes at the
damage locations.
C. Damage Detection on Mechanical Structures
Staszewski and Tomlinson [55] applied the wavelet transform to the problem of the detection of a
broken tooth in a spur gear. The fault detection algorithm was based on pattern recognition analysis. Features of
the pattern were the modulus of the wavelet transform. Spectral analysis and an orthogonal transform were used
to compress feature elements. The Mahalanobis distance of two patterns obtained from the normal (no fault)
condition and not normal (fault) condition was used as a fault detection symptom.Visual inspection of the
modulus and phase of the wavelet transform were used to localize the fault.
Wang and McFadden [56, 57] used the wavelet transform to detect abnormal transients generated by
gear damage. The early damage to a gear tooth usually caused a variation in the associated vibration signal over
a short time, initially less than one tooth meshing period, taking the form of modulated or unmodulated
oscillation. In later stages, the duration of the abnormal variation became longer, lasting more than one tooth
meshing period. Other distributed faults, such as eccentricity and wear, might cover the most part of the whole
revolution of the gear. Changes in the vibration signals therefore could be analyzed to provide an indicator of
gear condition. When the size and shape of a wavelet were exactly the same as a section of the signal, the
transform gave a maximum absolute value of wavelet coefficients. Therefore, the abnormal signal caused by a
gear fault could be displayed by the wavelet transform, which could be regarded as a procedure for comparing
the similarity of the signal and the chosen wavelet.

54
Li et al. [58] applied neural networks to the detection of motor bearing conditions based on the
frequency features of bearing vibration. Five basic frequencies related to rolling bearing dynamic movement
were extracted by fast Fourier transform (FFT) technique. The basic frequency amplitude vectors were
constructed to represent different bearing vibrations. These vectors were created from the power spectrum of
the vibration signal and consisted of the five basic frequencies with varying amplitudes based on the defect
present. The network consisted of five input measurements corresponding to the amplitudes of the five basic
frequencies of interest, ten hidden nodes, and three output fault detectors (bearing looseness, defects on the inner
raceway, and defects on the rolling elements). The network was tested using the data generated by
MOTORSIM. The results show that neural network can be an effective agent in the detection of various motor
bearing faults through the measurement and interpretation of motor bearing vibration signals.
Liao et al. [59] developed a novel technique for monitoring the gearbox condition based on the Self-
Organizing Feature Maps (SOFM) network. Seven time-domain features parameters, i.e. standard deviation,
Kurtosis, root mean square value, absolute mean value, crest factor, clearance factor and impulse factor were
extracted from industrial gearbox vibration signals measured under different operating conditions. Trained with
the SOFM network and visualized using the U-matrix method, the feature data were mapped into a two-
dimensional space and formed clustering regions, each indicative of a specific gearbox work condition.
Therefore the gearbox operating condition with fatigue crack or a broken tooth compared with the normal
condition was identified clearly.
Kar and Mohanty [60] applied the multi-resolution Fourier transform (MFT) of vibration and current
signals to gearbox health monitoring. One and two teeth were artificially removed in one gear of the gearbox to
simulate actual fault condition. When the gearbox was operated under several loads, the vibration signals were
acquired from the tail-end bearing of the gearbox, and simultaneously the current drawn by the induction motor
is acquired. The vibration and current signals were decomposed into four levels using discrete wavelet transform
(DWT) with an orthogonal wavelet of „db8‟. Then a hanning window with 256 data points and 50% overlap was
applied to the scaled signals to find the MFT coefficients.The MFT coefficients of signals were used to classify
the types of defects by tracking the energy level possessed by the defect characteristic frequency.
IV. CONCLUSIONS
This paper provided an overview of signal-based methods to detect, locate, and characterize damage in
structure and mechanical systems. These methods examine changes in the features derived directly from the
measured time histories or their corresponding spectra through proper signal processing methods and algorithms
to detect damage.Based on different signal processing algorithms for feature extraction, these methods are
classified into time-domain methods, frequency-domain methods, and time-frequency (or time-scale)-domain
methods.Features derived by time-domain methods include auto-regressive model, auto-regressive moving
average model and auto-regressive with eXogenous input model. Features derived by frequency-domain
methods include frequency response functions, frequency spectra, and power spectral density and FFT
magnitudes. Features derived by time-frequency-domain methods include spectrogram, continuous wavelet
transform coefficients, wavelet packet energies and wavelet entropy.Many successful applications demonstrated
that different damage scenarios can be uniquely identified by these signal-based features and pattern recognition
techniques can enhance the accuracy and efficiency of damage detection.
REFERENCES
[1]. P.C. Chang and S.C. Liu, “Recent Research in Nondestructive Evaluation of Civil Infrastructures”,
Journal of Materials in Civil Engineering, Vol. 15, No.3, pp. 298, 2003.
[2]. P.J. Barr, B. Woodward, B. Najera and M.N. Amin, “Long-Term Structural Health Monitoring of the
San Ysidro Bridge”, Journal of Performance of Constructed Facilities, Vol. 20, No.1, pp. 14, 2006.
[3]. G. Cardinale, G. and M. Orlando, “Structural Evaluation and Strengthening of a Reinforced Concrete
Bridge”, Journal of Bridge Engineering, Vol. 9, No. 1, pp. 35, 2004.
[4]. S.W. Doebling, C.R. Farrar, M.B. Prime, and D.W. Shevitz, “Damage Identification and Health
Monitoring of Structural and Mechanical Systems from Change in Their Vibration Characteristics: A
Literature Review”, Report, LA-13070-MS, US-900: Los Alamos National Laboratory, NM, 1996.
[5]. H. Sohn, C.R. Farrar, F.M. Hemez, D.D. Shunk, D.W. Stinemates and B.R. Nadler, “A Review of
Structural Health Monitoring Literature: 1996-2001”, Report, LA-13976-MS: Los Alamos National
Laboratory, NM, 2003.
[6]. J.B. Kosmatka and J.M. Ricles, “Damage Detection in Structures by Modal Vibration
Characterization”, Journal of Structural Engineering, Vol. 125, No.12, pp. 1384, 1999.
[7]. W.X. Ren and G.D. Roeck, “Structural Damage Identification Using Modal Data. I: Simulation
Verification”, Journal of Structural Engineering, Vol. 128, No.1, pp. 87, 2002.

