Dynamic Analysis of Multispan Viaducts with Weak

Coupling between Adjacent Spans

Kai Liu1; Geert Lombaert2; and Guido De Roeck, M.ASCE3

Abstract: This paper deals with the dynamic analysis of multispan simply supported railway viaducts subjected to moving trains. In the anal-
Downloaded from ascelibrary.org by University of Central Florida on 11/25/20. Copyright ASCE. For personal use only; all rights reserved.

ysis, the Sesia viaduct is considered, which is a composite railway bridge consisting of seven isostatic spans. An experimental analysis shows
that the modes of adjacent spans are connected because of the weak coupling through the ballast and rails. In this case, a model of the entire
viaduct is needed for an accurate prediction of the dynamic response of the bridge during the passage of the train. To reduce the computational
cost, two alternatives are investigated: a single-span model with adjusted boundary conditions and a component mode synthesis (CMS)
reduced-order model comprising all seven spans. The two models are subsequently used for dynamic train-bridge interaction analysis. The
predicted dynamic responses of the viaduct during a train passage with both models are in good agreement with the experimental results. Com-
pared with the single-span model, the coupling between adjacent spans is better represented by the CMS model of the entire viaduct. In practical
engineering design, however, the single-span model with adjusted boundary conditions can be used to obtain a reasonably good representation
of the dynamic behavior of the bridge. DOI: 10.1061/(ASCE)BE.1943-5592.0000476. © 2014 American Society of Civil Engineers.
Author keywords: Multispan railway viaduct; Weak coupling; Component mode synthesis; Train-bridge interaction.

Introduction
Railway viaducts are often built as multispan simply supported or
continuous bridges, as illustrated in Figs. 1 and 2. For continuous
viaducts, a strong coupling exists between adjacent spans. For
simply supported multispan viaducts, a weak coupling between
adjacent spans occurs through the ballast and rails. In this case,
a model of the entire viaduct is needed for an accurate prediction of
the dynamic response of the bridge during the passage of the train.
The evaluation of the natural frequencies and eigenmodes of a de-
tailed model of the entire viaduct, however, can be cumbersome and
requires a large computational cost.
In case of weakly coupled spans, an attempt can be made to
approximate the coupling by an appropriate choice of the boundary
conditions at the ends of a single-span model. The boundary con-
ditions may strongly depend on the structural type and the particular
type of modes predominately excited during the train passage. For
the case of the Sesia viaduct, a composite railway bridge consisting
of seven spans (Figs. 1 and 3), an experimental analysis shows that
coupling between adjacent spans results in modes that can approxi-
mately be divided into symmetrical and antisymmetrical modes (Liu
et al. 2009). In this case, a choice has to be made between symmetrical
and antisymmetrical boundary conditions at the ends of a single-span
1
Research Assistant, Dept. of Civil Engineering, K.U. Leuven, B-3001
Heverlee, Belgium (corresponding author). E-mail: liukaithu@hotmail.com
2
Associate Professor, Dept. of Civil Engineering, K.U. Leuven, B-3001
Heverlee, Belgium.
3
Professor, Dept. of Civil Engineering, K.U. Leuven, B-3001 Heverlee,
Belgium.
Note. This manuscript was submitted on July 27, 2012; approved on
January 31, 2013; published online on February 2, 2013. Discussion period
open until June 1, 2014; separate discussions must be submitted for in-
dividual papers. This paper is part of the Journal of Bridge Engineering,
Vol. 19, No. 1, January 1, 2014. ©ASCE, ISSN 1084-0702/2014/
1-83–90/$25.00.
model to account for the coupling between adjacent spans (Liu et al.
2009). The experimental modal analysis of the Antoing bridge, which
is a prestressed concrete viaduct consisting of five simply supported
spans (Fig. 4), on the other hand, shows that hardly any coupling is
present between adjacent spans. This is because of the fact that the rails
and ballast are close to the center of gravity of the cross section and
do not provide a coupling in bending of adjacent spans.
Another approach is to use submodeling techniques to set up
a reduced-order model of the entire viaduct. Submodeling tech-
niques allow constructing efficient reduced-order models for an-
alyzing the dynamic behavior of large or complex structures. The
first dynamic substructuring ideas were developed by Hurty (1960,
1965) and have been extensively developed since then. These
methods are known as component mode synthesis (CMS). Some
major developments followed shortly, resulting in the classical
methods by Craig and Bampton (Bampton and Craig 1968; Craig
1989), Rubin (1975), MacNeal (1971), and Hintz (1975). In-
teresting reviews of CMS can be found in Craig (1987, 1995,
2000); Craig and Kurdila (2006); and De Klerk et al. (2008). The
CMS method subdivides a given structure into substructures,
which are analyzed independently for natural frequencies and
mode shapes. The CMS methods are classified into fixed-interface,
free-interface, and hybrid methods, for component modes fixed at
the boundaries, free at the boundaries, or with boundaries partially
fixed, respectively. The substructure mode shapes are then as-
sembled to compute the natural frequencies and mode shapes of the
entire structure. With this method, a large reduction in the com-
putational cost can be obtained. The CMS method has been used
in many areas: modal analysis (Yin et al. 1997; Kisa and Gurelb
2006), forced vibration analysis (Seongsoo et al. 2011), design
optimization (Hou et al. 1995; Masson et al. 2006), etc. The CMS
method has recently been applied to the dynamic analysis of
bridges under moving trains. Biondi et al. (2005) used the CMS
method to analyze the coupled train-track-bridge system, which is
composed of two continuous substructures (rails and bridge) and
one discrete subsystem (running train). With this approach, the
number of unknowns involved in the dynamic analysis of bridges is

