Beruflich Dokumente
Kultur Dokumente
www.elsevier.com/locate/compstruc
Abstract
In this paper, a spatial dynamic analysis model for the train±bridge system under random excitations was estab-
lished. The random excitations were analyzed and generated by the time series autoregressive model from the experi-
mentally measured wheel accelerations. The whole histories of the train running on the bridges were simulated on
computer. The responses of bridges were calculated and statistically studied. Based on the calculation results, the re-
inforcement schemes for steel girders were designed and some of them were applied to real bridges. The ®eld-tests were
carried out before and after the bridge reinforcement. The measured data from the tests proved the eectiveness of the
reinforcement schemes. Ó 2001 Elsevier Science Ltd. All rights reserved.
Keywords: Train±bridge system; Dynamic interaction; Computer simulation; AR model; Bridge reinforcement; Field test
0045-7949/01/$ - see front matter Ó 2001 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 4 5 - 7 9 4 9 ( 0 1 ) 0 0 1 1 5 - 8
1852 H. Xia et al. / Computers and Structures 79 (2001) 1851±1860
wheel-track relation. The vehicle model can be a train structure is dominantly in¯uenced by its several lowest
consisting of several locomotives, vehicles or their modes, this method has a very great advantage that an
combinations. The model is based on the following as- adequate estimation on the dynamic response can be
sumptions: obtained by considering only a few modes of vibration,
1. The eects of the elastic deformations of the car even for a spatial structure having hundreds of DOFs.
bodies and the wheel sets are neglected. Therefore, the computational eort is signi®cantly re-
2. All vehicles are simpli®ed into the suspension duced.
system of single-set-springs and dashpots. The spring The following assumptions have been taken into ac-
stiness and the dashpot damping coecient of a bogie count for the bridge model:
are divided over its wheels. 1. There is no relative displacement between the track
3. The con®guration of each vehicle body is speci®ed and the bridge deck. The elastic eects of the rail pads
by ®ve degrees of freedom (DOFs): lateral movement Yi , and fasteners are neglected.
rolling hi , yawing Wi , ¯oating Zi and nodding ui ); and 2. The modal analysis is based on the bridge as a
that of each wheel by three DOFs: lateral movement whole, therefore, the vibration modes of the girder nodes
Ywij , rolling hwij and ¯oating Zwij . The total analytical are regarded as the modes of the bridge deck. The mode
DOFs are respectively twenty three for a six-axle loco- shapes between the nodes are obtained by a Lagrange
motive and seventeen for a four-axle passenger coach or interpolation.
a freight car. The idealization of the vehicle model for a 3. The masses of the vehicle wheels are much smaller
six-axle locomotive is shown in Fig. 1. than the mass of the bridge, therefore, their in¯uence is
When considering only the responses of the bridge as neglected.
a whole, such as the accelerations and the displacements 4. The cross-section deformation of the girder is ne-
of bridges induced by moving trains, this model can give glected, thus its movement at any section can be deter-
good results, when comparing the calculated results and mined by the lateral displacement Yb , rotation hb and
the measured ones. If the responses such as the vehicle vertical displacement Zb (Fig. 1). When the lowest Nq
acceleration, wheel derail factor and lateral thrusts are modes are considered, the displacements can be ex-
needed from the calculation, a more detailed model has pressed by the superposition of the modes as:
to be used. 8
The modal superposition technique has been used by >
> PNq
>
> Yb
x qn /nu
x
many researchers [2,13±19] to analyze the dynamic in- >
>
>
<
n1
teraction of vehicles and bridges. In this paper, a modal PNq
hb
x qn /nh
x
1
superposition analysis is also performed for modelling >
>
>
>
n1
the track±bridge system. First, the system is solved for >
> PNq
>
: Zb
x qn /nv
x
free vibration frequencies. By the orthogonality of the n1
modes, the FEM equations coupled with each other can
then be decoupled, which makes the bridge behaviour where /nuij
x, /nhij
x and /nvij
x are respectively the
become the superposition of independent modal equa- function values of the lateral, rotational and vertical
tions. Owing to the fact that the dynamic response of a components of the nth bridge mode at the position of the
jth wheel of the ith vehicle; qn is the nth generalized It is obvious from the equation that the movements
coordinate, namely, the nth modal amplitude. When the of the wheel-sets can be expressed by the linear com-
modes are mass normalized on the basis of /Tn m/n 1, position of the generalized bridge modal coordinates
the dynamic equation of the girder related to the nth qn
n 1; 2; . . . ; Nq . Therefore they need not be incor-
mode becomes: porated into the train±bridge system as independent
equations.
