Vibration-Based Health Monitoring of an Existing Truss

Bridge Using Energy-Based Damping Evaluation

Samim Mustafa1; Yasunao Matsumoto2; and Hiroki Yamaguchi3
Downloaded from ascelibrary.org by "Indian Inst Of Technology Library, Kharagpur" on 03/04/18. Copyright ASCE. For personal use only; all rights reserved.

Abstract: This paper proposes an analytical framework for vibration-based structural health monitoring (SHM) by introducing an energy-
based damping evaluation (EBDE). The damage detection by the proposed EBDE was carried out by estimating the contribution of modal
damping ratios from different structural elements utilizing experimentally identified modal damping ratios, and estimating modal strain and
modal potential energies from an updated finite-element (FE) model of the structure under consideration. Model updating was performed using
modal frequencies and mode shapes that are generally not sensitive to local damage. The advantage of using damping as a damage indicator is
that the damping change in global modes affected by the local damage can be identified with a small number of sensors. A previous study
reported that the studied bridge with damage at the local diagonal member showed a significant increase in the damping of global vibration
mode of the structure. The present study utilized the EBDE to identify the cause of the modal damping increase by observing the change in the
contribution from different structural elements on the modal damping ratios. DOI: 10.1061/(ASCE)BE.1943-5592.0001159. © 2017
American Society of Civil Engineers.
Author keywords: Vibration-based health monitoring; Damping analysis; Damage detection; Steel bridges; Modal data.

Introduction
Because most of the current civil infrastructures, such as bridges,
are aging and structurally deteriorating due to reaching their theo-
retical design life, the need for vibration-based structural health
monitoring (SHM) has gained a significant amount of interest in
recent years. After the tragic collapse of the I-35 West Bridge in
Minneapolis, Minnesota, many serious incidents were recorded
in rigorous inspections of steel truss bridges (Fujino and
Siringoringo 2008). In all of these cases, the damage in the early
stage was not detected by periodic visual inspections, which may
imply the limitations of these inspections. The goal is to detect
damage in a structure, if damage occurs, to prevent catastrophic
failure of such infrastructures or to evaluate the condition and
performance of the same to ensure public safety. The basic idea
for vibration-based damage detection is that the damage-induced
changes in the physical properties will result in detectable
changes in modal properties (Nakamura et al. 1998; Nair et al.
2006; Salawu 1997; Vanik et al. 2000; Feng and Feng 2016).
Therefore, it is intuitive that damage can be identified by analyz-
ing the changes in vibration features of the structure.
1
Assistant Professor Dept. of Civil Engineering IIT Indore Simrol,
Indore 453552, India; Former Doctoral Student, Dept. of Civil and
Environmental Engineering, Saitama Univ., Saitama 338-8570, Japan (corre-
sponding author). ORCID: https://orcid.org/0000-0002-2303-3898. E-mail:
samim@alumni.iitg.ernet.in
2
Professor, Dept. of Civil and Environmental Engineering, Saitama
Univ., 255 Shimo-Okubo, Sakura-ku, Saitama 338-8570, Japan. E-mail:
ymatsu@mail.saitama-u.ac.jp
3
Professor, Dept. of Civil and Environmental Engineering, Saitama
Univ., 255 Shimo-Okubo, Sakura-ku, Saitama 338-8570, Japan. E-mail:
hiroki@mail.saitama-u.ac.jp
Note. This manuscript was submitted on December 12, 2016; approved
on July 12, 2017; published online on October 26, 2017. Discussion period
open until March 26, 2018; separate discussions must be submitted for
individual papers. This paper is part of the Journal of Bridge
Engineering, © ASCE, ISSN 1084-0702.
There are some serious issues with the practical application of
vibration-based SHM to detect local damage. The most important
concern for its practical application is the issue sensitivity of
modal parameters to local damage. Many vibration-based dam-
age-detection approaches have been proposed in the literature,
and an extensive review of those can be found in the work by
Doebling et al. (1998) and Fan and Qiao (2011). Most of these
studies on damage detection have been mainly based on change in
modal frequencies and/or change in mode shape–related indices.
However, the problem related to low sensitivity of damage fea-
tures due to local damages remains a concern for the practical
application of these methodologies. In contrast, it has been recog-
nized that damping is more sensitive to local damages, especially
due to cracks or some internal changes in the structural property
(Modena et al. 1999; Curadelli et al. 2008; Yoshioka et al. 2010).
Frizzarin et al. (2010) proposed a time-domain damage-detection
method for concrete structures based on the analysis of nonlinear
damping from measured vibration responses. In this study, a
strong correlation was observed between the increase in the non-
linear damping and decrease in the structural stiffness associated
with the increase in seismic damage severity. These are the main
motivations of considering damping as a damage indicator for
vibration-based SHM of steel truss bridges in the present study
study. Another issue with the practical application of vibration-
based SHM is the availability of a limited number of sensors due to
the budget constraint. In the case of large structures, such as steel
truss bridges, a number of sensors may be required to detect local
damage that would lead to a trade-off between monitoring cost and
ability of damage detection. Therefore, it would be worthy to de-
velop a method to detect damage with a limited number of sensors.
This study introduced a practical and effective vibration-based
SHM to overcome the aforementioned difficulties. The proposed
framework used an energy-based damping evaluation (EBDE) for
the detection of local damage. The advantage of using damping is
that the damping change in global mode, affected by the local dam-
age, can be identified with a small number of sensors (Yamaguchi
et al. 2015). There are only a few examples in the literature of

© ASCE 04017114-1 J. Bridge Eng.

J. Bridge Eng., 2018, 23(1): 04017114

 analytical damping evaluation using an energy-based approach
(Kawashima et al. 1994; Yamaguchi and Jayawardena 1992;
Yamaguchi et al. 1997; Dammika et al. 2015). The lack of compre-
hensive studies has initiated research in this field. Dammika et al.
(2015) investigated the analytical modal damping evaluation as a
complementary method to the experimental SHM of bridges. The
objective was to provide the theoretical basis for modal damping
evaluation, which can improve the reliability of the experimental
damping identification. However, the uncertainties associated with
the finite-element (FE) model of the bridge studied were not taken
Downloaded from ascelibrary.org by "Indian Inst Of Technology Library, Kharagpur" on 03/04/18. Copyright ASCE. For personal use only; all rights reserved.
into consideration in that study. Furthermore, the practical applica-
tion of EBDE to detect local damage was not carried out in that
study. Hence, this paper made an effort to fill in the aforementioned
research gaps by proposing an analytical framework for vibration-
based SHM. The evaluation of analytical modal damping has an
advantage in that the contribution of energy dissipations in sub-
structures on modal damping ratios can be estimated, which could
be useful knowledge in the field of vibration-based SHM. This
EBDE requires the estimation of modal strain energies and modal
potential energies of the structure, which can be obtained from a FE
model of the structure under consideration. Hence, to ensure higher
accuracy in the EBDE, these modal energies should be estimated
from a FE model that predicts the actual behavior of the field struc-
ture under consideration as closely as possible. Therefore, an effec-
tive method for FE model updating needs to be established to treat
the underlying uncertainties associated with the preliminary FE
model explicitly (Soize et al. 2008; Govers and Link 2010; Simoen
et al. 2015; Feng and Feng 2015). In this study, a Bayesian model
update was considered to obtain a probabilistic baseline using
experimentally identified modal data. The advantages of the
Bayesian method is that it is capable of incorporating all types of
available information, all types of uncertainties, and incomplete ex-
perimental data (Beck and Katafygiotis 1998; Sohn and Law 2000;
Yuen et al. 2004; Mustafa et al. 2015; Mustafa and Matsumoto
2017).
In this study, an existing steel truss bridge was considered as a
test structure. First, FE model updating was performed to obtain an
updated model utilizing identified modal frequencies and mode
shapes of global vibration modes from a car-running test. The
EBDE was then carried out to estimate the damping parameters of
the test structure utilizing experimentally identified modal damping
ratios. As an application, damage detection by the proposed frame-
work was carried out by observing the change in the contribution of
modal damping ratios from different structural elements. From the
experimental identification results of the damaged fourth span of
Fig. 1. View of steel truss bridge: (a) side view; (b) dimensions of diagonals; (c) cross section
© ASCE 04017114-2
J. Bridge Eng., 2018, 23(1): 04017114
the same truss bridge in a previous study by Yamaguchi et al.
(2015), it was understood that the effect of local damage has no sig-
nificant effect on the modal frequencies and mode shapes of the
global vibrational modes. Hence, it was expected that the model
updating using only global modes identified from a limited number
of sensors, which is the case in most practical applications, may not
be able to trace the damaged stiffness parameters. In contrast, the
studied truss bridge with damage at a local diagonal member
showed a significant increase in the damping of a global vibrational
mode that can be identified with a small number of sensors. An
objective of the study was to obtain theoretical insight into the cause
of the increase in the damping of a global mode, which would then
lead to the development of a practical and effective vibration-based
SHM by utilizing global vibrational characteristics that can be iden-
tified with a small number of sensors.
EBDE
It is common practice to express the damping of real structures in
terms of an equivalent viscous damping ratio, which is the ratio of
the given damping of an equivalent single-degree-of-freedom sys-
tem to its critical damping. Using an energy-based definition, this
damping ratio can be defined as the ratio of the dissipated energy
per cycle to the maximum potential energy in a cycle (Ungar and
Kerwin 1962; Yamaguchi and Nagahawatta 1995). Hence, the nth
modal damping ratio ( j n ) can be expressed as
Dn
jn
(1)
4p Un
where Dn and Un = nth modal dissipating energy and nth modal
potential energy per unit cycle, respectively. The nth modal dissi-
pating energy (Dn ) can be expressed as the summation of dissipat-
ing energies due to different sources, such as damping capacity of
materials, friction at structural connections, and energy dissipation
at supports.
Energy-Based Damping Model for Test Bridge
The bridge studied was a Warren-type steel truss bridge constructed
in 1965 over a river for road traffic, and it consists of five simply
supported main spans, each having a length of 70.77 m and a width
of 6.0 m. Fig. 1 shows the general view of the studied truss bridge.
The compression diagonal members are made of a box section, and
J. Bridge Eng.
 the tension diagonal members are an H-section with eight or nine el- D G UC TLB BLB
liptical holes in the web to reduce the weight of steel, as shown in
Fig. 1(b). The bridge has a composite deck of steel plate girders and
reinforced concrete slab of 160-mm thickness, as shown in Fig. 1(c).
The studied steel truss bridge was considered to be composed of
five substructures, as shown in Fig. 2: diagonal members (D), gird-
ers (G), upper chord (UC) members, top lateral bracings (TLBs),
and bottom lateral bracings (BLBs). The modal energy dissipation
from each substructure, which is referred to as the internal modal
energy dissipation (Di;n ) in this paper, was assumed to be propor-
Downloaded from ascelibrary.org by "Indian Inst Of Technology Library, Kharagpur" on 03/04/18. Copyright ASCE. For personal use only; all rights reserved.

