Drive-By Blind Modal Identification with

Singular Spectrum Analysis

Jiantao Li 1; Xinqun Zhu, M.ASCE 2; Siu-seong Law 3; and Bijan Samali 4
Downloaded from ascelibrary.org by University of Technology, Sydney on 05/05/19. Copyright ASCE. For personal use only; all rights reserved.

Abstract: Drive-by bridge parameter identification has been an active research area in recent years. An instrumented vehicle passing over a
bridge deck captures dynamic information of the bridge structure without bridge closure and on-site instrumentation. The vehicle dynamic
response includes components associated with the bridge surface roughness and the vehicle and bridge vibration. It is a challenge to separate
these components and extract the bridge modal parameters from the vehicle response. A novel drive-by blind modal identification with
singular spectrum analysis is proposed to extract the bridge modal frequencies from the vehicle dynamic response. The single-channel
measured vehicular response is decomposed into a multichannel data set using singular spectrum analysis, and the bridge frequencies
are then extracted via the blind modal identification. Numerical results showed that the proposed method is effective and robust to extract
the bridge frequencies from the vehicle response measurement even with Class B road surface roughness. The effects of the moving speed and
the vehicle parameters on the identification were studied. A vehicle–bridge interaction model in the laboratory was studied to further verify
the proposed method using one- and two-axle vehicles. DOI: 10.1061/(ASCE)AS.1943-5525.0001030. © 2019 American Society of Civil
Engineers.
Author keywords: Vehicle–bridge interaction; Drive-by blind modal identification; Singular spectrum analysis; Road surface roughness;
Instrumented vehicle.

Introduction response spectra of two successive vehicles to mitigate the impacts

Indirect bridge monitoring, also referred to as drive-by bridge
health monitoring, has been an active field of research in recent
years (Malekjafarian et al. 2015). Unlike the conventional direct
approach, the sensor is installed on the axle of a vehicle instead
of on the bridge deck. This indirect method is cost-effective and
convenient compared with the direct approach. There is great po-
tential for a quick scan of bridges in the road network using the
instrumented vehicle. The identification of bridge modal parame-
ters is a critical part of vibration-based structural health monitoring
(SHM). Yang et al. (2004) pioneered an indirect approach to extract
natural frequencies of bridge structures from the acceleration re-
sponse of a vehicle during its passage over the bridge deck. The
response of the moving vehicle contains dynamic information of
the structure. However, when the vehicle moves on a rough bridge
deck, the vehicular frequency usually dominates its response spec-
trum and masks the bridge frequency components. Yang et al.
(2012) used the technique of subtracting the responses or the
1 mated from modal response–related components. Recently, BSS
Ph.D. Student, School of Civil and Environmental Engineering,
Univ. of Technology Sydney, Broadway, Ultimo, NSW 2007, Australia.
Email: jiantao.li@student.uts.edu.au
2
Associate Professor, School of Civil and Environmental Engineering,
Univ. of Technology Sydney, Broadway, Ultimo, NSW 2007, Australia
(corresponding author). ORCID: https://orcid.org/0000-0001-5083-9320.
Email: xinqun.zhu@uts.edu.au
3
Professor, School of Civil Engineering, Chongqing Univ., Chongqing
400044, China. Email: siu-seong.law@connect.polyu.hk
4
Professor, School of Computing, Engineering and Mathematics,
Western Sydney Univ., Penrith, NSW 2751, Australia. Email: b.samali@
westernsydney.edu.au
Note. This manuscript was submitted on August 4, 2018; approved on
January 16, 2019; published online on May 6, 2019. Discussion period
open until October 6, 2019; separate discussions must be submitted for
individual papers. This paper is part of the Journal of Aerospace Engi-
neering, © ASCE, ISSN 0893-1321.
of road surface roughness on the identification of bridge frequen-
cies. However, the elimination or reduction of the road surface
roughness effect is still a challenge which needs further study
for practical application of the indirect method, especially with
only one instrumented vehicle (Zhu and Law 2015; Yang and Yang
2018).
Blind source separation (BSS) has been a promising tool for
output-only modal identification (Sadhu et al. 2017) in last decade.
BSS was originally used to recover special source components
from the measured data, and second-order blind identification
(SOBI) is used to solve the BSS problem (Antoni 2005). The math-
ematical equivalence between the modal expansion theorem and
the BSS methods has been studied (Kerschen et al. 2007; Poncelet
et al. 2007). The SOBI algorithm could produce components that
are mathematically equivalent to structural modal responses from
the measured data without any modifications (Zhou and Chelidze
2007). A framework for output-only blind modal identification
(BMID) was developed basing on SOBI (McNeill and Zimmerman
2008). Structural modal frequencies and damping ratios were esti-
has been modified and applied to nonstationary problems (Hazra
et al. 2010; Yang and Nagarajaiah 2012). It needs to have sufficient
independent observations, so the number of sensors should be
equal to or greater than the number of modes (McNeill and
Zimmerman 2008). In drive-by bridge modal parameter identifica-
tion using one instrumented vehicle, only one sensor is installed
on the vehicle and one single channel of measurement is available.
To solve the underdetermined problems in which the number of
observations is less than the number of active components, the
sparsity of sources is widely exploited in the time-frequency
domain (Zhen et al. 2017). Wang and Hao (2013) proposed a struc-
tural damage identification method based on compressive sensing
(CS). The application of CS relies on the sparsity of signals in
a transform domain. In this paper, to extract the independent
components for the bridge modal frequency identification with