Nv X
X Nbi
qn 2nxq_ n x2 qn
/nuij Fuij /nhij Fhij /nvij Fvij
i1 j1 2.2. Dynamic equations of train±bridge system
2
The dynamic equilibrium equations of the vehicle±
in which Fuij , Fhij and Fvij are respectively the lateral bridge system can be obtained by the composition of the
force, moment and vertical force acting on bridge vehicle equation, the bridge equation and the wheel-
through the jth wheel of the ith vehicle. See Fig. 2 for an track relations:
illustration of forces acting on bridge. 2 38 9
Mi 0 0 0 0 > > Yi >
>
The dynamic coupling between the vehicle and the 6 0 Jhi 0 0 0 7 >
> > >
6 7 hi =
<
bridge is shown in Fig. 3, which illustrates the relation 6 0 0 Jwi 0 0 7
6 7 > wi >
between wheel-set movements Ywij , hwij , Zwij and girder 4 0 0 0 Mi 0 5> > > >
: Zi >
> ;
displacements Yb , hb and Zb . Obviously, they satisfy the 0 0 0 0 Jhi ui
2 38 9
following relations: Cui hCui 0 0 0 > Y_ i >
>
> >
8 9 8 9 6 hCui Chi 0 0 0 7 > _ > >
< Ywij = < Yb
xij Hi hb
xij Ys
xij = 6 7< hi =
6
6 0 0 Cwi 0 0 7 w _
7> i >
hwij hb
xij hs
xij 4 0 0 0 Cv 0 5>> _ > >
:
Zwij
; :
Zb
xij Bhb
xij Zs
xij
; : Zi >
> ;
0 0 0 0 Cui u_ i
8 n n 9 2 38 9
XNq < qn
/uij Hi /hij Ys
xij = Kui hKui 0 0 0 > Yi >
>
> >
qn /n hs
xij
3 6 hKui K 0 0 0 7 > hi >
< >
=
: q
/n hijB/n Z
x ; 6 hi 7
n1 n vij hij s ij 66 0 0 Kwi 0 0 77> w i >
4 0 0 0 Kvi 0 5> >
> Zi >
>
>
: ;
where Ys , hs and Zs are respectively the lateral, rotational 0 0 0 0 Kui ui
8 9
and vertical displacements of the wheel relative to the >
> /nuij Hi /nhij
kui qn cui q_ n >
>
>
> >
>
track, representing the in¯uences of wheel hunting >
> n 2
Nq < 0:5/hij a
kvi qn cvi q _ n Hi /uij
kui qn cui q_ n >
n >
XNbi X =
movements, lateral, rotational and vertical rail irregu- n
gj Sij /uij
kui qn cui q_ n
larities of the track on the bridge. j1 n1 >
> >
>
>
> /nvij
kvi qn cvi q_ n >
>
>
> n >
>
: gj Sij /vij
kvi qn cvi q_ n ;
8 9
>
> kui Ys
xij cui Y_ s
xij >
>
>
> _ >
>
>
< 0:5a2
k h
x c h
x H k Y
x c _
Y
x >
=
X
N bi v s vi s i ui s ui s
_
gj Sij kui Ys
xij cui Ys
xij
j1 >
> >
>
>
>
> kvi Zs
xij cvi Z_ s
xij >
>
>
: ;
gj Sij kvi Zs
xij cvi Z_ s
xij
X
Nv X
Nbi
/nm
vi kvi qm g f
/nuij Hi /nhij kui
Yi hi hi
i1 j1
3.2. AR model analysis of system excitation Dx 1:0 m, were well ®tted by a 2-AR(12) model. Then
the estimated autoregressive matrices from dierent data
A time series model, a high order vector AR model, records were calculated and were ®nally averaged to
was adopted for the analysis of system excitation, which acquire the statistical autoregressive matrices H1 ±H12 .
