期刊论文全文 5 Empirical Wavelet Transform

International Journal of Structural Stability and Dynamics
Vol. 21, No. 4 (2021) 2150046 (25 pages)

#.c World Scienti¯c Publishing Company
DOI: 10.1142/S0219455421500462
by NANYANG TECHNOLOGICAL UNIVERSITY on 06/15/22. Re-use and distribution is strictly not permitted, except for Open Access articles.
Time-Varying Parameter Identi¯cation of Bridges Subject to

Moving Vehicles Using Ridge Extraction Based on
Empirical Wavelet Transform
Jiantao Li*, Jian Guo*,‡ and Xinqun Zhu†

Int. J. Str. Stab. Dyn. 2021.21. Downloaded from www.worldscientific.com
*Institute
of Bridge Engineering
School of Civil Engineering
Zhejiang University of Technology
Hangzhou 310000, Zhejiang, P. R. China
†
School of Civil and Environmental Engineering
University of Technology Sydney
Broadway, NSW 2007, Australia
‡
guoj@zjut.edu.cn
Received 10 August 2020

Accepted 26 November 2020
Published 18 January 2021
For a vehicle moving over a bridge, the vehicle-bridge interaction (VBI) embraces the time-
varying modal parameters of the system. The identi¯cation of non-stationary characteristics of
bridge responses due to moving vehicle load is important and remains a challenging task. A new
method based on the improved empirical wavelet transform (EWT) along with ridge detection
of signals in time-frequency representation (TFR) is proposed to estimate the instantaneous
frequencies (IFs) of the bridge. Numerical studies are conducted using a VBI model to inves-
tigate the time-varying characteristics of the system. The e®ects of the measurement noise, road
surface roughness and structural damage on the bridge IFs are investigated. Finally, the dy-
namic responses of an in-situ cable-stayed bridge subjected to a passing vehicle are analyzed to
further explore the time varying characteristics of the VBI system. Numerical and experimental
studies demonstrate the feasibility and e®ectiveness of the proposed method on the IF esti-
mation. The identi¯ed IFs reveal important time-varying characteristics of the bridge dynamics
that is signi¯cant to evaluating the actual performance of operational bridges in operation and
may be used for structural health assessment.
Keywords: Damage; vehicle-bridge interaction; instantaneous frequency; empirical wavelet

transform; time-frequency representation; ridge detection; surface roughness.
1. Introduction
The investigation on the bridge dynamic characteristics under operational tra±c
load is a signi¯cant part of the bridge structural health monitoring. Extensive study
‡
Corresponding author.
2150046-1
J. Li, J. Guo & X. Zhu
on the bridge dynamic characteristics has been conducted in a vehicle-bridge inter-

action (VBI) framework.1–4 Research shows that the VBI is a time-varying process
that the dynamic responses of the bridge are non-stationary.5,6 Li et al.7 theoretically
studied the natural frequency of railway girder bridges under vehicular load. The
bridge frequency was found varying periodically with the passage of the vehicle. Kim
et al.8 experimentally studied the e®ect of vehicle weight on the bridge natural fre-
quencies under tra±c-induced excitation. A non-negligible change 5.4% was noted in
the natural frequencies of a short span bridge when the mass ratio between the
vehicle and the bridge was 3.8%. Chang et al.9 conducted theoretical and experi-
mental studies to estimate the variability of bridge frequency due to a vehicle parked
on top. Yang et al.10 ¯rst identi¯ed the instantaneous frequencies (IFs) of bridge
from the moving vehicle responses. The literature11 presented the state-of-the-art
review of the vehicle-based methods for detecting the di®erent modal parameters of
highway bridges. Cantero et al.12 assessed the non-stationary and nonlinear features
of a scaled VBI test bed experimentally. The identi¯cation of the time-varying dy-
namic characteristics is critical to accurately assess the structural performance and
condition of target structure.13 The time-varying parameter identi¯cation of bridges
under moving vehicle loads is still a big challenge.
Non-parametric-based time-frequency analysis techniques, i.e. short time Fourier
transform (STFT), Winger–Ville distribution (WVD), Empirical Mode Decompo-
sition (EMD) and wavelet transform (WT), are extensively used to study dynamic
characteristics of non-stationary structural responses.14,15 Yang et al.16 brie°y
reviewed these methods and their engineering applications. For STFT, the time-
frequency representation (TFR) is determined by the window function and the
resolutions are ¯xed that may result in a de¯ciency in extracting accurate time-
frequency characteristics from non-stationary signals. The TFR based on continuous
WT is a trade-o® between time-frequency resolutions, while the results of WVD are
a®ected by the cross-terms in the instantaneous auto-correlation function. EMD and
the ensemble EMD become one of the most popular techniques to analyze non-
stationary and nonlinear dynamic response signals. It decomposes a signal into a
¯nite sum of Intrinsic Mode Functions (IMFs) and Hilbert transform (HT) is then
used to extract the instantaneous modal frequencies from IMFs. However, the EMD
su®ers from the problems of mode mixture and pseudo-IMFs and the issue of its lack
of mathematical theory. The synchrosqueezing transform (SST), an EMD-like tool,
was recently proposed to analyze the non-stationary signals.17,18 SST-based methods
have been developed for the time-frequency analysis and have presented better es-
timation results over EMD.19,20 However, the SST method uses a classic wavelet
analysis instead of a full adaptive WT. Gills21 developed the Empirical Wavelet
Transform (EWT) based on wavelet decomposition. EWT is an adaptive approach
for conveniently extracting the di®erent modes of a signal by constructing a wavelet
¯lter bank. The results have shown its feasibility and e®ectiveness compared to the
classic EMD. It has been successfully applied to machinery fault diagnosis.22–24
Kedadouche et al.25 compared the EWT and EMD methods in the application to the
2150046-2
Time-Varying Parameter Identi¯cation of Bridges Subject to Moving Vehicles
bearing e®ect diagnosis. The results showed that the EWT was much e±cient
compared to EMD and Ensemble EMD on mode estimates and computation time.
Recently, EWT was applied to identify modal parameters of civil structures.26 Xia
and Zhou27 adopted EWT to obtain the mono-component of structural response and
employed HT to extract the time-varying features for condition assessment of civil
structures.
The EWT approach is to extract the di®erent modes by designing the appropriate
wavelet ¯lter bank and provides a consistent decomposition. To the authors' best
knowledge, there is a little research on using or improving EWT method for time-
varying parameter identi¯cation, especially for VBIsystems. Xin et al.28 successfully
applied the improved EWT for operational modal identi¯cation of linear structures.
This method is enhanced using the time-frequency analysis based on Synchroex-
tracting transform to determine the ¯ltering boundaries of EWT.29 The results

