Sunteți pe pagina 1din 181

Three-dimensional Dynamic Models of a Railway

Track for High-speed Trains

Three-dimensional Dynamic Models of a Railway


Track for High-speed Trains

PROEFSCHRIFT

ter verkrijging van de graad van doctor


aan de Technische Universiteit Delft,
op gezag van Rector Magnificus prof. dr. ir. J.T. Fokkema,
door het College voor Promoties aangewezen,
op ______dag ___________ 2002 te _____ uur

door
Alexei Viktorovich Vostroukhov
Master of Science (Physics of Radiowaves), The Nizhny Novgorod State University
Geboren te Pavlovo, Russia

Dit proefschrift is goedgekeurd door de promotor:


Prof. Ir. A.C.W.M. Vrouwenvelder

Samenstelling promotiecommissie:
Rector Magnificus,
Prof. ir. A.C.W.M. Vrouwenvelder,
Prof. Ph.D. C.Eng. M.P. Cartmell
Prof. dr. ir. G. Degrande
Prof. dr. ir. C. Esveld,
Prof. Dr. Sc. L. Frba
Dr. Sc. A.V. Metrikine
Prof. dr. ir. A. Verruijt,

Voorzitter
Technische Universiteit Delft, promotor
University of Glasgow, Scotland, UK
Katholieke Universiteit Leuven, Belgium
Technische Universiteit Delft
Institute of Theoretical and Applied
Mechanics, Prague, Czech Republic
Technische Universiteit Delft
Technische Universiteit Delft

Acknowledgements

Contents
Introduction 1
Overview of models for a railway track .. 1
Objectives and Scope .. 7
Outline . 9
Chapter 1. Steady-State Response of Two- and Three-Dimensional Elastic
Systems to a Uniformly Moving Load . 11
1.1. Response of an elastically supported membrane 11
Subcritical motion V < c .13
Supercritical motion V > c ...14
Elastic drag for a single load .17
Elastic drag in the undamped case 18
Elastic drag in the damped case 19
1.2. Response of an elastic half-space to a normal constant load
uniformly moving over its surface . 21
1.3. Response of a visco-elastic half-space to a normal constant load
uniformly moving over its surface . 33
Chapter 2. Beam on a Visco-elastic Half-space as a Simple Three-Dimensional
Model for a Railway Track .. 39
2.1. Equivalent dynamic stiffness of a visco-elastic half-space 40
2.2. Steady-state response of the beam on the half-space to a moving load . 50
2.3. Elastic drag experienced by a high-speed train due to excitation
of ground vibrations ... 55
Chapter 3. Beam on a Visco-elastic Stratified Half-space as a Continual Model
for a Railway Track . 62
3.1. A three-dimensional continual model for a railway track . 62
3.2. The method of the dynamic flexibility matrix ... 65
The equivalent stiffness of a layered half-space . 72
Equivalent stiffness of a visco-elastic layer fixed at its bottom . 73
Equivalent stiffness of a layer on a half-space ... 77
3.3. Steady-state response of a beam on a layered half-space to
a moving load . 81
Layer fixed at its bottom 82
Layer on half-space 86
Chapter 4. Periodically Supported Beam on a Visco-elastic Layer
as a Model for Conventional Railway Track . 93
4.1. Equivalent dynamic stiffness of a visco-elastic layer
at the sleeper-layer interface ... 94
4.2. Steady-state response of the beam to a set of moving loads .111
4.3. Elastic drag experienced by a high-speed train due to excitation
of ground vibrations ...119
4.4. Comparison of responses of 3D and 1D models of the conventional
track 123

Chapter 5. Locomotive Detection of Derailment of a Wagon of a Freight Train:


Theory and Experiment .. 125
5.1. Steady-state response of a railway track to axle loading.
Normal regime ... 126
The amplitude spectrum of vibrations in the observation point 131
5.2. Steady-state response of a railway track to axle loading
in the emergency case. Comparison with normal regime .. 133
Comparison of the rail vibrations in the normal and
emergency cases 136
5.3. Effect of multi-axle loading on the track response .. 139
5.4. Comparison of experimental results to theoretical predictions 141
Experimental set-up .. 142
Measurement set-up .. 144
Data processing . 146
Conclusions .. 152
Main results ... 152
Discussion of the results 153
Future perspectives in modelling dynamics of the railway track ..154
References 156
Summary . 169
Samenvatting ... 172
.. 173
Curriculum Vitae 174

INTRODUCTION.
Modern means of railway transportation in Western Europe are being currently
developed to further reduce the travelling time for passengers. Cruise velocities of such
high-speed trains as French TGV, German ICE, Swedish X-2000, etc. are nowadays in
the range of 200-300 km/h and increase continuously. This velocity increase brought a
new problem to railway engineering, namely the problem of significant amplification of
the train-track vibrations at high train speeds. This amplification occurs when a train
moves with a velocity that is close to the Rayleigh wave velocity in the subsoil of a
railway track, which varies from 150 to 800 km/h, depending on the soil type.
Obviously, modern high-speed trains can easily reach the lower threshold.
There exist quite a few papers that report observed and measured amplification
of the railway track vibrations as the train speed reaches the Rayleigh wave velocity in
the ground. Among the recent ones, one can mention papers of Esveld [34] (1997),
Kaynia et al. [68] (2000) and Degrande & Shillemans [26] (2001), which report that the
amplification was measured in The Netherlands, Sweden and Belgium, respectively.
The amplification of the train and track vibrations at high-speed is a threatening
phenomenon, which may lead to a rapid deterioration of the track structure and might
cause derailment of a train, see Fig. I.

Fig. I. Derailment of EUROSTAR 9047 at 290 km/h near Croisilles


(south of Arras, France) due to anomalous vibrations, 5 June 2001.

Thus, railway tracks should be designed in such a way as to prevent the


amplification from happening. To enable this design, a dynamic, three-dimensional
model is to be developed that takes into account the track-subsoil interaction.
Development of such a model is the main objective of this thesis.

Overview of models for a railway track.


Until the beginning of 90th, researchers used mainly 1D models for a railway
track. The first models were developed by Winkler in 1867 [168] and by Timoshenko in
1915 [151]. Winkler has modelled the track as an elastically supported Euler-Bernoulli
1

beam and studied its static response. Timoshenko extended this study to the case of the
dynamic loading.
Various Winkler-type models have been investigated by great number of
authors. One ought to emphasize the papers of Criner [24], Kenney [69], Mathews
[108], Achenbach [2], Kerr [70], Choros and Adams [20].
Criner [24], in 1953, considered an Euler-Bernoulli beam resting on linear and
non-linear foundation and subjected to a uniformly moving and accelerating load. In his
paper, one of the first electric-analog-computer techniques for calculation of the beam
response has been developed and presented. Thus, the beam response has been
numerically examined for the first time.
Later, in 1954, Kenney [69] found an analytical solution for the steady-state
response of an Euler-Bernoulli beam on Kelvin foundation to a uniformly moving load
and investigated the effect of damping on wave propagation and critical velocities. He
showed that there is a critical damping that removes the amplification of the beam
response at critical velocities.
In 1958, Mathews [108] considered the steady-state response of an EulerBernoulli beam on Winkler foundation to a moving, harmonically varying in time load.
He showed that there exist five qualitatively different solutions in the speed-frequency
plane of the load parameters. In 1989, this model has been extended by Bogacz et al.
[15], who examined a Timoshenko beam instead of the Euler-Bernoulli one. In [15] it
has been shown that for the Timoshenko beam there may be 22 qualitatively different
solutions.
In 1965, Achenbach and Sun [2] first investigated the dynamic response of
Timoshenko beam laying on the Kelvin foundation to a constant, uniformly moving
load. The solution for the undamped case resulted in three critical velocities at which
the displacement grows beyond bounds. The first critical velocity is smaller than the
shear wave velocity in the beam and equals to the minimum phase velocity of waves
propagating in the beam. The second critical velocity is related to the shear wave
velocity in the beam, whereas the third one equals to the largest possible group velocity
of waves in the beam. In this paper, the effect of damping in the beams foundation has
been investigated as well.
In 1972, Kerr [70] studied the effect of an axial force that can arise in a
continuously welded rail due to the temperature extension. For this purpose he
considered an axially compressed Euler-Bernoulli beam subjected to a moving
concentrated load. He found out that the critical velocity of the load tends to zero as the
axial stress approaches the buckling magnitude.
In 1979, Choros and Adams [20] considered the dynamic response of an EulerBernoulli beam resting on a tensionless Winkler foundation (non-linear problem) to a
uniformly moving load. The method for finding the contact and non-contact regions of
the beam and the foundation has been developed and presented by the authors.
In 1970 A.P. Filippov [37] and then, in 1972, L. Frba [39] published their
monographs devoted to the dynamic analysis of railway models known by then. The
book of the latter author has been supplemented and undergone the third edition in 1999
[40].
It should be also mentioned that in 1985-1988 Krysov, Holuev and Bychenkov
[17]-[18], [88]-[90] published a series of papers devoted to a new problem related to the
train motion at high velocities. They investigated a question concerning the wave drag
experienced by a rigid wheel rolling over a railway track and showed that this drag can
rapidly increase when the load velocity is close to the critical velocity. They studied this

question using a 1D model for railway track, namely an Euler-Bernoulli beam resting
on Winkler and Kelvin foundations.
In early 80th, researchers started to account for periodic character of the railway
structure that is introduced by the sleepers. Jezequel [64] was the first to publish in 1981
a paper in which the steady-state dynamic response of an Euler-Bernoulli beam laying
on periodically located supports to a uniformly moving force has been considered. A
technique based on application of the Fourier series has been employed in this paper.
The paper of Jezequel [64] was followed by a number of papers among which
the papers of Popp and Muller [131], 1982; Vestnitskii and Metrikine [158], 1993,
Bogacz, et. al. [16], 1993; Belotserkovskiy [10], 1996, etc. In these papers, the steadystate dynamic response of a periodically inhomogeneous structure was considered by
employing the Floquet theorem or related periodicity condition.
In 1997, Krzyzynski and Popp [84] published a paper in which a twodimensional problem has been stated and studied. The authors considered the dynamic
behaviour of two Timoshenko beams mounted to a visco-elastic foundation by
periodically spaced sleeper. The beam was excited by two harmonically varying loads,
moving uniformly along the right and left rails. In this paper, a study was carried
out of a general case in which the loads move not necessarily in-phase. It was found
out that there are two modes of wave propagation in such a two-dimensional periodic
structure. The first mode corresponds to the in-phase propagation of waves in the rails,
whereas the second one represents the case of the half-wave-length phase difference
between the waves. Recently, in 1998 an overview on dynamics of periodic railway
structures has been published in [82] by Kruse, Popp and Krzyzynski.
It is worth mentioning here a paper of Nordborg [128], 1999, in which he
considered a 1D periodical model with track-support irregularities to model a lowfrequency response of the track to a slowly moving load. The results of the theoretical
study, being compared to laboratory experiments and onboard ICE train measurements
on ballast and slab track showed a good resemblance with the measurements.
All papers that were mentioned above dealt with linear models of infinite length
and were focused on the steady-state response. To model a structure of a finite length, a
structure with significant non-linearities or to calculate the transient response of a
structure, it is preferable to use the finite element methods (FEM) or finite difference
methods (FDM). Discussion on these methods is beyond the scope of this thesis. We
ought to mention that in 1999, an overview of known 1D and 2D railway models was
published by K. Knothe [78].
One may call the beginning of 90th years as an origin of intensive investigations
of 3D models for railway track. The basic reason for developing such models has been a
rapid introduction of high-speed trains and inapplicability of 1D models at train speeds
that are comparable to the wave speeds in the ground. One should mention, however,
that there was a number of pioneering works before the beginning of 90th, in which
some 3D-models for a railway track were considered.
First 3D modelling of a railway track was accomplished in 1961 by Filippov
[36]. He considered the steady-state response of an Euler-Bernoulli beam resting on an
elastic half-space to a uniformly moving over the beam load and showed that the
vertical deflection of the beam becomes infinite if the load velocity is equal to the
Rayleigh wave speed.
In 1975, Labra [93] extended the model of Filippov. He combined the Filippovs
model with that of an axially compressed track proposed by Kerr [70]. Labra found out
that the axial stress decreases the critical velocity in the 3D model at a much smaller
rate than it does in the related 1D model.

In 1995, Krylov [86] published a paper in which he proposed a model for


analysis of the level of track vibration generated by a high-speed train. Representing the
ground by elastic half-space, he used an approximate expression for the Greens
function of elastic half-space that takes into account only the Rayleigh wave
contribution. The train loading has been modelled by a set of concentrated loads moving
with a constant velocity and applied to the half-space at discrete points, which are
defined by the sleeper span. The deflection of the track under applied vertical axle
forces have been calculated in quasi-static approximation using a simple model of the
track as an Euler-Bernoulli beam laying on Winkler foundation. The author has studied
spectrum of ground vibration and shown that the level of vibration excited by a very fast
train, whose velocity exceeds the Rayleigh wave velocity, is huge with respect to the
vibration level for conventional trains (1000-2000 times larger). In 1998 [87], Krylov
applied this method to consider the ground vibrations generated by an accelerating train.
In 1996, Dieterman and Metrikine [29] investigated a so-called equivalent
stiffness of a half-space interacting with a beam. They extended the model of Filippov
by considering sub-seismic, trans-seismic and super-seismic speed ranges of the train.
In [29] it has been shown that the equivalent stiffness is a complex function of
frequency and wave number of waves in the beam. Authors also found out that there
exist two critical velocities for a uniform motion of a constant load. The first one equals
to the Rayleigh wave speed (as it has been found in [36]) and the second one is slightly
smaller than this speed.
Suiker et al. 1998 [139] considered the response of a Timoshenko beam lying on
a half-plane to a moving load. The authors used a finite element technique to obtain the
solution.
In 1998, Lieb and Sudret [101] considered the response of an Euler-Bernoulli
beam laying on a visco-elastic half-space to a harmonic load by using the Wavelet
technique. The results have shown a very good agreement with results obtained by
Metrikine and Dieterman in 1996 [29].
In 1999, Popp et al. [130] enlightened problem of theoretical description of
vehicle-track dynamic modelling in mid-frequency range. They discussed various
possibilities and difficulties that researchers can face creating algorithms for calculation
of dynamics of train-track models.
In 1999, Metrikine and Popp [116] published a paper, in which the first
analytical study of a 3D, periodically inhomogeneous model for a railway track was
presented. The authors considered an Euler-Bernoulli beam mounted to elastic halfspace by periodically spaced sleepers. The main attention has been paid to evaluation of
the equivalent stiffness of the half-space against the sleepers. It has been shown that the
original 3D problem can be reduced to a 1D model, by replacing the half-space by a
system of identical springs placed under the sleepers. The stiffness of these springs is
then a complex function of the frequency and phase shift between vibrations of
neighbouring sleepers. The authors showed that the Rayleigh wave speed plays a crucial
role in dynamics of the model. No computations on the beam response to the load have
been presented in the paper.
In 2001, Kruse and Popp [83] proposed a general modular algorithm for
investigation of train-track models. The main restrictions of the models are that they are
to be linear and only the steady-state solution (uniform motion) can be studied.
It should be noticed that in all above-mentioned 3D models of the track
homogeneous elastic half-space (or half-plane) was used as a model for the soil. In
practice, the ground is inhomogeneous over its depth. The first extension for 3D models
of the ground originates from 1981, when Kausel and Rosset [67] first proposed to use

the method of stiffness matrix for layered soil developed by Thompson 1950 [150] and
Haskel 1953 [60]. The solution technique is based on the application of Fourier and
Hankel transforms to wave equations in each layer to reduce them to a system of
ordinary differential equations. But this method at first time has been used mainly to
discuss surface wave modes, to fit measured dispersion curves (Rosset et al. 1991
[135]), and especially to solve problems of the soil-structure interaction.
The first attempt to analyse a wave field in a horizontally layered soil was
presented by Auersch 1994 [6]. Auersch investigated and compared the response to a
point harmonic load of a homogeneous half-space to that of a soft and stiff layer
overlying a half-space and to that of a soil with continuously varying with depth
stiffness and damping. He studied the dependence of the displacement field on the load
frequency and found out that this amplitude strongly depends upon the soil structure.
In 1997, Dieterman and Metrikine [30] studied the steady-state response of an
elastic layer fixed at its bottom to a moving harmonically varying point load applied at
the layer surface. The authors have determined the critical velocities of the load as a
function of the load frequency. The critical layer depths have been derived for subseismic, trans-seismic and super-seismic ranges. In 2000, Metrikine and Popp [117]
extended this model for a railway track modelling it by an Euler-Bernoulli beam resting
on a visco-elastic layer. They showed that decrease of the layer depth results in increase
of the critical velocity and decrease of the vibration level.
In 1999, Sheng et al. [136]-[137] published a couple of papers devoted to the
ground vibration generated by a fixed-in-position load and a moving along a railway
track harmonic load. They studied the dynamic response of a sandwich beam-structure
resting on a layered half-space to the afore-mentioned loads. The investigation showed
that the dynamic response of the track depends on the subgrade structure crucially.
Grundmann et al. 1999 [58] considered the response of a half-space and a
layered half-space to a set of localized loads (train) with the help of technique of
Wavelet transform. They investigated the static and dynamic parts of the ground
response and showed the crucial role of the soil stratification.
In 2000 Kaynia et al [68] presented results of measurements carried out by
Swedish Geotechnical Institute at a site of the West Coast Line between Gteborg and
Malm where subgrade soil is extremely soft. The test runs were performed using a
Sweden X-2000 passenger train composed of locomotive and four cars. The
experiments showed that significant amplification occurs of the railway track vibrations
at velocities close to the Rayleigh wave speed, which turned out to be very low in this
part of the track about 40 m s . The authors proposed a method for simulation of
ground vibration from high-speed train based on the Kausel-Rosset Greens function
[67]. They modelled the ground as a layered visco-elastic half-space and the track as an
Euler-Bernoulli beam. Forces, uniformly moving at fixed distances from each other,
modelled the bogies of the train. Authors found out a very good agreement between the
measurements and their computational simulations. On the basis of the developed
model they studied the effect of stiffer embankment (beam) and showed that increasing
of the embankments bending stiffness results in substantial reduction of the vibration
level.
In 2001, Degrande and Schillemans 2001 [26] published results of experimental
investigations of the ground vibrations induced by Thalys high-speed train at a speed
varying between 223km h and 314 km h on the track between Brussels and Paris.
They studied the dynamic soil and track characteristics. According to their
measurements the Rayleigh wave speed in the subgrade can be estimated as 80 m s . But