© ASCE 04019050-1 J. Aerosp. Eng.

J. Aerosp. Eng., 2019, 32(4): 04019050

 one moving sensor in time domain, a preprocess to construct multi-
channel data sets from the single channel measurement is required
before applying the BMID method.
Singular spectrum analysis (SSA) is a data analysis technique
that can decompose a set of time series data into a finite number
of interpretable components in the time domain ordered by their
corresponding singular values (Liu et al. 2014). The obtained
components represent trends, oscillatory components, noise, and
so on. When the vehicle moves over a rough deck surface, the
Fig. 1. Vehicle–bridge system.
spectrum of the vehicle response contains a dominant component
related to the vehicular frequency. This component is taken as the
trend which masks the bridge-related frequencies in the spectrum
Downloaded from ascelibrary.org by University of Technology, Sydney on 05/05/19. Copyright ASCE. For personal use only; all rights reserved.

(Yang et al. 2013). Yang et al. (2013) used SSA to filter the ve-
hicle response component to improve the visibility of bridge re-
sponse components. However, underfiltering or overfiltering may
happen with the grouping. In this study, the vehicular response is
decomposed into a number of components as a multichannel
data set which is analyzed with BSS to identify the bridge modal
frequencies.
Some research has been conducted on the effect of some
influential factors on drive-by bridge frequency identification,
i.e., road surface roughness, vehicle properties, speed, ongoing
traffic, and measurement noise (Chang et al. 2010; Malekjafarian
and Obrien 2017). To the best knowledge of the authors, there is
little research on the component analysis of the measured vehicu-
lar response for the bridge modal frequency identification. In this
paper, the single-channel measurement–based blind modal identi-
fication method is proposed to extract the bridge modal frequen-
cies from dynamic responses of the vehicle passage over the
bridge deck. The SSA technique is used to separate the vehicle
response into multiple independent components, which are then
input into BSS to extract the modal responses for the indirect
identification of the bridge modal frequency. Numerical and ex-
perimental studies with a vehicle–bridge interaction model in the
laboratory are conducted to verify the proposed method. The ef-
fects of some influential system parameters on the identification
are also investigated.
Ÿ i þ 2ξ i ωi Ẏ i þ ω2i Y i ¼ Pi ðtÞ ð3Þ
where ωi and ξ i ¼ ith modal frequency and damping ratio
of the bridge. The modal force is given by Pi ðtÞ ¼
∫ L0 Fðx; tÞϕi ðxÞdx=M i , where Fðx; tÞ is the traffic excitation and
M i ¼ ith modal mass of the bridge.
Equation of Motion for Instrumented Vehicle
The instrumented vehicle is assumed to move over the deck at a
constant speed v (Fig. 1). The vehicle is modeled as a quarter-
car model with a single-degree-of-freedom (SDOF). The equation
of motion for the vehicle is
mv d̈v ðtÞ þ cv ḋv ðtÞ þ kv dv ðtÞ
¼ fcv ½ḋb ðx; tÞ þ vr 0 ðxÞ þ kv ½db ðx; tÞ þ rðxÞgx¼vt ð4Þ
where mv , kv , and cv = mass, stiffness, and damping of the vehicle,
respectively; dv ðtÞ, ḋv ðtÞ, and d̈v ðtÞ = vertical displacement, veloc-
ity, and acceleration of the vehicle, respectively; db ðx; tÞ and
ḋb ðx; tÞ = vertical displacement and velocity of the bridge at contact
point x and time t; rðxÞ = road surface roughness function;
and r 0 ðxÞ ¼ drðxÞ=dx.
The right-hand-side of Eq. (4) can be rewritten