showed that the EWT is more reliable and accurate in identifying the time-varying
components of the signal than that by the variational mode decomposition.30 The
experimental results showed the potential application to track the passage of the
heavy vehicles on the bridge.29 There is no detailed study for time-varying char-
acteristics of bridges subjected moving vehicles. This study is to develop a new
method by integrating the wavelet ridge extraction and improved EWT to identify
the time-varying parameter of the vehicle-bridge system.
This paper proposes a hybrid framework based on an improved EWT for the
analysis of non-stationary bridge responses under operational vehicle load and
extractingIFs. The framework involves two steps. In the ¯rst step, an improved
EWT is used to decompose the measured bridge responses into a number of IMFs.
The second step is to extract IFsfrom the TFR of IMFs by ridge detection. Section 2
presents the formulation of the VBI model for analysis. This study focuses on the
popular beam bridge. The 2D VBI model is considered with the road surface
roughness. In Sec. 3, the background of the EWT is brief introduced and the pro-
posed method for the estimation of time-dependent characteristic is described.
Section 4 conducts numerical analysis on the estimation of instantaneous modal
frequencies to verify the feasibility of proposed framework. Experimental analysis on
an actual cable-stayed bridge is conducted in Sec. 5 followed by the conclusions in
Sec. 6.
2. Formulation of the VBI Model

2.1. Equation of motion of a bridge subjected to a moving vehicle
A VBI model, as shown in Fig. 1, is considered. The vehicle is modelled as a 2-DoF
quarter car and the bridge as a simply-supported beam with length L. The vehicle
parameters, mv and m1 ; are the masses of vehicle body and axle, respectively. ks and
cs are the sti®ness and damping of suspension spring and damper, respectively. kt is
the sti®ness of the tire. The vehicle is assumed to move along the bridge deck at a
2150046-3
Fig. 1. VBI model.
constant velocity v. The equation of motion for the vehicle can be expressed as
" #( ) ( ) " #
mv 0 y€v cs cs y_ v ks ks yv
þ þ
0 m1 y€1 cs cs y_ 1 ks ks y1
( ) ð1Þ
0
¼ ;
ðmv þ m1 Þg Pint ðtÞ
where yv and y1 are the displacement responses of the vehicle body and axle, re-
spectively. The interaction force between the bridge and the vehicle
Pint ðtÞ ¼ ðmv þ m1 Þg þ kt ðy1 yðxðtÞÞ rðxðtÞÞÞ. yðxðtÞÞ is the displacement re-
sponse of the bridge, and rðxðtÞÞ is the road surface roughness of the bridge deck at
the location xðtÞ ¼ vt. Equation (1) is rewritten as
:
€ v ðtÞ þ Cv dv ðtÞ þ Kv dv ðtÞ ¼ DðFv Pint Þ:
Mv d ð2Þ
As shown in Fig. 1, the planar Euler–Bernoulli beam bridge can be modeled with n
¯nite elements. The equation of motion of the bridge can be expressed as31
:
€ b ðtÞ þ Cb db ðtÞ þ Kb db ðtÞ ¼ Hc ðtÞPint ðtÞ;
Mb d ð3Þ
where db denotes the vertical displacement vector of the bridge. Mb , Cb , Kb are the
mass, damping and sti®ness matrices of the bridge, respectively. Hc ðtÞ ¼
f0; 0; . . . ; Hi ðtÞ; . . . ; 0g T is a function of time and Hi ðtÞ is the vector of shape function
in the ith element on which the moving vehicle is located at time instant t, and it can
be expressed as Hi ðtÞ ¼ f1 3 2 þ 2; ð 2 2 þ 3 Þle ; 3 2 2 3 ; ð 2 þ 3 Þle g,
with ¼ ðxðtÞ xi Þ=le , xi ¼ ði 1Þle ; where le is the length of the element.
Combining Eqs. (2) and (3), the coupled VBI system can be obtained as
:: :
MDðtÞ þ CDðtÞ þ KDðtÞ ¼ FðtÞ; ð4Þ
8 9 2 3 2 3 2 3
< db ðtÞ = Mb Hc ðtÞmv Hc ðtÞm1 Cb 0 0 Kb 0 0
6 ks 7
where DðtÞ ¼ : yv ;, M ¼ 4 0 mv 0 5, C ¼ 4 0 cs cs 5, K ¼ 4 0 ks 5,
y1 0 0 m1 0 cs cs H Tc ðtÞkt ks ks þ kt
8 9
< Hc ðtÞðmv þ m1 Þg =
FðtÞ ¼
:
0
;
.
kt rðxðtÞÞ
The system matrices of the coupled interaction system in Eq. (4) are noted time-
dependent according to the location of the interacting force and the frequencies of the
system is time-varying.
2150046-4
3. Theoretical Background of the Hybrid Analysis Method