tional to its modal strain energy (Nashif et al. 1985; Yamaguchi and
Fig. 2. Three-dimensional FE model with substructures considered for
Ito 1997). Therefore, the internal modal energy dissipation of the
damping analysis
truss bridge for the nth mode (Di;n ) can be expressed as

Di;n ¼ 2p h d Vd;n þ 2p h g Vg;n þ 2p h uc Vuc;n þ 2p h tlb Vtlb;n

foundations, which was assumed to be included in the energy dissi-
þ 2p h blb Vblb;n (2) pation represented by Eq. (3).
After calculating Dn using Eqs. (2) and (3), the nth modal
where Vd;n , Vg;n , Vuc;n , Vtlb;n , and Vblb;n = nth modal strain energies damping ratio for the steel truss bridge can be evaluated using
of diagonal members, girders, UC members, TLBs, and BLBs, Eq. (1)
respectively; and h d , h g , h uc , h tlb , and h blb = equivalent loss fac-
2p h d Vd;n 2p h g Vg;n 2p h uc Vuc;n 2p h tlb Vtlb;n
tors of substructures described earlier. In the present study, it was jn ¼ þ þ þ
assumed that the equivalent loss factors accounted for the material 4p Un 4p Un 4p Un 4p Un
damping as well as the damping at structural connection for simplic-
2p h blb Vblb;n 8As;n m s R
ity. It would be better if the energy dissipation at the nodes was mod- þ þ (4)
eled separately; however, this would require some issues to be inves- 4p Un 4p Un
tigated further, such as modeling of frictional damping at nodes, and
measurement and estimation of relative motion between members at Therefore, as can be seen from the previous equation, it is neces-
the nodes, which could be addressed in further study. The modal sary to know the equivalent loss factor of each substructure and fric-
energy dissipation at supports (S) (Ds;n ) was expressed as tion coefficient of support to evaluate the modal damping ratios of a
X steel truss bridge. The unknown loss factors and friction coefficient
Ds;n ¼ 4As;n m s R (3) can be evaluated by using experimentally identified modal damping
ratios and corresponding energy ratios obtained from the updated
where As;n , m s and R = nth modal amplitude at the movable support, FE model. To proceed, the Eq. (4) was expanded to a set of equa-
dynamic friction coefficient, and support’s reaction due to vertical tions of m number of modes by assuming that the unknown damp-
load, respectively. Energy dissipation may occur at the support due ing parameters are independent of the vibration mode (Dammika et
to transmission of energy from supports to piers, abutments, and al. 2015)

8 9
2 3>
> hd > >
> >
2p Vd;1 2p Vg;1 2p Vuc;1 2p Vtlb;1 2p Vblb;1 8As;1 R >> >
8 9
6 4p U
>
> hg > >
>
>
>
>
j1 >>
> 6 4p U1 4p U1 4p U1 4p U1 4p U1 7 >
7>
>
>
>
>
7> >
1
< = 6 < h uc =
.. 6 .. .. .. .. .. .. 7 7
> . >¼6 . . . . . . 7> (5)
>
> >
>
6
6 7>
> h tlb >
>
>
: ; 4 2p V 2p Vg;m 2p Vuc;m 2p Vtlb;m 2p Vblb;m 8As;m R 5>
> >
>
jm >
> >
h blb >
d;m
> >
4p Um 4p Um 4p Um 4p Um 4p Um 4p Um > >
>
>
>
>
: ;
ms

It is important to note that the number of modes considered (m)
should be equal to or greater than the number of unknown damping
parameters to be able to solve the previous equation. Eq. (5) can be
solved by the nonnegative least-squares (NNLS) method to obtain
the unknown damping parameters (i.e., equivalent loss factor of
each substructure and friction coefficient of support). After obtain-
ing the loss factors and friction coefficient, the analytical modal
damping ratio then can be evaluated using Eq. (4), and the contribu-
tion of each substructure to the modal damping ratio can also be
evaluated. Using the same principle, the contribution of each inter-
nal element to the modal damping ratio ( j l;n ) can also be obtained
by using the strain energy of that element and the corresponding
loss factor
2p h l Vl;n
j l;n ¼ (6)
4p Un
where h l and Vl;n = equivalent loss factor and nth modal strain
energy of lth element. Because of the practical limitation of the
number of identified modes, the equivalent loss factor of the lth ele-
ment ( h l ) can be assumed to be the same as the loss factor of the
substructure to which it belongs. By looking at these contributions

© ASCE 04017114-3 J. Bridge Eng.

J. Bridge Eng., 2018, 23(1): 04017114

 from all elements, one can have a clear understanding about those where Ke , KG , KT , and /n = updated elastic stiffness matrix, geo-
elements that make a dominant contribution to damping for a partic- metric stiffness matrix, total stiffness matrix, and nth updated mode
ular mode, which can be useful information from the SHM point of shape of the structure, respectively; Ke;l and /l;n = updated elastic
view. Another good aspect of this elemental damping evaluation is stiffness matrix and nth updated mode shape of the lth element,
that it may enable one to detect damage at the local level by observ- respectively; and Nl = number of elements in the FE model. Ke;l can
ing the change in elemental damping parameter, such as loss factor be calculated using updated stiffness parameters for the lth element
or friction coefficient, before and after possible damage. and corresponding subsystem stiffness matrices. Once the modal
strain energy of each element is calculated, the contribution of each
Evaluation of Modal Energies element to the modal damping ratio can be evaluated using Eq. (6).
Here, KG was considered in the evaluation of total potential energy
Downloaded from ascelibrary.org by "Indian Inst Of Technology Library, Kharagpur" on 03/04/18. Copyright ASCE. For personal use only; all rights reserved.

