Dynamic analysis of railway bridges by means of the spectral method

Giuseppe Catania, Silvio Sorrentino

DIEM, Department of Mechanical Engineering,
University of Bologna, Viale del Risorgimento 2, 40136 Bologna, Italy

ABSTRACT

This study investigates the dynamic behaviour of railway bridges crossed by travelling trains. A simplified formulation was
adopted in order to perform a direct analysis of the effects of the parameters involved in the problem. The bridge is modelled
as a rectangular plate, while the trains are modelled as travelling inertial distributed loads. The formulation is accomplished
by the use of the Rayleigh-Ritz method, yielding a low order model with time-dependent coefficients. Several numerical
examples are presented and discussed, aimed at investigating the effects of each of the model-governing parameters.

Keywords

Railway bridges, moving loads, Rayleigh-Ritz method.

Nomenclature

C damping matrix u unit step distribution

D stiffness of the plate V potential of applied loads
E Young’s modulus v velocity of the load
g gravity acceleration w displacement
H velocity response function x spatial coordinate
h thickness of the plate y spatial coordinate
K stiffness matrix D velocity ratio
L length E frequency parameter
l width H relative error
M mass matrix ) modal matrix
n modal index P length ratio
p load per unit area Ȟ Poisson’s ratio
r mass ratio U mass per unit area
q modal coordinate Ȧn modal natural frequency
t time [ moving coordinate
U potential of strain energy ] damping factor

1 Introduction

In the analysis of dynamic effects of railway vehicles on bridges, simplified models are suggested and usually used, taking
into account only some aspects, such as deterministic, vertical effects, and the influence of moving forces and masses [1-2].
Inertial effects of both bridge and vehicle can be influential, and not negligible, since the mass of the external load introduces
a coupling effect between the load and the structure. Other important aspects, such as dynamic properties of travelling
vehicles and track irregularities, are not considered in the present study.
The railway bridge model most commonly used is a continuous Euler Bernoulli beam [2], or a Timoshenko beam [3],
traversed by either concentrated [4] or distributed moving loads [5]. Possible applications of lumped vibration absorbers have
also been investigated [6]. However, in the present study a homogeneous Kirchhoff plate is considered, allowing the analysis
of lateral vibrations due to trains travelling on double-track bridges. Structure damping is included in the model, as it may
play an important role.

T. Proulx (ed.), Civil Engineering Topics, Volume 4, Conference Proceedings of the Society for Experimental Mechanics Series 7, 21
DOI 10.1007/978-1-4419-9316-8_2, © The Society for Experimental Mechanics, Inc. 2011
22

The train is simply modelled by means of a continuous load in the form of a moving strip, an idealization which can be
adopted when the span of the bridge is large in comparison with the distance between axles [2]. A constant speed of motion
of vehicles along the bridge is assumed.
The formulation is accomplished by the use of the Rayleigh-Ritz method [7], and the solution is expressed in terms of a linear
combination of functions, which in the present study are selected as tensor products of eigenfunctions of prismatic
pinned-pinned and free-free beams in flexural vibration. This approach yields a reduced order model with time-dependent
coefficients, allowing a parametric analysis of plates loaded by travelling distributed masses.
Different example cases are presented and discussed in detail, analyzing the effects of velocity, mass and length of the train
on the plate’s dynamic response with respect to the mass, stiffness and damping of the plate itself.

2 Theoretical model

A homogeneous isotropic Kirchhoff plate is considered, simply supported on two opposite sides, free on the other two sides
and crossed by a travelling distributed load. The load per unit area p over the plate may be expressed as:

ª d 2 w( x, y, t ) º
p ( x, y , t )  Ut f ([ , y ) « g  », [ x  vxt , (1)
¬ dt 2 ¼

where w is the vertical displacement of a point of the plate or of the load (Fig.1),Ut is the equivalent mass per unit area of the
load, g is the gravity acceleration, vx is the travelling speed in the x direction, [ is a moving coordinate in the same direction
[1] and f models the translating strip representing the instantaneous position of the load:

f ([ , y ) >u([  Lt )  u([ )@>u( y  G )  u( y  G  lt )@ . (2)