for IF Estimations
3.1. The empirical wavelet transform
A dynamic response signal of a bridge sðtÞ can be assumed to consist of N mono-

components fsi ðtÞ; i ¼ 1; 2; . . . ; NÞg, which correspond to modes centered on a spe-
ci¯c frequency of the spectrum. The EWT divides the di®erent modes mainly in three
steps. First, the frequency spectrum of the signal sðtÞ is obtained with the frequency
range ½0; by using Fourier transform. Second, the spectrum along the Fourier axis
is segmented into N number of contiguous portions according to the local maxima of
spectrum. The individual segments correspond to di®erent modes that are centered
around each local maximum.
The boundary between each segment is denoted as !n and each segment is ¯ltered
by an interval ½!n1 ; !n with ð!0 ¼ 0 and !N ¼ Þ. A transition phase Tn of width
2 n is de¯ned for each !n . The n can be de¯ned as
n ¼ !n ð5Þ
and is chosen by 0 < < minn !!nþ1 !n
nþ1 þ!n
. At last, the ¯lter bank for each segment is
de¯ned by the empirical scaling function and the empirical wavelets expressed as
8
> 1; if j!j ð1 Þ!n ;
>
>
< 1
^n ð!Þ ¼ cos ðj!j ð1 Þ!n Þ ; if ð1 Þ!n j!j ð1 þ Þ!n ;
>
> 2 2!n
>
:
0; otherwise;
ð6Þ
8
>
> 1; if ð1 þ Þ!n j!j ð1 Þ!nþ1 ;
>
>
>
> 1
>
< cos 2 2!
> ðj!j ð1 Þ!nþ1 Þ ; if ð1 Þ!nþ1 j!j ð1 þ Þ!nþ1 ;
nþ1
^ n ð!Þ ¼
>
> 1
>
> sin ðj!j ð1 Þ! n Þ ; if ð1 Þ!n j!j ð1 þ Þ!n ;
>
> 2 2!n
>
>
:
0; otherwise;
ð7Þ
where the function ðxÞ is an arbitrary function de¯ned on [0,1] that can be
expressed as
8
< 0; if x 0;
ðxÞ ¼ ðxÞ þ ð1 xÞ ¼ 1; 8x 2 ½0; 1; ð8Þ
:
1; if x 1
and the most used one in the literatures21,32 is
ðxÞ ¼ x 4 ð35 84x þ 70x 2 20x 3 Þ: ð9Þ
2150046-5
After the scaling function and empirical wavelets are derived, the detailed coe±cients
are given by the inner products with the empirical wavelets ^ n
Ws ðn; tÞ ¼ hsðtÞ; ^ n i ¼ ðsð!Þ ^ n ð!ÞÞ v ð10Þ
and the approximation coe±cients are given by the inner product with the scaling
function:
Ws ð0; tÞ ¼ hsðtÞ; ^1 i ¼ ðsð!Þ ^1 ð!ÞÞ v ; ð11Þ

v
where denotes the complex conjugate, ð:Þ denotes the inverse Fourier transform,
and h:; :i denotes the inner product.
The empirical mode sn is given by
s0 ðtÞ ¼ Ws ð0; tÞ 1 ðtÞ; ð12aÞ

sn ðtÞ ¼ Ws ðn; tÞ n ðtÞ; ð12bÞ
where denotes a convolution.
3.2. The improved EWT approach

Using Fourier spectrum for determining the boundaries associated with EWT
analysis of noisy and non-stationary signals may be challenging or may lead to false
modes.28 The AR power spectrum is smoothed with a lower level variance compared
with Fourier spectrum. It can better de¯ne the boundaries of EWT than using
Fourier spectrum. Therefore, the AR power spectrum is employed to improve the
e®ectiveness in de¯ning the boundaries for using EWT to perform the signal de-
composition. The most common form of parametric spectral density estimate uses an
autoregressive model AR (p) of order p. The signal sequence sðtÞ obeying a zero mean
AR (p) process can be expressed as33
sðtÞ ¼ 1 sðt 1Þ þ 2 sðt 2Þ þ þ p sðt pÞ þ ðtÞ; ð13Þ
where p denotes the order of the AR model; the 1 ; . . . ; p are the model coe±cients
and ðtÞ is a white noise process with zero mean and innovation variance 2p . The
power spectrum of the process PAR ðe jw Þ is
2
p
PAR ðe jw Þ ¼ Pp : ð14Þ
jwk 2
1 k¼1 k e
There are a number of approaches to determine the order q, i.e. Singular

Value Decomposition, Akaika Information Criterion, and Final Prediction Error
Criterion.34 Akaike's Final Prediction Error is used to get the optimal q of AR model.
When the optimal order is identi¯ed, the model parameters and thus the spectral
density can be estimated. The Burg algorithm is used to calculate the AR power
spectrum in this study.
2150046-6
3.3. IFs estimation of the decomposed modes

Once the component is extracted by EWT, the instantaneous frequency can be
estimated. In this study, the technique for the extraction of IFs from ridges in TFR of
signals35 is adopted. First, WT is used to obtain the TFRof the response compo-
nent,36 and the obtained TFR is Hs ð!; tÞ. Second, the fast path optimization ap-
proach employing a support function is developed to optimally extract the ridge of
the TFR corresponding to the IFs. If the ridge frequencies at each time is denoted as
vm ðtÞ, TFR amplitudes as Qm ðtÞ, and their numbers as Np ðtÞ, the following relation
can be obtained:
(
½@ ! jHs ð!; tÞj!¼vm ðtÞ ¼ 0;
vm ðtÞ : ð15aÞ
½@ 2! jHs ð!; tÞj!¼vm ðtÞ < 0;
Qm ðtÞ jHs ðvm ðtÞ; tÞj: ð15bÞ