unlike theoretical expectations the authors have not found large increase of the track
vibrations at high velocities.
Finalizing a literature overview of 3D models for a railway track and the
methods for their analysis, we should mention one method more, which, although
beyond the scope of this thesis, is being widely used nowadays.
Thanks to rapid development of computational technologies in 90th it became
possible to use the boundary element method (BEM) and hybrid BE-FE methods in
investigation of the dynamic response of the railway track. The BEM for the ground
response, applicable for earthquake engineering problems, was developed in 70th.
The first attempts to describe a 3D ground by technique of BEM may be related
to report of Dominguez 1978 [32] and paper of Karabalis and Beskos 1984 [65]. The
former author was the first to use the BEM to obtain the dynamic stiffness of the ground
in the frequency domain. The latter authors proposed a time domain BEM that allowed
evaluating the dynamic response of 3D foundation of an arbitrary shape. They indicated
that the proposed method could be a basis for extensions to non-linear soil-structure
interaction problems.
Gaitanaros and Karabalis 1988 [47], and Auersch and Schmid 1990 [5] extended
the approach developed in [65] by using combined FEM-BEM.
The first application of FDM-BEM to railway track is related to the paper of
Triantafyllidis 1991 [153]. The author examined the steady-state response of the track
modelled by an Euler-Bernoulli beam interacting with an elastic half-space by means of
ties, which were considered as rigid rectangular foundations.
From recent publications one should mention papers of Degrande and co-authors
[102]-[104] and paper of Grundman and Trommer [58].
In papers [102]-[103], BEM was applied to the problem of traffic-induced
vibrations. Comparison of theoretical prediction to experimental data showed a good
agreement of the computer simulations and the practice. In [104], the authors
investigated the influence of the soil stratification on free field traffic-induced vibrations
by BEM. They showed that the soil stratification has a considerable influence on the
dynamic interaction between the road and the soil on the free field vibrations.
Grundman and Trommer 2001 [57] published an overview concerning
applicability of integral transform methods (ITM) to complicated models of railway
track. Having given examples of different models for illustration of applicability of
different integral transforms, authors emphasized that, although ITM are applicable only
if the considered system obeys some severe restrictions, they can give new
understanding of the problem, for the main features and effects of the system become
visible in the transform domain. The authors indicated also that ITM could be applied
for boundary element method on the basis of Parsevals theorem. This can allow for
describing the fundamental solution to complicated problems and for reducing the time
of evaluation to the original domain.
Summarizing information retrieved from the above-mentioned references one
can conclude that it is conventional nowadays to use 3D models to describe the dynamic
behaviour of the high-speed railway track. These models include
1. three-dimensional visco-elastic solid (half-space, layer or multilayered halfspace) as a model for the ground;
2. beams as a model for the rails;
3. lumped elements as a model for sleepers and pads;

Although the type of models that have to be used for dynamic analysis of a
railway track has become clear, there remain many problems to study on the hand of
these models. Some of those problems, which form the objectives of this thesis, are
discussed in the next section.

Objectives and Scope.


The study presented in this thesis is devoted to the steady-state dynamic
response of various three-dimensional models for a railway track to uniformly moving
loads of a constant magnitude (train). The thesis, aiming to reveal the physical
parameters of the track that could significantly affect the track response to a high-speed
train, is constructed according to the principle: from simple to sophisticated. In
particular, the thesis has the following objectives:
1. Analysis of the influence of physical parameters and stratification of the subsoil
on the dynamic response of the track;
2. Investigation of the elastic drag that is a measure of energy loss of a highspeed train on excitation of elastic waves in the track;
3. Development of a 3D analytical model for railway track, which would account
for inhomogeneity of the track along its length (due to sleepers), and study of the
dynamic response of such a model to a set of constant, uniformly moving loads;
4. Study a possibility to detect derailment of a wagon of a long freight train by
measuring vibrations of the rails under locomotives wheels.
Let us first discuss the problem of the influence of physical parameters and
stratification of the subsoil on the dynamic response of a railway track. As was already
mentioned in the previous section, properties of the ground on which the track is laid
play an important role in the dynamics of high-speed trains. It was proven (see, for
instance [36], [29] or [117]) that the critical velocity of the train conditioned by its dead
weight (constant loading) is approximately equal to the Rayleigh wave speed, which is
determined by the physical parameters of the ground (the shear modulus, the Poisson
ratio and the mass density). However, in these papers effect was shown on the hand of
quite simple models of the subsoil. The authors of [36] and [29] considered a pure
elastic homogeneous half-space, whereas the authors of [117] analysed a visco-elastic
layer fixed at its bottom as the model for the subsoil. Actually, it is well known that the
subsoil has a layered structure each stratum of which has its own visco-elastic
properties, and it is important to know how these properties influence the critical
velocity. It should be mentioned, that the first attempt to include a layered ground into a
model for railway track was done in the papers of Sheng X. et al. [136]-[137], where a
system composed of a layer overlying a half-space of stiffer material was considered.
The authors, however, focused their attention on the track response only and did not
study the effect of the ground parameters on the critical velocity. In this thesis, a more
complete analysis of the influence of physical properties and stratification of the subsoil
on the track response is carried out to reveal the main peculiarities of the response that
could be conditioned by variations in parameters of the soil layers.
Amongst the physical parameters of the track subsoil that are tested in frames of
this work there is one parameter more, about which we would like to speak separately,
namely the material damping. The effect of the material damping is studied very well on
the basis of one-dimensional models (Kelvins foundation). For three-dimensional
7

ground the knowledge on this issue is not systematic. Usually, to describe the material
damping in the ground, researchers use rheological Voigts model (see, for instance,
Kononov A.V. and Wolfert A.R.M. [80] 2000, Metrikine A.V. and Popp K. [117] 2000)
or the hysteretic damping (Sheng X. et al. [136]-[137] 1999, Verruijt A. and Cornejo
Crdova C.J. [157] 2001, Cornejo Crdova C.J [23] 2002). The former model describes
the damping that linearly increases with the frequency. The latter one corresponds to a
frequency independent damping. Kononov and Wolfert [80] were the first to study the
effect of the material damping on the steady-state response of the beam overlying a
visco-elastic half-space. Authors of [117] and of [136]-[137] although included the
damping effect in the considered models but did not investigate it. They paid their
attention to the analysis of critical velocities of the load ([117]) and to the study of the
track response ([136]-[137]). In this thesis, the effect of material damping is studied for
various 3D models of the railway track and of the ground. Voigts rheological model is
chosen to describe material damping in the ground. In our perception, this model is
more suitable for the low-frequency dynamics of a railway track, which is addressed in
this development.
The second central issue of the thesis is the analysis of energy loss of a highspeed train on excitation of elastic waves in the track. If the train perturbs waves in the
track subsoil, these waves consume some energy. It implies that a part of energy of the
locomotive engine can be lost due to generation of waves and it is interesting to
estimate this energy. To get confidence in importance of this problem, one can draw a
parallel between a high-speed train moving faster than waves in the ground and a supersonic plane. It is well known that supersonic planes have very high fuel consumption
due to radiation of sound waves into Mach cone, which forms behind the plane. The
phenomenon of the same kind arises if velocity of a high-speed train exceeds the
Rayleigh wave speed in the subsoil. In this case, the train is accompanied by powerful
surface waves propagating within a Mach angle the edges of which are determined by
the Rayleigh wave speed. This effect was predicted by Lansing [97] in 1966, who
analytically studied a motion of a point load over a surface of a pure elastic half-space
and showed that the surface deflection field has hyperbolic discontinuity at the edges of
the Mach angle, which arises when the load moves with a velocity larger than the
Rayleigh wave speed. These waves might consume a considerable energy, especially if
the train speed is close to the Rayleigh wave speed. That is why the analysis of the
energy loss on wave excitation by a high-speed train seems to be of high significance.
In this thesis, this energy loss is analysed in terms of elastic drag, which the moving
train experiences when generates waves in the track.
The third goal of the thesis is to develop an analytical 3D model for a railway
track, which would allow to account for such an important feature of a conventional
track as inhomogeneity of the track along its length (due to sleepers). The first attempt
to develop such a model was done in paper of A.V. Metrikine and K. Popp 1999 [116].
In this thesis the model proposed in this paper is extended, critical velocities of the load
are determined and the effect of physical parameters of the model on the track response
is investigated.
The fourth objective of the thesis is to study whether it is possible to detect
derailment of a wagon of long freight train by measuring the spectrum of the rail
vibration under locomotives wheels. The wagon derailment can happen because of
different reasons. Very often, however, the wagon returns back to the track thanks to
rigid joints between the wagons. In this case, the derailed wagon leaves behind a
considerably demolished piece of the track. To reduce destruction of the railway track,
the moment of derailment must become known to the trains driver as soon as possible.

The easiest solution for the problem is to mount to each bogie of the train an
accelerometer, which could register the derailment. However, such a way is very
expensive for companies, which exploit a huge number of wagons. For instance, in
Russia, there are more than 600000 wagons and about 2000 locomotives. For such kind
of companies, a possibility to detect the derailment with the help of an accelerometer
mounted to the locomotives bogie could be undoubtedly advantageous. Theoretical and
experimental analysis of such a possibility is performed and presented in the last chapter
of this thesis.

Outline.
The thesis consists of introduction, five chapters and conclusions.
In Introduction, the relevance of carried out theoretical investigations is
founded. The literature overview on the main topic is given, the objectives and scope of
the study are formulated and the contents of the thesis are briefly enounced.
In Chapter 1, the dynamic response of an elastically supported membrane and a
visco-elastic half-space to a uniformly moving constant load is investigated. The main
objective of the chapter is to demonstrate specific features of the dynamic response of
simplest two- and three-dimensional elastic systems to a moving load. The second
objective is to study the effect of material damping in the elastic systems on the
dynamic response and on the drag experienced by the load due to excitation of waves.
Analysis of afore-mentioned problems is carried out analytically by means of
application of the integral Fourier transforms and the contour integration method. For
both models, regimes of subcritical (Vload < cwaves ) and supercritical (Vload > cwaves )
motion are studied to show crucial influence of waves in the support on the dynamic
behaviour of both systems.
In Chapter 2, a simplest 3D model of railway track accounting for interaction
between the ground and other components of the railway track is investigated. This
model is composed of an Euler-Bernoulli beam and a visco-elastic half-space. The beam
is employed as a model for the rails, sleepers and ballast, whilst the half-space
represents the ground.
The chapter pursues the following objectives. First, the crucial role of the
Rayleigh waves in the ground on the dynamic response of the railway track is exposed.
Second, effect of shear stresses in the interface between the ground and the overlying
track structure on the dynamic response of the track is studied. And third, elastic drag
experienced by a high-speed train (French TGV) due to excitation of waves in the rails
and surrounding soil is estimated for different velocities and compared to results
obtained experimentally for aerodynamic and rolling drag. The investigation is carried
out by means of the method of equivalent dynamic stiffness of the ground developed by
Dieterman and Metrikine 1996 [29].
In Chapter 3, the model considered in Chapter 2 is improved by accounting for
stratification of the ground and visco-elastic and inertial properties of the rail-supporting
structure overlying the ground (sleepers, pads and ballast). The model allows to
describe low-frequency dynamics of conventional track and can be employed to study
dynamics of modern embedded tracks, which are truly continuous being built without
sleepers. The main objectives of the chapter are to study the effect of ground
stratification and to investigate the effect of material damping in the soil on the dynamic
response of the railway track.

In Chapter 4, a model for conventional track taking into account


inhomogeneity of railway track along its length, which is due to sleepers, is considered.
The model consists of two Euler-Bernoulli beams, periodically spaced supports and
layer that describe rails, sleepers and subsoil, respectively. There are two main
objectives of this chapter. First, a method should be developed that could allow for an
accurate calculation of the steady-state response of a three-dimensional, periodically
inhomogeneous structure to a moving load. Second, employing this method, it has to be
clarified whether the effect of a three-dimensional ground in modelling a conventional
track is as crucial as that in modelling an embedded track. These two problems are
solved in the chapter by employing the method of equivalent dynamic stiffness for the
layer against sleepers and periodicity condition.
In Chapter 5, possibility to detect moment of derailment of a wagon of a long
freight train with the help of accelerometer mounted to the locomotives bogie is
investigated. To this end, vibrations of the rails that are perturbed by a derailed wheelset
and then measured under the locomotive are studied both theoretically and
experimentally.
In the theoretical analysis, a Timoshenko beam on discrete, periodically spaced
supports is employed as a model for the railway track. The beam and supports model
the rails and sleepers, respectively. Winklers model is chosen to describe reaction of
the ground, since a slow train motion justifies modelling the ground reaction by a spring
with a constant stiffness. To model the axle loading of the train, a constant load is used
that uniformly moves along the track. If there is no derailment, e.g. the normal motion
of the train takes place, the load is applied to the beam (rails). In the case of derailment,
which is referred to as an emergency motion, the load is applied to the supports
(sleepers).
Results of theoretical analysis are compared to experimental data, collected
during measurements carried out by Administration of Gorky Railway (Russia) next to
the station Ingenernaya in August and December of the year 2000. A method of
spectral analysis of the data that allows to increase sensitivity of the derailment
detection is proposed. Perspectives of using the locomotive detection for the wagon
derailment are discussed.
Finally, the main results of the research and future perspectives in modelling
dynamics of railway track are addressed to Conclusions.

10

CHAPTER 1
Steady-State Response of Two- and Three-Dimensional
Elastic Systems to a Uniformly Moving Load
In this chapter the dynamic response of an elastically supported membrane and a
visco-elastic half-space to a uniformly moving constant load is investigated.
The main objective of the chapter is to demonstrate specific features of the
dynamic response of simplest two- and three-dimensional elastic systems to a moving
load. The second central issue, which is discussed in the chapter, is the effect of material
damping in elastic systems on the dynamic response and on the drag experienced by the
load due to excitation of waves.
In section 1.1, the response of an elastically supported membrane is considered.
The load is assumed to be point-like, constant and move along the membrane uniformly
and rectilinearly. It is also assumed that the contact between the load and the membrane
is frictionless. The steady-state solution to the problem is found analytically under the
assumption that there is no material damping in the membrane. Accounting for this
damping, the solution is found in the form of a single integral, which is then evaluated
numerically. In both cases a crucial difference is shown between the response in the
subcritical case, in which the load moves slower than the waves in the membrane
(Vload < cwaves ) , and that in the supercritical case Vload > cwaves . Further, an elastic drag
is introduced as a measure of the energy loss on excitation of waves. A general
expression for the elastic drag is obtained. This is done by employing the energy
variation low. Making use of the obtained solutions, dependence of this drag on the load
velocity is analysed.
In section 1.2, the steady-state response of the half-space to a normal surface
load is studied. This response is analysed neglecting the material damping in the halfspace. Main attention is paid to the crucial effect of the Rayleigh waves on the forced
deflection of the half-space surface.
In section 1.3, the material damping is taken into consideration according to the
rheological model of Voigt. The effect of this damping on the response of the half-space
is studied. Finally, the elastic drag experienced by the load is investigated. Results
obtained in this section are compared to that found on the hand of the membrane in
section 1.1.
Results presented in sections 1.1 and 1.3 are original. They are published in
[162]-[164]. In section 1.2, the results of Lansing [97] are used, who has studied the
response of a purely elastic half-space to a surface load.

1.1. Response of an elastically supported membrane.


Before starting with a study of sophisticated models of a railway track, it is
worthwhile to gain an insight into the dynamic response of elastic systems to a moving

11

load. This should be done, however, employing models that capture some properties of
the realistic track, which are significant for its dynamic behaviour.
Let us first consider one of the simplest models in which 2D wave field can exist
(as on the surface of the ground), namely an elastically supported membrane. To perturb
the membrane, a point load P that moves uniformly along the membrane is employed.
It is assumed that the contact between the load and the membrane is frictionless.

Fig. 1.1. Moving load on elastically supported membrane.

Equation for small transversal vibrations of the membrane in this case reads:

wtt N ( wxx + wyy ) + kw = P ( x Vt ) ( y )

(1.1)

In this equation, w [ m ] is the transversal deflection of the membrane,

kg/m 2 and N [ N/m ] are the mass of the unit area of the membrane and the tension
in the membrane, k N/m3 is the stiffness of the foundation per unit area, V is the
load velocity, (...) is the Dirac delta-function.
In the steady-state regime the deflection field in the membrane is stationary in
the reference system = x Vt that moves together with the load. In this reference
system, equation (1.1) reduces to the following time-independent equation:

(V

c 2 ) w c 2 wyy + h 2 w =

( ) ( y )

(1.2)

with c = N the wave propagation speed in the membrane and h = k the cut-off
frequency. To solve equation (1.2), the following integral Fourier transforms with
respect to co-ordinates and y is applied:

W ( k1 , k2 ) =

w ( , y ) e

i ( k1 + k2 y )

d dy .