55
[8]. Z.Y. Shi, S.S Law and L.M. Zhang, “Structural Damage Detection from Modal Strain Energy Change”,
Journal of Engineering Mechanics, Vol. 126, No.12, pp. 1216, 2000.
[9]. J.T. Kim, Y.S. Ryu, H.M. Cho and N. Stubbs, “Damage Identification in Beam-Type Structures:
Frequency-Based Method vs Mode-Shape-Based Method”, Engineering Structures, Vol. 25, pp. 57,
2003.
[10]. W.J. Staszewski, “Identification of Non-Linear System Using Multi-Scale Ridges and Skeletons of the
Wavelet Transform”, Journal of Sound and Vibration, Vol. 214, No.4, pp. 639, 1998.
[11]. T. Kijewski and A. Kareem, “Wavelet Transforms for System Identification in Civil Engineering”,
Computer-Aided Civil and Infrastructure Engineering, Vol. 18, pp. 339, 2003.
[12]. J.N Yang, Y. Lei, S. Lin and N. Huang, “Hilbert-Huang Based Approach for Structural Damage
Detection”, Journal of Engineering Mechanics, Vol. 130, No.1, pp. 85, 2004.
[13]. C.S. Huang, S.L. Huang, C.I. Lin and W.C. Su, “A Wavelet-Based Approach to Identifying Structural
Modal Parameters From Seismic Response and Free Vibration Data”, Computer-Aided Civil and
Infrastructure Engineering, Vol. 20, pp. 408, 2005.
[14]. Z. Hou, A. Hera and A. Shinde, “Wavelet-Based Structural Health Monitoring of Earthquake Excited
Structures”, Computer-Aided Civil and Infrastructure Engineering, Vol. 21, pp. 268, 2006.
[15]. H.G. Chen, Y.J. Yan and J.S. Jiang, “Vibration-Based Damage Detection of Composite Wingbox
Structures Using Improved Hilbert-Huang Transform”, Key Engineering Materials, Vols. 324-325, pp.
539, 2006.
[16]. B. Yan and A. Miyamoto, “A Comparative Study of Modal Parameter Identification Based on Wavelet
and Hilbert-Huang Transform”, Computer-Aided Civil and Infrastructure Engineering, Vol. 21, pp. 9,
2006.
[17]. H. Sohn, J.A. Czarnecki and C.R. Farrar, “Structural Health Monitoring Using Statistical Process
Control”, Journal of Structural Engineering, Vol. 126, No.11, pp. 1356, 2000.
[18]. H. Sohn and C.R. Farrar, “Damage Diagnosis Using Time-Series Analysis of Vibration Signals”, Smart
Materials &Structures, Vol. 10, No.3, pp. 446, 2001.
[19]. H. Sohn, K. Worden and C.R. Farrar, “Statistical Damage Classification Under Changing
Environmental and Operational Conditions”, Journal of Intelligent Material Systems and structures,
Vol. 13, No.9, pp. 561, 2002.
[20]. J.B. Bodeux and J.C. Golinval, “Application of ARMAV Models to the Identification and Damage
Detection of Mechanical and Civil Engineering Structures”, Smart Materials and Structures, Vol. 10,
pp. 479, 2001.
[21]. K.K. Nair, A.S. Kiremidjian and K.H. Law, “Time Series-Based Damage Detection and Localization
Algorithm with Application to the ASCE Benchmark Structure”, Journal of Sound and Vibration, Vol.
291, pp. 349, 2006.
[22]. Y. Liu, A. Li, Q. Fei and Y. Ding, “Feature Extraction and Damage Alarming Using Time Series
Analysis”, Journal of Southeast University, Vol. 23, No.1, pp. 86, 2007.
[23]. H. Tang, J.Z. Cha, Y. Wang and C. Zhang, “The Principal of Cepstrum and Its Application in
Quantitative Fault Diagnostics of Gears”, ASME Design Technical Conference, Miami, FL, De-Vol.
38, pp. 141, 1991.
[24]. S. Kamarthi, and S.Pittner, “Fourier and Wavelet Transform for Flank Wear Estimation – A
Comparison”, Mechanical Systems and Signal Processing, Vol. 11, No.6, pp. 791, 1997.
[25]. J. Lee and S. Kim, “Structural Damage Detection in the Frequency Domain Using Neural Networks”,
Journal of Intelligent Materials Systems and Structures, Vol. 18, pp. 785, 2007.
[26]. T.R. Fasel, H.Sohn, G. Park and C.R. Farrar, “Active Sensing Using Impedance-Based ARX Models
and Extreme Value Statistics for Damage Detection”, Earthquake Engineering and Structural
Dynamics, Vol. 34, pp. 763, 2005.
[27]. L. Qiao, A. Esmaeily and H. Melhem, “Structural Damage Diagnosis Using Signal Pattern-Recognition
intelligent Computing”, Proceedings of 15th Annual Workshop of the European Group for Intelligent
Computing in Engineering (EG-ICE), Plymouth, UK, July 2-4, 2008.
[28]. L. Qiao, A. Esmaeily and H. Melhem, “Signal Pattern Recognition for Damage Diagnosis in
Structures”, Computer-Aided Civil and Infrastructure Engineering, Vol. 27, pp. 699-710, 2012.
[29]. W.J. Staszewski, K. Worden and G.R. Tomlinson, “Time-Frequency Analysis in Gearbox Fault
Detection Using the Wigner-Ville Distribution and Pattern Recognition”, Mechanical Systems and
Signal Processing, Vol. 11, No.5, pp. 673, 1997.
[30]. C. Biemans, W.J. Staszewski, C. Boller and G.R. Tomlinson, “Crack Detection in Metallic Structures
Using Broadband Excitation of Acousto-Ultrasonics”, Journal of Intelligent Material Systems and
Structures, Vol. 12, pp. 589, 2001.