JOURNAL OF BRIDGE ENGINEERING © ASCE / JANUARY 2014 / 83

J. Bridge Eng., 2014, 19(1): 83-90


Downloaded from ascelibrary.org by University of Central Florida on 11/25/20. Copyright ASCE. For personal use only; all rights reserved.
Fig. 1. Sesia viaduct: a seven-span simply supported viaduct (image by
Kai Liu)
Fig. 2. Arroyo del Valle Viaduct: a 26-span continuous viaduct (image
by Guido De Roeck)
Fig. 3. Cross section of the Sesia viaduct
reduced, leading to a lower computational cost. De Salvo et al.
(2010) applied the CMS method to deal with the dynamic analysis
of multiple supported continuous beams subjected to moving
loads. The continuous beams are decomposed in primary and
secondary spans with restraints. The assumed local modes for each
84 / JOURNAL OF BRIDGE ENGINEERING © ASCE / JANUARY 2014
J. Bridge Eng., 2014, 19(1): 83-90
span are taken as the exact modal shapes of the homogenized span
with appropriate boundary conditions. The continuous beam is
then reassembled. This substructure approach provides an efficient
way for the dynamic analysis of the multiple supported continuous
beams under moving loads or moving masses.
In this paper, the dynamic response of the Sesia viaduct is an-
alyzed with both a single-span model and a CMS reduced-order
model. It is organized in the following manner. The “Experimen-
tal Analysis of the Sesia Viaduct” section presents the dynamic
experiments conducted at the Sesia viaduct and the modal analysis
based on the measured vibration data. “Numerical Model of the
Sesia Viaduct” describes the two numerical models developed for
the Sesia viaduct. In “Dynamic Train-Bridge Interaction Analy-
sis,” the bridge responses predicted are compared with the mea-
sured response. Finally, the conclusions are provided.
Experimental Analysis of the Sesia Viaduct
The Sesia viaduct, located on the new Italian high-speed line be-
tween Torino and Milano, is a composite railway bridge by seven
spans of 46 m and has a total length of 322 m (Fig. 1). Each simply
supported span consists of a girder of the same double box cross
section (Fig. 3). The bottom steel box is composed of a lower flange
and three webs. The concrete slab has a width of 13.6 m, a thickness
of 0.4 m, and is connected to the steel girder by studs.
Elaborate dynamic experiments have been performed on the
Sesia viaduct. Because of the large dimensions of the bridge, the
experiments were concentrated on the second span from Torino to
Milano. A small number of sensors was additionally placed on the
first and third span to evaluate the dynamic coupling between two
neighboring spans. The layout of the accelerometers is shown in
Fig. 5. In the experiments, seven reference channels were adopted,
consisting of six vertical and one horizontal acceleration mea-
surement (Fig. 5). All measurement channels are denoted by a label
xYzz, where x refers to the number of the span (1, 2, or 3), Y denotes
the main girder (A, B, or C) or the longitudinal stiffener (D or E), and
the number zz represents the cross section (1–15).
During the experiments, the response of the bridge to ambient
excitation (mainly caused by wind, traffic on the adjacent highway,
and human activity) and the response during a train passage have
been measured. For the measurements during ambient vibration,
a sampling frequency equal to 200 Hz and an acquisition time of
about 15 min have been used, whereas for the train passages,
a higher sampling frequency of 512 Hz has been used.
The modal analysis is performed using ambient vibration data,
data of free vibration after a train passage and data measured while
the train was on the bridge, respectively. The duration of the free
vibration data records is 3 s, whereas the records of the train passage
have a duration of about 4 s for a train speed of about 280 km/h.
From the ambient vibration data, eight modes were extracted us-
ing the reference-based stochastic subspace identification algorithm
(SSI) (Peeters and De Roeck 1999). Based on the dynamic behavior
of neighboring spans, the modes of the second span of the viaduct
can approximately be subdivided into quasi-symmetrical and quasi-
antisymmetrical modes (Fig. 6), which indicates that, although
girders of adjacent spans are not connected, the ballast and rails
realize a weak coupling between adjacent spans that is clearly
reflected in the dynamic properties of the considered span. The
mode shapes of the identified antisymmetrical and symmetrical
modes are presented in Figs. 7 and 8.
From the free vibration data after a train passage, three sym-
metrical modes are identified: the first vertical bending mode, the
second vertical bending mode, and the first torsional mode,
Downloaded from ascelibrary.org by University of Central Florida on 11/25/20. Copyright ASCE. For personal use only; all rights reserved.