Theoretical Background
fðtÞ ¼ fcv ½ḋb ðx; tÞ þ vr 0 ðxÞ þ kv ½db ðx; tÞ þ rðxÞgx¼vt ð5Þ
Equation of Motion for Bridge
With Duhamel’s integral, the dynamic response of the vehicle is
The equation of motion for the bridge is (Zhu and Law 2002)

Mb d̈b þ Cb ḋb þ Kb db ¼ F ð1Þ dv ðtÞ ¼ hv ðtÞ ⊗ fðtÞ ð6Þ

where Mb , Cb , and Kb = mass, damping, and stiffness matrices where hv ðtÞ = impulse response function of the vehicle system; and
of the bridge, respectively; F = vector of interaction forces acting ⊗ is the convolution operator.
on the bridge due to the traffic excitation; and db , ḋb , and d̈b = Ignoring the effect of the road surface roughness and the vehicle
vectors of displacement, velocity, and acceleration responses of damping and substituting Eqs. (2) and (5) into Eq. (6), the vehicle
the bridge, respectively. response can be written
The displacement of the bridge can be expressed as follows with
the modal superposition method (Clough and Penzien 1975):  XN 
dv ðtÞ ¼ hv ðtÞ ⊗ kv ϕi ðvtÞY i ðtÞ ð7Þ
X
N i¼1

db ðx; tÞ ¼ ϕi ðxÞY i ðtÞ ð2Þ

i¼1
where N = number of vibration modes considered; and ϕi ðxÞ and
Y i ðtÞ ¼ ith mode shape and modal response, respectively.
Substituting Eq. (2) into Eq. (1) and applying the orthogonality
conditions, Eq. (1) becomes
which is the convolution of the impulse response function and the
bridge response, and it includes the vehicle and bridge response
components. The vehicle response becomes more complicated
when the vehicle damping and the road surface roughness are con-
sidered, and there is a need for an effective tool to extract the bridge
response components from the vehicle response.

© ASCE 04019050-2 J. Aerosp. Eng.

J. Aerosp. Eng., 2019, 32(4): 04019050

 Drive-By Blind Modal Identification Using Singular
Spectrum Analysis
Drive-by blind modal identification with SSA mainly consists
of two steps. The first step is to decompose the vehicle response
into a set of independent time series data. The second step is to
extract the modal responses through BSS for the identification
of the bridge modal frequencies. Only the dynamic response meas-
urement of the vehicle when crossing the bridge deck is used in the
identification.
Grouping
All elementary matrices obtained in the previous step can be put
into N g groups using preset criteria. The grouping criteria depend
on the expected function of the SSA, e.g., denoising, smoothing,
harmonic component extracting, and so on. The elementary matri-
ces in the same group are summed, and N g resultant matrices,
i.e., Xg1 ; Xg2 ; : : : ; XgN g , can be obtained. The original trajectory
matrix X can then be expressed as
X ¼ Xg1 þ Xg2 þ · · · þXgN g ð11Þ

The groups can be formed based on the information contained in

Downloaded from ascelibrary.org by University of Technology, Sydney on 05/05/19. Copyright ASCE. For personal use only; all rights reserved.