It is understood from Eq. (5) that the evaluation of modal potential because the effect of KG was found to be significant for the local
energy and modal strain energy of each substructure is indispensable modes, as can be seen in Fig. 3. Here, the local modes are those with
for the estimation of damping parameters. To know the contribution frequencies ranging from 8.828 to 9.325 Hz as discussed in the fol-
of each element to the modal damping ratio, it is necessary to calcu- lowing section.
late the strain energy of each element. Hence, to ensure higher accu-
racy in the EBDE, these modal energies should be estimated from an
updated FE model, which predicts the actual behavior of the field Application
structure under consideration as closely as possible. The nth modal
potential energy (Un ) and nth modal strain energy (Vn ) are given by
Vibration Measurements and System Identification
1     1 T XNl
1 T In a separate study (Yamaguchi et al. 2015), the vibration accelera-
Un ¼ /T
n K e þ K G / n ¼ /n K T /
n ¼ / KT;l /l;n
2 2 l¼1
2 l;n tion measurement was conducted by a car-running test using two
1 X Nl
1 T   sensor layouts at the first span of the bridge, which did not have any
Vn ¼ /T  
n Ke /n ¼ / K / (7) damage. In Measurement Case 1, shown in Fig. 4(a), a total of 13
2 l¼1
2 l;n e;l l;n
vertical sensors were used to identify the global vibration modes. A
total of five data sets of 10-min measurements were collected at a
12 sampling rate of 100 samples per second. The sensitivities of the
piezoelectric accelerometers and the servo velocimeters were 0.4
Local modes pC/Gal and 5 mV/Gal, respectively. An example of full-length
Natural frequency(Hz)

10
recorded data measured at the U5 location is shown in Fig. 5(a). For
8 the identification of diagonal modes, another two sets of measure-
ments were carried out using the sensor setup in Measurement Case
6 2 shown in Fig. 4(b). Hence, a total of seven sets of measurements
were recorded. Free vibration (FV) responses were extracted from
4 full-length recorded data by checking the traffic-free time on the
Linear elastic bridge from the video of traffic recorded during measurements. Fig.
Geometric nonlinearity 5(b) shows one such FV response extracted from full-length
2
1 2 3 4 5 6 7 8 9 10 11 recorded data as shown in Fig. 5(a). Two system-identification
Mode No.
methods, stochastic subspace identification (SSI) (Overschee and
Fig. 3. Comparison of frequencies obtained from linear elastic and
Moor 1993) and eigensystem realization algorithm (ERA) (Juang
geometric nonlinear analysis and Pappa 1985), were used to identify the modal parameters from
the extracted FV responses. Seven global modes were identified

Downstream side Downstream side

10 10
D4d D6d D6d' D4d' D4d D6d D6d' D4d'
D2d D2d' D2d D2d'
5 5 D3d D5d
D1d D3d D5d D5d' D3d' D1d' D1d D5d' D3d' D1d'
0 0
0 10.11 20.22 30.33 40.44 50.55 60.66 70.77 0 10.11 20.22 30.33 40.44 50.55 60.66 70.77
Upstream side Upstream side
10 D6u D6u' 10
D2u D4u D4u' D2u' D4u D6u D6u' D4u'
D2u D2u'
5 D1u D3u D5u D5u' D3u' D1u'
5 D5u
D1u D3u D5u' D3u' D1u'
0 0
0 10.11 20.22 30.33 40.44 50.55 60.66 70.77 0 10.11 20.22 30.33 40.44 50.55 60.66 70.77
Downstream Downstream
L1 L2 L3 L4 L5 L6 L1 L2 L3 L4 L5 L6
-2 -2
0 0
2 2
0Base
10.11 20.22 30.33 40.44 50.55 60.66 70.77 0Base
10.11 20.22 30.33 40.44 50.55 60.66 70.77
U1 U2 U3 U4 U5 U6 U1 U2 U3 U4 U5 U6
Upstream Upstream
(a) (b)
Piezoelectric accelerometer Servo velocimeter Piezoelectric accelerometer Servo velocimeter

Fig. 4. Sensor layout at first span: (a) Measurement Case 1; (b) Measurement Case 2

© ASCE 04017114-4 J. Bridge Eng.

J. Bridge Eng., 2018, 23(1): 04017114

 0.3 0.1

Acceleration(m/s2)

Acceleration(m/s2)
0.2
0.05
0.1
0 0
-0.1
-0.05
-0.2
-0.3 -0.1
0 100 200 300 400 500 600 400 405 410 415
Time(sec) Time(sec)
Downloaded from ascelibrary.org by "Indian Inst Of Technology Library, Kharagpur" on 03/04/18. Copyright ASCE. For personal use only; all rights reserved.

(a) (b)

Fig. 5. Recorded vibration data at L2 location: (a) full-length recorded data; (b) extracted FV response

SSI ERA

B1: 2.569Hz T1: 4.605Hz B2: 5.288Hz B3: 7.302Hz B4: 9.376Hz B5: 9.915Hz T3: 11.42Hz

B1: 2.559Hz T1: 4.601Hz B2: 5.285Hz B3: 7.296Hz B4: 9.245Hz B5: 10.05Hz T3: 11.41Hz

Fig. 6. Identified global vibration modes by SSI and ERA (Note: B = bending modes; T = torsional modes)

from Measurement Case 1 and four diagonal modes were identified
from Measurement Case 2. Figs. 6 and 7 show the identified global
and diagonal modes, respectively. As observed in Fig. 6, the identi-
fied modal parameters from SSI and ERA showed a good agree-
ment, suggesting a high level of reliability in the identification
results.
Fig. 7 shows the bar plots of the identified mode shape vectors
corresponding to sensor locations as shown in Fig. 4(b). In Fig. 7,
the deck and the diagonal represent the mode shape vectors corre-
sponding to the sensors at deck level, such as at U1, U2, U4, U5,
L2, L4, and L6, and at diagonal members, such as at D5u, D5u0 ,
D3d, D5d, and D5d0 , respectively. It can be seen that the compo-
nents of mode shapes corresponding to diagonal members are
dominant compared to the components corresponding to the
deck. Fig. 8 shows the modal identification results for damping
and the corresponding natural frequency for all FV records by
taking the average of the identified values obtained from SSI and
ERA. There were slight fluctuations of the identified modal
damping ratios, as can be seen from Fig. 8. However, these fluctu-
ations in the identified modal damping ratios were not signifi-
cantly large, and this kind of fluctuation is typical for experiments
conducted to determine damping, mainly because of the variation
in amplitude of vibration owing to different traffic conditions
(Chen et al. 2016). It was observed that the fluctuation in modal
damping ratio for the first bending mode was higher than for the
other modes, which may be caused by the amplitude dependence
of frictional damping at the support.
Initial and Updated FE Model of Test Bridge
An initial FE model was developed in MATLAB for this study, as
shown in Fig. 2, by modeling only one span (first span) of the
bridge due to its geometric similarity. The structural members
were modeled as three-dimensional frame elements, and six
degrees of freedom (DOFs) were considered at each joint. The
concrete deck slab was modeled using an equivalent section con-
cept with added mass and stiffness properties to the stringer
beams and added stiffness to the cross beams with a Young’s
modulus ratio of 7. Young’s modulus, the modulus of rigidity, and
Poisson’s ratio of steel were 2.0  108 kN/m2, 7.7  107 kN/m2, and
0.3, respectively. The numerical model used was a 2,510-DOF FE
model consisting of 514 frame elements and 420 nodes. Lumped
mass matrix was considered for the dynamic analysis in which the
coefficients corresponding to translatory displacements were equal
to one-half of the total inertia of the frame element, whereas the
coefficients corresponding to flexural rotations were assumed to be
zero.
To ensure higher accuracy in the estimation of modal strain
energies and modal potential energies for the EBDE, an updated
FE model was obtained in this study using Bayesian model
updating (Mustafa and Matsumoto 2017) utilizing the seven
identified global modes with partial mode shape vectors of 12
components, shown in Fig. 6. An element-level parameterization
of stiffness matrix was considered in the model-updating frame-
work to estimate modal strain energy of each structural element
more accurately. This was necessary in evaluating the contribu-
tion of analytical modal damping ratio from each structural ele-
ment using Eq. (6). A detailed description of the model-updating
framework can be found in Mustafa and Matsumoto (2017).
Table 1 shows the updated results for modal frequencies and
mode shapes where a good agreement in frequencies can be
observed between the updated model and the measured ones.
The mode shapes were also successfully updated because there
were some improvements in modal assurance criteria values of
the updated FE model.

© ASCE 04017114-5 J. Bridge Eng.

J. Bridge Eng., 2018, 23(1): 04017114

 1 1
8.8282 Hz Deck 9.0159 Hz Deck
Diagonal Diagonal
0 0
U1 U2 U5u U4 U5u' U5
-1 -1
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Upper stream side Upper stream side
1 1
8.8282 Hz Deck 9.0159 Hz Deck
Diagonal Diagonal
0 0
Downloaded from ascelibrary.org by "Indian Inst Of Technology Library, Kharagpur" on 03/04/18. Copyright ASCE. For personal use only; all rights reserved.