Fig. 4. Cross section of the Antoing viaduct

Fig. 5. Layout of accelerometers in vertical direction (reference sensors are located at A6, B6, C4, C6, C8, and C12)

Numerical Model of the Sesia Viaduct

Fig. 6. Identified modes: (a) quasi-symmetrical; (b) quasi-antisymmetrical
whereas the second torsional mode is not identified. The first and
second symmetrical bending modes are also identified from
a modal analysis of the data during the train passage. This indicates
that the symmetrical modes are predominantly excited by the train
passage on the bridge. Table 1 compares the natural frequencies
identified from the different sets of vibration data. The lower value
of the first natural frequency for the train passage data is because
of the additional mass of the train. It should furthermore be noted
that both the train passage data and the free vibration data have
a short time duration, resulting in a lower resolution that makes
the identified frequencies less accruable compared with those
obtained from the ambient vibration data. The differences among
identified damping ratios are relatively small in view of the inherent
uncertainties.
Single-Span Model
As adjacent spans are coupled through the ballast and rails, a model
that includes all seven spans of the viaduct is required to study the
dynamic behavior of the bridge. However, because of the size of the
resulting finite-element (FE) model, this would lead to a large
computational cost. Therefore, a simplified model of one span is
used. As described in the previous section, the symmetrical modes
are predominantly excited when the train passes the bridge. There-
fore, symmetrical boundary conditions are adopted, where the con-
tinuity between adjacent spans is accounted for by restricting both the
rotation and longitudinal displacements of the ballast and the rails
(Fig. 9).
The numerical model of a single span of the Sesia viaduct is
created with the FE program ANSYS (Swanson Analysis Systems
2004) (Fig. 10). Table 2 presents the element types and the material
properties of the different components in the FE model. The steel
girder is modeled by shell63 elements with four nodes, each having
six degrees of freedom (DOFs), whereas the concrete slab is rep-
resented by solid45 elements with eight nodes having three DOFs.
The ballast is modeled by solid45 elements with material