(1.3)

This yields an algebraic equation with respect to w , the solution to which is given as

12

W =

(1.4)

( c k 2 k12 + h 2 )
2

2
2

with 2 = V 2 c 2 . Integrating expression (1.4), the solution to the differential equation


(1.2) is obtained:
w ( , y ) =

e ( 1 2 ) dk1dk2
c 2 k22 2 k12 + h2
4 2
P

i k + k y

(1.5)

As will be shown, the response of the membrane depends crucially on the ratio between
the load velocity and the wave speed in the membrane. Therefore, it is customary to
consider the subcritical motion V < c and the supercritical motion V > c separately.

Subcritical motion V < c .


Let us first evaluate the following auxiliary integral:
I =

1
2

e ik1 dk1
2 k12 c 2 k22 h2 .

(1.6)

This can be done by the method of contour integration [1]. For > 0 , in accordance
with Jordans lemma, the integration contour should be closed in the lower half-plane of
the complex variable k1 , whereas for < 0 the upper half-plane should be used, see Fig
1.2.

( k1 )

( k1 )

a)

b)

Fig. 1.2. Integration contours and relevant poles in the subcritical case for (a): > 0 and (b): < 0 .

In the case in question ( 2 < 0 ) , the integrand in (1.6) has two simple poles that
are located on the imaginary axis. These poles are complex-conjugated and,
consequently, lay symmetric with respect to the real axis. Applying now the residue
13

theorem [1] to evaluate integral (1.6) and substituting the obtained result into expression
(1.5), one obtains
w=

P
2 c c 2 V 2

cos ( k2 y )

k22 + ( h / c )

c
exp
c2 V 2

2
k22 + ( h / c ) dk2

(1.7)

The integral in (1.7) can be evaluated analytically by using the following formula [53]:

b 2 + x2

cos axdx

+x
2

= K0 a 2 + b2

(1.8)

in which Re b > 0 , Re > 0 and K 0 is the modified Bessel function of the second kind,
of the order zero. Using equation (1.8), expression (1.7) can be reduced to
w=

h
K0
2 + y 2 ( c2 V 2 ) c2
2
2
2 c c 2 V 2
c V

(1.9)

Expression (1.9) describes the steady-state displacement field in the membrane


in the subcritical case. This field moves together with the load and is axially symmetric
with respect to the loading point. In the loading point itself, the field is singular.

Supercritical motion V > c .


In this case the poles of the integrand (1.6) are located on the integration path,
i.e. on the real axis of the complex variable k1 . To bypass theses poles properly a small
damping should be introduced into the system [1]. We choose for the viscosity of the
foundation, which brings an additional term 2 ut ( 0 < 1 ) to the left-hand side of
equation (1.1).
Introduction of this viscosity modifies integral (1.6) into:
I =

1
2

e ik1 dk1
2 k12 2iVk1 c 2 k22 h2

(1.10)

The poles of the integrand in (1.10) are located in the upper half-plane. These poles lay
symmetrically with respect to the imaginary axis as shown in Figure 1.3.
In accordance with the Jordans lemma, as in the subcritical case, the contour
should be closed over the upper half-plane if < 0 and over the lower half-plane if
> 0 . Since the integrand in (1.10) has no singularities in the lower half-plane, integral
(1.10) is equal to zero if > 0 .
In the case < 0 , this integral can be evaluated with the help of the residue
theorem to give in the limit 0 :

14

( k1 )

Fig. 1.3. Integration contour and the poles in the supercritical case for < 0 .

w=

cos ( k2 y )

k22 + ( h / c )

H ( )

c 2
2
sin
k2 + ( h / c ) dk2

(1.11)

with H (...) the Heaviside step-function. The integral in (1.11) can be evaluated
analytically by using the following formula [53]:

sin p x 2 + a 2

x +a
2

) cos bxdx = 2 J ( a
0

p 2 b2

0, ( b > p > 0 )

) , ( p > b > 0)

(1.12)

with J 0 the Bessel function of the first kind, of the zero order. Using (1.12), expression
(1.11) can be reduced to
w=

h 2
2

H ( ) H y J 0
( / c) y2
2 c
c

(1.13)

Expression (1.13) describes the steady-state displacement field in the membrane


in the supercritical case. This field lays behind the load, within the angle y < ( c ) ,

( < 0 ) .

Outside this angle, the membrane is not perturbed. The character of the

membrane response is analogous to the Mach cone that forms behind a supersonically
moving plane. At the edges of the angle that limits the perturbed area of the membrane,
the deflection of the membrane has a discontinuity that is equal to P 2 c .
The singularities of the membrane response, which take place for both the
subcritical and the supercritical regimes, actually violate the basic assumptions of the
linear model of the membrane. Indeed, the equation for the membrane vibrations is
valid if the slopes of the membrane are small (violated in the vicinity of the load in the
subcritical case) and the displacement field in the membrane is continuous (violated in
the supercritical case). Thus, a modification of the model is necessary that would allow
to remove the singularities from the response. There are several modifications possible.
The simplest ones are

15

1) to replace the point load by a load that is applied to a finite area;


2) to introduce a material damping into the membrane.
We choose for the material damping in the membrane. The reason for this choice
is, as it will be shown in the following chapters, that the material damping is of crucial
importance for the dynamics of the railway track.
We introduce this damping according to the Voigts rheological model, by
adding to the left-hand side of (1.1) the term N1

wtt N + N1

( wxx + wyy ) :
t

( wxx + wyy ) + kw = P ( x Vt ) ( y )
t

(1.14)

with N1 [ N*s/m ] the coefficient characterizing the material damping in the membrane.
Introducing dimensionless variables = ht and

{ x, y} = ( h c ){ x, y}

and the

reference system that moves together with the load = x , where = V / c , we


obtain from (1.14):
N h
P
2 w 1 1 2
( w + wyy ) + w = 2 ( ) ( y )
c
c

(1.15)

with = V c the dimensionless velocity of the load.


Applying to equation (1.15) the following integral Fourier transform
W ( k1 , k2 ) =

w ( , y ) e

i ( k1 + k2 yn )

d dy ,

(1.16)

an algebraic equation with respect to W ( k1 , k2 ) can be obtained. Solving this equation


and then performing the inverse Fourier transform with respect to k2 (with the help of
the contour integration), the following expression that describes the steady-state
response of the membrane can be obtained:
iP
w=
4 c 2

{k ( 1 iAk ) 1} {1 + iAk } ) dk
(1 + iAk ) ( k ( 1 iAk ) 1)

exp ik1 + i yn

2
1

2
1

(1.17)

with A = ( N1h N ) . The results of numerical integration of expression (1.17) are


presented in Figure 1.4 (a,b), which shows patterns of the deflection field in the
membrane for the subcritical and the supercritical cases.
Figure 1.4 (a) shows that in the subcritical case, the deflection field in the
membrane is localized near the load, as in the absence of the material damping. The
deflection at the loading point, however, is finite now. Figure 1.4 (b) shows that in the
supercritical case, with introduction of the material damping, the deflection field
remains localized within the Mach angle. The behaviour of the deflection field

16

changes qualitatively only at the edges of the angle, where the membrane deflection
becomes a continuous function that tends smoothly to zero outside the angle.
Thus, the material damping indeed helped to remove all singularities from the
membrane response.

a)

b)

Fig. 1.4. Characteristic profiles of the membrane deflection in the (a) subcritical case and
(b) supercritical case.

Elastic drag for a single load.


We have shown that in the supercritical case the load perturbs waves in the
membrane. These waves consume some energy. Therefore, there must be an external
energy source that supplies the membrane with this energy. Keeping in mind that in
application to the railway track this source is just the engine of the train, it is interesting
to gain an idea on how large this energy can be and what in depends upon.
To answer this question, let us first obtain a mathematical expression for the
energy that is transferred into the membrane by the load. This can be done by
multiplying both the left- and the right-hand side of equation (1.1) by wt . This yields:

wtt wt N ( wxx + wyy ) wt + kwwt = Pwt ( x Vt ) ( y )

(1.18)

By adding and subtracting terms Nwx wxt and Nwy wyt on the left-hand side of equation
(1.18) and then collecting some terms, equation (1.18) can be rewritten in the form:
wtt wt + Nwx wxt + Nwy wyt + kwwt + N ( wx wxt + wxx wt ) N ( wy wyt + wyy wt ) =
(1.19)
= Pwt ( x Vt ) ( y )

time

The expression in the first square brackets in (1.19) is nothing but the partial
derivative of the generalized density of energy of the membrane

17

h = (1 2 ) wt2 + N ( wx2 + wy2 ) + kw2 . The second square brackets describe the

divergence of the flux of the generalized energy: S = Nwt w . The right-hand side of
equation (1.19) is the surface density of the power input by the load.
Thus, we obtain the generalized energy variation law for an element of the
membrane:
h
+ divS = Pwt ( x Vt ) ( y ) ,
t

(1.20)

Integration of equation (1.20) over x and y gives the global energy variation
law for the elastically supported membrane. The right-hand side of the obtained
expression describes the power input Q into the membrane, which is equal to the
energy per unit time that is spent on perturbation of waves in the membrane:

Q=P

w ( x Vt ) ( y ) dxdy .
t

(1.21)

In engineering practice it is conventional to work with various types of drag


(aerodynamic drag, rolling drag) that a train experiences while moving. Therefore, we
will proceed with our analysis in terms of what we will call elastic drag De that the
load undergoes by perturbing vibrations of the membrane. By definition, this drag is
given as

De =

Q P
= wt ( x Vt ) ( y ) dxdy
V V

(1.22)

Thus, the drag De experienced by the load can be straightforwardly calculated


making use of the previously obtained expressions for the membrane deflection.

Elastic drag in the undamped case.


Let us first perform this calculation in the undamped case.
To this end, the -function is represented in the form
1
( x ) = lim 2 2
0
x +

(1.23)

In the subcritical case, substituting representation (1.23) and the expression for the
membrane deflection (1.9) into (1.21) we obtain
De =

2
h
2 d dy
2
2

y
(1.24)

2
1 c 2 V 2
2
c
2 3 2 c 0

2
2
2
2
2
2

( + )( y + ) y
c2

Ph

lim

18

The integrand in (1.24) is an odd function of the variable . Because of this, the
integral over , as an integral with symmetric limits from an odd function, is equal to
zero. Therefore, if the load moves subcritically, the steady-state deflection field in the
membrane consumes no energy. Nor a drag is experienced by the load.
In the supercritical case (V > c ) , substituting expressions (1.13) and (1.23) into
(1.21), and taking into account that J 0 ( 0 ) = 1 , and J 0 ( 0 ) = J1 ( 0 ) = 0 , we obtain
1
d
1
dy
+
De = 3
lim 2
2
2
2
2
2
2

2 c
( ( / c ) y ) + ( + ) ( y + 2 )
+
P2

(1.25)

The integrand in expression (1.25) is a sum of two positive terms. As shown


below, integration of the first term gives infinity as the result:
De1 =

P2

lim
4 c 0

+ 2 )

dy
P2

=
lim 3 3 =
2
2

0
2
( y + ) 4 c

(1.26)

Integration of the second (positive) term can only increase the total result. This
allows us to conclude that in the supercritical case the load experiences infinitely large
drag due to excitation of waves in the membrane. Such a result is a consequence of the
physically unrealistic discontinuity of the membrane deflection at the edges of Mach
angle.

Elastic drag in the damped case.


Consider the damped case that is described by equation (1.14).
Substituting expression (1.17) for the displacement of the membrane into
expression (1.21) for De , we obtain
iP 2V
De =
4 c 2

with g ( k1 ) =


( ) ( y )

{k (
2
1

d dy

k12 ( 2 1 iAk1 ) 1

exp ( ik1 + i yn g ( k1 ) ) dk1

(1 + iAk1 ) (

(1.27)

} {1 + iAk } .

1 iAk1 ) 1

The easiest way to evaluate expression (1.27) would be to first take the
derivative of the integrand with respect to , then to use the properties of the deltafunctions to arrive to a single integral with respect to k1 and finally to evaluate this
integral numerically. Thus obtained single integral, however, would be divergent. This
does not imply that the original expression (1.27) is divergent. The point is that the
integral with respect to k1 in (1.27) is not uniformly convergent as and y tend to
zero. Consequently, the order of differentiation with respect to and integration with

19

respect to k1 in (1.27) cannot be interchanged. Thus, evaluation of De should be


performed in another way.
As follows from (1.21),
De =

P w
w
= P
V t x =Vt
=0
y =0

(1.28)

y =0

We have shown above that for the damped case the derivative w is limited at any
point of the membrane and, consequently, we can evaluate this partial derivative
numerically having calculated the deflection field. The latter can be calculated by
employing a standard program for numerical integration.

N N1 = 100

Dnorm

N N1 = 50

V c
Fig. 1.5. Normalized drag versus the load velocity.

Results of the numerical analysis are depicted in Figure 1.5, which shows the
dependence of the elastic drag De , normalized by P 2 h 4 c3 , on the load velocity for
two different values of the damping in the membrane. One can see from the figure that
in contrast to the undamped case, whatever the load velocity is, there is an elastic drag
experienced by the load. The drag increases as the load velocity approaches the wave
speed in the membrane. The effect of the damping is controversial in the subcritical and
the supercritical case. In the former case the increase of the damping leads to the
increase of De , while in the latter case the larger is the damping, the smaller is the drag
De (in the considered range of the damping coefficients).

20

1.2. Response of an elastic half-space to a normal constant


load uniformly moving over its surface.
The next step on the way towards realistic models of the railway track is to
consider not a two-dimensional but a three-dimensional model for the ground. The
simplest model of this kind is a purely elastic half-space, the dynamic response of which
is studied in this paragraph. This system allows for demonstration of the main features
of the subcritical and supercritical response of the ground surface to a moving load.
Consider the half-space that is perturbed by a point, vertical load, which moves
uniformly on the surface of the half-space, see Fig. 1.6.

V
P

x, u
y, v
z, w

Fig. 1.6. Constant normal load moving over the surface of elastic half-space.

Governing equations for the steady-state response of the half-space can be


written as follows:

Equation of motion for an inner element of the half-space


2u
u + ( + ) iu = 2 ,
t

(1.29)

where u ( x, y , z , t ) = {u ( x, y , z , t ) , v ( x, y , z , t ) , w ( x, y , z , t )}[ m ] is the displacement of


the half-space, kg/m3 is the mass density of the half-space, N/m 2 and

N/m 2 are the Lams constants, = x0 + y0 + z0


is the operator nabla,
x
y
z

2
2
2
+
+
is the Laplacian operator, u = iu u ,
x 2 y 2 z 2
) are scalar and vector products, respectively.

= i =

( i ) and (

Boundary conditions on the surface of the half-space

21

zz ( x, y, 0, t ) = P ( x Vt ) ( y )
xz ( x, y, 0, t ) = yz ( x, y, 0, t ) = 0,

(1.30)

where zz ( x, y, z , t ) is the normal stress, xz ( x, y, z , t ) and yz ( x, y, z , t ) are the shear


stresses, P [ N ] is the load magnitude, V [ m/s ] is the load velocity, (...) is the Dirac
delta-function.
Following the work of Lamb [94], we represent the solution to the Lams
equations (1.29) in the form
u=

2
2
2 1 2
+
, v=
+
, w=
+

.
x x z
y y z
z z 2 cT2 t 2

(1.31)

Scalar functions ( x, y, z , t ) and ( x, y, z , t ) in (1.31) are solutions to the wave


equations
1 2
= 2 2 ,
cL t
with cL =

( + 2 ) /

1 2
= 2 2 ,
cT t

(1.32)

and cT = / the velocities of the dilatational and shear

waves in the half-space.