U3d L2 U5d L4 U5u' L6

-1 -1
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Down stream side Down stream side
(a) (b)

1 1
9.0712 Hz Deck 9.3249 Hz Deck
Diagonal Diagonal
0 0

-1 -1
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Upper stream side Upper stream side
1 1
9.0712 Hz Deck 9.3249 Hz Deck
Diagonal Diagonal
0 0

-1 -1
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Down stream side Down stream side
(c) (d)

Fig. 7. Diagonal mode shapes identified from FV6: (a) first diagonal; (b) second diagonal; (c) third diagonal; (d) fourth diagonal

0.012
FV1 FV2 FV3 FV4 FV5 FV6 FV7
0.01
Modal damping ratio

0.008

0.006

0.004

0.002

0
0 1 2 3 4 5 6 7 8 9 10 11 12
Natural frequency (Hz)

Fig. 8. Modal identification results for damping and corresponding natural frequency for all FV records

Table 1. Model Updating Results for Frequencies and Modal Assurance Criterion (MAC) Values (Data from Mustafa and Matsumoto 2017)

Experimental Initial Updated

Experimental order of mode [f(Hz)] [f(Hz)] [f(Hz)] MAC initial MAC updated
First bending 2.569 2.605 2.569 0.9956 0.9998
First torsional 4.605 4.773 4.605 0.9987 0.9997
Second bending 5.288 5.454 5.288 0.9917 0.9986
Third bending 7.302 7.730 7.306 0.9502 0.9998
Fourth bending 9.376 9.315 9.384 0.7651 0.9983
Fifth bending 9.915 9.978 9.918 0.7599 0.9978
Third torsional 11.42 11.67 11.43 0.8743 0.9998

© ASCE 04017114-6 J. Bridge Eng.

J. Bridge Eng., 2018, 23(1): 04017114

 Identification of Loss Factors and Friction Coefficients dissipated owing to the diagonal members only. The mean value of
the estimated loss factors corresponding to different FV records
For the identification of equivalent loss factors and friction coeffi-
was taken as the equivalent loss factor of diagonal members and
cients, it is necessary to evaluate the strain energy corresponding to
used in the next step for the determination of other unknown damp-
each substructure. The strain energy of each element was calculated
ing parameters. At Step 2, the equivalent loss factors of girders, UC
using updated mode shapes and updated stiffness parameters and
members, TLBs, and BLBs and friction coefficient of supports were
the corresponding subsystem stiffness matrices. Then, the strain
estimated by considering six global modes. Table 3 shows the esti-
energy of each substructure was estimated by adding the strain ener-
mated equivalent loss factors for all FV records along with the aver-
gies of elements belonging to that substructure. Fig. 9 shows the dis-
age estimated values. Different damping parameters were estimated
tribution of the modal strain energy ratio of each substructure for
for different FV records owing to the fluctuation in the identified
first five bending modes (B1–B5), first three torsional modes (T1–
Downloaded from ascelibrary.org by "Indian Inst Of Technology Library, Kharagpur" on 03/04/18. Copyright ASCE. For personal use only; all rights reserved.

modal damping ratios. However, the reasonability of the estimated

T3), and the third diagonal mode (D3) in which the contribution of
strain energy from diagonal members was dominant compared to
the other diagonal modes. As seen in Fig. 9, the bending modes
have significant contribution of strain energies from upper chords,
girders, and diagonal members, whereas the torsional modes also
have contribution of strain energies from top and bottom lateral
bracing members as well in addition to those substructures, as
expected.
Equivalent loss factors and friction coefficients were then identi-
fied following the steps described in Table 2. At Step 1, the equiva-
lent loss factor of diagonal members was estimated by considering
diagonal dominant mode on the assumption that the energy
D G UC TLB
1 mentally identified ones for all FV records. The experimental and
Strain energy ratio
damping parameters can be judged by the fact that all estimated val-
ues are larger than the loss factor of steel, which ranges from 0.0001
to 0.0006 (Zoghi 2014), suggesting the assumed energy dissipation
in the substructures.
Evaluation of Analytical Modal Damping Ratios
After obtaining the equivalent loss factors, the analytical modal
damping ratios were evaluated using Eq. (4) by substituting the loss
factors and corresponding energy ratios. Fig. 10 shows the compari-
son between experimentally identified and analytically estimated
modal damping ratios for different FV records. In Fig. 10, Mode
Numbers 1–6 correspond to the modes considered for the identifica-
tion of damping parameters in Table 2. It can be seen that analyti-
BLB cally evaluated modal damping ratios agree well with the experi-
numerical damping ratios did not match exactly because the equiva-

0.8 lent loss factors and friction coefficient were evaluated by nonlinear
least-squares solution by considering more global modes than the
0.6 total number of unknown damping parameters. Fig. 11 shows the
comparison between analytical modal damping ratios estimated
0.4
using average loss factors and friction coefficient shown in Table 3
0.2 and the corresponding experimental damping by taking the average
of all FV records. Here also, it can be seen that analytically eval-
0 uated modal damping ratios agree well with the experimentally
B1 T1 B2 B3 T2 D3 B4 B5 T3
identified ones.
Fig. 9. Distribution of MERs The contribution of each substructure to the modal damping
ratios was then estimated by using the corresponding loss factor and

Table 2. Steps for Estimating Loss Factors and Friction Coefficient

Analysis
step Considered mode FV record Estimated parameter Analysis conditions
Step 1 Third diagonal (dominant diagonal mode) FV 1–6 hd Assumed that energy dissipated only from diagonal
members
Step 2 1: First bending FV 1–7 h uc , h g , h tlb , h blb , m s Average value of h d from Step 1 was used
2: Second bending NNLS was used to obtain equivalent loss factors
3: Third bending
4: Second torsional
5: Fourth bending
6: Third torsional

Table 3. Estimated Equivalent Loss Factors and Friction Coefficient

Analysis step Para-meter FV1 FV2 FV3 FV4 FV5 FV6 FV7 Avg.
Step 1 hd 0.0085 0.0068 0.0073 0.0100 0.0068 0.0073 — 0.0078
Step 2 h uc 0.0206 0.0245 0.0000 0.0277 0.0183 0.0136 0.0226 0.0212
hg 0.0085 0.0077 0.0121 0.0090 0.0088 0.0074 0.0075 0.0087
h tlb 0.0000 0.1532 0.0914 0.1041 0.0000 0.0346 0.0982 0.0963
h blb 0.0000 0.0196 0.0021 0.0000 0.0027 0.0211 0.0692 0.0229
ms 0.7688 0.0025 0.7753 0.0288 0.2545 0.4476 0.0000 0.3796

© ASCE 04017114-7 J. Bridge Eng.

J. Bridge Eng., 2018, 23(1): 04017114

 Modal damping ratio

Modal damping ratio

Modal damping ratio

0.015 Analytical 0.015 0.015
Analytical Analytical
Experimental Experimental Experimental
0.01 0.01 0.01

0.005 0.005 0.005

FV1 FV2 FV3
0 0 0
1 2 3 4 5 6 1 2 3 4 5 6 1 2 3 4 5 6
Mode No. Mode No. Mode No.
Modal damping ratio

Modal damping ratio

Modal damping ratio

Downloaded from ascelibrary.org by "Indian Inst Of Technology Library, Kharagpur" on 03/04/18. Copyright ASCE. For personal use only; all rights reserved.

0.015 0.015 0.015

Analytical Analytical Analytical
Experimental Experimental Experimental
0.01 0.01 0.01

0.005 0.005 0.005

FV4 FV5 FV7
0 0 0
1 2 3 4 5 6 1 2 3 4 5 6 1 2 3 4 5 6
Mode No. Mode No. Mode No.