has distinct advantage compared with the FFT analysis By substituting them into Eq. (6), the 2-AR(12)
in the frequency domain [12]. model of the system excitation becomes
( )
According to the time series theory, an ordered dis- Ym1
crete set Y
x0 ; . . . ; Y
xi ; . . . ; Y
xn is called the discrete
Ym2
digital time series, which can be obtained by sampling ( ) ( )
the signal at a series of discrete time points x0 ; x1 ; . . . ; xn 2:68 0:23 Ym1 1 3:22 0:69 Ym1 2
(where x is the independent variable, and x0 < x1 < 0:02 2:65 1 Ym2 0:0 3:12 Ym2 2
< xi < < xn ). When Y is a vector, the set becomes ( ) ( )
2:87 1:09 Ym1 2:54 1:32 Ym1
a multivariate vector time series Yx Y1;x Y2;x 3
4
YN ;x T , which can be expressed by a vector AR model as 0:07 2:76 Ym2 3 0:07 2:47 Ym2 4
( 1 ) ( )
2:18 0:99 Ym 1:73 0:41 Ym1
Yx H1 Yx 1 HP Y x P ex
6 5
6
0:08 2:21 Ym2 5 0:07 1:99 Ym2 6
where Yx is the zero mean stationary random series of N ( 1 ) ( )
1:30 0:08 Ym 7 0:94 1:86 Ym1 8
variates, H1 ; . . . ; HP are autoregressive matrices of N
0:06 1:75 Ym2 0:08 1:52 Ym2
N orders, ex e1;x e2;x eP ;x T is an N-variate 7 8
( 1 ) ( )
white noise series that satis®es Eex 0, Eex eTs Q for 0:62 0:25 Ym 9 0:39 0:06 Ym1 10
x s, Q is the variance matrix and Eex eTs 0 for x 6 s. 0:11 1:26 Ym2 9 0:09 0:99 Ym2 10
Suppose a series of L P 1 observations with zero ( 1 ) ( 1 )
0:27 0:23 Ym 0:12 0:08 Ym 12
mean have been sampled from the wheel vibration re- 11
sponses at times t0 ; t1 ; . . . ; tLP 1 , and are to be ®tted by 0:03 0:66 Ym2 11 0:01 0:23 Ym2 12
( )
an N -AR
P model, the autoregressive matrix H can be e1m
estimated by least squares minimizing
e2m
9
Te trace
eeT trace
U HV
U HV T
The random excitation series groups can then be gen-
To minimize Te , we can dierentiate it with respect to H erated by continuously inputting random numbers with
and make the result equal to zero: white noise properties into the established AR model. At
each time t m, a group of random displacements
oTe Ym1 ; Ym2 T are determined by the white noise e1m ; e2m T at
0 UV T UV T 2HVV T 0
7
oH HH^ the same time and the displacements Ym1 k ; Ym2 k T ,
k 1; 2; . . . ; 12 at previous times. The frequency
^
In this way the discrete autoregressive matrix H properties of the random excitation series are deter-
2N 2P order) can be estimated as: mined by the parameter matrices of the AR model.
The power and phase spectra of the excitation are
^ H
H ^1 ^2
H ^ P
UV T
VV T
H
1
8 calculated and shown in Fig. 5. The frequency property
can be seen from the auto-power spectra in Fig. 5A: its
where: outstanding frequency is slightly higher than that of the
common track irregularities; the corresponding wave
U X
P X
P 1 X
P L ;
length is about 22 m, longer than the theoretical hunting
2 3 wave length. It is obvious that under the action of track
X
P 1 X
P 2 X
P 1 L irregularity, the principal frequency of the system exci-
6 X
P 2 X
P 1 X
P 2 L 7
6 7 tation becomes lower than the theoretical one.
V 6 .. 7
4 . ...
..