The stresses in the half-space expressed in terms of functions ( x, y, z , t ) and

( x, y, z, t ) have the form

zz =

2 3
2
2

+
2+ 3
cL2 t 2
z
z

3
2

z t 2

2
3
3

2
x t 2
x z x z

zx = 2

(1.33)

2
3
3
,

2
y t 2
y z y z

zy = 2

We will solve the problem (1.29)-(1.30) by using the following integral Fourier
transforms over time t and horizontal co-ordinates x and y :

22

f ( k1 , k2 , z , ) =

( x, y, z, t ) exp ( i ( t k x k y ) ) dxdydt
1

g ( k1 , k2 , z , ) =

( x, y, z, t ) exp ( i (t k x k y ) ) dxdydt
1

u ,k1 ,k2 ( k1 , k2 , z , ) =

u ( x, y, z, t ) exp ( i ( t k x k y ) ) dxdydt
1

(1.34)

v ,k1 , k2 ( k1 , k2 , z , ) =

v ( x, y, z, t ) exp ( i ( t k x k y ) ) dxdydt
1

w , k1 ,k2 ( k1 , k2 , z , ) =

w ( x, y, z, t ) exp ( i ( t k x k y ) ) dxdydt
1

Applying these transforms to (1.30), (1.32) and (1.33), one can obtain the following
equations in the Fourier domain:

Equations of motion for the inner element of the half-space (from (1.32)):

2 f 2
+ 2 k12 k22 f = 0
2
z
cL

2 g 2
+ 2 k12 k22 g = 0
2
z cT

(1.35)

Boundary conditions on the half-space surface z = 0 (from the equations (1.30)


using (1.33)):

2 f 3g
g
2 f + 2 2 + 3 + 2 2
= 2 P ( k1V )
z
cL
z z
f 2g
+ 2 + 2 g = 0,
2
z z

(1.36)

The general solution to equations (1.35) is

f = A ( k1 , k2 , ) exp ( zRL ) ,
g = B ( k1 , k2 , ) exp ( zRT ) ,

(1.37)

RL ,T = k12 + k22 2 cL2,T ,


In expressions (1.37), the branch of the radicals is fixed by inequality Re ( RL ,T ) > 0 ,
which provides that the displacements of the half-space vanish at z .
Substituting expressions (1.37) into the boundary conditions (1.36), one obtains
a system of two linear algebraic equations with respect to A and B . When solved, this
system gives

23

2 P ( k1V ) 2 RT2 + 2 cT2


A=
,

( , k1 , k2 )
2 P ( k1V )

B=

with ( , k1 , k2 ) = 2 ( k12 + k22 ) 2 cT2

(1.38)

2 RL
,
( , k1 , k2 )

4 RL RT ( k12 + k22 ) .

As it follows from (1.31) and (1.34), the Fourier components of the half-space
displacement may be expressed as
u ,k1 ,k2 = ik1 f + ik1
w , k1 ,k2

g
,
z

v , k1 , k2 = ik2 f + ik2

g
z

f 2g 2
g
=
+
+
z z 2 cT2

(1.39)

Substituting solutions (1.37) into these expressions and using (1.38), the following
expressions for the Fourier components of the displacement of the half-space surface
can be obtained:
2
2
2
2 P ( k1V ) ik1 ( 2 RT + cT ) 2 RL RT
u ,k1 ,k2 ( k1 , k2 , 0, ) =

( , k1 , k2 )
2
2
2
2 P ( k1V ) ik2 ( 2 RT + cT ) 2 RL RT
v ,k1 , k2 ( k1 , k2 , 0, ) =

( , k1 , k2 )

w , k1 ,k2 ( k1 , k2 , 0, ) =

2 P ( k1V )

(1.40)

2 RL
cT2 ( , k1 , k2 )

Applying now the inverse Fourier transforms, integrating with respect to by


using the following property of the Dirac delta function:

F ( x ) ( x x ) dx = F ( x ) ,
0

(1.41)

and introducing the notation = x Vt , we obtain the following expression for the
components of the half-space surface displacement:

u ( x, y, 0, t ) = uS ( , y ) =

k1eik1 +ik2 y
( 2 RT2 + k1 T2 ) 2 RL RT dk1dk2
2

4 ( k1 , k2 )

v ( x, y, 0, t ) = vS ( , y ) =

k2 eik1 +ik2 y
( 2 RT2 + k1 T2 ) 2 RL RT dk1dk2
2

4 ( k1 , k2 )

iP

w ( x, y, 0, t ) = wS ( , y ) =

iP

PV 2
4 2 cT2

k12 RL eik1 +ik2 y


( k1 , k2 ) dk1dk2

24

(1.42)

In (1.42), ( k1 , k2 ) = 2 ( k12 + k22 ) k12 T2

4 RL RT ( k12 + k22 ) , L = V cL , T = V cT

and RL ,T = k12 + k22 k12 L2,T .


Expressions (1.42) show that the half-space displacement depends upon time
through the variable only, i.e. in the steady-state regime the displacement field in the
half-space remains unchanged in the reference system that moves together with the
load.
The main attention hereafter will be paid to the analysis of the vertical vibrations
of the half-space surface. The other components of the half-space displacement can be
analysed analogously.
To study wS ( , y ) , it is customary to introduce a new variable = k2 k1 and
divide the obtained integral in two parts:
wS ( , y ) =

RL ( ) eik1 +ik1 y
RL ( ) eik1 +ik1 y
PV 2
d

dk
d dk1
+
1
2
2

( )
( )
4 cT
0

with ( ) = 2 (1 + 2 ) T2

(1.43)

4 RL ( ) RT ( ) (1 + 2 ) , RL ,T ( ) = 1 + 2 L2,T .

Due to the symmetry of the problem at hand with respect to y = 0 (the load
path), the surface displacement of the half-space should be symmetric with respect to
this line. Therefore, it is sufficient to consider the problem assuming that y > 0 and
than, in the final result, replace y by y .
To simplify (1.43), it is customary to consider the following auxiliary integral
RL ( ) exp ( ik1 y ) d
,
( )

I=

(1.44)

which we will calculate by the method of contour integration. The integrand in (1.44) is
a multi-valued function because of the presence of radicals RL and RT . Besides this, the
integrand becomes infinite at so-called Rayleigh poles, which are simple zeros of the
equation ( ) = 0 . Therefore, to apply the contour integration, it is necessary to modify
the integration contour in the vicinity of the poles that are located on the integration
path and to define the branch of the radicals.
Branch points. The branch points of the integrand in (1.44) are determined by
the zeroes of radicals RL and RT . Their location depends upon the ratio between the
load velocity V and the wave speeds cL and cT . Since the argument of both radicals
has the same form 1 + 2 2 (where = L in RL and = T in RT ), it is sufficient
to consider both radicals simultaneously.
In the complex plane ( ) , the branch points are located at the positions

= i 1 2 if < 1 and = 2 1 if > 1 .


In the case < 1 , the branch points are located on the imaginary axis as it is
shown in Figure 1.7. Fixing the branch of the radicals, it is convenient to cut the
complex plane along the imaginary axis from 1 to +i and from 2 to i . The

25

radicals are then assumed to be real and positive on the path of integration (along the
real axis).

Im ( )

Branch cut

1 = i 1 2
Re ( )

2 = i 1 2

Branch cut

Fig. 1.7. Branch points and branch cuts in the case < 1 .

In the case > 1 , the branch points are located on the integration path, i.e. on
the real axis. To make a decision on how to deform the path in the vicinity of these
points, one should take into account a small material damping in the half-space, as
explained in [3]. The simplest way to account for this damping is to perform in
expressions (1.40) the following replacement

2 2 + i .

(1.45)

Parameter << 1 in this expression is real and positive. Such a replacement, obviously
will cause a small imaginary part of the parameter :

2 =

2
k12 c 2

=
= k1V

V2
c2

2 =

2 + i
k12 c 2

=
= k1V

V 2 iV
+
c 2 k1c 2

(1.46)

Expression (1.46) shows that the sign of the imaginary part of 2 depends on the sign
of k1 . Let us first consider k1 > 0 . In this case the imaginary part of 2 is positive,
which implies that introduction of the material damping can be accounted for by the
replacement

2 2 + i , > 0

(1.47)

We assume the material damping to be small enough to provide the inequality


2 / . If this is the case, then the location of the branch points is given as

1,2

1 i /
= 1 1 +

2
2 1
2

26

(1.48)

Thus, introduction of a small material damping in the elastic half-space results in the
shift of the branch points, which is shown in Figure 1.8 (for k1 > 0 ). Consequently, in
the limit 0 (vanishing damping), the integration path should be chosen above the
left branch point and below the right branch point.

Im ( )
Branch cut

1 = 2 1
Re ( )

2 = 2 1

Branch cut

Fig. 1.8. Branch points and branch cuts in the case > 1 ( k1 > 0 ) .

We cut the plane of the complex variable along the real axis from 1 to 0 + i 0
and then from 0 + i 0 to +i for 1 and from 2 to 0 i0 and then from 0 i 0 to i
for 2 (see Fig. 1.8). These cuts ensure that the real part of the radicals does not change
over the complex plane (we assume this part to be positive).
Rayleigh poles. The Rayleigh poles are the roots of the equation ( ) = 0 .
Introducing the variable 2 = 1 + 2 , this equation can be reduced to

( 2

+ T2 ) 4 2 2 L2 2 T2 = 0
2

(1.49)

This equation is studied in the book of Achenbach, [3]. As it is shown there, if both
square roots in this equation have a positive real part, then this equation has two real
roots, which can be defined as = R , where R = V cR and cR is the Rayleigh wave
speed.
Thus, there are two Rayleigh poles in the complex plane ( ) . We will introduce
them as 3 and 4 . These poles are located in the points i 1 R2 when R < 1 and in
the points R2 1 when R > 1 . To get physically sensible results, we should again
take into account the material damping. As in the case with the branch points, the pole

+ R2 1 will then shift to the upper half-plane while the pole R2 1 will move to
the lower one (provided that k1 > 0 ). Consequently, the integration path should be
chosen between these two poles likewise in the case with the branch points.

27

It is easy to show that in the case k1 < 0 , the poles and the branch points located
in the point + R ,T , L 1 shift to the lower half-pane while the singularities located at
the point R ,T , L 1 move to the upper one.
As follows from the discussion on the branch cuts and the poles, there are four
possible, qualitatively different distributions of the singular points over the complex
plane ( ) . These configurations depend on the load velocity and turn out to be different
in the following four cases:
1.
2.
3.
4.

V < cR
cR < V < cT
cT < V < cL
V > cL .

It is not within the scope of this development to discuss all these cases in detail.
This is done by Lansing [97], although for a particular case of the Poisson ratio
= 0.25 only. We will consider two first regimes that are of primary importance for
practice.
Case V < cR . In this case R , L and T are smaller then 1. If k1 > 0 and
y > 0 , then in accordance with Jordans Lemma to evaluate integral (1.44) the
integration contour should be closed over the upper half-plane of ( ) . The integration
contour, the poles and the branch points for this case are shown in Figure 1.9.
Taking into account that
1) the integrals over the large quarter-circles C1 C2 vanish as their radii tend to
infinity and
2) the integrals around the branch points vanish as the radii of the circles
surrounding these points tend to zero,
employing the residue theorem, one may write the following equation:

R ( )

f ( ) d = + + + + f ( ) d = 2 i L
exp ( ik1 y ) ,
d d

1
2
3
4

where f ( ) = ( RL ( ) ( ) ) exp ( ik1 y ) , and integral

the Cauchy principle sense.

28

(1.50)

is to be understood in

( )

i 1 L2

C2

C1

i 1 T2
i 1 R2

Fig. 1.9. Integration contour in the case V < cR , k1 > 0 , y > 0 .

After some simplifications, integral (1.44) may be reduced to:


I = Ze k1 y

2 I1 ( k1 y ) 8 I 2 ( k1 y )

1 R2

( k1 > 0, y > 0 )

(1.51)

with
Z=

R2 L2

2B

1 R2

I1 ( k1 y ) =

I 2 ( k1 y ) =

R2 L2 R2 T2

L2 + 2 1 exp ( k1 y ) d

1 L2

2 R2 L2 T2

, B = 2 ( 2 R2 T2 ) 2 R2 L2 R2 T2 R2

( 2 (1 ) )
2

2
T

+ 4 (1 2 ) L2 + 2 1 T2 + 2 1

(1 ) + 1 (1 ) exp ( k y ) d .

( 2 (1 ) ) + 16 (1 ) ( + 1)( + 1)

1 L2

1 T2

2
T

2
T

2 2

2
L

2
L

2
T

In the case k1 < 0 , integral (1.44) can be evaluated analogously to give


I = Ze k1 y

1 R2

2 I1 ( k1 y ) 8 I 2 ( k1 y ) .

(1.52)

Obviously, expression (1.52) follows from expression (1.51) once k1 in the latter
expression is replaced by k1 .
Substituting now (1.51) and (1.52) into (1.43), we obtain the following
expression for the vertical displacement of the half-space surface:

29

0
PV 2 k1 y
wS ( , y ) = 2 2 Ze
4 cT

+ Ze

k1 y 1 R2

2 I1 ( k1 y ) 8 I 2 ( k1 y ) eik1 dk1 +

1 R2

2 I1 ( k1 y ) 8 I 2 ( k1 y ) e dk1 ,

ik1

(1.53)

( y > 0)

This expression can be simplified to


wS ( , y ) =

PV 2
Ze k1 y
2
2
2 cT 0

1 R2

2 I1 ( k1 y ) 8 I 2 ( k1 y ) cos ( k1 ) dk1 ,

( y > 0)

(1.54)

Employing the following formula [53]

cos ax exp ( bx ) dx = a

b
,
+ b2

(1.55)

integration over k1 can be carried out in (1.54). This yields


w ( , y , 0 ) =

1 R2 y
PV 2

2
,

8
,
Z
J
y
J
y

) 2 ( ) ,
1(
2 2 cT2 (1 R2 ) y 2 + 2

( y > 0)

(1.56)

with
J1 ( , y ) =

L2 + 2 1d

y
2 2 + y 2 2
2 (1 2 ) T2
1 L

+ 4 (1

+ 1 + 1
2
L

2
T

1 2 ) T2 + 2 1 (1 2 L2 ) d
(
y
.
J 2 ( , y ) = 2
+ y 2 2 2 (1 2 ) 2 4 + 16 (1 2 )2 ( 2 + 2 1)( 2 + 2 1)
1 T2
T
L
T
1 L2

As aforementioned, this solution can be generalized for the case of arbitrary y with the
help of replacement y y to give

1 R2 y
PV 2

2
,

8
,
w ( , y , 0 ) = 2 2 Z
J
y
J
y

(
)
(
)
1
2
2 cT (1 R2 ) y 2 + 2

(1.57)

Case cR < V < cT . In this case L and T are smaller than 1, while R > 1 .
Consequently, in comparison to the previous case, the branch points remain their
positions, while the Rayleigh poles shift to the real axis. Therefore, the integration
contour, the branch points and the poles for k1 > 0 assume the positions shown in
Figure 1.10.

30

( )

i 1 L2
C2

C1

i 1 T2

i 1 R2

Fig. 1.10. Integration contour in the case cR < V < cT , k1 > 0 , y > 0 .

The integration steps here are analogous to that explained in the previous case. Taking
advantage of this, we will write the result of integration of the auxiliary integral (1.44)
at once, skipping the intermediate steps:
I = iZ1e ik1 y
I = iZ1eik1 y
with Z1 =

R2 L2

2B

R2 1

R2 1

R2 1

2 I1 ( k1 y ) 8 I 2 ( k1 y ) ,

2 I1 ( k1 y ) 8 I 2 ( k1 y ) ,

( k1 > 0 )

(1.58)

( k1 < 0 )

(1.59)

, and the remaining notations the same as in (1.51).

Substituting (1.58) and (1.59) into (1.43), we obtain:


PV 2
wS ( , y ) = 2 2
2 cT

Z1 sin ( k1 ) eik1 y

R2 1

2 I1 ( k1 y ) + 8 I 2 ( k1 y ) cos ( k1 ) dk1 (1.60)

Expression (1.60), though looking simple, does not determine the vertical
displacement of the half-space surface. The problem is that the integral

ik y
Z1 sin ( k1 ) e 1

R2 1

dk1

is undefined. Such a situation is common in dynamics of solids under moving load, see
[40]. To overcome this uncertainty, the displacement of the half-space surface should be
calculated as the limit of z 0 from u ( x, y, z , t ) . This is accomplished below.
As follows from (1.37)-(1.39), the vertical ( z -dependent) displacement of the
half-space reads

31

w ( , y , z ) =

RL eik1 +ik2 y
( 2 RT2 + k12 T2 ) e RL z 2 ( RT2 + k12 T2 ) e RT z dk1dk2 ,
2

4 ( k1 , k2 )
P

( y > 0)
(1.61)

Introducing a new variable = k2 k1 in (1.61), one obtains


w ( , y , z ) =

0
k1RL ( ) z
k R z
2
2
2 ( RT2 ( ) + T2 ) e 1 T ( ) +
( 2 RT ( ) + T ) e
2
4

+ ( 2 RT2 ( ) + T2 ) e

k1 RL ( ) z

2 ( RT2 ( ) + T2 ) e

with ( ) = 2 (1 + 2 ) T2

k1RT ( ) z

RL ( ) eik1 +ik1 y
d dk1 ,
( )

(1.62)

( y > 0)

4 RL ( ) RT ( ) (1 + 2 ) , RL ,T ( ) = 1 + 2 L2,T .

We first evaluate the following auxiliary integral by employing the method of


contour integration:

I=

k R z
k R z
RL ( ) ( 2 RT2 ( ) + T2 ) e 1 L ( ) 2 ( RT2 ( ) + T2 ) e 1 T ( ) exp ( ik1 y ) d

( )

. (1.63)

For small positive values of z this integral can be evaluated in exactly the same manner
as integral (1.44). In the limit z 0 this gives
k1 > 0 :
lim I =
z 0

ik y
= lim iZ1e 1
z 0

R2 1

( 2 2 2 ) e k1
R
T

R2 L2 z

2 ( R2 T2 ) e

k1 R2 T2 z

2 I ( k y ) 8I ( k y ) ,
1
1
2
1

(1.64)
k1 < 0 :
lim I =
z 0

ik y
= lim iZ1e 1
z 0

R2 1

2 2 2 ) e k1
T
( R

R2 L2 z

2 ( R2 T2 ) e k1

R2 T2 z

2 I ( k y ) 8I ( k y ) ,
1
1
2
1

(1.65)
with Z1 , I1 and I 2 defined in (1.51).
Substitution of (1.64) and (1.65) into (1.62) yields
lim w ( , y, z ) =
z 0

2 0

P ik y
lim 2 Z1e 1
z 0 2
0

( I ( k y ) + 4I ( k y ) ) cos k ydk
1

R2 1

sin ( k1 ) ( 2 R2 T2 ) e

k1

32

R2 L2 z

2 ( R2 T2 ) e

k1

R2 T2 z

dk
1

(1.66)

The second integral in (1.66) can be taken analytically. This leads to the following
expression:

( 2 R2 T2 )
Z1 lim
lim w ( , y, z ) =

z 0
2 2 z 0 y 2 ( R2 1) 2 ( R2 L2 ) z 2 2 y R2 1 R2 L2 z

2 ( R2 T2 )
(1.67)
+

y 2 ( R2 1) 2 ( R2 T2 ) z 2 2 y R2 1 R2 T2 z
P

( I ( k y ) + 4 I ( k y ) ) cos k ydk
1

Calculating now the limit z 0 , we obtain


wS ( , y ) =

PV 2

2
J
,
y

8
J
,
y

) 2 ( )
1
1(
2 2 cT2 y 2 ( R2 1) 2

(1.68)

Analysing solutions (1.57) and (1.68), the following conclusions can be drawn
with respect to the steady-state response of the half-space. In the subcritical case
(V < cR ) the vertical displacement of the half-space in the loading point is infinite, as in
the case of elastically supported membrane. In the supercritical case (V > cR ) the
vertical displacement of the half-space surface experiences a hyperbolic discontinuity
along the edges of the Mach angle. Besides this, this displacement tends to infinity
everywhere, if V = cR .
To remove these singularities we will introduce a material damping in the halfspace, analogously to that what has been done in 1.1 for the membrane.