Decomposition of Vehicle Response Using SSA

the singular vectors (Hassani 2007). Because the singular values are
In this step, the single-channel measured vehicle response is decom- arranged in descending order, the first few elementary matrices
posed into a multichannel data set using SSA. There are two stages contribute more than others to the trajectory matrix. Therefore each
for performing the SSA, decomposition and reconstruction. The major elementary matrix forms one group for the reconstruction in
first stage decomposes the time series into a set of elementary ma- the proposed method.
trices based on two separate steps: embedding and singular value
decomposition. The second stage extracts the constituting compo- Skew Diagonal Averaging
nents based on the diagonal averaging and grouping steps. The Each resultant matrix in the previous step is converted into a new
SSA adopted in this study is briefly described subsequently; more set of time series data with the same length as the original data set.
details were given by Liu et al. (2014). A skew diagonal averaging procedure is adopted to recover the time
series. Let Y be any of the resultant matrices Xgl , with the elements
Embedding denoted yij , i ¼ 1; 2; : : : ; N L , j ¼ 1; 2; : : : ; L. For N L < L, the
A measurement data vector dv ðtÞ ¼ ½d0 ; d1 ; d2 ; : : : ; dN−1  with recovered time series data dl ¼ ½dl0 ; dl1 ; : : : dlN−1  are given by
length N can be divided into L lagged vectors X i as fX i ¼
½di−1 ; di ; diþ1 ; : : : ; diþN L −2 T ; i ¼ 1; 2; : : : ; Lg, where N L is the 8
>
> 1 X kþ1
window length, which is an integer between 1 and N, and L ¼ >
> y for 0 ≤ k < N L − 1
>
> k þ 1 m¼1 m;k−mþ2
N − N L þ 1. These L vectors can further be formed into a >
>
>
>
trajectory matrix X < 1 X NL
ðlÞ
2 3 dk ¼ ym;k−mþ2 for N L − 1 ≤ k ≤ L ð12Þ
>
> N L m¼1
d0 d1 ··· dL−1 >
>
6 7 >
>
6 d1 7 >
> 1 X
N−Lþ1
6
d2 ··· dL
7 >
> for L < k ≤ N − 1
X ¼ ½X1 ; X2 ; : : : XL  ¼ 6 . 7 ð8Þ : y
N − k m¼k−Lþ2 m;k−mþ2
6 .. ... ..
. ... 7
4 5
dN L −1 dN L ··· dN−1 For N L > L, the length N L should be switched with L in the
preceding expressions. There are N g sets of time series data
The ði; jÞth element of X in Eq. (8) is xij ¼ diþj−2 . Hence the fdðlÞ ; l ¼ 1; 2; : : : ; N g g obtained from dv ðtÞ and the new data
trajectory matrix X ∈ RN L ×L is a Hankel matrix. PN g ðlÞ
vector dðtÞ after the SSA becomes dðtÞ ¼ l¼1 d .
Singular Value Decomposition
Let S ¼ XXT , which is a N L × N L square matrix; the N L Blind Modal Identification with SSA
eigenvalues and the corresponding eigenvectors of matrix The multichannel data set from SSA is used as the input into BSS
S are denoted λ1 ; λ2 ; : : : ; λN L (λ1 > λ2 > · · · > λN L ) and for modal parameter identification. Suppose the data set includes
U1 ; U2 ; : : : ; UN L , respectively. If N s is the number of positive N g sets of time series data d ¼ fdðlÞ ; l ¼ 1; 2; : : : ; N g gT . Each
eigenvalues (N s ≤ Np L ),
ffiffiffiffiffiffiffithe square roots of these eigenvalues, set of time series data dðlÞ is a linear mixture of n components
pffiffiffiffiffi pffiffiffiffiffi
i.e., λ1 ; λ2 ; : : : ; λN s , are referred to as the singular values fsi ; i ¼ 1; 2; : : : ; ng. The relation between the components and
of the trajectory matrix X. Extremely small singular values are the measured data can be written
ignored in the decomposition process, and this does
pffiffiffiffi
ffi not affect
the accuracy. The elementary matrix Xsi for a λi is d ¼ As ð13Þ

pffiffiffiffiffi where A ¼ N g × n mixing matrix, s ¼ fs1 ; s2 ; : : : ; sn gT ; d ¼