JOURNAL OF BRIDGE ENGINEERING © ASCE / JANUARY 2014 / 85

J. Bridge Eng., 2014, 19(1): 83-90

 characteristics as found in the literature (Zhai et al. 2004). The
diagonals and bracings are represented by beam4 elements having
six DOFs at each node. The headed shear studs are represented by
linear spring elements combin14 and connect the corresponding
nodes in the longitudinal direction (Queiroz et al. 2007). The track is
modeled as a longitudinally invariant track, where the rail pads and
the mass of the sleepers are uniformly distributed along the track.
The track model consists of two Euler-Bernoulli beams, each rep-
resenting the two rails, and mass elements representing the sleepers.
The rail pads are represented by springs (Krs ) and dampers (Crs )
between the beams and mass elements (Ilias 1999). The mass per unit
length of the UIC60 rail is rr Ar 5 60 kg=m, and the vertical bending
stiffness is Er Ir 5 6:4 3 106 Nm2 .
Downloaded from ascelibrary.org by University of Central Florida on 11/25/20. Copyright ASCE. For personal use only; all rights reserved.
Fig. 7. Identified antisymmetrical mode shapes: (a) first vertical
bending mode at 3.62 Hz; (b) first torsional mode at 8.35 Hz; (c) second
vertical bending mode at 10.00 Hz; (d) second torsional mode at
11.26 Hz
Fig. 8. Identified symmetrical mode shapes: (a) first vertical bending
mode at 4.14 Hz; (b) first torsional mode at 9.00 Hz; (c) second vertical
bending mode at 10.44 Hz; (d) second torsional mode at 14.28 Hz
Table 1. Comparison of Identified Natural Frequencies from Different Vibration Data
Free vibration
Symmetrical modes Frequency (Hz) Damping ratio, j
First vertical bending mode 3.90 2.30
First torsional mode 9.13 1.25
Second vertical bending mode 10.41 2.38
86 / JOURNAL OF BRIDGE ENGINEERING © ASCE / JANUARY 2014
J. Bridge Eng., 2014, 19(1): 83-90
The computed natural frequencies are compared with the fre-
quencies of the symmetrical modes as identified from the ambient
vibration data (Table 3). A good agreement is obtained between the
calculated and measured natural frequencies. As an indication of the
level of agreement between the measured and predicted mode
shapes, the modal assurance criterion value (MAC) for each mode is
computed as given in Table 3. The MAC measures the least-squares
fit between two different mode shapes, with a MAC value of 1
indicating perfectly consistent modes and a value of 0 indicating
orthogonality. As shown in Table 3, a very good agreement is found
between the experimental and predicted mode shapes. The com-
parisons of the natural frequencies and mode shapes show that the
numerical model accurately represents the dynamic behavior of the
bridge.
Component Mode Synthesis Reduced-Order Model of
the Entire Viaduct
In this section, the CMS method is used to set up the model of
the entire viaduct. The CMS method involves three basic steps:
(1) the original structure is divided into several components, (2)
component modes are defined in terms of the degrees of freedom
of the interface and modal degrees of freedom, and (3) the
component mode models are coupled to form the reduced-order
model of the entire system (Craig 2000). For the multispan
simply supported viaducts, each span is considered as a single
component of the entire viaduct; the interface between adjacent
spans contains the nodes of the rail and ballast at the span ends
(Fig. 11).
Fig. 9. Boundary conditions for the single-span model
Fig. 10. FE model of the Sesia viaduct
Train passage Ambient vibration
Frequency (Hz) Damping ratio, j Frequency (Hz) Damping ratio, j
3.67 2.81 4.14 2.17
— — 9.00 1.84
10.54 4.27 10.44 2.64
 Table 2. Element Types and Material Characteristics of Numerical Model
Bridge
components Element type E ð3106 N=m2 Þ r ðkg=m3 Þ n Mass (kg) Stiffness ð3106 N=mÞ Damping ð3103 N×s=mÞ
Steel box shell63 205,600 7,850 0.3 — — —
Ballast solid45 280 1,700 0.28 — — —
Concrete slab solid45 31,000 2,500 0.17 — — —
Stud combin14 — — — — 450 —
Lateral panel solid45 15,500 2,300 0.17 — — —
Rail beam4 205,600 7,850 0.3 — — —
Sleepers mass21 — — — 290 — —
Krs combin14 — — — — 500 —
Crs combin14 — — — — — 200
Downloaded from ascelibrary.org by University of Central Florida on 11/25/20. Copyright ASCE. For personal use only; all rights reserved.