1.3. Response of a visco-elastic half-space to a normal


constant load uniformly moving over its surface.
We will introduce the damping according to the rheological model of Voigt. In
this case the equation of motion for the half-space takes the form:
2u
u + + ( iu ) = 2

(1.69)

where = + / t and = + / t are operators, which are used instead of the


Lams constants to describe the visco-elastic material in accordance with the Voigts
model.
The solution to equation (1.69), likewise in the undamped case, can be
represented in the form

33

u=

2
2
2 2
+
, v=
+
, w=
+

.
x x z
y y z
z z 2 t 2

(1.70)

The scalar functions ( x, y, z , t ) and ( x, y, z , t ) in (1.70) are now solutions to


equations

+ 2 = 2 ,
t

2
.
t2

(1.71)

The stresses in terms of functions ( x, y, z , t ) and ( x, y, z , t ) are

( + 2 )

zz

2
2
3
3
+ 2 + 2 + 2 + 2
=
2

t2
z3
z t 2
z
2
3
3
+

zx = 2

2
x t 2
x z x z

(1.72)

2
3
3
+

2
y t 2
y z y z

zy = 2

Applying the integral Fourier transforms (1.34) to the equations of motion (1.71) and to
the boundary conditions (1.30) that remain unchanged in the damped case, one obtains:

Equations of motion for the inner element of the half-space:

2 f 2
+ 2 k12 k22 f = 0
2
z
cL

2 g 2
+ 2 k12 k22 g = 0
2
z cT

(1.73)

where cL2 = cL2 i ( / + 2 / ) , cT2 = cT2 i / , cL and cT are the propagation


speeds of the dilatational and shear waves, respectively.

Boundary conditions at the surface of the half-space:


2 f 3g
2 g

+
= 2 P ( k1V )
f
2
2 + 3 + 2
2
z
cL
z z
f 2g
+ 2 + 2 g = 0,
2

z
z

(1.74)

with = i / , = i / .
The general solution to equations (1.73), accounting for the proper behaviour for
large positive values of z , is

34

( )
g = B ( k , k , ) exp ( zR ) ,

f = A ( k1 , k2 , ) exp zRL ,
1

(1.75)

RL ,T = k12 + k22 2 cL2,T ,

provided that the branches in the complex domain are chosen such that the radicals have
a positive real part, i.e. Re RL ,T > 0 .

Substituting (1.75) into the boundary conditions (1.74), we obtain a system of


two linear algebraic equations with respect to A and B . This system can be easily
solved to give analytical expressions for A and B . Using then representations (1.70),
one can obtain the following expression for the vertical component of the displacement
vector at the surface of the half-space:

wS ( , y ) =

PV 2
4 2

k12 RL eik1 +ik2 y


cT2 ( k1 , k2 )dk1dk2

(1.76)

with

( k1 , k2 ) = 2 ( k12 + k22 ) k12 V 2 cT2

4 RL RT ( k12 + k22 ) , RL ,T = k12 + k22 k12V 2 cL2,T ,

= ik1V / , cT2 = cT2 ik1V / , cL2 = cL2 ik1V ( / + 2 / ) .

Thanks to the damping, the integrand in (1.76) has no singularities and decays
proportionally to (1 ki3 ) as ki . Therefore, the double integral in (1.76) can be
evaluated numerically. Results of the numerical integration are schematically shown in
Figures 1.11 and 1.12.

x Vt

Fig. 1.11. Vertical displacement of the half-space surface in the case V < cR .

35

x Vt

Fig. 1.12. Vertical displacement of the half-space surface in the case cR < V < cT .

Figure 1.11 demonstrates that in the subcritical case the displacement field in the
half-space is localized around the load. In the supercritical regime of motion that is
shown in Figure 1.12, the displacement of the surface varies rapidly only at the edges of
the Mach angle. Inside the angle the displacement tends quickly to zero. The flat
surface inside the Mach angle is the direct consequence of the half-space model,
which, in the absence of damping, is non-dispersive. The material damping introduces
dispersion, but, being assumed as small in our calculations, does not change the
response dramatically (except at the edges of the Mach angle).
Having found the response of the half-space that is free of singularities, we can
analyse the drag experienced by the load due to excitation of waves in the half-space
The power input into the half-space, by definition, is given as [3]

Q = ti ui ds

(1.77)

with ti the components of the surface traction vector and ui the components of the
displacement vector. The repeated subscript implies summation. Taking into account
that the load is vertical and employing the Cauchys stress formula

ti = ij n j ,

(1.78)

with ij the components of the stress tensor and n j the components of the unit normal
vector, expression (1.77) can be rewritten as

Q=

zz

w
dxdy
t z =0

(1.79)

Substituting into this expression the boundary conditions (1.30) that relates the vertical
stress on the surface to the load intensity, we obtain the following expression for the
elastic drag
36

De =

Q
P wS
=
V
V t

(1.80)
x =Vt

As follows from (1.80), the elastic drag experienced by the load is fully
determined by the vertical displacement of the elastic half-space at the loading point,
which is given by (1.76).
Results of numerical evaluation of expression (1.80) are depicted in Figure 1.13
for two different values of the material damping in the half-space. Calculations were
carried out using the following parameters of the half-space:

= 1.8 103 kg m3 , = 6 106 N m 2 , = 0.3 ,

(1.81)

with the Poissons ratio. Parameters (1.81) describe a very soft soil (peat) with
relatively small speeds of the Rayleigh and shear waves ( 193 km/h and 210 km/h ,
respectively).

120

Dnorm

/=0.001

80

/ = 0.01

40

V , km/h
0
0

100

200

Fig. 1.13. Normalized elastic drag Dnorm = De

(P V
2

300

4 2 ) versus the load velocity.

Figure 1.13 shows that the elastic drag experienced by the load increases as the
load velocity approaches the Rayleigh wave velocity ( 193 km/h ). In the subcritical
case, the larger is the damping, the larger is the drag De . The increase of the damping in
the supercritical case leads to the decrease of the drag De .
Comparing the response of the half-space to that of the membrane, the following
conclusions can be drawn:
1. In the subcritical case, the deflection field in both the half-space and the
membrane is localised in the vicinity of the load and is almost symmetrical
with respect to the loading point (if the material damping is small).
2. In the supercritical case, both models show the Mach angle on the surface.
Within this angle, the membrane displacement is wavy, whereas the surface of
the half-space is almost flat.
37

3. The velocity dependence of the drag experienced by the load is similar for both
models. The material damping results in the increase of the drag in the
subcritical case and in the reduction of the drag in the case of supercritical
motion.
Although the half-space can better catch properties of the ground than the membrane, it
remains a single component of the railway track model. The other components are the
rails, the sleepers and the ballast. In the next chapter the latter three components will be
accounted for in a simplified manner.

38

CHAPTER 2
Beam on a Visco-elastic Half-space as a Simple
Three-Dimensional Model for a Railway Track
In this chapter, a simple model that accounts for the dynamic interaction of the
ground with the other components of the railway track is studied. This model is
composed of an Euler-Bernoulli beam and a visco-elastic half-space. The beam is
employed as a model for the rails, sleepers and ballast, whilst the half-space represents
the ground.
The chapter has the following objectives:
1. To expose the crucial role of the surface waves in the ground on the dynamic
response of the railway track.
2. To study the effect of shear stresses in the interface between the ground and the
overlying track structure on the dynamic response of the track.
3. To estimate the drag experienced by a high-speed train due to excitation of
vibrations and waves in the rails and surrounding ground.
The chapter consists of three sections. In section 2.1, an equivalent stiffness of
the half-space eq is considered [29]. It is shown that the original three-dimensional
(3D) model can be reduced to a one-dimensional (1D) model by replacing the halfspace by a 1D visco-elastic foundation. This replacement is exact provided that the
foundation stiffness is a function of angular frequency and wave number of waves in the
beam. Introduction of the equivalent stiffness is customary for further analyses.
Additionally, it allows for demonstration of a fundamental role of the Rayleigh waves in
the dynamics of railway tracks.
In section 2.2, the steady-state response of the beam on the half-space is
analysed to a vertical load that uniformly moves on the beam. The critical velocities of
the load are determined and the beam deflection for different load velocities is studied.
Section 2.3 is devoted to the analysis of an elastic drag experienced by a highspeed train due to excitation of vibrations and waves in the ground and railway track.
First, vibrations perturbed by a single load are considered to study the effect of the load
velocity and material damping on this drag. Second, by studying vibrations of the
system under two loads, the effect of the distance between these loads to the drag is
analysed. Finally, total elastic drag that is experienced by a TGV train is calculated and
compared to the rolling drag and the aerodynamic drag.
The results presented in sections 2.2 and 2.3 are original. They are published in
[119] and [165]. In section 2.1, the results of Dieterman and Metrikine [29], and
Kononov and Wolfert [80] have been used. The former authors were the first to
introduce the dynamic equivalent stiffness of the ground underlying the railway track
and to investigate it in the case of an Euler-Bernoulli beam resting on purely elastic
half-space. The latter authors have studied the dynamic response of the same beam on a
visco-elastic half-space to a uniformly moving load.

39

2.1. Equivalent dynamic stiffness of a visco-elastic half-space.


In this section, the first step in the analysis of the steady-state response of the
beam on visco-elastic half-space to a moving load is accomplished, namely the
equivalent dynamic stiffness of the half-space is calculated and studied.
The concept of equivalent dynamic stiffness of the ground underlying the
railway track was introduced in 1996 by Dieterman and Metrikine [29]. Having
considered vibrations of a beam on elastic half-space, they showed that the half-space
could be replaced exactly by a 1D foundation with a complex-valued stiffness that
depends on the angular frequency of the beam vibration and wavenumber of waves
propagating in the beam.
Following the method proposed by Dieterman and Metrikine [29] and using
results of Kononov and Wolfert [80], in this section we determine and then study the
equivalent stiffness of the visco-elastic half-space. The new element of the present study
with respect to the aforementioned papers is that not only vertical but also longitudinal
stresses at the beam half-space interface are taken into account. Thus, the equivalent
stiffness becomes a tensor of rank two.
As shown by Dieterman and Metrikine [29], the equivalent stiffness can be
calculated by assuming a harmonic sinusoidal wave propagates in the beam. We will
introduce the equivalent stiffness in a different way, starting from a problem statement
that describes the steady-state dynamic response of the beam half-space structure to a
set of moving loads. This approach is customary in the current development, for its final
goal is to calculate the steady-state structural response.
Consider an infinitely long Euler-Bernoulli beam of a finite width 2a overlying
a visco-elastic half-space, subjected to a set of moving loads of a constant magnitude P
as shown in Figure 2.1.

V
P

2a

x, u
y, v
z, w

Fig. 2.1. Bernulli-Euler beam on visco-elastic half-space under moving loads.

The loads are vertical, line-like, uniformly distributed over the beam width, and
move at a fixed distance from each other. The beam performs the vertical motion only
and its vertical displacement wbeam ( x, t ) does not depend on the lateral co-ordinate y .
The contact between the beam and the half-space is described approximately by
assuming that the stresses at the interface are uniformly distributed over the beam

40

width. The vertical contact is accounted for kinematically by


displacement equal to the vertical displacement of the half-space
centreline of the beam. The shear contact in the x direction is
manner that is schematically depicted in Figure 2.2, which presents
section of the system by the plane y = 0 .

setting the beam


surface along the
considered in the
the vertical cross-

z=0

wbeam
u

z=0

Fig. 2.2. Vertical cross-section y = 0 with magnified beam half-space interface.

Figure 2.2 shows that the contact takes place through the shear springs with
stiffness per unit length K , which are assumed to be uniformly and continuously
distributed beneath the beam. The end of the springs that is attached to the beam is
immovable in the x direction whereas the end, which contacts the half-space,
undergoes a displacement equal to the horizontal displacement of the half-space surface
along the centreline of the beam. The lateral stress yz at the interface is neglected, for
in the chosen approximation this stress does not influence the beam vertical response
[112].
With these assumptions, the governing equations of motion for the model can be
written as follows.

Equation of motion for an inner element of the half-space:

2u
u + + iu = 2 ,
t

) (

(2.1)

where u ( x, y , z , t ) = ( u ( x, y , z , t ) , v ( x, y , z , t ) , w ( x, y , z , t ) ) [ m ] is the displacement


vector of the half-space, kg m3 is its mass density, = + * / t and
= + * / t are operators that are used instead of Lams constants N m 2 and

N m 2 to describe the half-space according to the Voigts phenomenological


model.

Boundary conditions at the surface of the half-space z = 0 :

41

2a zz ( x, y, 0, t ) = ( S beam w,beam
+ Ebeam Iw,beam
tt
xxxx P ( x Vt ) P ( x Vt + d ) ) H ( a y )
2a xz ( x, y, 0, t ) = Ku ( x, 0, 0, t ) H ( a y )

(2.2)

yz ( x, y, 0, t ) = 0
where zz ( x, y, z , t ) is the normal stress, xz ( x, y, z , t ) yz ( x, y, z , t ) are the shear
stresses, wbeam ( x, t ) [ m ] is the vertical beam displacement, S beam [ kg m ] and
Ebeam I N m 2 are the mass per unit length and the bending stiffness of the beam,

d [ m ] is the distance between the loads, K N m 2 is the stiffness per unit length of
the shear springs at the beam - half-space interface, H (...) is the Heaviside step

function.

Compatibility condition for the vertical motion:


w ( x, 0, 0, t ) = wbeam ( x, t ) .

(2.3)

To analyse the system of equations (2.1)-(2.3) we use the Helmholtz


decomposition of the displacement vector:

u = + ,

i = 0 .

(2.4)

In terms of the potentials and the equations of the half-space motion read [3]:
cL2 2 =

with cL2 = + 2

2
,
t 2

cT2 2 =

2
,
t 2

(2.5)

, cT2 = .

The components of the half-space displacement vector and the stresses in the
half-space may be written in terms of these potentials as [3]:
u=

z y
z x
y x
, v=
, w=
,
+

+
+

x
y
z
y
x
z
z
x
y

zz = 2 + 2

y x

z x
y

(2.6)
(2.7)

2 z y y x
+

z x x
y
xz z y

(2.8)

2 z x y x

+
z y x
y
yz z x

(2.9)

xz = 2
yz = 2

We will solve the problem (2.1)-(2.3) by employing integral Fourier transforms with
respect to time t and horizontal co-ordinates x and y . These transforms are defined as

42

f ( k1 , k2 , z , ) =

g ( k , k , z, ) =

wbeam
, k1 ( , k1 ) =

( x, y, z , t ) exp ( i ( t k1 x k2 y ) ) dxdydt

( x, y, z , t ) exp ( i ( t k1 x k2 y ) ) dxdydt

(2.10)

w ( x, t ) exp ( i t ik x ) dxdt.
beam

Application of theses transforms to equations (2.5), (2.2) and (2.3) leads to the
following statement of the problem in the Fourier domain:

Equations of motion for the inner element of the half-space (from (2.4) and (2.5)):

2 f 2
+ 2 k12 k22 f = 0
2
z
cL

2 g 2
+ 2 k12 k22 g = 0 ,
2
z cT

(2.11)

g x g y g z
+
+
=0
x
y
z
with
cL =

c~L2 = c L2 i * + 2 * / ,

b + 2 g /

and

c~T2 = cT2 i * /

being

complex

values,

the velocity of the dilatational waves (P-waves) and cT = / the

velocity of the shear waves (S-waves).

Boundary conditions at the surface of the half-space z = 0 (from (2.2), using (2.6)(2.9)):

2ik1

f

sin k 2 a
+ ik 2 g z y + ik1 ( ik1 g y ik 2 g x ) = Ku , k1
z z
z
k2

f

ik1 g z y + ik 2 ( ik1 g y ik 2 g x ) = 0
2ik 2
z z
z

2
2

2 k12 k 22 f + 2 2 + ( ik1 g y ik 2 g x ) =

z
z
z

ik1d
= wbeam
)
, k1 D , k1 2 P ( k1V ) (1 + e

) sink ka a
2

(2.12)

where D ,k1 = S beam 2 + EIk14 is the dispersion relation for the vertical individual
vibration of the beam and

u ( x, 0, 0, t ) exp ( i ( t k x ) ) dxdt
= w ( x, 0, 0, t ) exp ( i ( t k x ) ) dxdt.

u ,k1 =

w , k1

For the compatibility condition (from (2.3)):

43

(2.13)

beam
, k1

1
( , k1 ) =
2

( k1 , k2 , 0, ) dk2 = w ,k

, k1 , k2

(2.14)

The general solutions of the first two equations of the system (2.11), accounting
for the proper behaviour for the large positive values of z , are

f = A ( k1 , k2 , z ) exp ( zRL )
g = B ( k1 , k2 , z ) exp ( zRT )

(2.15)

with RL,T = k12 + k22 2 / c~L2,T , Re ( RL,T ) > 0 .