Xsi ¼ λi Ui V Ti ð9Þ fd1 ; d2 ; : : : ; dN g gT ; and both A and s are unknown.
pffiffiffiffiffi Assuming that the components s ¼ fs1 ; s2 ; : : : ; sn gT are statis-
where V i ¼ XT Ui = λi ; and Ui and V i = left and right singular tically independent, they can be determined by second-order
vectors, respectively. The trajectory matrix X can then be expressed blind identification for the overdetermined case of N g > n as
as the summation of the N s elementary matrices (Belouchrani et al. 1997)

X ¼ Xs1 þ Xs2 þ · · · þXsN s ð10Þ s ¼ Wd ð14Þ

where the demixing matrix W is the inverse of the mixing matrix A,

The trajectory matrix X is then decomposed into N s elementary and needs to be estimated. There are two steps in the SOBI algo-
matrices of rank 1 with a norm equal to the singular value. This is rithm: data whitening and estimation of the mixing and demixing
the singular value decomposition of the trajectory matrix X. matrices. For the observed data dðtÞ, the time-shifted covariance

© ASCE 04019050-3 J. Aerosp. Eng.

J. Aerosp. Eng., 2019, 32(4): 04019050

 The first component X: 5.628 The first component
0.4
0.4 Y: 0.2409

Amplitude
v1 v2 v3
Amplitude
m1 m2 m3
0.2
0.2
X: 5.666 X: 5.427
X: 5.296
Y: 0.2424 0 Y: 0.1902
0 Y: 0.1467
0 5 10 15 20 25 30 35
0 5 10 15 20 25 30 35
The second component The second component
0.4 0.4 X: 8.426
v1 v2 v3
Y: 0.1987
Amplitude

X: 8.64

Amplitude
m1 m2 m3
0.2 X: 8.466 0.2 Y: 0.3899
Y: 0.3119
X: 8.355
0 0 Y: 0.154
0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35
Downloaded from ascelibrary.org by University of Technology, Sydney on 05/05/19. Copyright ASCE. For personal use only; all rights reserved.

The third component X: 32.44 The third component

0.4 0.4
m1 m2 m3 X: 31.65 Y: 0.2502
Amplitude

Amplitude
Y: 0.1593
0.2 v1 v2 v3
0.2
X: 33.92 X: 32.44
Y: 0.1996 0 Y: 0.2502
0
0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35
(a) Frequency [Hz] (b) Frequency [Hz]

Fig. 19. Spectra of decomposed components under different test conditions: (a) different weights of two-axle vehicle; and (b) different moving
speeds.

Table 2. Bridge and vehicle frequencies from dynamic response of the two-axle vehicle (Hz)
v1 v2 v3
Frequency m1 m2 m3 m1 m2 m3 m1 m2 m3
Bridge — — — — — — — — —
First 4.94 5.30 5.19 5.68 5.64 5.67 5.86 5.41 5.56
Second 8.37 8.35 8.35 8.47 8.47 8.47 8.57 8.62 8.64
Vehicle 33.79 32.19 32.15 33.92 32.44 31.65 33.33 32.57 32.37

3 15
First component
Second component
2

1 10
Frequency [Hz]
Amplitude

-1 5

-2

-3 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
(a) Time [Sec] (b) Time [Sec]

Fig. 20. (a) First two response components; and (b) instantaneous frequencies.

method is insensitive to the window length in the SSA compared that the proposed method could be used to identify the instanta-
with the direct SSA. The vehicle and bridge modal frequencies neous frequency of the vehicle-bridge interaction system.
can be separated easily with the proposed method even with Class
B road surface roughness. The effect of vehicle parameters on the
identification was investigated numerically. Results showed that a Acknowledgments
heavier vehicle with a lower speed can more accurately identify
the frequency, and the vehicle stiffness does have a large effect on This research is supported by research funding of the Australian
the identified results. The proposed method is also robust to meas- Research Council Discover Project (DP160103197). The financial
urement noise. Further analysis with the Hilbert transform showed aid is gratefully acknowledged.