. 5 The phase spectrum indicates the phase relation be-
X
0 X
1 X
L tween the wheels on the front and the rear bogies, as
shown in Fig. 5B. The phase dierence at the out-
In order to obtain the interrelation of the system ran- standing frequency is about p=4. The random excita-
dom excitations (especially the phase relation between tions were generated by the 2-AR(12) model, one of
the wheels at the front and the rear bogies of a vehicle), a them is shown in Fig. 5C.
two-variate high order AR model was adopted in the Based on the analytical model and the generated
analysis. random excitations, the whole histories of trains running
The wheel acceleration data recorded at various train on the bridges were simulated on computer. In the cal-
speeds, normalized by sampling in a standard interval culation, each freight train consists of two locomotives
1856 H. Xia et al. / Computers and Structures 79 (2001) 1851±1860
and 10 freight cars with the speed range of 60±90 km/h, Twenty ®ve groups of random series were used for
and each passenger train consists of one locomotive and calculation of each train speed. By analysis on the peak
eight passenger cars with the speed range of 80±160 km/ amplitude of each history, the statistic parameters such
h. The main parameters of the locomotives, freight cars as the maximums, means and root-mean-square devia-
and passenger cars are given in Table 1. tions (RMSs) at every train speed were obtained for
Table 1
Main parameters of vehicles used in the case study
Parameter Unit Locomotive Freight car Passenger car
Full length of a vehicle m 23.10 16.44 26.60
Distance between two bogies m 11.90 11.50 18.00
Distance between two wheel-sets m 2.0 1.75 2.40
Sprung mass T 89.82 71.70 43.03
Unsprung mass t 6.03 2.04 5.25
Roll mass moment of car body t m2 192 138 345
Pitch mass moment of car body t m2 3160 1630 2540
Yaw mass moment of car body t m2 3160 1630 2540
Roll mass moment of wheel-set t m2 4.52 1.31 3.41
Vertical spring stiness per wheel kN/m 5930 21800 650
Lateral spring stiness per wheel kN/m 2325 16900 387
Vertical dashpot per wheel kN s/m 4920 678 100
Lateral dashpot per wheel kN s/m 262 678 78
H. Xia et al. / Computers and Structures 79 (2001) 1851±1860 1857
Table 3
Comparison of reinforcement results
No. Passenger train Freight train
Max. Ave. RMS Max. Ave. RMS
0 100% 100% 100% 100% 100% 100%
(6.59) (5.41) (0.73) (4.71) (3.22) (0.54)
1 50.9 44.2 74.0 64.7 69.1 76.1
2 51.7 42.8 80.7 64.0 67.2 77.8
3 50.6 42.4 79.0 61.3 63.9 74.2
4 50.3 42.3 80.2 60.9 63.4 74.3
5 52.7 42.1 79.3 60.6 63.0 73.7
6 53.5 42.3 79.2 60.5 62.9 74.1
Table 4
Measured results before and after reinforcement
Bridge Original Reinforced
Speed Amplitude Speed Amplitude
(km/h) (mm) (km/h) (mm)
Passenger
Max. 65.4 1.45 81.0 1.08
Ave. 55.0 0.95 65.2 0.72
RMS 9.03 0.29 10.5 0.14
Freight
Max. 58.0 3.88 72.0 2.33
Ave. 51.2 2.75 53.2 1.75
RMS 8.60 0.75 6.69 0.48
Table 5
Comparison of theoretical and measured results
Type of trains Passenger Freight
Calculated (mm) Original 5.41 3.22
Reinforced 2.28 2.03
R/O 42.1% 63.0%
Measured R/O 75.8% 63.6%
Fig. 11. Measured vibration curves of bridge at mid span.
1860 H. Xia et al. / Computers and Structures 79 (2001) 1851±1860
case of freight trains, the theoretical results are rather [3] Matsuura AA. Study of dynamic behaviors of bridge
close to the test ones. girders for high-speed railway. J JSCE 1976;256:35±47.