Since the Hs ð!; tÞ has a time span tn ¼ ðn 1Þt for n ¼ 1; . . . ; N, the path opti-
mization can be described as
arg max
fmc ðt1 Þ; . . . ; mc ðtN Þg ¼ F½t ; Q ðt Þ; v ðt Þ; fvm1 ðt1 Þ; . . . ; vmN ðtN Þg;
fm1 . . . ; mN g n m n m n
ð16Þ
Fig. 2. Flowchart of the IF estimation approach.
2150046-7
where mc ðtn Þ is the sequence of peak indices to be extracted as the ridge at tn , and F
is the chosen support function for the optimization. Last, an adaptive support
function is constructed, and the optimization problem is solved with the fast path
optimization. The detailed procedure of the scheme can be referred to literature.35,37

The dynamic response of the bridge is calculated using Newmark-ß method.
White noise is added to the calculated responses to simulate the measured signal
sðtÞ as
sðtÞ ¼ acccal þ Ep Noise ðacccal Þ; ð17Þ
where acccal is the calculated acceleration response, Ep is the noise level, N oise is a
standard normal distribution vector with zeros mean and unit variance, and ðacccal Þ
is the standard deviation of the calculated acceleration response. Based on the above
analysis procedures, the °owchart of the proposed IF estimation method is shown in

Fig. 2.
4. Numerical Study on Time-Varying Parameter Identi¯cation of the

Bridge Subject to a Moving Vehicular Load
The VBI system, as shown in Fig. 1, is studied. The select properties of the bridge
are: L ¼ 30 m, ¼ 6000 kg/m, EI ¼ 2:3 e10Nm. Rayleigh damping is assumed with
Cb ¼ 1 Mb þ 2 Kb and 1 ¼ 0:344, 2 ¼ 0:0002, where Mb , Cb and Kb are the
mass, damping and sti®ness matrices of the bridge, respectively. The ¯rst three
theoretical bridge modal frequencies are 3.42, 13.68 and 30.77 Hz, respectively.
4.1. Moving mass case

A vehicle moving over the bridge is simpli¯ed as a moving mass in this case. The
mass can be viewed as extra weight to the bridge system that would change the mass
property of the bridge structure. Since the location of the extra mass on the bridge is
time varying, the physical property of the bridge system is also time dependent. As a
result, the modal frequencies of the bridge system, are time varying. Therefore, the
IFs, instead of the ¯xed modal frequencies, are more suitable to describe the oper-
ational dynamic performance of the bridge system. A mass of 6000 kg is assumed to
move over the bridge with a speed of 5 m/s. The dynamic response measured at the
3/8 span of the bridge is analyzed using the proposed method to get the IFs under the
moving mass. Figure 3(a) shows the AR spectrum of the signal and the identi¯ed
boundaries. The ¯rst three bridge modal frequencies are identi¯ed. The EWT is used
to decompose the response and the obtained three modes corresponding to the bridge
vibration are shown in Fig. 3(b). The TFR of each mode is calculated and the three-
dimensional plotting of the results is shown in Fig. 4. The ¯gures show that the ¯rst
few modes are the bending modes. The ¯nal results of the analysis are the ridge point
of the TFR in Fig. 4 corresponding to the IFs of bridge vibration mode. In order to
evaluate the accuracy of the proposed method on the identi¯cation, the identi¯ed IFs
2150046-8
(a) The obtained AR spectrum and identi¯ed boundaries.
(b) Extracted ¯rst three components.
Fig. 3. Application of the improved EWT.
2150046-9
2150046-10
(a) TFR of ¯rst mode.
(b) TFR of second mode.
Fig. 4. TFR of obtained vibration mode.

2150046-11
Fig. 4. (Continued )
(c) TFR of third mode.
Fig. 5. Comparison between the identi¯ed IFs and theoretical values.

are compared to the theoretical values which are obtained via eigenvalue analysis.
The comparison of the results is given in Fig. 5. It can be seen that the variation
trend of the frequencies identi¯ed is consistent to the theoretical values. The IF of the
¯rst vibration mode identi¯ed using the proposed method is very close to the theo-
retical values. The obvious di®erence at the beginning and the end of the results is
due to the very small dynamic amplitude of the bridge when the mass is close to the
supports and the end e®ect of the TFR. The values of the second and third modes are
a bit smaller than the theoretical values. The theoretical values are calculated based
on the eigenvalues of M 1 K and M and K are the mass and sti®ness matrices in
Eq. (4), respectively. The damping has not been considered in the calculation. The
identi¯ed results using the proposed method are obtained from the bridge responses
based on Eq. (4) and the damping is included. The identi¯ed results are smaller than
the theoretical values due to the damping e®ect. Besides, some local °uctuations of
the IFs are observed for those two modes. The reason can be due to the system
frequencies modulated by the driving frequency n v L , where n is the mode order, v is
the moving speed of vehicle and L is the bridge length.38
4.2. Moving vehicle case