Note that within the present studyUt is assumed to be constant; however, piecewise-constant or other distributionsUt([ ) may
be considered and adopted in the following developments.
Equation 2 contains the unit step distribution u(˜), Lt and lt are the length and the width of the strip modelling the train, and G
is the distance between the side of the strip and the edge y = 0 of the plate, as shown in Fig. 1. The second term on the
right-hand side of Eq. 1 describes the inertial action of the load. The total acceleration may be expressed in the following
general form:

d 2w w2 w w2 w w2w w2w w2w w2w ww ww

 v 2x 2  2v x v y  v 2y 2  2v x  2v y  ax  ay (3)
dt 2 wt 2
wx wxwy wy wxwt wywt wx wy

where vx, vy, ax, ay express the velocities and accelerations of the travelling load in the x and y directions respectively [1].
Considering a train travelling at constant speed v in the x direction, Eq. 3 reduces to:

d 2w w2 w w2 w 2 w w
2
 2v  v (4)
dt 2 wt 2 wxwt wx 2

The first term of the right-hand side of Eq. 4 expresses the influence of vertical acceleration of the moving load, the second
term the influence of Coriolis acceleration, and the third term the influence of track curvature [1].

y
Lb

Lt
vx lb
lt
G
0 x

Fig. 1 Model scheme

 23

The functional of the total potential energy of the coupled system can be written as the sum of a term U due to the strain
energy plus a term V representing the potential of all applied loads (including the inertial forces):

3 U V (5)

The potential of the strain energy can be written in terms of second order derivatives of the out-of-plane displacement w:

l L
1b b Eh3
2 ³0 ³0 ¬
U D ª wxx2  wyy2  2Q wxx wyy  2(1  Q ) wxy2 º¼ dxdy , D (6)
12(1  Q 2 )

where the subscripts denote differentiation with respect to the spatial variables and D is the flexural stiffness of the plate,
expressed as a function of Young’s modulus E, Poisson’s ratio Q and thickness h [7]. In the adopted formulation the inertial
forces are included in the potential of applied loads V as follows:

lb Lb

V ³ ³ ( U ww  wp) dxdy

0 0
b (7)

where Ub is the mass per unit area of the plate and p is the load in Eq. 1. The out-of-plane displacement w is expressed by
means of a linear combination of shape functions, selected as products of homogeneous uniform prismatic beam
eigenfunctions I:

N
w ¦q
n 1
n In ([ ,K ) Ÿ w I7 q (8)

where q is the generalized coordinate vector. Introducing the displacement expansion in the quadratic functional 3, and
imposing its stationarity, yields the following algebraic eigenproblem:

[M  r 'M ] q
  2r Įȕ['C] q  ȕ 2 [K  r Į 2 'K ] q  rg f (9)

with:

Ut v 1 D
r , Į , ȕ (10)
Ub Lbȕ L2b Ub

where E is a frequency parameter and D is a dimensionless parameter depending on the speed v. The matrices in square
brackets in Eq. 9 can be regarded as dimensionless quantities, and they can be computed according to the following integrals:

lb Lb į  lt x1 į  lt x1

³ ³ (I I ) dxdy , 'M ³ ³ (I I ) dxdy , 'C ³ ³ (I I

7 7 T
M Lb x ) dxdy
0 0 į x0 į x0
lb Lb

K L4b ³ ³ [Ixx IxxT  Iyy IyyT  Q (Ixx IyyT  Iyy IxxT )  2(1  Q )(Ixy IxyT )] dxdy (11)
0 0
į  lt x1 į  lt x1

'K 2 L2b ³ ³ (I Ixx ) dxdy , f ³ ³ I dxdy

T

į x0 į x0

In Eq. 11 the integration interval [x0, x1] is time-dependent.

Introducing the ratio between the lengths Lt and Lb:

Lt
ȝ (12)
Lb
then x0 and x1 vary according to Tab. 1.
24

Table 1 Time-dependent interval of integration

case ȝ  1 case ȝ 1 case ȝ ! 1

­ x1 vt ­ x1 vt ­ x1 vt
0 d vt  Lt Ÿ ® 0 d vt  Lb Ÿ ® 0 d vt  Lb Ÿ ®
¯ x0 0 ¯ x0 0 ¯ x0 0

­ x1 vt ­ x1 Lb ­ x1 Lb
Lt d vt  Lb Ÿ ® Lb d vt  2 Lb Ÿ ® Lb d vt  Lt Ÿ ®
¯ x0 vt  Lt ¯ x0 vt  Lb ¯ x0 0