Substituting (2.15) into the third equation of the system (2.11) and into the
boundary conditions (2.12), the following system of linear algebraic equations with
respect to unknowns A and B can be obtained:

2ik1 RL A
+

2ik2 RL A
+

2
2
2
2 k1 + k2 2 +
cT

(R

ik1 Bx

k1k2 Bx

(R

2
T

+ k22 Bx

2ik2 RT Bx

RT Bz

( ik2 RT ) Bz

( k1k2 ) By

ik1 RT Bz

( 2ik1RT ) By

ik2 By
2
T

k12

= Hx
=

(2.16)

= Hz

with

K u , k1 sin k2 a

H x =
k2 a

H = 1 sin k2 a D wbeam 2 P ( k V ) (1 + eik1d )


1

z k2 a ,k1 ,k

(2.17)

The system of algebraic equations (2.16) can be readily solved to give analytical
expressions for A and B . These expressions can be used to obtain the Fourier
displacements of the half-space surface in the x direction and in the z direction. In
accordance with the general representation (2.4) and solutions (2.16), these
displacements read
g y

u ,k1 ,k2 ( k1 , k2 , 0, ) = ik1 f + ik2 g z


= ik1 A + ik2 Bz + RT By
z z =0

.
(2.18)
f

w , k1 , k2 ( k1 , k2 , 0, ) = + ik1 g y ik2 g x = RL A + ik1 By ik2 Bx


z
z =0
Substitution of expressions for A and B that were obtained from system (2.16) into
(2.18) yields

44

u ,k1 ,k2 ( k1 , k2 , 0, ) = a11 H x + a13 H z


w , k1 ,k2 ( k1 , k2 , 0, ) = a31 H x + a33 H z

(2.19)

with
a11 =

1
2k12 RT2 ( RT2 + k22 ) q + 4k22 RL RT

RT 0

ik1
2 RL
a13 =
( q 2 RL RT ) , a31 = a13 , a33 = 2
cT 0
0

(2.20)

0 = q 2 4 ( k12 + k22 ) RL RT

q = k12 + k22 + RT2 ,

Applying the inverse Fourier transform with respect to k2 to equations (2.19),


substituting expressions (2.17), and setting y to zero we find

K
I13

ik1d
beam
u , k1 = 2 I11u , k1 + 2 D , k1 w , k 2 P ( k1V ) (1 + e )

w = K I u + I 33 D wbeam P ( k V ) (1 + e ik1d )
1
, k1 2 31 , k1 2 , k1 , k

with

I ij =

aij

sin k 2 a
dk 2 .
k2a

(2.21)

(2.22)

Using the compatibility condition (2.14), the system of equations (2.21) can be rewritten
in the following matrix form
K
2 I11 1

K
2 I 31

I13

D , k1
u , k
2
I
1 = P ( k1V ) (1 + e ik1d ) 13
I 33

w
I 33
D , k1 1 , k1
2

(2.23)

Introducing now the equivalent stiffness matrix that describes the dynamic stiffness
of the half space to a harmonic wave propagating in the beam, we finally obtain the
following matrix-equation:

( K + M + ) u ,k

=F

(2.24)

with K the stiffness matrix of the structure that overlies the half-space, M the mass
matrix of this structure, F the external force-vector, u ,k1 the displacement vector in
the Fourier domain. These quantities in the case at hand are given as

45

u ,k
u ,k1 = 1 ,
w , k1
K
K=
0

0
,
EIk14

,
F=
ik1d
2 P ( k1V ) (1 + e )
I11
2
0
0
1
M=
, =
2

m
0
I 31

I13
2
.
I 33

(2.25)

If the shear interaction between the beam and the half-space is disregarded, as it was
done by Dieterman and Metrikine [29], the stiffness matrix becomes a scalar and the
matrix equation (2.24) reduces to the following algebraic equation

( K 22 + M 22 + 22 ) w ,k

= F2 ,

(2.26)

which can be decoded as ( D ,k1 = S beam 2 + EIk14 )

2
ik1d
D , k1
w , k1 = 2 P ( k1V ) (1 + e ) .
I
33

(2.27)

The term 2 I 33 was referred to by Dieterman and Metrikine [29] as equivalent


stiffness of the half-space:
2
2 cT2
eq ( , k1 ) =
=
I 33
2

sin ( ak2 )
RL
dk2

( , k1 , k2 ) ak2

-1

(2.28)

The reason for such a reference is as follows. The term in the parenthesis on the lefthand side of equation (2.27) can be rewritten as
D , k1

2
= S beam 2 + EIk14 + eq ( , k1 ) .
I 33

(2.29)

Being set to zero, this term defines the dispersion equation for bending waves in the
beam on the half-space:
S beam 2 + EIk14 + eq ( , k1 ) = 0

(2.30)

Comparing the dispersion equation (2.30) to that for the beam on Winklers foundation,
which would look as
S beam 2 + EIk14 + 0 = 0,

(2.31)

with 0 the Winklers constant, it is easy to see that eq ( , k1 ) is nothing but the
(vertical) stiffness of a foundation. The reaction of this foundation is equivalent to the

46

reaction of the visco-elastic half-space. Thus, the equivalent 1D model for vibrations of
the beam on the visco-elastic half-space can be depicted as shown in Figure 2.3.

eq ( , k1 )

wbeam ( x, t )

Fig. 2.3. Equivalent model for the beam lying on the visco-elastic half-space.

In contrast to the Winklers constant, the equivalent stiffness eq ( , k1 ) is a


complex-valued function of the angular frequency vibration and wavenumber of waves
propagating in the beam.
The other elements of the matrix also describe the dynamic stiffness of the
half-space. As it will be seen from the following analyse, these elements are of
secondary importance for the vertical vibrations of the beam. Therefore, in what follows
the attention is focused on the properties of the vertical dynamic stiffness of the half
space eq ( , k1 ) .
To study eq ( , k1 ) , it is customary to introduce in Eq. (2.28) the following new
variables:

Variable of integration = k2 k1 which represents the ratio between the wave


numbers of waves propagating in the lateral and longitudinal directions;
Phase velocity of waves propagating along the beam V ph = k1 ;

Ratio of V ph and complex velocities of longitudinal and shear waves in the half-

space L ,T = V ph cL ,T .
In terms of these variables, expression for the equivalent stiffness (2.28) can be
rewritten as

eq ( , k1 ) = eq ( k1V ph , k1 ) = 2 cT2

ak1
1
,
2
V ph I ( k1V ph , k1 )

R ( ) sin ( ak1 )
I ( k1V ph , k1 ) = L
d ,
( )

with

( ) = 2 (1 + 2 ) T2

Re ( RL ,T ( ) ) > 0 .

4 RL ( ) RT ( ) (1 + 2 ) ,

(2.32)

RL ,T ( ) = 1 + 2 L2,T ,

The equivalent stiffness in (2.32) is represented as a function of the phase


velocity and wavenumber of waves in the beam. Introduction of the phase velocity

47

instead of the frequency is convenient in the light of the following analysis of the forced
response of the beam. Indeed, in the steady-state regime, a constant load that uniformly
moves on the beam perturbs in the latter waves, whose phase velocity is equal to the
load velocity. Therefore, analysing the dependence of the equivalent stiffness upon the
phase velocity of waves in the beam we can think in terms of dependence of the system
response on the load velocity.
Thanks to the viscosity in the half-space, the integrand in (2.32) has no
singularities and tends to zero at quite fast ( 1 4 ) . Therefore, integration of
(2.32) can be carried out numerically with the help of any standard integration routine.

Re ( eq *108 ) , N m 2

Im ( eq *108 ) , N m 2

V ph cT

k1

V ph cT

k1

a)

b)

Fig. 2.4. Equivalent stiffness of the half-space versus the phase velocity and wave number of waves in the
beam: a) real part, b) imaginary part.

1.5

eq ( k1V ph , k1 ) *10 8 , N m 2

1.0

Re ( eq )

0.5
V ph / cT

0.0
1

-0.5
cR / cT

-1.0

Im ( eq )

-1.5

Fig. 2.5. The equivalent stiffness of the half-space as a function of phase velocity at fixed value of k1 .

48

The results of numerical analysis of expression (2.32) are represented in Figures


2.4 (a,b) and 2.5. Figures 2.4(a,b) show the real and imaginary parts of the equivalent
stiffness versus the phase velocity and wave number. In Figure 2.5, cross-sections of the
previous figures by the vertical plane k1 = 0.5 m -1 are depicted.
The calculations were carried out for the following parameters of the system:

= 2 107 N m 2 , * = 104 N*s m 2 , = 1960 kg m3 , = 0.3 , 2a = 2.6 m , (2.33)


where is the Poissons ratio of the half-space (corresponding ratio between the
Rayleigh wave and the shear wave speeds is cR 0.927cT ).
Figure 2.5 shows clearly that the real part of the equivalent stiffness has a
minimum at a certain velocity V ph* . This velocity depends on the material damping in
the half-space. If this damping were absent, this velocity would be exactly equal to the
Rayleigh wave velocity. The higher the damping, the larger is V ph* . However, for
realistic values of the damping, this velocity always belongs to the interval
cR < V ph* < cT .
The imaginary part of the equivalent stiffness, being always smaller than zero
(for the parameters under consideration), rapidly grows as V ph becomes larger than V ph* .
To understand such a behaviour, the physical sense of the imaginary part of the
equivalent stiffness eq should be clarified. Whilst the real part of eq is responsible for
the elastic and inertial properties of the beam foundation, the imaginary part of eq
reflects the viscous properties of this foundation. This equivalent viscosity may be
caused by either material damping in the half-space or by so-called radiation damping
that is associated with waves in the half space. These waves, being perturbed, transfer
energy away from the beam, thereby consuming a part of vibrational energy of the
beam. Now, we can easily explain why the imaginary part of eq exhibits a sharp
increase as soon as the phase velocity exceeds V ph* . Indeed, as long as V < V ph* , the
waves in the beam do not perturb waves in the half space, and, therefore, the imaginary
part of eq is caused by the material damping only. As soon as the phase velocity of
waves in the beam exceeds V ph* , waves are getting generated in the half-space, leading

to the sharp growth of Im ( eq ) . In correspondence with the aforementioned reasoning,


the imaginary part for V < V ph* would be equal to zero if the material damping were
absent, see [29].
It is worth noting that generation of waves in the half-space can lead to
instability of vibrations of a mechanical object moving along the beam, see Metrikine
and Popp [115]. This instability may arise in the case that the object moves with a
velocity larger that the minimum phase velocity of waves in the beam half-space
structure. This velocity is in the order of the Rayleigh wave velocity for a realistic
railway track. Thus, if the track is laid on a soft soil, the instability can be a real danger.
Fortunately, by choosing properly parameters of the train bogies, this instability can be
avoided. In this development it is assumed that such a choice has been made and the
vibrations of a moving train are stable.
Thus, we have obtained and studied the expression for the equivalent stiffness of
the visco-elastic half-space. Introduction of this stiffness reduces the original 3D

49

problem to a 1D problem. Hereafter we will profit from this reduction while calculating
the beam deflections under a uniformly moving load. Note again that the equivalent
stiffness is a complex-valued function that depends on both the frequency and
wavenumber of waves in the beam. This is in contrast to the idea of Timoshenko that
the ground can be replaced by a visco-elastic foundation with constant stiffness and
damping coefficient.

2.2. Steady-state response of the beam on the half-space to a


moving load.
Train wheels act on the rails with a force that has a complicated spectrum. A
capital amount of energy in this spectrum, however, belongs to the constant constituent,
which is conditioned by the train weight. In connection with this fact, it is important to
know
a) the steady-state deflection of the track due to the constant part of the loading;
b) critical (resonance) velocities of the train that are related to this loading.
This section treats both aforementioned issues.
To calculate the vertical beam deflection, we return to the system of equations
(2.21). By introducing the following notations
b11 = 1

K
I11 = 1 + K 111 ,
2

K
b21 =
I13 =K 211 ,
2

b12 =

D , k1

I13 = 121 D , k1 ,

2
D ,k1
1
,
b22 = 1
I 33 = 1 + D ,k1 22
2

(2.34)

this system may be rewritten as


b11u , k + b12 w , k = 2 P 121 ( k1V ) (1 + e ik1d )
1
1

ik d
1
b21u , k1 + b22 w , k1 = 2 P 22 ( k1V ) (1 + e 1 )

(2.35)

Solving (2.35) with respect to w , k1 , which in accordance with the compatibility


condition (2.14) is equal to wbeam
, k1 , and then applying the inverse Fourier transforms with
respect to and k1 , we obtain the following expression for the vertical displacement of
the beam:

50

wbeam ( x, t ) =
=

P
2

P
=
2

1
4 2

w ( k , ) exp ( i t + ik x ) d dk
, k1

b11 b21 121


b11b22 b21b12 ( k1V ) (1 + exp ( ik1d ) ) exp ( i t + ik1 x ) d dk1 = (2.36)

1
22

1
b11 22
b21 121
exp ( ik1 ( x Vt ) ) (1 + exp ( ik1d ) ) dk1
b11b22 b21b12

=k V
1

Let us first consider the case that only one load moves along the beam.
Assuming that the load is positioned at x = Vt , we can formally neglect the term
exp ( ik1d ) in (2.36) to get the following expression that describes the beam response to
a single load:
beam
single load

P
( x, t ) =
2

1
b11 22
b21 121
exp ( ik1 ( x Vt ) ) dk1
b11b22 b21b12

=k V

(2.37)

beam
First, we will study the beam displacement at the loading point wsingle
load (Vt , t )
beam
as a function of the load velocity. In Figures 2.6 (a,b) and 2.7 (a,b), wsingle
load (Vt , t ) is

plotted versus the load velocity for four different values of the material damping in the
half-space. The solid lines in the graphs correspond to the case K = 0 (smooth shear
contact between the beam and the half-space), whilst the dashed curves correspond to
the case K = 108 N m . The following parameters of the beam and half-space were used
in calculations:

= 2 107 N m 2 , = 1960 kg m3 , = 0.3, cR = 337 km h ,


Abeam = 760 kg/m, Ebeam J = 1.29*107 N*m 2 , 2a = 2.6 m,

P = 2.1 105 N

(2.38)

The vertical dashed line in the figures points out the value of the Rayleigh wave
velocity in the half-space.
Analysing the figures, the following conclusions may be drawn:
1. There is only one critical velocity of the load. This velocity is close to the Rayleigh
wave speed as long as the material damping is small. The reason for this is that the
equivalent stiffness of the half-space has a pronounced minimum as V ph cR . The
elasticity and inertia of the beam do not influence the critical speed perceptibly. The
latter is only true for the chosen parameters of the beam, which are realistic for
railway tracks. If the beam parameters were chosen differently, in no relation to the
railway tracks, the beam could influence the critical velocity significantly.
2. The larger the material damping in the half-space, the smaller is the displacement at
the loading point. For high (not realistic) values of the material damping the critical
velocity disappears.
3. The effect of the shear stiffness K is negligible for the chosen parameters of the
system. Basically, the shear interaction leads to a slight decrease of the beam
deflection in the sub-critical case and to a slight increase of that in the super-critical
case.

51

0.006

beam
wsingle
load

x =Vt

0.006

K =0

,m

K = 10 N m
8

beam
wsingle
load

K =0

,m

x =Vt

K = 108 N m 2

0.005

0.005

0.004

0.004

0.003

0.003

0.002

0.002

0.001

0.001

0
0

0.2

0.4

0.6

0.8

1.2

1.4

1.6

0.0

0.2

0.4

0.6

V cT

0.8

1.0

1.2

1.4

1.6

V cT

a)

b)

Fig. 2.6. Beam deflection at the loading point versus the load velocity for (a) * = 103 N*s m 2 and
(b) * = 104 N*s m 2 .

0.006

beam
wsingle
load

x =Vt

0.006

K =0

,m

K = 10 N m
8

beam
wsingle
load

0.005

0.005

0.004

0.004

0.003

0.003

0.002

0.002

0.001

0.001

x =Vt

K =0

,m

K = 108 N m 2

0
0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

V cT

0.0

0.2

0.4

0.6

0.8
V cT

1.0

1.2

1.4

1.6

Fig. 2.7. Beam deflection at the loading point versus the load velocity for (a) * = 8 104 N*s m 2 and
(b) * = 106 N*s m 2 .

It should be mentioned that the displacement under the load is not always the
maximum displacement of the beam. It is clearly seen from Figures 2.8-2.10, where the
beam deflection is depicted versus the distance = x Vt from the loading point for
three different values of the load velocity. Figure 2.8 is plotted for
V = 50 km h (V = 0.15 cR ) and shows the beam pattern in the sub-critical case. Figure
2.9 shows the beam pattern in near-critical case in which the load velocity is

52

V = 340 km h (V = 1.008 cR ) . Figure 2.10 shows the beam pattern in the super-critical

case V = 500 km h (V = 1.48 cR ) .


The load in Figures 2.8-2.10 is located at the point = 0 . Figures () are plotted
for K = 0 (smooth shear contact), whilst for Figures (b) K was taken equal 108 N m .
The solid lines in the figures correspond to a small material damping 1* = 103 N*s m 2 ,
whereas the dashed lines are related to a moderately high damping
2* = 8 104 N*s m 2 .

-0.001

-0.001

-0.002

-0.002

-0.003

-0.003

1
2

-0.004

-0.004

-0.005

-0.005
-4

-3

-2

-1

-4

-3

-2

-1

a)

b)

Fig. 2.8. Steady-state deflection of the beam in the case V = 50 km h for (a) K = 0 and (b)
K = 108 N m 2 .

-0.001

-0.001

-0.002

-0.002

-0.003

-0.003
1

-0.004

-0.004

-0.005

-0.005

-10

-8

-6

-4

-2

10

-10

a)

-8

-6

-4

-2

b)

Fig. 2.9. Steady-state deflection of the beam in the case V = 340 km h for (a) K = 0 and (b)
K = 108 N m 2 .

