Analysis of High-speed Train Related Ground

Vibrations by a Hybrid Method

TORBJRN EKEVID
NILS-ERIK WIBERG


Paper presented at IABSE Symposium "Structures for high-speed railway transportation", Antwerp,
Belgium, August 27-29, 2003


Department of Structural Engineering and Mechanics
Computational Mechanics
CHALMERS UNIVERSITY OF TECHNOLOGY
Gteborg, Sweden, 2003


Analysis of high-speed train related ground vibrations by a hybrid method

Torbjrn EKEVID
Ph. D.
Chalmers University of
Technology, Gteborg
SWEDEN

Torbjrn Ekevid, born 1973,
received his Ph. D. in Structural
Mechanics from Chalmers
University of Technology.

Nils-Erik WIBERG
Professor
Chalmers University of
Technology, Gteborg
SWEDEN

Nils-Erik Wiberg, born 1938,
received his Ph. D. in Structural
Mechanics from Chalmers
University of Technology.


Summary
On actual railway lines, particular large vibrations amplitudes have been noticed at sites with soft
clay deposits as high-speed trains have begun to operate on the lines. To understand the
phenomenon and to propose countermeasures to limit the effects of the propagating waves, detailed
dynamic analyses are necessary. A displacement based finite element method is not very suitable for
dynamic analysis of this type of problem since the method not fulfils the radiation condition. We
propose a hybrid approach combining conventional finite element techniques and the Scaled
Boundary Finite Element Method (SBFEM). A number of advantages can be identified; detailed
modelling of track components including nonlinear effects is possible; generated waves are
transmitted through the boundary and further the size of the model can substantially be reduced.
Keywords: Wave propagation, High-speed trains, Railway mechanics, Hybrid method.
1. Introduction
In Sweden, the research dealing with wave propagation phenomena related to high-speed trains was
initiated by field observations of extreme vibration amplitudes at Ledsgrd
[1]
south of Gteborg.
The phenomenon of high-speed train vibrations has later also been observed at other locations in
Sweden but also at locations in Holland and Germany where the high-speed trains operate at much
higher velocities. However, compared to the vibrations observed in Ledsgrd, the amplitudes at
those locations are comparably small.
Vibrations and propagating waves are generated by railway traffic even at low train speeds since
imperfections at the train wheel surface and at the steel rail cause variations in the contact forces
between the rail and the wheel, which in addition to the quasi-static deflection of the ground
induces vibrations (propagating waves) originating at the contact point. These high-frequent waves
but also the transmitted sound are normally what people recognise as disturbances from railway
traffic. However, as the speed of the train reaches or exceeds the critical wave velocity of the
compound track-ground structure, the response dramatically changes. Although no imperfections
are present the moving loads generate propagating shock waves. Theoretically, if no damping was
present in the ground and the materials were perfectly homogenous, there exist points where
displacements and stresses become infinitely large. However, since such conditions do not exist in
reality, the displacement and stress amplitudes are limited but high amplitudes could be expected in
comparison to normal conditions at subsonic speed. Solutions of an idealised model consisting of a
moving point load on elastic half-space have been studied by example Frba
[2]
. In Ekevid
[3]
a more
detailed problem description is given.
2. The hybrid method
The coupling between the conventional finite element domain and the unbounded domain

approximated by SBFEM
[4]-[5]
are based on dynamic equilibrium. In the program used for the
simulations, the approach based on Lagrange multipliers for handling the constraint equations has
been adopted. The approach gives a very clear and consistent implementation of both linear and
non-linear constraint equations. However, compared to the conventional approach where the
constraint equations are condensed in advance, the major drawback is the enlarged system of
equations. The dynamic equilibrium for the FE-domain could be expressed as:
( )
ext a u i
+ + + = F F F F R 0
(1)
= 0 (2)
where
ext
F are the external forces and , ,
a u
F F F are forces from inertia, internal and constraint
equations defined as
:
t
T
a
u ext
d
d d d

= =

+ = =

F u u F
u
F F u b u t
&&
% %
%
% % % %

(3)
Further,
i
R are interaction forces from the unbounded domain at the interface region, see Fig. 1,
obtained from
( ) ( )
i
t
i
i
0
t d d

=

R M u&&
%

(4)
where
i
is the surface representing the unbounded domain and

M the unit impulse response

matrix computed by the SBFEM. Moreover, represents a set of algebraic equations either from
constraint equations or boundary conditions.
x
z
y
Scaled Boundary FEM
FEM
i
R

Fig. 1. Interaction forces in the interface
region from the unbounded domain.
At the present stage, the well-established Newmark scheme has been used for the time stepping
procedure of the governing ordinary differential equations (1) subjected to the constraint equations
(2). By introducing approximations and corresponding test functions, linearisation of (1) and (2)
with respect to the primal unknowns
n 1 +
a
%
and
n 1 +