Table 3. Comparison of the Computed Natural Frequencies and Natural Dynamic Train-Bridge Interaction Analysis
Frequencies Identified from Ambient Vibration Data
Natural frequency (Hz) The numerical models of the Sesia viaduct are now used to study
the dynamic response of the bridge under a train passage. The in-
Mode Measured Computed MAC teraction between a bridge and a train traveling over the bridge is
First vertical bending mode 4.14 4.15 0.99 a coupled dynamic problem (Xu et al. 2004; Xia and Zhang 2005;
First torsional mode 9.00 9.01 0.98 Liu et al. 2009; Huang 2012). The bridge and the train interact with
Second vertical bending mode 10.44 10.27 0.99 each other through the contact points between the wheel sets and
Second torsional mode 14.28 14.56 0.99 the rails. In this study, a vehicle model with 15 DOFs is used (Liu
et al. 2009). The interaction between the vehicle and the bridge is
considered assuming a perfect contact between the wheel sets and
the rail (Lee et al. 2005; Zhang et al. 2008). The track unevenness is
not considered in this study. The dynamic response of the system is
obtained by solving the following coupled equations:

€ v þ Cv V_ v þ Kv Vv ¼ Pv
Mv V
(1)
€ b þ Cb V_ b þ Kb Vb ¼ Pb
Mb V

where Mv , Cv , and Kv 5 mass matrix, damping matrix, and

Fig. 11. Components for a multispan simply supported railway bridge
The Sesia viaduct is divided into seven components, labeled
a1 , a2 , . . . , a7 , each corresponding to one span. The interface nodes
between two spans include two nodes for the rail, with each rail
representing one track, and 42 nodes for the ballast, resulting in a total
of 138 DOFs for each interface. It is commonly agreed to keep all
substructure modes with natural frequency f # 1:5fmax to ensure
convergence, where fmax is the maximum frequency of interest of
the complete structure (Tournour et al. 2001). In this paper, a maxi-
mum frequency fmax of 30 Hz is considered for the bridge [European
Committee for Standardization (CEN) 2002], leading to the inclusion of
the first 339 fixed-interface modes with a natural frequency up to 45 Hz.
This results in a total of 3,201 DOFs (339 DOFs 3 7 1 138 DOFs
3 6 5 3,201 DOFs). Compared with the full model that has more
than 300,000 DOFs, the computational cost of the modal analysis is
greatly reduced. However, the full model of the Sesia viaduct is set
up as well for a verification of the reduced-order model. The natural
frequencies of the bridge predicted by both models are com-
pared with the experimental frequencies in Table 4. Both quasi-
antisymmetrical and quasi-symmetrical modes of the second span
are obtained for the full FE model and the CMS reduced model. The
good agreement of the natural frequencies for the two approaches
shows that the reduced-order model provides an accurate approx-
imation of the full model.
The computed mode shapes of the first antisymmetrical bending
mode, the first symmetrical bending mode, the second antisym-
metrical bending mode, and the second symmetrical bending mode
for the two beams representing the tracks of the second span are
presented in Fig. 12.
stiffness matrix of the vehicle, respectively; Vv , V_ v , and V €v
5 displacement, velocity, and acceleration of the vehicle, re-
spectively; vector Pv 5 force transferred to the vehicle through the
wheel sets and depends on the displacements and velocities of the
wheel sets at each time step; Mb , Cb , and Kb 5 mass matrix,
damping matrix, and stiffness matrix of the bridge, respectively; Vb ,
V_ b , and V€ b 5 displacement, velocity, and acceleration vectors of
the bridge, respectively; and Pb 5 force transferred to the bridge and
depends on the motion of the vehicle and wheel sets at each time step
(Liu et al. 2009).
In this work, the two coupled equations are solved with an
iterative method at each time step (Liu et al. 2009). The New-
mark-b time integration scheme is used (Newmark 1959). A
modal superposition is applied for the bridge model, considering
the modes of the bridge with natural frequencies up to 30 Hz
(CEN 2002). A damping ratio of 2.5%, corresponding to ap-
proximately a mean value of the identified damping ratio, is used
for all the modes considered. In Clough and Penzien (1975),
a time step Dt # Tb =10 is suggested to achieve a reasonable in-
tegration accuracy, where Tb is the lowest vibration period of the
bridge. The considered upper frequency of the bridge is 30 Hz;
therefore, a time step Dt 5 1=300 s is chosen.
Fig. 13 presents vertical displacement at the midsection of the
first span and the midsection of the seventh span from Torino to
Milano for an ETR500Y train running from Milano to Torino at
a speed of 288 km/h predicted by the CMS model. More information
on the vehicle characteristics can be found in Liu et al. (2009). The
vertical lines at t 5 t1 and t 5 t2 indicate the times at which the train
enters and leaves the seventh span from Torino to Milano (Fig. 5), as
illustrated in Fig. 14. Because of the dynamic coupling between