­ x1 Lb ­ x1 Lb
Lb d vt  Lb  Lt Ÿ ® Lt d vt  Lt  Lb Ÿ ®
¯ x0 vt  Lt ¯ x0 vt  Lt

To model energy dissipation within the structure, a dimensionless damping matrix C may be defined by means of the plate
modal matrix ) (mass normalized) and eigenvalues Zn (computed from the M and K matrices), and considering a modal
damping ratio ] equal for all modes:

§Z · § Z ·
2ȗC 2ȗĭ T diag ¨ n ¸ ĭ1 ĭ T diag ¨ 2ȗ n ¸ ĭ1 (13)
© ȕ ¹ © ȕ ¹

Introducing Eq. 13 in Eq. 9 yields:

[ M  r 'M ] q
  2ȕ[ȗC  r Į 'C] q  ȕ 2 [K  r Į 2 'K ] q  rg f (14)

Equation 14 is a reduced order discretized model with time-dependent coefficients, which can be solved numerically.

3 Numerical results

Some numerical examples are presented for studying the dynamic behaviour of the model described in section 2. The
influence of parameters v, r, P, E, ] governing Eq. 14 is highlighted by studying time responses w(t) and dynamic response
functions H of the dimensionless frequency D (playing the role of ‘frequency response functions’) defined according to:

max t > w(t ) @

H ( x , y; D ) (15)
ws x , y ;D

where ws is the static deflection due to the load centered in Lb/2.

Numerical solutions of Eq. 14 are computed using the Runge-Kutta algorithm, expanding the solution w (Eq. 8) with 4 u 2
beam eigenfunctions (4 pinned – pinned eigenfunctions along x direction and 2 free – free eigenfunctions along y direction).
Realistic values for parameter E are computed by means of the empirical expression:

2S a
ȕ LbJ (16)
O12

based on large collections of experimental data [2], where a and J are parameters depending on the kind of bridge considered,
as reported in Tab. 2. The values (in Hz) of the first natural frequency and of parameter E for different kinds of bridges are
reported as functions of the length Lb in Fig.2.
 25

25 18
General bridges General bridges
Steel truss bridges 16 Steel truss bridges
20 Steel plate girder bridges with ballast 14 Concrete bridges with ballast
Concrete bridges with ballast
Concrete bridges without ballast 12
15

E [rad/s]
f1 [Hz]

10

8
10
6

5 4

0 0
10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100
Lb [m] Lb [m]

Fig. 2 First natural frequency f1 [Hz] (left) and frequency parameter E [rad/s] (right)
as functions of the length Lb for different kinds of bridges

Table 2 Parameters in Eq.16, as reported in [2]

Kind of bridge a J
General bridges (average case) 133 0.9
Steel truss bridges 307 1.1
Steel plate girder bridges with ballast 59 0.7
Steel plate girder bridges without ballast 208 1
Concrete bridges with ballast 190 1.1
Concrete bridges without ballast 225 1.2

Effect of the speed of the load

As a reference case study, the following values for the parameters are assumed:
Plate: Lb = 50 m, lb = 10 m, E = 5 rad/s, ] = 0.05.
Moving load: P = 1.4, lt = 2.5 m, G = 1.5 m, r = 0.5.
Time responses w(t) are computed at coordinate x = Lb/2, y = lb/2 with speed v varying from 30 m/s to 50 m/s (108 Km/h to
180 Km/h), as shown in figure 4.

Maximum deflection at different points

Parameter values are assumed as in the reference case. Response functions H(D) are computed at different points (x, y) along
the structure, as reported in Fig. 3. The frequency parameter varies from 0.1 up to 1, i.e. v varies from 40 Km/h up to 400
Km/h. Functions H(x, y; D) show a peculiar undulating trend, not significantly affected by the choice of coordinate x.

Effect of the mass of the load

Parameter values are assumed as in the reference case with v = 40 m/s = 144 Km/h, varying r from 0.1 to 1. Response
functions w(t) and H(D) are computed in x = Lb/2, y = lb/2, as reported in Fig. 4.