For more practical study, a vehicle model with parameters mv ¼ 4500 kg,
m1 ¼ 1500 kg, kt ¼ 2:00 e4N/m, ks ¼ 5 e5N/m and cs ¼ 5000 N/m/s is adopted. The
bounce frequency of vehicle body vibration is 0.33 Hz and 2.96 Hz is for the axle mass
vibration, respectively. The moving speed is 10 m/s with sampling frequency 200 Hz.
The feasibility of the proposed study is illustrated with numerical results with con-
sideration of di®erent possible e®ects in practice, e.g. measurement noise, road sur-
face roughness, etc. The values selected for these parameters are realistic with
reference to similar published research studies in VBI.39 The e®ects of the road
surface roughness and damage on the time varying modal properties of the bridge are
considered in the VBI model, as shown in Eq. (4), where rðxðtÞÞ is the road surface
roughness and the damage is simulated as the sti®ness reduction in bridge sti®ness
matrix in Eq. (3). The white noise is added to the calculated bridge responses from
Eq. (4) to simulate the measurements using Eq. (17). These factors will a®ect the
bridge dynamic responses sðtÞ in Eqs. (10) and (11). Therefore, the measurement
noise, surface roughness and damage are the key factors to consider their e®ects on
the time varying properties of the VBI system. The dynamic response of the bridge is
calculated, and 10% white noise is added to simulate the measurement. By using the
constructed ¯lter bank, the empirical modes in dynamic responses sðtÞ are extracted.
4.2.1. Comparison of IF estimation using di®erent methods

The dynamic response of the bridge under a moving vehicle is analyzed when the
bridge road surface is smooth. The IFsare extracted from the bridge response using
the proposed method and compared with that by SST and HT. Figure 6 shows the
signal decomposition results using EWT. In Fig. 6(a), three dominant peaks in the
2150046-12
(a) Spectrum of the response and identi¯ed boundaries.
(b) First three components.
Fig. 6. Signal decomposition using improved EWT.
2150046-13
2150046-14
(a) IFs of ¯rst mode.
Fig. 7. Signal decomposition using improved EWT.
Fig. 8. Estimated IFs of bridge response using three di®erent methods.

2150046-15
(b) IFs of second mode.
Fig. 9. Estimated IFs considering di®erent measurement noise.

Fig. 10. The class A bridge surface roughness.
spectrum are corresponding to the bridge modal frequencies, while the vehicle-related
frequency is invisible. The local maxima in the spectrum can be successfully used to
detect the boundary. The signal is decomposed by EWT and the bridge vibration
related three components are extracted as shown in Fig. 6(b). The IFs identi¯ed for
the moving vehicle case are shown in Fig. 7 and the results are compared with those
of moving mass case. It can be seen that the amplitude of bridge frequency variation
under moving vehicle is smaller than that of moving mass. Figure 8 compares the
identi¯ed IFs using the proposed method with that by SST and HT. For comparison,
only the identi¯ed ¯rst and second bridge modal frequencies are presented in Fig. 8.
The results indicate that the identi¯ed IFs from three di®erent methods present
a similar variation trend and the proposed method outperforms the other two
methods in regard to the resolution. The rest of the study focuses on the identi¯ed
IFs results.
4.2.2. E®ect of measurement noise

The e®ects of measurement noise on the identi¯cation of the IFs using the proposed
method are studied. Other two additional noise levels, i.e. 15% and 20%, are included
to the calculated bridge responses. The IF trajectories identi¯ed from the responses
are given in Fig. 9. In this ¯gure, it can be seen that the identi¯ed IF trajectories
related to vehicle frequency and ¯rst bridge modal frequency are very close for
di®erent noise levels. For the IF trajectory of the second bridge modal frequency,
2150046-16
Fig. 11. IFs estimation when bridge surface roughness is considered.
there is an obvious di®erence in the identi¯ed results. The di®erence can be caused by
the e®ect of the driving frequency. Thus, the modal frequency of higher modes may
present higher di®erence due to the e®ect of noise.
Fig. 12. IFs estimation when bridge is damaged.
2150046-17
4.2.3. E®ect of bridge surface roughness

A Class A bridge road surface roughness,40 as shown in Fig. 10, is introduced in the
numerical model and the bridge response is calculated. The proposed method is used
to extract the IFs of the response and the identi¯ed results are given in Fig. 11. It is
illustrated that the bridge surface roughness changes the variation trend of the
response and ampli¯es the variation amplitude. Therefore, for the VBI based bridge
health monitoring, the e®ects of the roughness should be considered carefully. By
considering the case when 20% measurement is included, it can be seen that the e®ect
of road surface roughness on the IFs results is more prominent than that of mea-
surement noise.
4.3. Beam with crack damage

A crack is introduced to the bridge model to simulate the bridge structural damage.
According to Sinha et al.,41 the crack damage causes a linearly varying decrease in
°exural sti®ness in its vicinity. The crack is assumed at the 1/3 span of the bridge
with the crack level 10% and 20% of the overall bridge cross-section depth,
(a) The cable-stayed bridge.
(b) Sensor location.
Fig. 13. Long-term monitoring of a cable-stayed bridge.
2150046-18
Table 1. The identi¯ed bridge modal frequencies via modal analysis.
Mode number Frequency (Hz) Vibration mode

1 2.032 First vertical bending mode of the deck

2 3.548 Double mode that corresponds to the second vertical bending mode
3 3.649
4 5.584 Third bending mode
5 6.136 A mixture of torsion and bending modes
respectively. Bridge response is analyzed to extract the IFs. From the results, as
shown in Fig. 12, the bridge frequencies decrease due to the crack damage. The
results con¯rmed that the change of the frequency can be used for damage detection.
However, the identi¯ed IFs do not give information on the damage location.
5. IF Estimation of the Response of an Actual Bridge

A long-term monitoring system has been installed on a cable-stayed bridge, as
shown in Fig. 13(a). It is a single lane highway bridge with a span of 46 m and a
width of 5 m. The bridge connects the South and North campuses of Western
Sydney University in New South Wales. There are 24 accelerometers installed on
Fig. 14. A set of acceleration response and the response spectrum.
2150046-19
2150046-20
(b) IFs estimation using proposed method.
(a) AR spectrum and identi¯ed boundaries.
Fig. 15. The estimation of IFs of bridge response.