Fig. 10. Comparison of experimentally and analytically estimated modal damping ratios

0.015 energy of a structural component to the modal potential energy of

Analytical the whole structure. The assumption here was that all of the ele-
Modal damping ratio

0.01
0.005
0 nant contribution to damping, as can be seen from Fig. 13(b). The
1 2
Mode No.
Fig. 11. Comparison of average experimental modal damping ratios
and corresponding analytical modal damping ratios
energy ratio. Fig. 12 shows the contribution of damping from each
substructure to the modal damping ratios estimated analytically
with the average damping parameters shown in Table 3. Fig. 12(a)
shows that there is a significant damping contribution from the sup-
port for the first bending mode as expected due to the slip of the
movable support in the longitudinal direction. There is also a domi-
nant damping contribution from the UC members due to bending of
these members in this mode. For other bending modes, there are sig-
nificant damping contributions from the UCs, girders, and diagonal
members; however, the damping contribution from supports is
much less than the first bending mode. In Fig. 12(b), the significant
contribution of modal damping from the TLBs and BLBs for the
first torsional mode express the torsional effect in that mode.
Similar trends can also be found in the second and third torsional
modes, as can be seen in Figs. 12(e) and 12(i). In the diagonal
mode, the dominant damping contribution from diagonal members
can be observed in Fig. 12(f).
Contribution of Modal Damping from Each Element
Using the same energy-based damping definition, the contribution
of modal damping ratios from each element was estimated by using
the corresponding loss factor and modal energy ratio (MER) in Eq.
(6). In this paper, the MER was defined as the ratio of modal strain
Experimental ments in a substructure had the same loss factor, which was equal to
the loss factor of that substructure. Fig. 13 shows the elements con-
tributing largely to the modal damping ratio for the first bending
mode, first torsional mode, and third diagonal mode, respectively.
The dissipation of energy occurs mainly from the UC members due
to the bending in the vertical direction for the first bending mode, as
can be seen from Fig. 13(a). In case of the first torsional mode, the
elements from UCs, TLBs, and BLBs and diagonals have a domi-
3 4 5 6
diagonal elements, which have a dominant damping contribution,
can be observed in Fig. 13(c). From similar plots for other modes,
one should be able to identify the elements contributing largely to
the energy dissipation for a particular mode. This kind of informa-
tion can be useful for the identification of damage in the case of
large structures, such as bridges, especially when the problem
related to low sensitivity of frequencies and mode shapes due to
local damage remains a concern, and model updating using these
two modal parameters is inadequate to trace the changes in stiffness
parameters due to local damage. In such cases, increases in modal
damping due to damages can be attributed to increases in damping
contribution from elements that have damage. Those increases in
damping contribution from damaged elements will be identified as
increases in the loss factors of the elements. It should be noted that,
if this is the case, the loss factors identified will differ from the
actual loss factors of the elements due to errors in the stiffness pa-
rameters and the estimation of strain energy.
Application of Proposed Framework to SHM
Problem Description
A large crack at the lower end of the D5u diagonal member
(Fig. 14) was observed in the fourth span of the same bridge. As an
emergency measure, reinforcing plates were installed inside and
outside of the flange and web. Fig. 15 shows the damaged D5u
member and its reinforced condition. Field vibration measurements
of this span were carried out before and immediately after reinforce-
ment using five accelerometers only, as shown in Fig. 14. The three
accelerometers on the diagonal member were placed at the quarter

© ASCE 04017114-8 J. Bridge Eng.

J. Bridge Eng., 2018, 23(1): 04017114

 -3 -3 -3
x 10 x 10 x 10

Modal damping ratio

Modal damping ratio

Modal damping ratio

4 4 3
2.629 Hz 4.825 Hz 5.463 Hz
3 3 2
2 2
1
1 1
0 0 0
D G UC TLBBLB S D G UC TLBBLB S D G UC TLBBLB S
(a) (b) (c)
Downloaded from ascelibrary.org by "Indian Inst Of Technology Library, Kharagpur" on 03/04/18. Copyright ASCE. For personal use only; all rights reserved.

-3 -3 -3
x 10 x 10 x 10
Modal damping ratio

Modal damping ratio

Modal damping ratio

3 3 3
7.756 Hz 8.974 Hz 9.183 Hz
2 2 2

1 1 1

0 0 0
D G UC TLBBLB S D G UC TLBBLB S D G UC TLBBLB S
(d) (e) (f)

-3 -3 -3
x 10 x 10 x 10
Modal damping ratio

Modal damping ratio

Modal damping ratio

3 4 3
9.367 Hz 10.01 Hz 11.70 Hz
2 3 2
2
1 1
1
0 0 0
D G UC TLBBLB S D G UC TLBBLB S D G UC TLBBLB S
(g) (h) (i)

Fig. 12. Contribution of each damping source to modal damping ratios for: (a) first bending; (b) first torsional; (c) second bending; (d) third bending;
(e) second torsional; (f) third diagonal; (g) fourth bending; (h) fifth bending; (i) third torsional mode

Fig. 13. Elements having significant analytical modal damping ratio (shown in solid lines): (a) >0.3% of modal damping for first bending; (b)
>0.1% of modal damping for first torsional; (c) >1% of modal damping for third diagonal mode

Upstream side
10
D2u D4u D6u D6u' D4u' D2u'
5
D1u D3u D5u D5u' D3u' D1u'
0
0 10.11 20.22 30.33 40.44 50.55 60.66 70.77
Damage
Damaged part
part
Downstream L1 L2 L3 L4 L5 L6

Upstream U1 U2 U3 U4 U5 U6
Uniaxial accelerometer(vertical) Tri-axial accelerometer

Fig. 14. Placement of sensors at fourth span

point of the damaged D5u member to measure vibrations in three identified parameters corresponding to nine sets of recorded data
orthogonal axes. Table 4 shows the modal identification results for before and after reinforcement conditions of the D5u diagonal
by ERA for frequencies and damping that were averaged over member. It was observed that the changes in natural frequencies

© ASCE 04017114-9 J. Bridge Eng.

J. Bridge Eng., 2018, 23(1): 04017114

 of the global modes due to damage are much smaller than the
change in modal damping ratios. However, significant changes in
both natural frequency and modal damping ratio were observed in
the case of diagonal mode. It can also be observed that the fre-
quency of the diagonal mode came very close to the frequency of
the third bending mode due to damage. The standard deviations
of the identified modal damping ratios, shown in parentheses,
were not large, so the average values identified could be consid-
ered reliable. More details of the measurement at the fourth span
and discussion on the modal parameters identified experimentally
Downloaded from ascelibrary.org by "Indian Inst Of Technology Library, Kharagpur" on 03/04/18. Copyright ASCE. For personal use only; all rights reserved.
have been described by Yamaguchi et al. (2015).
To simulate the damage in the FE model, the sectional properties
of all the elements of the D5u diagonal member (Elements 436–
443) were reduced to half. The frequencies of the FE model show a
reasonable agreement with the corresponding experimental identifi-
cation, as observed in Table 4. The change in the natural frequency
of the diagonal mode due to damage, which resulted in close natural
frequencies for the diagonal and third bending modes, can also be
confirmed from the simulated damaged FE model. Because of this
closeness in frequency, dynamic coupling (Yamaguchi et al. 2015)
of the D5u diagonal member was observed in the third bending
mode, as can be seen in Fig. 16(b). It is important to note that the
proposed model-updating framework was not applied here to iden-
tify the damaged stiffness parameters corresponding to the damaged
D5u diagonal member because of the limitation of the number of
identified modes and too few components of experimentally
Fig. 15. Damaged and reinforced condition of D5u diagonal member
(images by Yasunao Matsumoto)
Table 4. Modal Identification Results of Frequencies and Damping for Fourth Span
Natural frequency (Hz)
Experimental
Identified mode Before After Rate (%)
First bending 2.577 2.604 –1
Second bending 5.254 5.313 –1
Third bending 7.143 7.295 –2
Diagonal 7.135 9.783 –27
Fig. 16. Modes obtained from damaged FE model of fourth span: (a) diagonal; (b) third bending
© ASCE
J. Bridge Eng., 2018, 23(1): 04017114
identified mode shapes. In a situation where the location of damage
is not known, the identification of damaged stiffness parameters by
the proposed FE model-updating framework requires a large array
of sensors at both deck and diagonal members, leading to an
increase in monitoring cost and time. In that case, identifying the
change in updated stiffness parameters would be sufficient to detect
damage at the local level without going into the analysis of damping
discussed in this section. However, in many practical situations,
model updating using only global modes identified with a limited
number of sensors may not be able to trace the damaged stiffness
parameters, as the changes in the frequencies for the global modes
due to local damage are not significant. The advantage of using the
damping, as used in this study, is that the damping change in global
modes affected by local damage can be identified with a small num-
ber of sensors (Yamaguchi et al. 2015). For the identification of
local damage using a global mode, only a small number of sensors
is required to identify the damping parameter of a particular global
mode for monitoring the condition, although at a certain occasion,
there is a need for comprehensive measurement with a large number
of sensors to understand the modal characteristics fully.
Identification of Loss Factors for Damaged Span
To understand possible causes of the significant change in experi-
mentally identified modal damping ratios for global modes due to
damage in the D5u diagonal member, the analytical modal damping
ratios were evaluated for this span using the same energy-based
damping model. Because of the limitation of the number of identi-
fied modes shown in Table 4, only the loss factors of diagonal mem-
bers, girders, and UC members, which appeared to be the dominant
contributors of damping in those modes as observed in Fig. 12,
were estimated using the experimentally identified modal damping
ratios at the fourth span. Loss factors of other substructures were
assumed to be the same as those of the first span, which are given in
Table 3. First, the analytical modal damping ratios corresponding to
the after-reinforcement (AR) condition were evaluated. Table 5
shows the analysis steps for the identification of loss factors for the
AR condition. The values of the equivalent loss factors of diagonal
members, girders, and UC members obtained from the previous
analysis were 0.0115, 0.0130, and 0.0078, respectively. For the
before-reinforcement (BR) condition, the equivalent loss factor of
the damaged D5u member ( h d5 ) was estimated by considering the
Modal damping ratio
FE model Experimental
Before After Before (SD) After (SD) Rate (%)
2.615 2.629 0.0093 (0.0020) 0.0069 (0.0007) 35
5.394 5.463 0.0072 (0.0014) 0.0061 (0.0009) 18
7.604 7.756 0.0106 (0.0009) 0.0060 (0.0007) 77
7.164 9.184 0.0055 (0.0008) 0.0039 (0.0017) 41
04017114-10 J. Bridge Eng.
 D5u diagonal in-plane mode shown in Fig. 16(a). In this step, the
equivalent loss factor for other diagonal members was taken as
0.0115, which corresponds to the AR condition. In the second step,
the loss factors of girders and UC members for the BR condition
were taken to be the same as the AR condition, as the damage was
observed in the D5u diagonal member only. Table 6 shows the iden-
tified damping parameters for the fourth span for the AR and BR
conditions. In the case of the BR condition, significant increase in
equivalent loss factor of the damaged D5u diagonal member was
observed.
Downloaded from ascelibrary.org by "Indian Inst Of Technology Library, Kharagpur" on 03/04/18. Copyright ASCE. For personal use only; all rights reserved.
elemental modal damping ratios evaluated analytically between AR
and BR conditions. The changes in elemental modal damping ratios
of Elements 436–443 that correspond to the D5u diagonal member
are much higher than the changes in other elements for the coupled
third bending mode. This result suggests that the elemental modal
damping ratios can be used to detect damage at the local level by
observing change in elemental modal damping ratios of the coupled
bending mode identified with a small number of sensors, which
could lead to an effective SHM based on vibration measurement. It
is worth noting here that to identify the damaged stiffness parameters