%
, corrections
i
a
%
,
i

%
are computed from
( ) ( )
( )
( )
*
i i T
i d
a
i i
d

| | (
| |
| ( =
|
|
(
\ .
\ .
R
a M

R
0
%
%

(5)
Here,
d
is the Jacobian of the constraint equations (2) and
( )
*
i
M is the equivalent mass matrix
defined as

( ) ( )
*
2
i i
t n
n 1
t
t
2

= + + M M M K (6)
where M and
( )
i
t
K are conventional mass and tangent stiffness matrices and
1

M is the first
discretizised unit impulse response matrix. The right-hand side of (5) represents the residual forces
and violation of the constraint equations according to.
( ) ( )
i
i
ext a u
a n 1 n 1 n 1 n 1

+ + + +
= + + R F F F F (7)
( )
i
i
n 1
2
n
2
t

+
=

R
(8)
As the corrections
i
a
%
and
i

%
have been computed accelerations, velocities, displacements and
Lagrange multipliers could be are updated from
;
;
i 1 i i i 1 i i
n 1 n 1 n 1 n 1
2
i 1 i i i 1 i i n
n 1 n 1 n n 1 n 1
t
t
2


+ +
+ + + +
+ +
+ + + +
= + = +

= + = +
a a a
v v a u u a
% % % % % %
% % % % % %

(9)
The equations (5) are for intermediate problem sizes solved by a direct solver. For larger problems
an iterative solver is necessary and in Ekevid
[6]
an efficient iterative solver based on a special
preconditioner of the indefinite system (5) and multigrid technique is proposed.
3. Numerical simulations of high-speed train induced vibrations
To study the propagating waves generated by high-speed trains, a model recovering the conditions
at Ledsgrd has been used. The result also demonstrates the performance of the proposed hybrid
method.
3.1 The computational model
The model of the railway section (total length 80 m), see Fig. 2, consists of rail, sleeper, ballast,
subballast, dry crust and the in-situ clay material.
X
Y
Z
B
A
C

Fig. 2. Model used for simulation of vibrations
caused byhigh-speed railway traffic.
Clay3
Clay2
Clay1
Dry crust
Clay2
Clay1
Dry crust B2
B1
Superelement


Fig. 3. Material layer definition.

Foundations and cantilever beams for the electrical power supply system have also been included to
the model. In order to reduce the size of the model, a symmetry plane is assumed along the track
section such that just one half has to be considered, see Fig. 2. The material in the model is assumed

to be elastic isotropic and the parameters have either been estimated from field measurements
conducted by J&W
[1]
, from laboratory tests made by NGI
[1]
or taken from Hall
[7]
. More advanced
constitutive laws could have been used for the track and its vicinity since finite elements are used to
model this part. However, the usage of advanced material models becomes meaningless if the
material models and the corresponding parameters are not estimated from material tests and for the
moment, no such tests have been conducted and evaluated for the Ledsgrd case. Fig. 3 shows the
material layers in the subground and the embankment.
F
1
F
3
F
2
F
3
F
3
F
3
F
3
F
3
F
4
F
5
15 18 7 18 7 18 7 10 7
3
(m)
F = 181 kN
1
F
3
= 122 kN
F
2
= 180 kN F
4
= 117 kN
F
5
= 160 kN

Fig. 4. Amplitudes and distances between
axle loads of a X2000 train.
To reduce the model to a reasonable size and avoid reflections at the boundary, SBFEM are used
along some of the boundaries. The SBFEM discretization is in the example restricted to 191 nodes
and 159 elements, see Fig. 5. Fig. 3shows the geometrical extension of the material layers for scaled
boundary discretization due to the position of the similarity centre. For further information
regarding the definition of similarity centre, see Ekevid
[8]
.
Fig. 5. Scaled boundary finite element
discretization.
Fig. 6. Conventional finite element
discretization.
The SBFEM discretization is treated as a superelement and can be reused repeatedly to reduce the
computational effort when examining the effect of different train speeds or modifications of the
track structure. The analysis procedure for constructing the SBFEM superelement requires a fixed
time step. As a consequence, adaptation of the time step size is not possible using the hybrid
method.
3.2 Result obtained in the analysis
The FE-discretization shown in Fig. 6 consists of 14584 nodes and 9654 elements; mostly 8-node
brick elements but also some beam elements. To couple the rail to the sleepers, the sleepers to the
ballast layer, the cantilever beams to the foundations etc., the facility to automatically generate