2150046-21
(c) IFs estimation using SST.
(d) IFs estimation fromRef. 43.

the bridge deck, and Fig. 13(b) indicates the sensor locations. A data acquisition
system continuously records the data from sensors with a sampling rate of 600 Hz.
Sun et al.42 conducted modal identi¯cation using a series of bridge responses under
ambient excitation. The identi¯ed ¯rst eight modal frequencies are summarized in
Table 1.
The vehicle-induced response of the bridge is studied. Figure 14 shows the
acceleration response at sensor A10 and its spectrum when a vehicle is passing over
the bridge. Both the wavelet SST and the proposed methods are used to analyze
the signal. Figure 15(a) shows the AR power spectrum and the identi¯ed
boundaries. The identi¯ed IFs using the proposed method are shown in Fig. 15(b)
and the IFs extracted using SST are given in Fig. 15(c). For comparison, the
estimated IFs results from a recent study43 are also presented in Fig. 15(d). As
shown in Fig. 15(b), the ¯rst ¯ve IFs are estimated by the proposed method. The
associated bridge vibration mode of the IFs is indicated in the ¯gure. The IFs of
¯rst vertical bending mode and the coupled Mode 2 and 3 are very similar to the
results from SST and reference, as shown in Figs. 15(c) and 15(d), respectively.
The identi¯ed IFs of coupled 6 and 7 mode from three methods are also compa-
rable. Moreover, the results of the proposed method and that from the reference
give the time-varying dynamic information related to the coupled Mode 4 and 5. It
is demonstrated that the proposed method successfully reveals more time-varying
characteristics related to the vehicle-induced bridge response than that from the
other methods.
6. Conclusions
The identi¯cation of time-varying characteristics of vehicle-induced bridge dynamic
response is a complex yet important task. This paper focuses on the VBIsystem and
proposes a hybrid method for the analysis of the non-stationary bridge responses.
The method combines the improved EWT and a TFR ridge detection technique to
extract the IFs. Numerical and experimental studies are conducted, and their results
veri¯ed the feasibility and e®ectiveness of the proposed method comparing with some
commonly used methods. The implementation of the proposed method to the re-
sponse of an in-situ bridge reveals the important time-dependent evolution infor-
mation of the bridge dynamic characteristics due to the moving operational vehicle
load. In this study, the beam bridge is the main focus and it could be simpli¯ed as a
2D VBI model. A 3D VBI model is the future work to be conducted when the
torsional mode of the bridge deck is critical.
Acknowledgments
This research is supported in part by research funding of the National Natural
Science Foundation of China (U1709207, 52078461), the Key R&D program of
Zhejiang (2019C03098) and the Postdoctoral Research Project of Zhejiang Province
2150046-22
(ZJ2020024). The ¯nancial aid is gratefully acknowledged. DATA61 is greatly ac-

knowledged for the instrumentation of the ¯eld bridge for monitoring.
References
1. J. H. Kim and J. P. Lynch, Experimental analysis of vehicle-bridge interaction using a
wireless monitoring system and a two-stage system identi¯cation technique, Mech. Syst.
Signal Process. 28 (2013) 3–19.
2. A. Elhattab, N. Uddin and E. J. O'Brien, Drive-by bridge damage monitoring using
bridge displacement pro¯le di®erence, J.Civil Struct. Health Monitor. 6(5) (2016)
839–850.
3. Y. B. Yang and J. P. Yang, State-of-the-art review on modal identi¯cation and damage
detection of bridges by moving test vehicles, Int. J. Struct. Stab, Dyn. 18(2) (2017)
1850025.
4. Z. Chen, Z. Xie and J. Zhang, Measurement of Vehicle-Bridge-Interaction force using
dynamic tire pressure monitoring, Mech. Syst. Signal Process. 104 (2018) 370–383.
5. X. Q. Zhu and S. S. Law, Structural health monitoring based on vehicle-bridge interac-
tion: Accomplishments and challenges, Adv. Struct. Eng. 18(12) (2015) 1999–2015.
6. F. Xiao, G. S. Chen, J. L. Hulsey and W. Zatar, Characterization of non-stationary
properties of vehicle–bridge response for structural health monitoring, Adv. Mech. Eng.
9 (2017) 1687814017699141.
7. Li, J. Z., Su, M. B. and Fan, L. C., Natural frequency of railway girder bridges under
vehicle loads, J. Bridge Eng. ASCE 8(4) (2003) 199–203.
8. C. Y. Kim, D. S. Jung, N. S. Kim, S. D. Kwon and M. Q. Feng, E®ect of vehicle weight on
natural frequencies of bridges measured from tra±c-induced vibration, Earthquake Eng.
Eng. Vib. 2(1) (2003) 109–115.
9. K. C. Chang, C. W. Kim and S. Borjigin, Variability in bridge frequency induced by a
parked vehicle, Smart Struct. Syst. 13(5) (2014) 755–773.
10. Y. B. Yang, Y. C. Li and K. C. Chang, Constructing the mode shapes of a bridge from a
passing vehicle: A theoretical study, Smart Struct. Syst. 13(5) (2014) 797–819.
11. Y. B. Yang, Z. L. Wang, K. Shi, H. Xu and Y. T. Wu, State-of-the-art of the vehicle-based
methods for detecting the various properties of highway bridges and railway tracks, Int. J.
Struct. Stab. Dyn. 20(13) (2020) 2041004.
12. D. Cantero, P. McGetrick, C. W. Kim and E. J. O'Brien, Experimental monitoring of
bridge frequency evolution during the passage of vehicles with di®erent suspension
properties, Eng. Struct. 187 (2019) 209–219.
13. W. C. Su, C. Y. Liu and C. S. Huang, Identi¯cation of Instantaneous modal parameter of
time-varying systems via a wavelet-based approach and its application, Comput.-Aided
Civil Infrastruct. Eng. 29(4) (2014) 279–298.
14. N. E. Huang, Z. Shen, S. R. Long, M. C. Wu, H. H. Shih, Q. A. Zheng, N. C. Yen, C. C.
Tung and H. H. Liu, The empirical mode decomposition and the Hilbert spectrum for
nonlinear and nonstationary time series analysis, in Proc. R. Soc. London, A, Vol. 454
(The Royal Society, London, 1998), pp. 903–955.
15. Z. C. Wang, W. X. Ren and G. D. Chen, Time-frequency analysis and applications in
time-varying/nonlinear structural systems: A state-of-the-art review, Adv. Struct. Eng.
21(10) (2018) 1562–1584.
16. Y. Yang, Z. K. Peng, W. M. Zhang and G. Meng, Parameterised time-frequency analysis
methods and their engineering applications: A review of recent advances, Mech. Syst.
Signal Process. 119 (2019) 182–221.
2150046-23
17. I. Daubechies, J. Lu and H. T. Wu, Synchrosqueezed wavelet transforms: An empirical