by FE model updating, a large number of sensors is required; how-

Damage Detection Using Change in Analytical Modal ever, with a small number of sensors, one can identify the change in
Damping Ratios modal properties for lower-order global modes and seek a possibility
to identify damage from that information along with a relevant anal-
The modal damping ratios of the identified global modes were then ysis instead of updating the model with data from a large array of
analytically evaluated using the damping parameters listed in Table 6 sensors at deck and diagonal members.
and the corresponding MER of substructures obtained from the FE It is understood from Eq. (4) that the change in modal damping
model. Table 10 shows the experimentally identified and analytically ratios is only possible if there is a change in either equivalent loss
evaluated modal damping ratios for the AR and BR conditions. factors or MER of substructures, or both. Hence, to investigate fur-
The percentage changes in analytically evaluated modal damping ther, the distributions of MER of each substructure in AR and BR
ratios are also listed in the last column of Table 7. It can be seen conditions are compared for the first three bending modes, as listed
that analytically evaluated modal damping ratios agree well with in Table 8. From this comparison, it can be seen that the distribution
the corresponding experimental ones for the AR condition of MER of diagonal members, girders, and UC members remains
because the damping parameters for three substructures were almost the same for global modes due to the local damage at the
identified with the experimental data for the AR condition, as D5u diagonal member. In contrast, some changes in the distribution
shown in Table 5. However, for the BR condition, such types of of MER of TLBs and BLBs and supports were observed, as can be
agreements between the analytically evaluated and experimen-
tally identified modal damping ratios were not observed. Table 8. Values of MER of All Substructures at AR and BR Conditions
Very few changes were observed in analytically evaluated  
MER of substructures Vn;i =Un
modal damping ratios of the first two bending modes due to damage
in the D5u diagonal member, although a significant number of First bending Second bending Third bending
changes were observed in experimentally identified damping ratios Members AR BR AR BR AR BR
of these two modes. In case of the third bending mode, an increment
of approximately 17% was observed in the analytical modal damp- D 0.3728 0.3791 0.5151 0.5136 0.5256 0.5054
ing ratio, although this change was much smaller than the change in G 0.2246 0.2230 0.3246 0.3152 0.3524 0.3583
UC 0.3693 0.3650 0.1204 0.1214 0.0995 0.1063
experimental modal damping ratio.
TLB 0.0006 0.0007 0.0009 0.0055 0.0004 0.0024
For the third bending mode, the modal damping contribution
BLB 0.0129 0.0128 0.0321 0.0379 0.0148 0.0134
from each element was estimated using Eq. (6) for both AR and BR
S 0.0633 0.0647 0.0090 0.0072 0.0042 0.0261
conditions. Fig. 17 shows the bar diagram plots of change in

Table 5. Analysis Steps for Identification of Equivalent Loss Factors for AR Condition

Analysis step Considered mode Estimated parameter Analysis conditions

Step 1 Diagonal hd Assumed that energy dissipated only from diagonal members
Step 2 1: First bending h uc , h g Same values of h tlb , h blb , m s from reinforced first span were used here
2: Second bending
3: Third bending NNLS was used to obtain equivalent loss factors

Table 6. Damping Parameters Corresponding to AR and BR Conditions

Damping parameter h d5 hd hg h uc h tlb h blb ms

AR 0.0115 0.0115 0.0130 0.0078 0.0963 0.0229 0.3796
BR 0.0172 0.0115 0.0130 0.0078 0.0963 0.0229 0.3796

Table 7. Experimentally Identified and Analytically Evaluated Modal Damping Ratios

Mode Experimental AR Analytical AR Experimental BR Analytical BR % Change in Analytical j

First bending 0.0069 0.0070 0.0093 0.0071 1.4
Second bending 0.0061 0.0061 0.0072 0.0064 4.9
Third bending 0.0060 0.0059 0.0106 0.0069 16.9

© ASCE 04017114-11 J. Bridge Eng.

J. Bridge Eng., 2018, 23(1): 04017114

 Table 9. Experimentally Identified and Analytically Evaluated Modal Damping Ratios after Reanalysis of Loss Factors

Mode Experimental AR Analytical AR Experimental BR Analytical BR % Change in analytical j

First bending 0.0069 0.0070 0.0093 0.0092 31.4
Second bending 0.0061 0.0061 0.0072 0.0078 27.9
Third bending 0.0060 0.0059 0.0106 0.0097 64.4

Table 10. MER of Different Components of Girders for AR and BR Conditions

Downloaded from ascelibrary.org by "Indian Inst Of Technology Library, Kharagpur" on 03/04/18. Copyright ASCE. For personal use only; all rights reserved.

MER of different components of girder

First bending Second bending Third bending
Member AR BR % AR BR % AR BR %
CB 0.0275 0.0272 –0.8 0.0656 0.0661 0.7 0.1225 0.1176 –4.0
St 0.0601 0.0599 –0.3 0.0949 0.1035 9.0 0.1739 0.1704 –2.0
LC 0.1371 0.1358 –1.0 0.1640 0.1456 –11.3 0.0560 0.0702 25.4

seen in Table 8, although their contributions are much smaller com- -5

x 10
pared to the contributions from diagonals, girders, and UC mem- 10

Change in elemental ξ
bers. From the energy-based damping model as given in Eq. (4), it
is understood that the modal damping ratio is proportional to the
equivalent loss factors and friction coefficient for a constant MER 5
of each substructure. Hence, there must be some increase in equiva-
lent loss factor(s) of one or several substructures as well due to dam- 0
age at the D5u diagonal member, as the experimental identification
results for the BR condition clearly indicated a significant increase
in modal damping ratios of the first three bending modes. Some -5
50 100 150 200 250 300 350 400 450 500
changes in loss factors of the substructures that appear not to have Element number
damage could be possible if the modal damping ratios are found to
be dependent on the amplitude of vibration, which is discussed in Fig. 17. Bar diagram of change in elemental modal damping ratios for
more detail in the next subsection. third bending mode