53

10

-0.0004

-0.0004

-0.0008

-0.0008

-0.0012

-0.0012

-0.0016

-0.0016

-0.002

-0.002

-10

-8

-6

-4

-2

10

-10

a)

-8

-6

-4

-2

10

b)

Fig. 2.10. Steady-state deflection of the beam in the case V = 500 km h for (a) K = 0 and
(b) K = 108 N m 2 .

Figures 2.8 shows that if the load velocity is small, the beam deflection is
localised in the vicinity of the loading point and is almost symmetric with respect to the
loading point the load. The small asymmetry of the profile regards the presence of
material damping in the half-space. The variation of this damping and the account for
the shear stiffness of the interface influences the result very slightly.
If the load velocity is close to the critical velocity, see Figure 2.9, the beam
pattern becomes visible asymmetric with respect to the loading point. This is a
consequence of radiation of waves into the beam and the half-space. Because of the
Doppler effect, the wavelength of waves behind the load is larger that that in front of the
load. Accordingly, the attenuation of waves in the beam caused by the material damping
and radiation of waves into the half-space is different for waves propagating behind and
in front of the load.
In the super-critical case that is depicted in Figure 2.10, the beam pattern is
similar to that in the near-critical case. The difference between these two cases contains
in a much smaller beam deflection under the load in the super critical case and in a
slight change of the wavelength of the perturbed waves. The latter is a direct
consequence of the Doppler effect.
Finalising this section, we would like to emphasise its main result: the critical
velocity of the train conditioned by its dead weight (constant load) is approximately
equal to the Rayleigh wave speed in the ground surrounding the railway track. The
value of this velocity can be in the order of 200-500 km/h, if the ground is soft. Thus,
nowadays-operated high-speed trains can easily reach the critical speed and a careful
study is necessary on what would happen if such a regime of motion were to take place.
To accomplish such a study, a model should be employed that includes a threedimensional solid as a model for the ground.

54

2.3. Elastic drag experienced by a high-speed train due to


excitation of ground vibrations.
In Chapter 1 we have introduced the term elastic drag, explained its relevance,
and studied this drag in the case of a single load moving along an elastically supported
membrane and a visco-elastic half-space. These studies, of course, have an academic
value, only.
In this section, we study the drag, which would be experienced by a high-speed
train, if the railway structure could be modelled by a beam on a visco-elastic half-space,
as it is introduced in sections 2.1 and 2.2. This drag reflects the energy loss of the trains
engine because of excitation of vibrations and waves in the railway track structure. To
get confidence in that this energy loss can be substantial, let us draw a parallel between
a super-critically moving high-speed train and a super-sonic plane. It is well known that
the supersonic motion of a plane results in very high fuel consumption. This happens
because of radiation of sound waves into the Mach cone forming behind the plane as it
flies supersonically. Drawing an analogy, one may suppose that phenomenon of the
same kind would arise if a high-speed train travelled with a velocity larger than the
Rayleigh wave speed cR . In this case, the train would be accompanied by powerful
surface waves propagating within a Mach angle the edges of which are determined by
the Rayleigh wave speed ([97], [164]). These waves might well consume a considerable
energy, especially if the train speed is close to the Rayleigh wave speed. That is why the
analysis of the elastic drag that may be experienced by a high-speed train seems to be of
high significance.
By definition, the energy per unit time Q that is inserted into an elastic system
by a point load (the power input) is equal to the scalar product of the load magnitude
and the resulting velocity of the loading point. In accordance with this definition, for the
model under consideration (see Eqs.(2.1)-(2.3)), Q reads
wbeam
Q = P
t

+
x =Vt

wbeam
t

.
x =Vt d

(2.39)

The elastic drag De , by definition, is the ratio between the power input Q and
the load velocity V :
De =

Q P wbeam
=
V V t

+
x =Vt

wbeam
t

x =Vt d

(2.40)

Note that this definition is valid only if the load is constant, e.g. P is independent of
time.
Expression (2.40) shows that to find De one has to determine the beam
displacement, which is given by (2.36). Substituting this expression into Eq.(2.40), we
obtain the following representation for the drag:
iP 2
De =
2

k1

1
b11 22
b21 121
(1 + cos ( k1d ) ) dk1 .
b11b22 b21b12 = k V
1

55

(2.41)

Expression (2.41) shows that the drag De depends not only on the magnitude of
the loads P and on their velocity but on the distance d between the loads also. It is not
surprising since the individual elastic fields generated by each load are interfering and,
therefore, the energy of the total field must be influenced by the spatial shift between
the individual fields.
The further investigation will proceed as follows. First, we consider the viscoelastic drag due to a single load to clarify the effect of the damping coefficient * , of
the velocity V and of the shear stiffness K at the beam-half-space interface. Next, the
elastic drag experienced by two loads will be studied with emphasis on the effect of the
distance between the loads. Employing results of this study, finally, the elastic drag, the
rolling drag and the aerodynamic drag will be compared for a TGV train.
The expression for the elastic drag faced by a single load can be retrieved from
(2.41) by removing the term cos ( k1d ) . This yields
wheel
e

iP 2
= De =
2

b11 221 b21 121


dk1
b11b22 b21b12 = k V

k1

(2.42)

Evaluation of expression (2.42) can be carried out numerically. For this evaluation, the
following set of the system parameters was used:
E = 5.2*107 N/m 2 , = 0.3, = 1.96*103 kg/m3 , cR = 340 km h
Abeam = 760 kg/m, Ebeam I = 1.29*108 N/m 2 , 2a = 2.6 m P = 105 N

200

Dewheel ( N )

200

160

Dewheel ( N )

1
2

(2.43)

160

120

120

80

80

40

40

0
0

200

V ( km h )

400

600

a)

200

400

V ( km h )

b)

Fig. 2.11. Visco-elastic drag versus the load velocity for (a) K = 0 and (b) K = 108 N m 2 .

56

600

In this set, the half-space parameters describe a realistic, though arbitrary soft
ground, the beam parameters represent a system composed of two rails and sleepers,
and the load is taken to describe the axle loading (one quarter of the passenger car
weight).
Results of numerical integration of equation (2.42) are presented in Figure 2.11
(a,b) that shows the dependence of the elastic drag Dewheel on the load velocity V . Both
figures present four curves that are plotted for different values of the damping
coefficient * , namely for 1* = 103 Ns/m 2 , 2* = 104 Ns/m 2 , 3* = 0.8*105 Ns/m 2 ,

4* = 106 Ns/m 2 . The difference between figures (a) and (b) is that the first one is plotted

for K = 0 , i.e. in the absence of the shear contact springs, while the second presents
results calculated for K = 108 N m .
Analysing the figures, the following conclusions may be drawn:

1. For realistic values of the damping coefficient, the elastic drag grows significantly
as the train approaches the Rayleigh wave velocity.
2. Below the Rayleigh wave velocity, the only cause for the elastic drag is the material
damping in the subsoil. In this range of velocities, the smaller the damping
coefficient, the smaller is the elastic drag.
3. At velocities higher than the Rayleigh wave velocity, the elastic drag is caused both
by the damping and by the radiation damping that is related to the energy loss on
excitation of elastic waves in the beam and the half-space. When the material
damping is small, the radiation damping plays a dominant role in causing the drag.
4. Concerning the effect of the shear interaction between the beam and the half-space
that is taken into account by the shear springs K , one can see that this effect is
noticeable when the damping is relatively small and the load velocity is close to the
Rayleigh wave velocity. Otherwise, the shear springs have almost no influence on
the drag.
The final objective of this section is to calculate the total elastic drag that is
experienced by a high-speed train. To accomplish this, one may not just multiply elastic
drag obtained for a single wheel by the number of wheels in the train. Such approach
would give a wrong result, for every wheel is dragged not only by its own elastic field
but also by elastic fields perturbed by other wheels. Mathematically, the effect of a
neighbouring wheel on the visco-elastic drag is presented by cos ( k1d ) in the expression
(2.41) for the drag experienced by two loads.
To explore the effect of this cross-drag, we numerically integrate equation
(2.41) by employing the parameter set (2.43) and K = 108 N m 2 . The result is shown in
Figure 2.12(a) and Figure 2.12(b). Each figure presents two curves: the solid curve
corresponds to the supercritical motion with V = 400 km h whereas the dashed curve
corresponds to the subcritical motion with V = 200 km h . The figures differ due to the
damping coefficient used in the calculations Figure 2.12(a) is plotted for
* = 104 Ns/m 2 , while Figure 2.12(b) depicts results obtained with
* = 0.8*105 Ns/m 2 .

57

1000

Debogie ( N )

1000

V = 400 km h

Debogie ( N )

V = 400 km h

V = 200 km h

V = 200 km h

100

100

10

10

1
0

d (m)

10

12

a)

d (m)

10

12

b)

Fig. 2.12. Visco-elastic drag versus distance between two loads for (a) * = 104 * 2 and
(b) * = 8 104 * 2 .

The figures demonstrate the following:


1. Elastic drag in the case of supercritical motion is substantially larger than that in the
case of subcritical motion. The difference, however, decreases when the material
damping becomes larger.
2. Elastic drag depends upon the distance between the loads if this distance is
relatively small: d < 5 6 m . For larger distances, the drag remains unchanged and
could be calculated by simple doubling of the drag calculated for the single load.
3. The dependence of the elastic drag upon the distance has a minimum located
approximately at d = 2 m . This implies that the cross-drag can be negative, i.e. it
can push the load set forward. We would like to note that the existence of this
minimum might be used in practice, since by slightly varying the bogie wheelbase
one can reduce the drag.
Based upon the information retrieved from Figure 2.12, one can calculate the
total elastic drag to the high-speed train motion. Indeed, since the distance between the
bogies of a train is larger than 6 m , the contribution of each bogie may be calculated
separately and then multiplied by the number of bogies on the train
Let us consider a TGV train consisting of two power cars and eight passenger
cars. We use this example; for there is information available on the rolling drag and the
aerodynamic drag that such a train experiences, see [61]. According to this paper, the
weight of the power car is about 80000 kg , the weight of the passenger car is 40000 kg
and the bogie wheelbase is 3 m . Thus, we have four heavy bogies with the axle
loading of 20000 kg and sixteen light bogies with the axle loading 10000 kg . Taking
the parameter set corresponding to Figure 2.12(a) and, as an example, choosing
supercritical motion with V = 400 km h , one obtains the following value for the total
visco-elastic drag:

58

Detotal = 4 Debogie
= 4 4 Debogie
= 32 Debogie

d = 3m, P = 200kN

d = 3m, P =100kN

d = 3m, P =100kN

+ 16 Debogie

+ 16 Debogie

d = 3m, P =100kN

d = 3m, P =100kN

(2.44)

32 120 N=3.84 kN

Evaluating equation (2.44), we have taken into account that the elastic drag is
P2
(see,
for
example
(2.41))
so
that
proportional
to
bogie
bogie
De
= 4 De
. Additionally we have used Figure 2.12(a),
d =3m, P = 200kN

d =3m, P =100kN

bogie
e
d =3m, P =100 kN

which shows that D

is approximately equal to 120 N .

In the same manner the total drag may be calculated for any value of * and V .
The result of this calculation is depicted in Figure 2.13 that shows the velocity
dependence of the total elastic drag for * = 104 Ns/m 2 (bold dashed line) and for
* = 0.8*105 Ns/m 2 (bold solid line). For comparison, the aerodynamic drag, and the
rolling drag are plotted in accordance with results presented in [61].

D ( kN )

100

10

Da
Dr

De(1)
De(2)

0.1

0.01
0

200

400

600

V ( km h )
Fig. 2.13. Visco-elastic drag for * = 104 */ 2

( D ) , visco-elastic drag for


(1)
e

* = 0.8 105 */ 2 ( De(2) ) , rolling drag ( Dr ) and aerodynamic drag ( Da ) versus the load velocity.

Figure 2.13 shows that the largest drag that the train experiences while moving
with a relatively high velocity ( > 200 km h ) is the aerodynamic drag. The rolling drag
and the elastic drag are much smaller. For example, if the train moved with
V = 400 km h , then the aerodynamic would be about 80 kN , the rolling drag would be

59

in the order of 16 kN and the elastic drag would be 3.84 kN . Thus, as one can see, the
elastic drag at this velocity is about 24% of the rolling drag and about 5% of the
aerodynamic drag. In the case of subcritical motion with V = 200 km h , these values
would respectively turn into 1.5% and 0.5% for * = 104 Ns/m 2 and into 10% and 3.8
% for * = 0.8 *105 Ns/m 2 .
Thus, the elastic drag is very small with respect to the aerodynamic drag. With
respect to the rolling drag, the elastic drag is not that small and the higher the velocity
of the train, the larger is the relative contribution of the elastic drag.
We should keep in mind, however, that in reality the contact force between the
train wheels and the rails is not given by the dead weight alone but also depends on the
aerodynamic resistance to the train motion. Thanks to the special form of high-speed
trains, this resistance presses the train, especially the first wagon (locomotive) against
the rails. In a rough way, this pressure can be accounted for by the following
modification of formula (2.41):
De =

iP 2 (1 + V 2 )
2

k1

1
b11 22
b21 121
(1 + cos ( k1d ) ) dk1 ,
b11b22 b21b12 = k V

(2.45)

where is a coefficient, which accounts for the fact that the aerodynamic drag
influences the contact force between the train wheels and the rails. This coefficient
depends on the shape of high-speed train. Unfortunately, we are not armed with
information about what the coefficients could be for the nowadays-operated highspeed trains. Thus, we will estimate the elastic drag for some sensible values of .

D ( kN )

1000

100

10

Da
Dr

0.1

De(1)
De(2)

0.01

0.001
0

200

V ( km/h )

400

600

Fig. 2.14. Effect of the aerodynamic pressure on the elastic drag (to be compared with Figure 2.13).

60

Figure 2.14 represents the velocity dependence of the elastic drag for two
different values of the material damping in the half-space, according to expression
(2.45). The aerodynamic and rolling drag are plotted as well for comparison. Since it is
evident that the highest aerodynamic drag falls on the locomotive, the wagons of the
train were disregarded in calculation of the elastic drag. The coefficient has been
taken as = 0.001 .
Figure 2.14 shows that the effect of the aerodynamic pressure on the elastic drag
is truly crucial. The elastic drag becomes comparable and, starting from the velocities
slightly smaller than the Rayleigh wave velocity, even exceeds the aerodynamic drag.
Besides, the larger is the material damping, the earlier the elastic drag exceeds the
aerodynamic drag.
Thus, in our opinion, one of the design criteria for high-speed trains, should be
the energy loss of the train because of all possible types of resistance to the train
motion. The elastic drag, assessed in this section, should be also included in this
criterion as a source of perceptible energy loss.

61

CHAPTER 3
Beam on a Visco-elastic Stratified Half-space as a
Continual Model for a Railway Track
In this chapter, the model for a railway track that was considered in the previous
chapter is improved by accounting for
a)
b)

stratification of the ground;


visco-elastic and inertial properties of the rail-supporting structure that overlies the
ground.

Having these improvements implemented, the model becomes capable of describing the low-frequency dynamics of a conventional railway track. A higher frequency dynamics cannot be handled accurately, since the model does not take into account the discrete character of the sleeper-ballast interaction. The model under consideration can be also employed to study dynamics of modern embedded tracks that are
truly continuous, being built without the sleepers.
The main objective of this chapter is to study the effect of ground stratification
on the dynamic response of a railway track.
The chapter consists of three sections. In section 3.1, a model is formulated that
accounts for a three-dimensional layered ground and a continualized rail-sleeper-ballast
structure. In section 3.2, the method of dynamic flexibility matrix for layered ground is
presented. Examples of this matrix for various layered grounds are adduced. Then, the
equivalent stiffness of a layered ground is analysed. First, a visco-elastic layer, fixed at
its bottom, is considered. Second, a layer that overlies a half-space is investigated paying particular attention to the effect of the relative stiffness of the layer and the halfspace.
In section 3.3, the steady-state response of a beam overlying a layered ground to
a uniformly moving load is studied. Attention is focused on the study of: a) the critical
velocities of the load; b) the effect of damping in the ground on the structural response;
c) the influence of the upper ground layer depth on the structural response.
In this chapter, results of works by X. Sheng et al. [136]-[137] have been used.
Original results of the chapter are the analysis of the equivalent stiffness of different
layered grounds and the study of critical velocities of the load that are introduced by the
ground stratification.

3.1. A three-dimensional continual model for a railway


track.
To model a railway track successfully, the track structure should be considered
in more detail in comparison to how it was done in the previous chapter. First, to model

62

the ground realistically, a variation of the ground properties over the depth has to be accounted for. Second, the rail-sleeper-ballast structure is to undergo a more detail description.
In this section, we introduce a model that consists of a layered half-space and a
beam that is mounted to its surface trough a distributed visco-elastic, inertial elements,
see Figure 3.1.

2a

x, u
y, v
z, w

Fig. 3.1. Beam on a layered half-space as a continual model for the embedded track.

A vertical cross-section of the model by the plane y = 0 is plotted in Figure 3.2.


This Figure shows a system of continuous elements that connects the beam to the layered half-space.

P
Rails

kB
Ballast
Ground surface

wbeam

Pads
kP
Sleepers mS

B
mB

z=0 x
Ground layers
Elastic half-space

Fig. 3.2. Vertical cross-section of the system by plane y = 0 .

In Figure 3.2, the visco-elastic element with the stiffness per unit length
k P N m 2 and the viscosity per unit length P N s m 2 is a continualized representation of the pads. The distributed mass-element mS [ kg m ] represents the inertial prop-

63

erty of the sleepers. The visco-elastic, inertial element with the stiffness per unit length
k B N m 2 , viscosity per unit length B N s m 2 and mass per unit length
mB [ kg m ] represents the ballast.