constraint equations in the FEM90 program by giving candidates for the coupling in node sets has
been used. The model involves approximately 50000 unknowns and the model is analysed for 3.5 s
where the time step size is kept fixed equal to 0.004 s. A direct solver has been used to solve the
system of equations. Two speeds, 50 and 70 m/s, of the train are considered in the analysis.
Fig. 7 Deformation of the railway section at
train speed 50 m/s.
Fig. 8 Deformation of the railway section at
train speed 70 m/s.
For the sub-critical speed (50 m/s) the deformation patterns are concentrated to the bogies and the
response of the model could be said to be more or less quasi static. For the trans-critical speed (70
m/s), the response changes and waves originating at the position of the bogies, forming a V-shape
behind the position of the bogie are propagating in the ground material. In Fig. 7 and Fig. 8,
snapshots of the deformation pattern (deformations are magnified 200 times) for each speed of the
train are available. The figures show that the responses at the two speeds are substantially different.
In order to make the response easier to interpret, the model and result are mirrored in the symmetry
plane and the positions of the bogies are visualized by means of arrows.
Time [s]
D
i
s
p
l
a
c
e
m
e
n
t
[
m
m
]
2
0
-2
-8
-6
-4
-10
-12
0 0.5 3.0 2.5 1.5 2.0 1.0 3.5

Fig. 9 Time history of the deflection of the
railhead at train speed 50 m/s.
D
i
s
p
l
a
c
e
m
e
n
t
[
m
m
]
Time [s]
0 0.5 3.0 2.5 1.5 2.0 1.0 3.5
2
0
-2
-8
-6
-4
-10
-12

Fig. 10 Time history of the deflection of the
railhead at train speed 70 m/s.
Fig. 9a and b show the time-history response of the railhead deflection in point C, see Fig. 2, at the
different train speeds. The vibration amplitudes (difference between the maximum and the
minimum value during the train passage) are approximately twice as large for the transcritical speed
(70 m/s) compared to the subcritical speed (50 m/s). The difference in vibration amplitude seems to
be even more pronounced if a comparison between the results in Fig. 7 and Fig. 8 is made.

However, it is likely the principal change of the response that makes propagating waves present at
the transcritical speed contributes to this impression. From Fig. 9 and Fig. 10 it is also possible to
notice the static deflection of the model, which indicates that the SBFEM works properly.
4. Discussion, Conclusions and Acknowledgements
Detailed modelling of railroad components and methods taking the unbounded media into account
are fundamental for accurate and reliable simulations of high-speed train induced vibrations. As
expected, the response close to critical speed is much higher than for subcritical speed. It should
also be noted that if the unsprung masses were taken into account, the response would increase to
large extent
[9]
.
The SBFEM in combination to conventional FEM results in an approach fulfilling these
requirements. In comparison to BEM, the SBFEM does not require fundamental solutions and since
the method is displacement-based, coupling to finite elements is straightforward. Since the method,
like BEM, is global in both space and time, convolutions and dense matrices are features included
in the solutions procedure. This makes the method expensive from a computational point of view
and limits the applicability. On the other hand the method can handle unbounded domains, which is
necessary when using FEM with artificial boundaries.
Financially the work has been supported by Banverket (Swedish National Rail Administration and
has been part of the work within CHARMEC; the national centre of excellence in railway
mechanics research.
5. References
[1] BANVERKET, High speed lines on soft ground: evaluation and analyses of measurements
from the West Coast line, Swedish National Rail Administration, Borlnge, 1999.
[2] FRBA L., Vibration of solids and structures under moving loads, Thomas Telford, London
1999, 494 pp.
[3] EKEVID T., On computational wave propagation in solids with emphasis on high-speed
train related ground vibrations, Lic. thesis, Department of Structural Mechanics, Chalmers
University of Technology, Gteborg, 2000, 99 pp.
[4] WOLF J.P., The scaled Boundary Finite Element Method, Wiley, Chichester 2003, 378 pp.
[5] SONG C. and WOLF J.P., The scaled boundary finite-element method-alias the infinitesimal
finite-element cell method- for elastodynamics, Comput. Methods Appl. Mech. Engrg., Vol.
147, 1997, pp 329-355.
[6] EKEVID T., Computational Solid Wave Propagation Numerical Techniques and Industrial
Applications, Ph. D. thesis, Department of Structural Mechanics, Chalmers University of
Technology, Gteborg, 2002, 170 pp.
[7] HALL L., Simulations and analyses of train-induced ground vibrations A comparative study
of two- and three dimensional calculations with actual measurements, Ph. D. thesis, Division
of Soil and Rock Mechanics, Royal Institute of Technology, Stockholm (2000).
[8] EKEVID T. and WIBERG N-E., Wave Propagation Related to High-Speed Train - a Scaled
Boundary FE-approach for Unbounded Domains, Comput. Methods Appl. Mech. Engrg, Vol.
191, No. 36, pp 3947-3964, 2002.
[9] M.X.D. LI, EKEVID T. and WIBERG N-E., An integrated vehicle-track-ground model for
investigating the wheel/rail dynamic forces due to high axle loads, To be presented at
CM2003, Gothenburg, Sweden, June 10-13, 6pp.