mode decomposition-like tool, Appl. Comput. Harmonic Anal. 30(2) (2011) 243–261.
18. G. Thakur, E. Brevdo, N. S. Fuckar and H. T. Wu, The Synchrosqueezing algorithm for
time-varying spectral analysis: Robustness properties and new paleoclimate applications,

Signal Process. 93 (2013) 1079–1094.
19. J. S. Luo, S. H. Zhang, M. G. Zhong and Z. S. Lin, Order spectrum analysis for bearing
fault detection via joint application of synchrosqueezing transform and multiscale chirplet
path pursuit, Shock Vib. 2016 (2016) 2976389.
20. Q. T. Jiang and B. W. Suter, Instantaneous frequency estimation based on syn-
chrosqueezing wavelet transform, Signal Process. 138 (2017) 167–181.
21. J. Gilles, Empirical wavelet transform, IEEE Trans. Signal Process. 61(16) (2013)
3999–4010.
22. J. L. Chen, J. Pan, Z. P. Li, Y. Y. Zi and X. F. Chen, Generator bearing fault diagnosis for
wind turbine via empirical wavelet transform using measured vibration signals, Renewab.
Energy 89 (2016) 80–92.

23. Y. Hu, X. T. Tu, F. C. Li, H. G. Li and G. Meng, An adaptive and tacholess order analysis
method based on enhanced empirical wavelet transform for fault detection of bearings
with varying speeds, J. Sound Vib. 409 (2017) 241–255.
24. Z. C. Qiao, Y. Q. Liu and Y. Y. Liao, An improved method of EWT and its application in
rolling bearings fault diagnosis, Shock Vib. 2020 (2020) 4973941.
25. M. Kedadouche, M. Thomas and A. Tahan, A comparative study between empirical
wavelet transforms and empirical mode decomposition methods: Application to bearing
defect diagnosis, Mech. Syst. Signal Process. 81 (2016) 88–107.
26. J. P. Amezquita-Sanchez, H. S. Park and H. Adeli, A novel methodology for modal
parameters identi¯cation of large smart structures using MUSIC, empirical wavelet
transform, and Hilbert transform, Eng. Struct. 147 (2017) 148–159.
27. Y. X. Xia and Y. L. Zhou, 2019. Mono-component feature extraction for condition as-
sessment in civil structures using empirical wavelet transform, Sensors 19(19) (2019)
4280.
28. Y. Xin, H. Hao and J. Li, Operational modal identi¯cation of structures based on im-
proved empirical wavelet transform, Struct. Control Health Monitor. 26(3) (2019) e2323.
29. Y. Xin, H. Hao and J. Li, Time-varying system identi¯cation by enhanced empirical
wavelet transform based on synchroextracting transform, Eng. Struct. 196 (2019)
109313.
30. P. H. Ni, J. Li, H. Hao, Y. Xia, X. Y. Wang, J. M. Lee and K. H. Jung, Time-varying
system identi¯cation using variational mode decomposition, Struct. Control Health
Monitor. 25(6) (2018) e2175.
31. X. Q. Zhu, S. S. Law, L. Huang and S. Y. Zhu, Damage identi¯cation of supporting
structures with a moving sensory system, J. Sound Vib. 415 (2018) 111–127.
32. I. Daubechies, Ten lectures on wavelets. Society for industrial and applied mathematics,
CBMS-NSF Regional Conference Series in Applied Mathematics, 1992.
33. D. B. Percival and A. T. Walden, Spectral Analysis for physical Applications: Multitaper
and Conventional Univariate Techniques, Cambridge University Press, Cambridge, UK.
34. L. Ljung, System Identi¯cation: Theory for the User, 2nd edn. (Prentice-Hall PTR, Upper
Saddle River, New Jersey, 1999).
35. D. Iatsenko, P. V. E. McClintock and A. Stefanovska, Extraction of instantaneous fre-
quencies from ridges in time-frequency representations of signals, Signal Process. 125
(2016) 290–303.
2150046-24
36. Z. Feng, M. Liang and F. Chu, Recent advances in time-frequency analysis methods for
machinery fault diagnosis: A review with application examples, Mech. Syst. Signal Pro-
cess. 38(1) (2013) 165–205.
37. H. Huang, N. Baddour and M. Liang, Multiple time-frequency curve extraction Matlab
code and its application to automatic bearing fault diagnosis under time-varying speed
conditions, MethodsX 6 (2019) 1415–1432.
38. Y. B. Yang, C. W. Lin and J. D. Yau, Extracting bridge frequencies from the dynamic
response of a passing vehicle, J. Sound Vib. 272 (2004) 471–493.
39. Y. B. Yang, M. C. Cheng and K. C. Chang, Frequency variation in vehicle–bridge in-
teraction systems, Int. J. Struct. Stab. Dyn. 13(2) (2013) 1350019.
40. ISO-8608, Mechanical Vibration-Road Surface Pro¯les-Reporting Of Measured Data,
International Organization for Standardization (ISO), Geneva, Switzerland (1995).
41. J. K. Sinha, M. I. Friswell and S. Edwards, Simpli¯ed models for the location of cracks in
beam structures using measured vibration data, J. Sound Vib. 251(1) (2002) 13–38.
42. M. Sun, M. M. Alamdari and H. Kalhori, Automated operational modal analysis of a