(2) There are a lot of undeterminable random factors [4] Yang YB, Yau JD. Vehicle-bridge interaction element for
which contributes to the lateral vibration of bridges, dynamic analysis. J Struct Engng ASCE 1997;123(11):
1512±8.
including wheel hunting, rail irregularities, initial con-
[5] Xia H, Xu YL, Chan THT. Dynamic interaction of long
ditions, member assembly errors, track eccentricity and suspension bridges with running trains. Int J Sound vibr
other unexpected factors. These factors, however, can 2000;237(2):263±80.
hardly be taken all into account in the theoretical model. [6] Fafard M, et al., Dynamics of bridge±vehicle interaction.
In consideration of the above-mentioned uncertain Proc EURODYN'93 Rotterdam: Balkema; 1993. p. 951±
factors, the reinforcement results are rather satisfactory. 60.
[7] Tanabe M, Yamada Y, Wakui H. Modal method for
interaction of train and bridge. J Comput Struct
6. Conclusions
1987;27:119±27.
[8] Sridharan N, Mallik AK. Numerical analysis of vibration
The following conclusions have been drawn from the of beams subjected to moving loads. J Sound and
above analysis: Vibration 1979;65:147±50.
(1) The analytical model for the spatial vibration [9] Yang YB, Lin BH. Vehicle±bridge interaction analysis by
system of the train and steel plate girders under the dynamic condensation method. J Struct Engng ASCE
random excitations and the corresponding computer 1987;121:1636±42.
simulation method can rather well re¯ect the main vi- [10] Wiriyachai A, Chu KH, Garg K. Bridge impact due to
bration characteristics of the train±bridge interaction sheel and track irregularities. J Engng Mech ASCE
system. The calculated bridge responses are in accor- 1982;108:648±67.
[11] Xia H, Chen YJ. Dynamic analysis of steel truss bridges
dance with those from the ®eld tests, when excluding the
under moving train loads. Adv Struct Engng Beijing: CRP
in¯uence of train speed dierence between them.
House; 1995. p. 48±54.
(2) The simulation of the random excitations of the [12] Xia H, De Roeck G. System identi®cation of mechanical
train±bridge system by the time series AR model is a structures by a high-order multivariate auto-regressive
suitable method. The statistical parameters of the bridge model. J Comput Struct 1997;64(1±4):341±51.
amplitudes in the whole train speed ranges instead of [13] Richardson HH, Wormley DN. Transportation vehicle/
any single value at certain train speed should be used to beam-elevated guideway dynamic interactions, a State-of-
evaluate the results of reinforcement schemes. the-art-Review. J Dynamic Systems Measurement and
(3) The reinforcement schemes proposed by this Control 1974;96:169±79.
paper behaved very well in reducing railway bridge [14] Ting EC, Genin J. Dynamics of bridge structures. Struct
Mech Arch 1980;5:217±52.
amplitudes. The theoretical analysis provides a sound
[15] Huang T. Vibrations of bridges. Shock and Vibration
basis for predicting the in¯uence of the reinforcement.
Digest 1976;8:61±76.
[16] Fryba L. The dynamic in¯uence of wheel ¯ats on railway
Acknowledgements bridges. Intl Railway Congr Assoc Bullet, 1967. p. 477±
512.
The research of this project is partly supported by the [17] Genin J, Ting EC, Vafa Z. Curved bridge response to
moving vehicle. J Sound and Vibration, 1982;81(4):469±75.
Natural Science Foundation of China (no. 50078001).
[18] Alessandrini L, Brancaleoni F, Petrangeli MF. On the
dynamic response of cable stayed bridges under moving
References loads. Costrzioni Metalliche, vol. 2. 1984.
[19] Bruno D. On the Dynamic Behavior of Long Span Cable-
[1] Diana G, Cheli F. Dynamic interaction of railway systems Stayed Bridges Under Moving Loads. Costrzioni Metalli-
with large bridges. Vehicle Syst Dynamics 1989;18:71±106. che, 2, 1985 [C2b].
[2] Fryba L. Vibration of solids and structures under moving [20] Klasztorny M. Vertical vibration of a multi-span bridge
loads. Groningen: Noordho International Publishing; under a train moving at high speed. Proc EURODYN'99
1972. Rotterdam: Balkema; 1999. p. 651±56.