JOURNAL OF BRIDGE ENGINEERING © ASCE / JANUARY 2014 / 87

J. Bridge Eng., 2014, 19(1): 83-90

 Table 4. Comparison of Natural Frequencies and Mode Shapes of the Second Span of the Sesia Viaduct as Identified from the Experiments and as Computed
from the Full Model and the Reduced Model
Natural frequency (Hz) MAC
Mode Experiment Full model Reduced model Experiment-reduced model Full-reduced model
First antisymmetrical bending mode 3.62 3.63 3.63 0.96 0.99
First symmetrical bending mode 4.14 4.15 4.15 0.99 0.99
First antisymmetrical torsional mode 8.35 8.39 8.40 0.95 0.98
First symmetrical torsional mode 9.00 8.95 8.99 0.99 0.97
Second antisymmetrical bending mode 10.00 9.99 9.99 0.94 0.98
Second symmetrical bending mode 10.44 10.23 10.24 0.99 0.96
Second antisymmetrical torsional mode 11.26 11.52 11.58 0.93 0.95
Second symmetrical torsional mode 14.28 14.30 14.32 0.98 0.98
Downloaded from ascelibrary.org by University of Central Florida on 11/25/20. Copyright ASCE. For personal use only; all rights reserved.

Fig. 13. Predicted displacement at the midsection of the first span

(dashed line) and the midsection of the seventh span (solid line) from
Torino to Milano

spans. This phenomenon is well represented by the CMS model that

Fig. 12. Computed mode shapes for the Sesia viaduct bridge: (a) first
antisymmetrical bending mode at 3.63 Hz; (b) first symmetrical bending
mode 4.15 Hz; (c) second antisymmetrical bending mode at 9.99 Hz;
(d) second symmetrical bending mode at 10.24 Hz
adjacent spans, the vibratory energy of the seventh span slowly
decreases after the train leaves the span. The vertical lines at t 5 t3
and t 5 t4 indicate the time at which the train enters and leaves the
first span from Torino to Milano. It can be observed that the vibration
starts before the train enters the span because of the dynamic
coupling between adjacent spans.
Figs. 15 and 16 compare the predicted and experimental vertical
acceleration at two measured Points 2b06 and 2c12 on the second
span from Torino to Milano (Fig. 5). A low-pass filter of 30 Hz has
been applied to both results. The vertical lines at t 5 t5 and t 5 t6
indicate the times at which the train enters and leaves the second
span from Torino to Milano. Comparing the results of the single-
span model to those of the CMS reduced model shows that the re-
duced model of the full viaduct obtains much better results. The
experimental results show that the spans start to vibrate before the
train enters, which is because of the coupling between adjacent
rigorously takes into account coupling between adjacent spans.
When the train is on the bridge, there is an excellent match between
the experiments and results predicted by the CMS model. When the
train leaves the considered span, the level of free vibration is slightly
underestimated by the CMS model. From Figs. 15(b) and 16(b), it
can be observed that the dominant frequencies are well predicted by
the CMS model, especially for frequencies up to 15 Hz. Around the
frequency peak at 4 Hz, a better agreement is obtained between the
experimental result and the result predicted by the CMS model than
for the result predicted by the single-span model.
Furthermore, the contribution of the antisymmetrical modes and
symmetrical modes to the response of the bridge is verified. Fig. 17
presents the contribution of the first 50 modes to the deflection
at Points 2b06 and 2c12. It is clearly observed that the first
symmetrical bending mode dominates the displacement of the
bridge, whereas the contribution of the first antisymmetrical bend-
ing mode is relatively small. This justifies the choice of symmet-
rical boundary conditions when a single-span model is used as a
simplification.
Conclusions
In this paper, two methodologies are presented for analyzing the
dynamic response of multispan viaducts with weak coupling