56
[31]. Z. Hou, M. Noori and R.S. Amand, “Wavelet-Based Approach for Structural Damage Detection”,
Journal of Engineering Mechanics, Vol. 126, No.7, pp. 677, 2000.
[32]. A. Hera and Z.K. Hou, “Application of Wavelet Approach for ASCE Structural Health Monitoring
Benchmark Studies”, Journal of Engineering Mechanics, Vol. 130, No.1, pp. 96, 2004.
[33]. A.V. Ovanesova and L.E. Suarez, “Application of Wavelet Transforms to Damage Detection in Frame
Structures”, Engineering Structures, Vol. 26, pp. 39, 2004.
[34]. H. Melhem and H. Kim, “Damage Detection in Concrete by Fourier and Wavelet Analyses”, Journal of
Engineering Mechanics, Vol. 129, pp. 571, 2003.
[35]. L. Qiao, A. Esmaeily and H. Melhem, “Structural Damage Detection Using Signal Pattern-
Recognition”, Key Engineering Materials, Vol. 400-402, pp. 465, 2009.
[36]. I.K. Kim and Y.Y. Kim, “Damage Size Estimation by the Continuous Wavelet Ridge Analysis of
Dispersive Bending Waves in A Beam”, Journal of Sound and Vibration, Vol. 287, pp. 707, 2005.
[37]. G.G. Yen and K.C. Lin, “Wavelet Packet Feature Extraction for Vibration Monitoring”, IEEE
Transactions on Industrial Electronics, Vol. 147, pp. 650, 2000.
[38]. Z. Sun and C.C. Chang, “Structural Damage Assessment Based on Wavelet Packet Transform”,
Journal of Structural Engineering, Vol. 128, No.10, pp. 1354, 2002.
[39]. Z. Sun and C.C. Chang, “Statistical Wavelet-Based Method for Structural Health Monitoring”, Journal
of Structural Engineering, Vol. 130, No.7, pp. 1055, 2004.
[40]. L.H. Yam, Y.J. Yan and J.S. Jiang, “Vibration-Based Damage Detection for Composite Structures
Using Wavelet Transform and Neural Network Identification”, Composite Structures, Vol. 60, pp. 403,
2003.
[41]. J.G. Han, W.X. Ren and Z.S. Sun, “Wavelet Packet Based Damage Identification of Beam Structures”,
International Journal of Solids and Structures, Vol. 42, pp. 6610, 2005.
[42]. Y.S. Diao, H.J. Li and Y. Wang, “A Two-Step Structural Damage Detection Approach Based on
Wavelet Packet Analysis and Neural Network”, Proceeding of the 5th International Conference on
Machine Learning and Cybernetics, August 13-16, Dalian, China, 2006.
[43]. Y.L. Ding, A.Q. Li and T. Liu, “A Study on the WPT-Based Structural Damage Alarming of the ASCE
Benchmark Experiments”, Advances in Structural Engineering, Vol. 11, No.1, pp. 121, 2008.
[44]. W.X. Ren and Z.S. Sun, “Structural Damage Identification by Using Wavelet Entropy”, Engineering
Structures, Vol. 30, No. 10, pp. 2840, 2008.
[45]. P. Omenzetter, J.M.W. Brownjohn and P. Moyo, “Identification of unusual events in multi-channel
bridge monitoring data”, Mechanical Systems and Signal Processing, Vol.18, pp. 409-430, 2004.
[46]. P. Omenzetter and J.M.W. Brownjohn, “Application of time series analysis for bridge monitoring”,
Smart Materials and Structures, Vol. 15, pp. 129-138, 2006.
[47]. Y.L. Ding and A.Q. Li, “Structural health monitoring of long-span suspension bridges using wavelet