© ASCE 04019050-15 J. Aerosp. Eng.

J. Aerosp. Eng., 2019, 32(4): 04019050

 References
Antoni, J. 2005. “Blind separation of vibration components: Principles
and demonstrations.” Mech. Syst. Signal Process. 19 (6): 1166–1180.
https://doi.org/10.1016/j.ymssp.2005.08.008.
Belouchrani, A., K. Abed-Meraim, J.-F. Cardoso, and E. Moulines. 1997.
“A blind source separation technique using second-order statistics.”
IEEE Trans. Signal Process. 45 (2): 434–444. https://doi.org/10
.1109/78.554307.
Chang, K. C., F. B. Wu, and Y. B. Yang. 2010. “Effect of road surface
roughness on indirect approach for measuring bridge frequencies from
a passing vehicle.” Interact. Multiscale Mech. 3 (4): 299–308. https://
doi.org/10.12989/imm.2010.3.4.299.
Downloaded from ascelibrary.org by University of Technology, Sydney on 05/05/19. Copyright ASCE. For personal use only; all rights reserved.
Clough, R. W., and J. Penzien. 1975. Dynamics of structures. New York:
McGraw-Hill.
Golyandina, N. 2010. “On the choice of parameters in singular spectrum
analysis and related subspace-based methods.” Stat. Interface 3 (3):
259–279. https://doi.org/10.4310/SII.2010.v3.n3.a2.
Harmouche, J., D. Fourer, F. Auger, P. Borgnat, and P. Flandrin. 2018. “The
sliding singular spectrum analysis: A data-driven nonstationary signal
decomposition tool.” IEEE Trans. Signal Process. 66 (1): 251–263.
https://doi.org/10.1109/TSP.2017.2752720.
Hassani, H. 2007. “Singular spectrum analysis: Methodology and compari-
son.” J. Data Sci. 5 (2): 239–257.
Hassani, H., R. Mahmoudvand, and M. Zokaei. 2011. “Separability
and window length in singular spectrum analysis.” C.R. Math.
349 (17–18): 987–990. https://doi.org/10.1016/j.crma.2011.07.012.
Hazra, B., A. Roffel, S. Narasimhan, and M. D. Pandey. 2010. “Modified
cross-correlation method for the blind identification of structures.”
J. Eng. Mech. 136 (7): 889–897. https://doi.org/10.1061/(ASCE)EM
.1943-7889.0000133.
ISO. 1995. Mechanical vibration—Road surfaces profiles—Reporting of
measured data. ISO S. 8608. Geneva, Switzerland: ISO.
Kerschen, G., F. Poncelet, and J.-C. Golinval. 2007. “Physical interpreta-
tion of independent component analysis in structural dynamics.” Mech.
Syst. Sig. Process. 21 (4): 1561–1575. https://doi.org/10.1016/j.ymssp
.2006.07.009.
Kim, C.-W., R. Isemoto, P. McGetrick, M. Kawatani, and E. J. Obrien.
2014. “Drive-by bridge inspection from three different approaches.”
Smart Struct. Syst. 13 (5): 775–796. https://doi.org/10.12989/sss
.2014.13.5.775.
Liu, K., S. S. Law, Y. Xia, and X. Q. Zhu. 2014. “Singular spectrum analysis
for enhancing the sensitivity in structural damage detection.” J. Sound
Vib. 333 (2): 392–417. https://doi.org/10.1016/j.jsv.2013.09.027.
Malekjafarian, A., P. J. McGetrick, and E. J. Obrien. 2015. “A review of
indirect bridge monitoring using passing vehicles.” Shock Vib.
2015: 16. http://doi.org/10.1155/2015/286139.
Malekjafarian, A., and E. J. Obrien. 2017. “On the use of a passing vehicle