Effect of the length of the load

Parameter values are assumed as in the reference case with v = 40 m/s = 144 Km/h, varying P from 0.1 to 2. Response
functions w(t) and H(D) are computed in x = Lb/2, y = lb/2, as reported in Fig. 5. Parameter P (related to Lt) is able to
significantly affect the behaviour of H(D). Note that the plots of H(D) in the case P t 1 are superimposed.
26

-3
x 10
0.5 1.1
x = Lb/2; y = lb/2
0
x = Lb/4; y = lb/2
1.08
-0.5 x = 3Lb/4; y = lb/2
x = Lb/2; y = 0
-1 1.06 x = Lb/2; y = lb
w [m]

-1.5

H
-2 1.04

-2.5
v = 40 m/s 1.02
-3 v = 50 m/s
v = 30 m/s
-3.5 1
0 1 2 3 4 5 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t [s] D

Fig. 3 Effect of parameter v on w(t) (left); H(D) at different points (x, y) (right)
-3
x 10
1 1.15
r = 1.0
0 r = 0.8
r = 0.5
-1 r = 0.1
1.1
-2
w [m]

-3
H

-4
1.05
-5
r = 0.5
r = 1.0
-6
r = 0.1
-7 1
0 1 2 3 4 5 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t [s] D

Fig. 4 Effect of parameter r on w(t) (left); effect of parameter r on H(D) (right)

-3
x 10
0.5 1.45
P = 2.0
0 1.4 P = 1.0
P = 0.5
1.35
-0.5 P = 0.1
1.3
-1
1.25
w [m]

-1.5
H

1.2
-2
P = 0.2 1.15
P = 0.6
-2.5 1.1
P = 1.0
-3 P = 1.4
1.05
P = 1.8
-3.5 1
0 1 2 3 4 5 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t [m] D

Fig. 5 Effect of parameter P on w(t) (left); effect of parameter P on H(D) (right)

 27

1.1 1.1
E=5
0.5
rad/s ]=0

E=3
0.5
rad/s ] = 0.01
1.08 1.08
E=8
0.5
rad/s ] = 0.05
] = 0.10
1.06 1.06 ] = 0.20
] = 1.00
H

H
1.04 1.04

1.02 1.02

1 1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
D D

Fig. 6 Effect of parameter E on H(D) (left); effect of parameter ] on H(D) (right)

-3
x 10
0.5 1.35
F = 1/2
1.3 F = 1/6
0
F = 1/48
1.25 F = 1/480
-0.5
1.2
w [m]

-1
H

1.15

-1.5
F = 1/480 1.1
F = 1/48
-2 F = 1/6 1.05
F = 1/2
-2.5 1
0 0.5 1 1.5 2 2.5 3 3.5 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t [s] D

Fig. 7 Effect of partially distributed load on w(t) (left); effect of partially distributed load on on H(D) (right)

Effect of structural stiffness

Parameter values are assumed as in the reference case, varying E from 3 rad/s to 8 rad/s. Response functions H(D) are
computed in x = Lb/2, y = lb/2, as reported in Fig. 6. The plots are almost superimposed.

Effect of structure damping

Parameter values are assumed as in the reference case, varying ] from 0 to 1. Response function H(D) is computed in x = Lb/2, y =
lb/2, as reported in Fig. 6. Raising ] reduces the amplitude of oscillation of H(D), until its behaviour becomes monotonic (however
this is not the case for real bridge structures).

Effect of partially distributed load

Parameter values are assumed as in the reference case, with Lt = 24 m and v = 40 m/s = 144 Km/h. Different loading
distributions are compared: the continuous one (as represented in Fig. 1) and ‘partial’ distributions consisting of two shorter
sections in which the load is distributed.
The assumed ‘partial’ distributions are given by:
Lt § L ·
0d[ d and ¨ Lt  t ¸ d [ d Lt (17)
F © F¹
28

with F  0.5 (F = 0.5 yields the ‘continuous’ distribution). Since for the continuously distributed load it is assumed r0 = 0.5,
for the ‘partially distributed’ load described by Eq. 17 r0 increases to r = 1/(2F) u r0. Response functions w(t) and H(D) are
computed in x = Lb/2, y = lb/2 for different values of F (1/6, 1/48, 1/480) as reported in Fig. 7. Load distribution variations
such as that described in Eq. 17 may dramatically affect the behaviour of the response function H(D).

Effect of time dependent matrices

Parameter values are assumed as in the reference case, with v = 40 m/s = 144 Km/h. The effect of neglecting the time
dependent matrices 'M, 'C, 'K on the solution w(t) is evaluated by introducing a relative error, according to:
w(t )  > w(t ) @'M
HM 0
(18)
max t w(t )

where [w(t)]'M = 0 refers to the solution computed assuming 'M = 0 in Eq. 14.
Similarly, HC and HK can be defined, considering 'C = 0 and 'K = 0.
-3
x 10
6 1.1
' M = 'C = 'K = 0
4 Exact
1.08 'C = 'K = 0
2 'C = 0