packet analysis”, Earthquake Engineering and Engineering Vibration, Vol. 6, No.3, pp. 289-294, 2007.
[48]. Q.W. Zhang, “Statistical damage identification for bridge using ambient vibration data”, Computers
and Structures, Vol. 85, pp. 476-485, 2007.
[49]. Q. Wang and X.M. Deng, “Damage detection with spatial wavelets”, International Journal of Solid and
Structures, Vol. 36, pp. 3443-3468, 1999.
[50]. C. Biemans, W.J. Staszewski, C. Boller and G.R. Tomlinson, “Crack detection in metallic structures
using broadband excitation of acousto-ultrasonics”, Journal of Intelligent Material Systems and
Structures, Vol. 12, pp. 589-597, 2001.
[51]. Y.J. Yan, H.N. Hao, and L.H. Yam, “Vibration-based construction and extraction of structural damage
feature index”, International Journal of Solid and Structures, Vol. 41, pp. 6661-6676, 2004.
[52]. C.C Chang and L.W. Chen, “Detection of the location and size of cracks in the multiple cracked beam
by spatial wavelet based approach”, Mechanical Systems and Signal Processing, Vol. 19, pp. 139-155,
2005.
[53]. U.P. Poudel, G. Fu and J. Ye, “Structural damage detection using digital video imaging technique and
Wavelet transformation”, Journal of Sound and Vibration, Vol. 286, pp. 869-895, 2005.
[54]. U.P. Poudel, G. Fu and J. Ye, “Wavelet transformation of mode shape difference function for structural
damage location identification”, Earthquake Engineering and Structural Dynamics, Vol. 36, pp. 1089-
1107, 2007.
[55]. W.J.Staszewski and G.R. Tomlinson, “Application of the wavelet transform to fault detection in a spur
gear”, Mechanical Systems and Signal Processing, Vol. 8, No.3, pp.289-307, 1994.
[56]. W.J. Wang and McFadden, “Application of orthogonal wavelets to early gear damage detection”,
Mechanical Systems and Signal Processing, Vol. 9, No.5, pp. 497-507, 1995.
[57]. W.J. Wang and McFadden, “Application of wavelets to gearbox vibration signals for fault detection,
Journal of Sound and Vibration, Vol. 192, No.5, pp. 927-939, 1996.

57
[58]. B. Li, G. Goddu and M.Y. Chow, “Detection of common motor bearing faults using frequency-domain
vibration signals and a neural network based approach”, Proceedings of the 16th American Control
Conference, Philadelphia, PA, June, 1998.
[59]. G. Liao, S. Liu, T. Shi and G. Zhang, “Gearbox condition monitoring using self-organizing feature
maps”, Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical
Engineering Science, Vol. 218, pp. 119-130, 2004.
[60]. C. Kar and A.R. Mohanty, “Technical note: gearbox health monitoring through multiresolution fourier
transformation of vibration and current signals”, Structural Health Monitoring, Vol. 5, pp. 195-200,
2006.

Signal-Based Damage Detection Methods – Algorithms and Applications

Recommended

More Related Content

What's hot (15)

Similar to Signal-Based Damage Detection Methods – Algorithms and Applications (20)

More from IJERD Editor (20)

Recently uploaded (20)

Signal-Based Damage Detection Methods – Algorithms and Applications