cable-stayed bridge, J. Bridge Eng. 22(12) (2017) 05017012.
43. J. T. Li, X. Q. Zhu, S. S. Law and B. Samali, Time-varying characteristics of bridges
under the passage of vehicles using synchroextracting transform, Mech. Syst. Signal
Process. 140 (2020) 106727.
2150046-25

期刊论文全文 5 Empirical Wavelet Transform

Uploaded by

Document Information

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

期刊论文全文 5 Empirical Wavelet Transform

Uploaded by

Copyright:

Available Formats

International Journal of Structural Stability and Dynamics

Vol. 21, No. 4 (2021) 2150046 (25 pages)

Time-Varying Parameter Identi¯cation of Bridges Subject to

Jiantao Li*, Jian Guo*,‡ and Xinqun Zhu†

Received 10 August 2020

Keywords: Damage; vehicle-bridge interaction; instantaneous frequency; empirical wavelet

on the bridge dynamic characteristics has been conducted in a vehicle-bridge inter-

tracting transform to determine the ¯ltering boundaries of EWT.29 The results

2. Formulation of the VBI Model

Fig. 1. VBI model.

3. Theoretical Background of the Hybrid Analysis Method

A dynamic response signal of a bridge sðtÞ can be assumed to consist of N mono-

Ws ðn; tÞ ¼ hsðtÞ; ^ n i ¼ ðsð!Þ ^ n ð!ÞÞ v ð10Þ

Ws ð0; tÞ ¼ hsðtÞ; ^1 i ¼ ðsð!Þ ^1 ð!ÞÞ v ; ð11Þ

s0 ðtÞ ¼ Ws ð0; tÞ  1 ðtÞ; ð12aÞ

3.2. The improved EWT approach

There are a number of approaches to determine the order q, i.e. Singular

3.3. IFs estimation of the decomposed modes

Qm ðtÞ  jHs ðvm ðtÞ; tÞj: ð15bÞ

Fig. 2. Flowchart of the IF estimation approach.

optimization. The detailed procedure of the scheme can be referred to literature.35,37

analysis procedures, the °owchart of the proposed IF estimation method is shown in

4. Numerical Study on Time-Varying Parameter Identi¯cation of the

4.1. Moving mass case

(a) The obtained AR spectrum and identi¯ed boundaries.

(b) Extracted ¯rst three components.

Fig. 3. Application of the improved EWT.

J. Li, J. Guo & X. Zhu

(b) TFR of second mode.

Fig. 4. TFR of obtained vibration mode.

Fig. 5. Comparison between the identi¯ed IFs and theoretical values.

4.2. Moving vehicle case

4.2.1. Comparison of IF estimation using di®erent methods

(a) Spectrum of the response and identi¯ed boundaries.

(b) First three components.

Fig. 6. Signal decomposition using improved EWT.

J. Li, J. Guo & X. Zhu

Fig. 8. Estimated IFs of bridge response using three di®erent methods.

Fig. 9. Estimated IFs considering di®erent measurement noise.

Fig. 10. The class A bridge surface roughness.

4.2.2. E®ect of measurement noise

Fig. 11. IFs estimation when bridge surface roughness is considered.

Fig. 12. IFs estimation when bridge is damaged.

4.2.3. E®ect of bridge surface roughness

4.3. Beam with crack damage

(a) The cable-stayed bridge.

(b) Sensor location.

Fig. 13. Long-term monitoring of a cable-stayed bridge.

Table 1. The identi¯ed bridge modal frequencies via modal analysis.

Mode number Frequency (Hz) Vibration mode

1 2.032 First vertical bending mode of the deck

5. IF Estimation of the Response of an Actual Bridge

Fig. 14. A set of acceleration response and the response spectrum.

J. Li, J. Guo & X. Zhu

Fig. 15. The estimation of IFs of bridge response.

(d) IFs estimation fromRef. 43.

(ZJ2020024). The ¯nancial aid is gratefully acknowledged. DATA61 is greatly ac-

17. I. Daubechies, J. Lu and H. T. Wu, Synchrosqueezed wavelet transforms: An empirical

time-varying spectral analysis: Robustness properties and new paleoclimate applications,

Energy 89 (2016) 80–92.

42. M. Sun, M. M. Alamdari and H. Kalhori, Automated operational modal analysis of a

You might also like

Jiantao Li, Jian Guo,‡ and Xinqun Zhu†

Ws ð0; tÞ ¼ hsðtÞ; ^1 i ¼ ðsð!Þ ^1 ð!ÞÞ v ; ð11Þ

s0 ðtÞ ¼ Ws ð0; tÞ 1 ðtÞ; ð12aÞ

Qm ðtÞ jHs ðvm ðtÞ; tÞj: ð15bÞ