Reanalysis of Loss Factors for Fourth Span

with Damage
The loss factors of substructures were reidentified using the experi-
mentally identified modal damping ratios of the first three bending
modes for the BR condition. However, due to the limitation of the
number of identified modes, only the loss factors of girders and UC
members were re-estimated following the same steps as described
in Table 5. The values of equivalent loss factors of girders and UC
members were 0.0258 and 0.0078, respectively. Comparing these
values with the corresponding values in Table 6, there was a signifi-
cant increase in the equivalent loss factor of girders from 0.0130 to
0.0258, although the value of the loss factor of UC members
remained unchanged at 0.0078. Table 9 shows the experimentally
identified and analytically evaluated modal damping ratios for both
AR and BR conditions. A much better agreement between experi-
mentally identified and analytically evaluated modal damping ratios
was observed for the BR condition when there was an increase in
loss factor of girders along with an increase in loss factor of the
damaged diagonal member (Tables 7 and 9).
To investigate that much change in the loss factor of girders,
the distributions of MER of different components of girders, such
as crossbeams (CBs), stringers (St), and lower chords (LCs),
were studied. Table 10 shows the values of the MER of different
components of girders for AR and BR conditions for the first three
bending modes. The percentage change in the MER of each com-
ponent with respect to the AR condition is also listed in Table 10.
Significant changes were observed in the distribution of MER of
lower chords for the second and third bending modes due to dam-
age at the D5u diagonal member. Some changes in the
distribution of MER of other components were observed as well
for the second and third bending modes. However, the distribu-
tion of MER of the first bending mode was almost unaffected by
the damage.
To understand the change in the distribution of modal damping
ratios of various components of girder, the distributions of MER of
elements near the damaged D5u member were studied. Fig. 18
shows the elements of girders near the damaged D5u member.
Table 11 shows the MER of elements near the damaged D5u mem-
ber for the first three bending modes. The results suggest a signifi-
cant increase in MER for the third bending mode in Elements 9 and
10, which are the elements of lower chords, due to damage. For the
second bending mode, the MERs of Elements 9 and 10 were
reduced, whereas the MERs of Elements 1,306 and 1,307, which
are the elements of stringers, were increased significantly. These
changes at the elemental level are consistent with the change in dif-
ferent components of the girder observed in Table 10. In contrast,
the MERs from all of these elements almost remain the same for the
first bending mode. From these results, it is clear that the damage in
the D5u diagonal member has some effect on the distribution of
MERs for the second and third bending modes and is more promi-
nent for the third bending mode due to the coupling with the local
motion of the damaged D5u diagonal member. The change in the
loss factors of the girder substructure observed between AR and BR
conditions may be attributed to the change in modal energy distribu-
tion, as discussed earlier, and the nonhomogeneity in the loss factors
of different components of girders. While considering the same loss
factor for a nonhomogeneous substructure, such as girders, due to
the limitation of number of identified modes in the present study,

© ASCE 04017114-12 J. Bridge Eng.

J. Bridge Eng., 2018, 23(1): 04017114

 715
1307
1306
Downloaded from ascelibrary.org by "Indian Inst Of Technology Library, Kharagpur" on 03/04/18. Copyright ASCE. For personal use only; all rights reserved.

716
14
13
10
9

Fig. 18. Elements of girder near to damaged D5u diagonal member

Table 11. MER of Elements Near the Damaged D5u Member

 
MER of elements Vn;i =Un
First bending Second bending Third bending
Element
number AR BR AR BR AR BR
−3 −3 −3 −3 −4
9 6.22  10 5.93  10 2.84  10 1.98  10 8.71  10 1.63  10−3
10 6.03  10−3 5.82  10−3 2.84  10−3 1.93  10−3 9.37  10−4 1.73  10−3
13 7.15  10−3 7.11  10−3 3.03  10−4 3.71  10−4 3.83  10−3 4.03  10−3
14 7.14  10−3 7.09  10−3 3.11  10−4 3.55  10−4 3.94  10−3 4.16  10−3
715 1.47  10−3 1.37  10−3 2.04  10−3 2.42  10−3 9.08  10−3 8.73  10−3
716 1.33  10−4 1.09  10−4 2.24  10−4 1.95  10−4 1.61  10−3 2.05  10−3
1,306 1.77  10−3 1.76  10−3 1.81  10−3 2.62  10−3 7.25  10−4 6.39  10−4
1,307 2.11  10−3 2.29  10−3 8.04  10−4 1.60  10−3 9.83  10−3 1.11  10−2

some weighting factors can be introduced to express nonidentical
contributions of damping from different components of a nonhomo-
geneous substructure. The selection of weighting factors could be
based on the actual exposure conditions and material properties of
these components in a real structure. In this study, the concrete deck
slab was modeled by equivalent section concept with added mass
and stiffness properties to stringer beams and added stiffness to the
cross beams. Thus, higher weightage values can be assigned to
stringers and cross beams compared to the lower chords, as viscous
damping of RC is higher than the steel.
The increase in the loss factor of girders might also be attributed
to the fact that the same values of loss factors and friction coefficient
of the first span were used in the fourth span for top and bottom lat-
eral bracings and supports, although it was observed that the modal
damping ratios of the fourth span are different from the first span,
even for the AR condition. Although the contributions of damping
from these substructures were smaller individually, collectively
their contribution can be significant and could affect the results as
well.
Possible reasons for the increase in the loss factor of the girder
could include the amplitude dependence of the loss factor.
Yamaguchi et al. (2015) found that the modal damping ratios for
the lower-order modes, in particular the first bending mode, of the
bridge in the present study showed the dependence on amplitude in
which the modal damping ratios tended to increase with an increase
in the initial modal amplitude. It was also reported in that study that
the modal damping ratios for the higher-order modes including the
coupled bending mode did not show any clear sign of amplitude
dependence. To investigate this more, the initial modal amplitudes
of vibration at measurement locations obtained from the ERA are
listed in Table 12. Table 12 shows that the initial modal amplitudes
of girders for the first and second bending modes are much higher in
the BR condition compared to the AR condition, which might have
caused the increase in the modal damping for these two modes. For
the third bending mode, a significantly large amplitude of the D5u
member was observed in the BR condition. If the loss factor of D5u
is amplitude-dependent, then the loss factor of the D5u member for
the third bending mode in the BR condition would be much higher
compared to that for the damaged diagonal mode, and this increase
in the loss factor might have caused the significant increase in the
modal damping ratio for the third bending mode. Considering the
aforementioned hypothesis, the loss factor of the damaged D5u
member for the third bending mode was evaluated as 0.1047, keep-
ing the loss factors of all other substructures the same as for the AR
condition, which was clearly much higher than the corresponding
value of 0.0172 for the damaged diagonal mode. Here, considera-
tion of the amplitude-dependent loss factor could be reasonable, as
the damage discussed in this paper was due to cracks that could
induce a significant increase in frictional damping, which is
expected to be dependent on the amplitude of vibration. Hence, the
analytical model for damping using the energy-based approach
could be modified considering amplitude-dependent loss factors to
express the substantial change in the modal damping ratio of the
coupled global mode more reasonably in the case of damage in the
diagonal member of steel truss bridges, which could be future works
in this field.

© ASCE 04017114-13 J. Bridge Eng.