88 / JOURNAL OF BRIDGE ENGINEERING © ASCE / JANUARY 2014

J. Bridge Eng., 2014, 19(1): 83-90

Downloaded from ascelibrary.org by University of Central Florida on 11/25/20. Copyright ASCE. For personal use only; all rights reserved.
Fig. 14. Position of the train at the following times: (a) t 5 t1 ; (b) t 5 t2 ; (c) t 5 t3 ; (d) t 5 t4
Fig. 15. Measured acceleration at Point 2b06 (black line) and accel-
eration predicted by the single-span model (gray line) and the CMS
model (dark gray line): (a) time history; (b) frequency content
between adjacent spans. In the first approach, a single-span model
is created to represent the entire viaduct. Analysis of measured data
for the Sesia viaduct shows that symmetrical modes are pre-
dominantly excited during the train passage. Therefore, the weak
coupling between adjacent spans is taken into account by sym-
metrical boundary conditions. In the second approach, the CMS
method is adopted to set up a reduced-order model of the entire
structure. Hereto, the original structure is divided into several
components, and each component is reduced individually and
subsequently coupled to form the reduced order model of the
whole structure. A comparison with the experimental data shows
that both models give a good prediction of the dynamic response of
the viaduct during train passage, and the following conclusions can
be drawn:
J. Bridge Eng., 2014, 19(1): 83-90
Fig. 16. Measured acceleration at Point 2c12 (black line) and accel-
eration predicted by the single-span model (gray line) and the CMS
model (dark gray line): (a) time history; (b) frequency content
1. The CMS method allows for the evaluation of the modal
parameters of multispan bridges at a reasonable computational
cost. The better agreement between the measured and pre-
dicted results obtained with the CMS model proves that a full
model is necessary for an accurate response prediction of these
types of structure even when only weak coupling is present.
2. The single-span model with adjusted boundary conditions can
be used to obtain reasonably accurate results for the dynamic
response of the bridge under train loading. A verification of the
contribution of the antisymmetrical modes and symmetrical
modes to the response of the bridge confirms the choice of
symmetrical boundary conditions in the simplified single-
span model. This model has the advantage of being simple
to implement; therefore, in practical engineering design, the
single-span model with adjusted boundary conditions can be
used.
JOURNAL OF BRIDGE ENGINEERING © ASCE / JANUARY 2014 / 89