for the estimation of bridge mode shapes.” J. Sound Vib. 397 (June):
77–91. https://doi.org/10.1016/j.jsv.2017.02.051.
© ASCE 04019050-16
J. Aerosp. Eng., 2019, 32(4): 04019050
McNeill, S., and D. Zimmerman. 2008. “A framework for blind modal
identification using joint approximate diagonalization.” Mech. Syst.
Sig. Process. 22 (7): 1526–1548. https://doi.org/10.1016/j.ymssp
.2008.01.010.
Poncelet, F., G. Kerschen, J.-C. Golinval, and D. Verhelst. 2007. “Output-
only modal analysis using blind source separation techniques.” Mech.
Syst. Sig. Process. 21 (6): 2335–2358. https://doi.org/10.1016/j.ymssp
.2006.12.005.
Sadhu, A., S. Narasimhan, and J. Antoni. 2017. “A review of output-only
structural mode identification literature employing blind source separa-
tion methods.” Mech. Syst. Signal Process. 94 (Sep): 415–431. https://
doi.org/10.1016/j.ymssp.2017.03.001.
Wang, Y., and H. Hao. 2013. “Damage identification scheme based
on compressive sensing.” J. Comput. Civ. Eng. 29 (2): 04014037.
https://doi.org/10.1061/(ASCE)CP.1943-5487.0000324.
Yang, Y., and S. Nagarajaiah. 2012. “Time-frequency blind source separa-
tion using independent component analysis for output-only modal
identification of highly damped structures.” J. Struct. Eng. 139 (10):
1780–1793. https://doi.org/10.1061/(ASCE)ST.1943-541X.0000621.
Yang, Y. B., K. C. Chang, and Y. C. Li. 2013. “Filtering techniques for
extracting bridge frequencies from a test vehicle moving over the
bridge.” Eng. Struct. 48 (Mar): 353–362. https://doi.org/10.1016/j
.engstruct.2012.09.025.
Yang, Y. B., Y. C. Li, and K. C. Chang. 2012. “Using two connected ve-
hicles to measure the frequencies of bridges with rough surface: A theo-
retical study.” Acta Mech. 223 (8): 1851–1861. https://doi.org/10.1007
/s00707-012-0671-7.
Yang, Y. B., C. W. Lin, and J. D. Yau. 2004. “Extracting bridge frequencies
from the dynamic response of a passing vehicle.” J. Sound Vib. 272 (3):
471–493. https://doi.org/10.1016/S0022-460X(03)00378-X.
Yang, Y. B., and J. P. Yang. 2018. “State-of-the art review on modal
identification and damage detection of bridges by moving test vehicles.”
Int. J. Struct. Stab. Dyn. 18 (2): 1850025. https://doi.org/10.1142
/S0219455418500256.
Zhen, L., D. Peng, Z. Yi, Y. Xiang, and P. Chen. 2017. “Underdetermined
blind source separation using sparse coding.” IEEE Trans. Neural
Networks Learn. Syst. 28: 3102–3108. https://doi.org/10.1109/TNNLS
.2016.2610960.
Zhou, W., and D. Chelidze. 2007. “Blind source separation based vibration
mode identification.” Mech. Syst. Sig. Process. 21 (8): 3072–3087.
https://doi.org/10.1016/j.ymssp.2007.05.007.
Zhu, X. Q., and S. S. Law. 2002. “Dynamic load on continuous multi-lane
bridge deck from moving vehicles.” J. Sound Vib. 251 (4): 697–716.
https://doi.org/10.1006/jsvi.2001.3996.
Zhu, X. Q., and S. S. Law. 2015. “Structural health monitoring based on
vehicle-bridge interaction: Accomplishments and challenges.” Adv.
Struct. Eng. 18 (12): 1999–2015. https://doi.org/10.1260/1369-4332
.18.12.1999.
J. Aerosp. Eng.