J. Bridge Eng., 2018, 23(1): 04017114

 Table 12. Initial Modal Amplitudes at Measurement Locations for AR and BR Conditions
Amplitude at AR condition (m/s2)
Mode Amplitude at U2 Amplitude at L2 Amplitude at D5u
First bending 0.0058 0.0062 0.0001
Second bending 0.0126 0.0128 0.0033
Third bending 0.0053 0.0043 0.0021
Diagonal 0.0008 0.0014 0.0342
Downloaded from ascelibrary.org by "Indian Inst Of Technology Library, Kharagpur" on 03/04/18. Copyright ASCE. For personal use only; all rights reserved.
Conclusions
In this study, an analytical framework for vibration-based SHM
was presented. An EBDE was introduced for detecting damage at
the local level by using a minimum number of sensors. The main
conclusions of this paper can be summarized as follows:
1. The modal damping ratios can be analytically evaluated by
using an energy-based damping definition, and a method to
know the contribution of modal damping ratios from different
structural elements of a FE model was introduced.
2. The results of EBDE implied that substantial change found
experimentally in the damping for the global mode coupled
with the diagonal mode was caused by a significant increase in
the equivalent loss factor of the damaged diagonal member.
3. The proposed approach may be able to detect damage at the
local level by analytically investigating the contribution from
structural elements to the damping of the global mode coupled
with vibration of the local member in steel truss bridges.
The proposed vibration-based SHM approach could be prom-
ising in detecting damage at the local level when the problem
related to low sensitivity of frequencies and mode shapes due to
local damage remains a concern, and damage detection by change
in stiffness parameters using FE model updating utilizing data
from a large number of sensors is not practically feasible due to
the limitation of budget and time. However, real structures often
do not show purely viscous damping, and in the case of damage,
the energy dissipation mechanism depends on the types of dam-
age. Hence, further study is needed to model damping of a dam-
aged structure more appropriately, and the dependence of loss
factors on amplitude can be taken into account while evaluating
the contribution of damping using an equivalent viscous damping
model.
Acknowledgments
The authors express sincere gratitude to Dr. Tsutomu Yoshioka for
providing vibration measurement data for the bridge before
reinforcement and the initial model parameters.
References
Beck, J. L., and Katafygiotis, L. S. (1998). “Updating models and their
uncertainties. I: Bayesian statistical framework.” J. Eng. Mech., 10
.1061/(ASCE)0733-9399(1998)124:4(455), 455–461.
Chen, G. W., Beskhyroun, S., and Omenzetter, P. (2016). “Experimental
investigation into amplitude-dependent modal properties of an eleven-
span motorway bridge.” Eng. Struct., 107, 80–100.
Curadelli, R. O., Riera, J. D., Ambrosini, D., and Amani, M. G. (2008).
“Damage detection by means of structural damping identification.” Eng.
Struct., 30(12), 3497–3504.
Dammika, A. J., Kawarai, K., Yamaguchi, H., Matsumoto, Y., and
Yoshioka, T. (2015). “Analytical damping evaluation complementary to
© ASCE 04017114-14
J. Bridge Eng., 2018, 23(1): 04017114
Amplitude at BR condition (m/s2)
Amplitude at U2 Amplitude at L2 Amplitude at D5u
0.0255 0.0270 0.0050
0.0261 0.0246 0.0015
0.0053 0.0083 0.1488
0.0024 0.0011 0.0583
experimental structural health monitoring of bridges.” J. Bridge Eng.,
10.1061/(ASCE)BE.1943-5592.0000691, 04014095.
Doebling, S. W., Farrar, C. R., and Prime, M. B. (1998). “A summary
review of vibration-based damage identification methods.” Shock Vib.
Digest, 30(2), 91–105.
Fan, W., and Qiao, P. (2011). “Vibration-based damage identification meth-
ods: A review and comparative study.” Struct. Health Monit., 10(1),
83–129.
Feng, D., and Feng, M. Q. (2015). “Model updating of railway bridge
using in situ dynamic displacement measurement under trainloads.”
J. Bridge Eng., 10.1061/(ASCE)BE.1943-5592.0000765, 04015019,
1–12.
Feng, D., and Feng, M. Q. (2016). “Output-only damage detection using ve-
hicle-induced displacement response and mode shape curvature index.”
Struct. Control Health Monit., 23(8), 1088–1107.
Frizzarin, M., Feng, M. Q., Franchetti, P., Soyoz, S., and Modena, C.
(2010). “Damage detection based on damping analysis of ambient vibra-
tion data.” Struct. Control Health Monit., 17(4), 368–385.
Fujino, Y., and Siringoringo, D. M. (2008). “Structural health monitoring of
bridges in Japan: An overview of the current trend.” Proc., 4th Int. Conf.
on FRP composites in Civil Engineering, Empa, Structural Engineering
Research Laboratory, Switzerland.
Govers, Y., and Link, M. (2010). “Stochastic model updating-covariance
matrix adjustment from uncertain experimental modal data.” Mech. Syst.
Signal Process., 24(3), 696–706.
Juang, J. N., and Pappa, R. S. (1985). “An eigensystem realization algo-
rithm for modal parameter identification and model reduction.” J.
Guidance Control Dyn., 8(5), 620–627.
Kawashima, K., Nagashima, H., and Iwasaki, H. (1994). “Evaluation of
modal damping ratio based on strain energy proportional damping
method.” Proc., 9th U.S.–Japan Bridge Engineering Workshop, Public
Works Research Institute (PWRI), Public Works Research Institute,
Tsukuba, Japan, 211–226.
MATLAB [Computer software]. MathWorks, Natick, MA.
Modena, C., Sonda, D., and Zonta, D. (1999). “Damage localization in rein-
forced concrete structures by using damping measurements, Damage
assessment of structures.” Proc., Int. Conf. on Damage Assessment of
Structures, DAMAS 99, Trans Tech Publisher, Switzerland, 132–141.
Mustafa, S., Debnath, N., and Dutta, A. (2015). “Bayesian probabilistic
approach for model updating and damage detection for a large truss
bridge.” Int. J. Steel Struct., 15(2), 473–485.
Mustafa, S., and Matsumoto, Y. (2017). “Bayesian model updating and its
limitations for detecting local damage of an existing truss bridge.” J.
Bridge Eng., 10.1061/(ASCE)BE.1943-5592.0001044, 04017019.
Nair, K. K., Kiremidjian, A. S., and Law, K. H. (2006). “Time series-based
damage detection and localization algorithm with application to the
ASCE benchmark structure.” J. Sound Vib., 291(1–2), 349–368.
Nakamura, M., Masri, S. F., Chassiakos, A. G., and Caughey, T. K. (1998).
“A method for non-parametric damage detection through the use of neu-
ral networks.” Earthquake Eng. Struct. Dyn., 27(9), 997–1010.
Nashif, A. D., Jones, D. I., and Henderson, J. P. (1985). Vibration damping,
Wiley, New York, 45–50.
Overschee, P. V., and Moor, B. D. (1993). “Subspace algorithms for the sto-
chastic identification problem.” Automatica, 29(3), 649–660.
Salawu, O. (1997). “Detection of structural damage through changes in fre-
quency: A review.” Eng. Struct., 19(9), 718–723.
Simoen, E., De Roeck, G., and Lombaert, G. (2015). “Dealing with uncer-
tainty in model updating for damage assessment: A review.” Mech. Syst.
Signal Process., 56–57, 123–149.
J. Bridge Eng.
 Sohn, H., and Law, K. H. (2000). “Bayesian probabilistic damage detection
of a reinforced-concrete bridge column.” Earthquake Eng. Struct. Dyn.,
29(8), 1131–1152.
Soize, C., Capiez-Lernout, E., and Ohayon, R. (2008). “Robust updating of
uncertain computational models using experimental modal analysis.”
AIAA J., 46(11), 2955–2965.
Ungar, E. E., and Kerwin, E. M. (1962). “Loss factor of viscoelastic systems
in terms of energy concepts.” J. Acoust. Soc. Am., 34(5), 954–957.
Vanik, M. W., Beck, J. L., and Au, S. K. (2000). “Bayesian probabilistic
approach to structural health monitoring.” J. Eng. Mech., 10.1061
/(ASCE)0733-9399(2000)126:7(738), 738–745.
Downloaded from ascelibrary.org by "Indian Inst Of Technology Library, Kharagpur" on 03/04/18. Copyright ASCE. For personal use only; all rights reserved.
Yamaguchi, H., and Ito, M. (1997). “Mode-dependence of structural damping
in cable-stayed bridges.” J. Wind Eng. Indust. Aerodyn., 72, 289–300.
Yamaguchi, H., and Jayawardena, L. (1992). “Analytical estimation of
structural damping in cable structures.” J. Wind Eng. Ind. Aerodyn.,
43(1–3), 1961–1972.
Yamaguchi, H., Matsumoto, Y., and Yoshioka, T. (2015). “Effects of local
structural damage in a steel truss bridge on internal dynamic coupling
and modal damping.” Smart Struct. Syst., 15(3), 523–541.
© ASCE 04017114-15
J. Bridge Eng., 2018, 23(1): 04017114
Yamaguchi, H., and Nagahawatta, H. D. (1995). “Damping effects of cable
cross ties in cable-stayed bridges.” J. Wind Eng. Ind. Aerodyn., 54–55,
35–43.
Yamaguchi, H., Takano, H., Ogasawara, M., Shimosato, T., Kato, M., and
Kato, H. (1997). “Energy-based damping evaluation of cable-stayed
bridges and application to Tsurumi Tsubasa bridge.” Struct. Eng.
Earthquake Eng., 14(2), 201s–213s.
Yoshioka, T., Yamaguchi, H., and Matsumoto, Y. (2010). “Structural
health monitoring of steel truss bridges based on modal damping
changes in local and global modes.” Proc., 5th World Conf. on
Structural Control and Monitoring, International Association for
Structural Control and Monitoring, Earthquake Engineering
Program of the U.S. National Science Foundation, Oakland, CA,
167.
Yuen, K. V., Au, S. K., and Beck, J. L. (2004). “Two-stage structural health
monitoring approach for Phase I benchmark studies.” J. Eng. Mech., 10
.1061/(ASCE)0733-9399(2004)130:1(16), 16–33.
Zoghi, M. (2014). The international handbook of FRP composites in civil
engineering, CRC Press, Boca Raton, FL.
J. Bridge Eng.