Downloaded from ascelibrary.org by University of Central Florida on 11/25/20. Copyright ASCE. For personal use only; all rights reserved.
Fig. 17. Modal contribution to the displacement at (a) Point 2b06; (b)
Point 2c12
Acknowledgments
This research is conducted in the framework of the European
Commission Research Fund for Coal and Steel RTD Project No.
RFSR-CT-2006-00032 “DETAILS” [Design for optimal life cycle
costs (LCC) of high speed railway bridges by enhanced monitoring
systems]. The support of the EC is gratefully acknowledged.
References
Bampton, M., and Craig, R. (1968). “Coupling of structures for dynamic
analyses.” Amer. Inst. Aeronaut. Astronaut., 6(7), 1313–1319.
Biondi, A., Muscolino, G., and Sofi, A. (2005). “A substructure approach
tailored to the dynamic analysis of multi-span continuous beams under
moving loads.” Comp. Struct., 83(28–30), 2271–2281.
Clough, R., and Penzien, J. (1975). Dynamics of structures, McGraw Hill,
New York.
Craig, R. (1987). “A review of time-domain and frequency-domain
component-mode synthesis methods.” Int. J. Analyt. Exp. Modal
Anal., 2(2), 59–72.
Craig, R. (1989). “Component-mode synthesis for model order reduction of
non classically damped systems.” J. Guid. Control Dyn., 12(4), 577–584.
Craig, R. (1995). “Substructure methods in vibration.” ASME Trans.,
117(B), 207–213.
Craig, R. (2000). “Coupling of structures for dynamic analyses: An over-
view.” Proc., 41st AIAA/ASME/ASCE/AHS/ASC Structures, Structural
90 / JOURNAL OF BRIDGE ENGINEERING © ASCE / JANUARY 2014
J. Bridge Eng., 2014, 19(1): 83-90
Dynamics, and Materials Conf. and Exhibit, American Institute of
Aeronautics and Astronautics, Reston, VA.
Craig, R., and Kurdila, A. (2006). Fundamentals of structural dynamics,
Wiley, New York.
De Klerk, D., Rixen, D., and Voormeeren, S. (2008). “General framework
for dynamic substructuring: History, review, and classification of
techniques.” Amer. Inst. Aeronaut. Astronaut., 46(5), 1169–1181.
De Salvo, V., Muscolino, G., and Palmeri, A. (2010). “A substructure
approach for the dynamic analysis of trainctrackcbridge system.”
J. Sound Vibrat., 329(15), 3101–3120.
European Committee for Standardization (CEN). (2002). “Actions on
structures—Part 2: Traffic loads on bridges.” Eurocode ENV1991-2,
Brussels, Belgium.
Hintz, R. (1975). “Analytical methods in component modes synthesis.”
Amer. Inst. Aeronaut. Astronaut., 13(8), 1007–1016.
Hou, G., Maraju, V., and Yang, R. (1995). “Component mode synthesis-
based design optimization method for local structural modification.”
Struct. Multidisciplin. Optim., 10(2), 128–136.
Huang, D. (2012). “Vehicle-induced vibration of steel deck arch bridges and
analytical methodology.” J. Bridge Eng., 17(2), 241–248.
Hurty, W. (1960). “Vibrations of structural systems by component mode
synthesis.” J. Engrg. Mech. Div., 86(4), 51–69.
Hurty, W. (1965). “Dynamic analysis of structural systems using component
modes.” Amer. Inst. Aeronaut. Astronaut., 3(4), 678–685.
Ilias, H. (1999). “The influence of railpad stiffness on wheelset/track in-
teraction and corrugation growth.” J. Sound Vibrat., 227(5), 935–948.
Kisa, M., and Gurelb, M. (2006). “Modal analysis of multi-cracked beams with
circular cross section.” Amer. Inst. Aeronaut. Astronaut., 73(8), 963–977.
Lee, C., Kim, C., Kawatani, M., Nishimura, N., and Kamizono, T. (2005).
“Dynamic response analysis of monorail bridges under moving trains
and riding comfort of trains.” Eng. Struct., 27(14), 1999–2013.
Liu, K., Reynders, E., De Roeck, G., and Lombaert, G. (2009). “Experi-
mental and numerical analysis of a composite bridge for high speed
trains.” J. Sound Vibrat., 320(1–2), 201–220.
MacNeal, R. (1971). “Hybrid method of component mode synthesis.”
Comp. Struct., 1(4), 581–601.
Masson, G., Brik, B. A., and Cogan, S. (2006). “Component mode synthesis
(CMS) based on an enriched ritz approach for efficient structural op-
timization.” J. Sound Vibrat., 296(4–5), 845–860.
Newmark, N. (1959). “A method of computation for structural dynamics.”
J. Engrg. Mech. Div., 85(3), 67–94.
Peeters, B., and De Roeck, G. (1999). “Reference-based stochastic subspace
identification for output-only modal analysis.” Mech. Syst. Signal
Process., 13(6), 855–878.
Queiroz, F., Vellasco, P., and Nethercot, D. (2007). “Finite element
modelling of composite beams with full and partial shear connection.”
J. Constr. Steel Res., 63(4), 505–521.
Rubin, S. (1975). “Improved component-mode representation for structural
dynamic analysis.” Amer. Inst. Aeronaut. Astronaut., 13(8), 995–1006.
Seongsoo, L., Hyungsoo, M., and Chang-Wan, K. (2011). “On a component
mode synthesis on multi-level and its application to dynamics analysis of
vehicle system supported with spring-stiffness damper system.” J. Mech.
Sci. Technol., 25(12), 3115–3121.
Swanson Analysis Systems. (2004). ANSYS user’s manual, version 8.0.,
Houston, PA.
Tournour, M. A., Atalla, N., Chiello, O., and Sgard, F. (2001). “Validation,
performance, convergence and application of free interface component
mode synthesis.” Comp. Struct., 79(20–21), 1861–1876.
Xia, H., and Zhang, N. (2005). “Dynamic analysis of railway bridge under
high-speed trains.” Comp. Struct., 83(23–24), 1891–1901.
Xu, Y., Zhang, N., and Xia, H. (2004). “Vibration of coupled train and
cable-stayed bridge systems in cross winds.” Eng. Struct., 26(10),
1389–1406.
Yin, B., Wang, W., and Jin, Y. (1997). “The application of component mode
synthesis for the dynamic analysis of complex structures using ADINA.”
Comp. Struct., 64(5–6), 931–938.
Zhai, W., Wang, K., and Lin, J. (2004). “Modelling and experiment of
railwayballast vibrations.” J. Sound Vibrat., 270(4–5), 673–683.
Zhang, N., Xia, H., and Guo, W. (2008). “Vehicle–bridge interaction
analysis under high-speed train.” J. Sound Vibrat., 309(3–5), 407–425.