0 1.06

-2
H
H

1.04
-4
HC
-6 HK
1.02
HM
-8
H Tot
-10 1
0 0.5 1 1.5 2 2.5 3 3.5 4 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t [s] D

Fig. 8 Effect of neglecting time dependent matrices 'M, 'C, 'K on w(t) (left) and on H(D) (right)

The error functions HM, HC and HK are plotted versus time in Fig. 8, where HTot represents the total error, assuming 'M, 'C and
'K equal to 0 at the same time. The smallest, and negligible contribution to the error HTot appears to be HC, while the main
contribution is due to HM. Fig. 8 also shows the effect on H(D) of neglecting the time dependent matrices 'M, 'C, 'K.
Again, the effect of neglecting 'C is very small. The total error, on the contrary, can be significant.

4 Discussion

Function H(x, y; D) appears to be an effective tool for studying the dynamic behaviour of a structure crossed by travelling
loads with constant speed, in some way equivalent to a ‘frequency response function’ for time-varying coefficient systems.
This function shows peculiar undulating trends (Fig. 3), influenced by the parameters governing Eq. 14. The response can be
evaluated at any coordinate point (x, y) of the plate, making it possible to study the variation of structural deflection also
along the y coordinate (Fig. 3). Mass parameter r can produce important shifts in magnitude, but not in shape (Fig. 4). On the
contrary, length parameter P controls both shape and magnitude of H(D) (Fig. 5), but only in the case 0 < P d 1. The damping
parameter ] has the effect of progressively smoothing the oscillation of H(D), until it becomes monotonic (Fig. 6, though the
latter limit case is not realistic for actual bridge structures): in general, the reduction in amplitude becomes particularly
significant at high speed. Frequency parameter E, within the range of real bridges, scarcely affects the behaviour of H(D)
(Fig. 6), so H may be considered independent from E. Changes in the spatial distribution of the load can produce dramatic
variations in H(D) (Fig. 7): this result should highlight the importance of properly modelling the ballast, directly influencing
the load distribution on the actual structure.
The contribution to the solution of the time dependent matrices 'M, 'C and 'K is globally not negligible (Fig. 8), however
the effect of 'C is usually very small in comparison with the contributions of 'K, and especially of 'M.
 29

5 Conclusions and future work

In this study, the dynamical behaviour of railway bridges crossed by travelling trains was investigated by adopting a
simplified model, i.e. a plate loaded by a travelling distributed mass, solved by means of the Rayleigh-Ritz method.
The effects of each of the model governing parameters was studied introducing a dynamic function of the travelling speed,
equivalent to a ‘frequency response function’ for time-varying coefficient systems. This function can be an effective tool for
studying the dynamic behaviour of a structure crossed by travelling loads, since the travelling speed is the most important
parameter influencing the dynamic stresses in railway bridges, which in general increase with increasing speed.
In particular, it was shown how different spatial distributions of the load can deeply influence the dynamic response of the
structure, highlighting the importance of properly modelling the ballast. Future work will thus concern this significant
problem.

Acknowledgments

This study was developed within the INTERMECH laboratory with the contribution of the Regione Emilia Romagna -
Assessorato Attività Produttive, Sviluppo Economico, Piano telematico, PRRIITT misura 3.4 azione A Obiettivo 2.

References

[1] Fryba L., Vibration of Solids and Structures under Moving Loads, 3rd edition, Telford, 1999.
[2] Fryba L., Dynamics of Railway bridges, Telford, 1996.
[3] Lin Y.H., “Vibration analysis of Timoshenko beams traversed by moving loads”, Journal of Marine Science and
Technology 2 (4), pp. 25-35, 1994.
[4] Stancioiu D., Ouyang H., Mottershead J.E., “Vibration of a continuous beam excited by a moving mass and
experimental validation”, Journal of Physics, Conference series 181, 1999.
[5] Adetunde I.A., “Dynamical Behavior of Euler-Bernoulli Beam Traversed by Uniform Partially Distributed Moving
Masses”, Reasearch Journal of Applied Sciences 2 (4), pp. 476-483, 2007.
[6] Lin Y.H., Cho C.H., “Vibration suppression of beam structures traversed by multiple moving loads using a damped
absorber”, Journal of Marine Science and Technology 1 (1), pp. 39-48, 1993.
[7] Timoshenko S., Young D.H., Weaver W., Vibration problems in engineering, 4th edition, Wiley, 1974.
http://www.springer.com/978-1-4419-9315-1