The interaction between the beam, the distributed elements and the surface of
the layered half-space is assumed to take place in the vertical direction only. This simplification is accepted on the basis of the previous chapter, in which it was shown that
the shear interaction between the beam and the half-space has no significant influence
on the track response.
As for the layered half-space, we assume that it consists of n layers. The n -th
layer overlies a homogeneous half-space, which is termed the layer number ( n + 1) . The
material constants of layer j are the shear modulus, j , Poissons ratio, j , the density, j , the damping coefficient, *j , and the thickness, h j . For example, for ( n + 1) -th
layer, which is the half-space, its material constants are n +1 , n +1 , n +1 and n*+1

( hn+1 = ) .

Employing the aforementioned assumptions and notations, the governing equations for the model can be written as

Equations of motion for the layers (ground):

2u j
j
j

j u + j + j iu = j 2 ,
t

) (

(3.1)

where u j ( x, y , z , t ) = u j ( x, y , z , t ) , v j ( x, y , z , t ) , w j ( x, y , z , t ) [ m ] is the displacement vector for the layer number j , j kg m3 is its density, j = j + *j / t and
j = j + *j / t are operators, which are used instead of the Lams constants

j N m 2 and j N m 2 to describe the visco-elastic properties of the layer according to Voigts phenomenological model.

Equations of motion for the components of the ground-overlying structure:


for the beam (rails):
Ebeam I

4 wbeam
2 wbeam

+
+ k P + P ( wbeam wS ) = P0 ( x Vt )

S
beam
4
2
x
t
t

(3.2)

for the mass-spring element (sleepers and pads):


mS

2 wS

1
k P + P ( wbeam wS ) + F ( ) = 0
2
t
t

(3.3)

for the mass-spring-dashpot element (ballast) by using the consistent mass approximation, see [136]:

64

(1)
1 1 wS F
mB 2 1 2 wS t 2
k
+
+

B B

=0
6 1 2 2 wB t 2
t 1 1 wB F ( 2)

(3.4)

with wbeam ( x, t ) [ m ] , wS ( x, t ) [ m ] and wB ( x, t ) [ m ] the vertical displacements of the


rails, sleepers and ballast, respectively, Ebeam I N m 2 the beam bending stiffness,
S beam [ kg m ] its mass per unit length, F ( ) and F (
1

2)

the forces between the sleepers

and ballast, and in the contact ballast-ground, respectively.

Balance of forces at the surface z = 0 :

( x, t ) H ( a y )
xz ( x, y, 0, t ) = yz ( x, y, 0, t ) = 0
2a zz ( x, y, 0, t ) = F (

2)

(3.5)

with zz ( x, y, z , t ) the normal stress, xz ( x, y, z , t ) and yz ( x, y, z , t ) the shear


stresses, H (...) the Heaviside step-function.

Compatibility condition at the surface z = 0 :


w1 ( x, 0, 0, t ) = wB ( x, t ) .

(3.6)

Equations (3.1)-(3.6) govern the steady-state dynamic response of the system depicted
in Figure 3.1. Since the ground is modelled by a multi-layered half-space, to solve the
problem, it is customary to employ the method of the dynamic stiffness matrix. This
method, in application to the problem in question, is described in the next section.

3.2. The method of the dynamic flexibility matrix.


Suppose that a point harmonic load F0 e ( i t ) ( x ) ( y ) with F0 [ N ] being the
unit vector is applied to the surface of the layered half-space. The steady-state amplitudes of displacement of an arbitrary surface point ( x, y, 0 ) can be described by the matrix
11 12 13
( x, y, ) = 21 22 23 ,
31 32 33

(3.7)

which is called the dynamic flexibility matrix of the ground surface. In this matrix, 11 ,
12 , and 13 are the amplitudes of displacements that are caused by the x component
of the loading vector, 21 , 22 , and 23 are that caused by the y component and 31 ,

65

32 , and 33 are the amplitudes caused by the z component. In general, the dynamic
flexibility matrix is complex-valued.
Assume now that, on the surface of the ground, the harmonic load distributions
p x ( x, y ) e i t , p y ( x, y ) e i t , p z ( x, y ) e i t ,

(3.8)

are applied in x , y and z directions, respectively. The total steady-state vibration amplitudes of the surface point ( x, y, 0 ) in x , y and z directions can be then expressed as
u1 ( x, y, 0, )
p x ( r, s )
1

v ( x, y, 0, ) = ( x r , y s ) p ( r , s ) drds
w1 ( x, y, 0, )
p z ( r, s )

(3.9)

Equation (3.9) is a convolution integral. Applying to (3.9) the integral Fourier transform
with respect to the horizontal co-ordinates x and y (throughout this thesis, the sign contention for the Fourier transforms is the defined by Eq. (2.32)), one obtains the following relation:
u1 ,k1 ,k2 ( k1 , k2 , 0, )
px ,k1 ,k2 ( k1 , k2 )
1

v ,k1 , k2 ( k1 , k2 , 0, ) = , k1 ,k2 ( k1 , k2 , ) p ,k1 ,k2 ( k1 , k2 )


w1

pz

, k1 , k2 ( k1 , k2 , 0, )
,k1 ,k2 ( k1 , k2 )

(3.10)

The matrix ,k1 ,k2 ( k1 , k2 ) is the Fourier image of the flexibility matrix.
Denote the displacements for j -th layer in the frequency domain at point

( x, y, z ) ( z 0, h ) as u ( x, y, z ) e , v ( x, y, z ) e and w ( x, y, z ) e ,
where u ( x, y, z ) , v ( x, y, z ) and w ( x, y, z ) are complex functions. Their Fourier
images in the domain of wave numbers k and k we denote as u
(k , k , z ) ,
v
( k , k , z ) and w ( k , k , z ) , respectively. Introduce
i t

i t

i t

, k1 , k2

, k1 , k2

, k1 , k2

{u

} = {u
} = {u
t

, k1 , k2

{u

, k1 , k2

, k1 , k2
j

, k1 , k2

(k , k , z ), v

(k , k , z ),v

t
j

b
j

, k1 , k2

, k1 , k2

(k , k , z ), w

(k , k , z ), w

t
j

b
j

, k1 , k2

, k1 , k2

( k , k , z )}
1

t
j

( k , k , z )}
1

b
j

(3.11)
,

where {u , k1 ,k2 } stands for the Fourier displacement vector of the top interface of the
t

j -th layer, and {u , k1 ,k2 } for the displacement vector of the bottom interface of this
b
j

layer, z z = h j .
b
j

t
j

The three components of stresses in x, y, z directions at the top of j th layer we

t
denote as xzj
( x, y, z tj ) eit , yzjt ( x, y, z tj ) eit , zzjt ( x, y, z tj ) eit , and at the bottom as

66

b
xzj
( x, y, z bj ) eit , yzjb ( x, y, z bj ) eit , zzjb ( x, y, z bj ) eit . The Fourier images of the stress

vectors at the top and the bottom of the j th layer are then defined as

{ } j = { xzjt ( k1 , k2 , z tj ) , yzjt ( k1 , k2 , z tj ) , zzjt ( k1 , k2 , z tj )}

{ } j = xzjb ( k1 , k2 , z bj ) , yzjb ( k1 , k2 , z bj ) , zzjb ( k1 , k2 , z bj ) .


b

(3.12)

Finally, we introduce the following vectors that combine the Fourier images of displacements and stresses:

{s} j
t

t
b
{u

{u

b
, k1 ,k2 } j
, k1 ,k2 } j
=
, {s} j =
.
t
b
{ } j
{ } j

(3.13)

Our objective is to express vector { s }n , which is related to the lowest boundary


b

of the layered ground, through the vector { s }1 corresponding to the ground surface. For
t

this purpose we apply Helmholtz decomposition of the displacement vectors:


u j = j + j ,

i j = 0

(3.14)

In terms of the potentials j and j , the equations of motion for the layers
(3.1) in the frequency domain read
j =

cLj2 = j + 2 j

where

cLj =

2
c

2
Lj

j , j =

j , cTj2 = j j ,

2
2
Tj

j , i j = 0

j = i *j ,

(3.15)

j = i *j ,

with

+ 2 j ) / j the speed of the dilatational (P) waves and cTj = j / j the

velocity of shear (S) waves in the j th layer.


The components of the displacement vectors and the stresses can be expressed in
terms of these potentials as
uj =

j
j
j j z y
j j j
j y j x
+

, vj = z + x , wj = +

(3.16)
x
y
z
y
x
z
z
x
y

xzj
yzj

2j j z j y j y j x
= j 2
+

+

xz z y
z
x
x
y

2j j z j x j y j x
= j 2

+
yz z x
z
y
x

67

(3.17)
(3.18)

2j j y j x
+


z 2 z x
y

zzj = j j + 2 j

(3.19)

Application of the integral Fourier transforms with respect to the horizontal co-ordinates
x and y to (3.15) yields:
2 f j 2

2
2
2 + 2 k1 k2 f j = 0
z
cLj

2
2

g j
2
2
2 + 2 k1 k2 g j = 0
z
cTj

g
g
g
xj + yj + zj = 0
x
y
z

(3.20)

The general solutions to equations (3.20) are


f j = B1 j exp ( zRLj ) + B2 j exp ( zRLj )

g xj = B3 j exp ( zRTj ) + B4 j exp ( zRTj )

g yj = B5 j exp ( zRTj ) + B6 j exp ( zRTj )

g zj = B7 j exp ( zRTj ) + B8 j exp ( zRTj )

(3.21)

with RLj ,Tj = k12 + k22 2 / cLj2 ,Tj and Bij ( i = 1,8 ) unknown constants.
Expressing now the displacements and the stresses in terms of the potentials by
employing Equations (3.16)-(3.19) and excluding constants B7 j and B8 j by means of
the last equation in (3.20), vectors { s } j and { s } j may be written as
t

{ s } j = [ A ] j {b} j ,
t

{s} j = eR
b

L, jhj

[ A ] j {b} j ,
b

(3.22)
(3.23)

where integration constants Bij ( i = 1, 6 ) are composed in the vector {b} j C 6 , and

[ A] j

and [ A ] j are matrixes ( 6 6 ) , which have the following form:


t

matrix [ A ] j :
b

68


ik1

ik
2

RLj

2ik1 RLj j

2ik2 RLj j

j j

k1k2
RTj

ik1

R k
2
Tj

ik2

k1k2
RTj

2
1

RTj

R k
2
Tj

2
1

ik2

ik2

2ik1 RLj j

2 j k1k2

2 j k1k2

2ik2 RLj j

j j

2i j RTj k2

k1k2
RTj
ik1

j ( R + k12 k22 )
2
Tj

j ( RTj2 k12 + k22 ) j ( RTj2 k12 + k22 )

2 j k1k2

2i j RTj k2

RTj

k1k2

RTj

ik1

2
2
2
j ( RTj + k1 k2 )

2 j k1k2

2i j RTj k1

RTj2 k22

RTj

RTj

RLj

RTj2 k22

2i j RTj k1

(3.24)
with j = 2 RTj2 + 2 cTj2 .

matrix [ A ] j :
t

[ A ] j = [ A ] j [ D] j ,
b

(3.25)

with [ D] j = ( d kl ) , k , l = 1, 2,..., 6 the diagonal matrix with elements


d11 = 1 , d 22 = e

2 RLj h j

, d33 = d55 = e

( RTj RLj ) h j

, d 44 = d 66 = e

RTj + RLj h j

(3.26)

Equations (3.22) and (3.23) couple the displacements and stresses at the top of the layer
with those at the bottom:
t 1

{ s} j = eR h [ A] j [ A] j { s} j
b

Lj j

(3.27)

The requirement for continuity of displacements and equilibrium of stresses at the layer
interfaces are expressed by equations

{ s }1 = { s }2 , { s }2 = { s }3 , , { s }n1 = { s }n
b

(3.28)

Using (3.27) and (3.28), one can couple displacements and stresses at the surface of the
ground with those at the bottom of the n th layer, which adjoins to the half-space:
t 1

t 1

{ s }n = exp ( RLn hn ) [ A ]n [ A ]n { s }n = e R h [ A ]n [ A ]n { s }n1 =


b

t 1

= exp ( RLn hn + RLn 1hn 1 ) [ A ]n [ A ]n


b

Ln n

t
[ A ]n1 [ A ]n1 { s }n1 =
b

.................................................................................................
n

b
t 1
b
t 1
b
t 1
t
= exp RLj h j [ A ]n [ A ]n [ A ]n 1 [ A ]n 1 [ A ]1 [ A ]1 { s }1
j =1

The multiplication of matrices in the last line in Eq. (3.29) can be rewritten as

69

(3.29)

[ T ]
[ T]
b
t 1
b
t 1
b
t 1
[T] = T 11 T 12 = [ A ]n [ A ]n [ A ]n1 [ A ]n1 [ A ]1 [ A ]1 ,
[ ]21 [ ]22

(3.30)

where [ T]ij are ( 3 3) matrices. Thus, we can now express the displacements and the
stresses at the lowest interface in terms of those at the surface of the layered structure:
b
{u

RLj h j
[ T]11
,k1 , k2 }n
j =1
e
=

b
[ T]21
{ }n

[T]12 {u ,k ,k }1
[T]22 { }t
t

(3.31)

Now we are only to consider the foundation for the layered ground. Three cases
are possible.
1) Half-space. The equivalent of equations (3.22), for the half-space are

{u

, k1 , k2 n +1

= [ R ]{b}n +1

{ }n+1 = [S ]{b}n+1
t

(3.32)

where {b}n +1 is a vector composed of three constants, [ R ] and [S ] are the following

( 3 3)

matrices:

ik1c1

[ R ] = ik2c1

R
c
L( n +1) 1

c2
RT ( n +1)

k1k2

c2
RT ( n +1)

ik1c2

RT2( n +1) k22

kk
1 2 c2
RT ( n +1)

RT2( n +1) k12


RT ( n +1)

c2

ik2 c2

2ik1 n +1 R
c
L ( n +1) 1

[S ] = 2ik2 n+1RL( n+1)c1

n +1 n +1c1

n +1 RT2( n +1) k12 + k22 c2

2k1k2 n +1c2

ik2 RT ( n +1) c2

(3.33)

n +1 RT2( n +1) + k12 k22 c2

2
2
n +1 k2 + RT ( n +1)
(3.34)

RT ( n +1)

ik1 RT ( n +1) c2

with c1 = exp RL( n +1) hn and c2 = exp RT ( n +1) hn .


Equations (3.32) yield the following relation between the displacements and
stresses at the top of the half-space:

{u

, k1 , k2 n +1

= [ R ][S ]

70

{ }n+1
t

(3.35)

The condition of continuity of the displacements and the stresses at the interface of the
n th layer and the half-space requires

{u

= {u ,k1 ,k2 } , { }n+1 = { }n


b

, k1 , k2 n +1

(3.36)

Consequently,

{u

} = [ R ][S] { } .
b

, k1 , k2 n

(3.37)

Substitution of equation (3.37) into equation (3.31) yields

{u

, k1 , k2 1

= [ R ][S ]

[T]21 [T]11 ) ([ T]12 [ R ][S ] [ T]22 ) { }1


1

(3.38)

Denote
Q11 Q12
[Q ] = Q21 Q22
Q31 Q32

Q13
1
Q23 = [ R ][S ] [ T]21 [ T]11
Q33

) ([ T]
1

12

[ R ][S ]

[ T]22 ) .

(3.39)

Now, comparing equation (3.38) to equation (3.10), one obtains

,k1 ,k2 ( k1 , k2 , ) = [Q ]

(3.40)

The minus sign reflects the fact that the positive direction of normal stress on the surface of the ground is defined as opposite to the direction of the axis OZ.
Thus, by definition [Q ] is the dynamic flexibility matrix for the layered halfspace.
It should be mentioned here, that the method of the dynamic flexibility matrix
does not always help to avoid difficulties in numerical calculations as it was claimed in
the paper of Sheng et al. [136]. The problems could be caused by matrix [ D] j . Indeed,
it is clearly seen from expressions (3.26) that the even members of main diagonal of
matrix [ D] j tend to zero very fast as the argument of the exponent increases. This implies that corresponding matrices [ T]ij could be ill conditioned and difficulties in numerical calculations would then appear necessarily. Because of this, it is sensible to use
this method only for relatively small depths up to 10 meters.
2) Stiff foundation. In this case

{u

, k1 , k2 n +1

= {u ,k1 ,k2 } = 0 ,
b

and (3.31) yields


71

(3.41)

{u

} = [ T ] [ T ] { } = [ Q ] { } .
t

, k1 , k2 1

11

12

(3.42)

3) Ground presented by only half-space. When the ground is modelled by just a halfspace, i.e. n = 0 , one may use equation (3.35)

{u

} = [ R ][S] { } = [Q]{ } .
t

, k1 , k2 1

(3.43)

The equivalent stiffness of a layered half-space.


In Chapter 3, the equivalent dynamic stiffness of a half-space was introduced.
With the help of this stiffness, which is a complex-valued, frequency and wavenumber
dependent function, a reduction of a 3D model to a 1D model was carried out. The same
approach can be used for the layered half-space.
To find an expression for the equivalent stiffness of a layered half-space, the
Fourier transform with respect to time co-ordinate x (the form of this) should first be
applied to Eqs.(3.2)-(3.4). This yields the following system of algebraic equations:
E Jk 4 m 2 + k i
beam

( kP iP )
0
R
P
P
beam 1
w,k1 2 P0 ( k1V )

2
2
S

kP iP + kB iB ( mS + mB 3) ( kB iB + mB 6) w,k1 =
0
kP

( 2)

0
( kB iB + mB2 6)
( kB iB mB2 3) w,k1 F,k1

(3.44)
The term F( ,k)1 on the right-hand side of Eq.(3.44) is the Fourier image of the contact
2

force between the ballast and the ground. It is this force that has to be replaced by the
equivalent stiffness of the ground. To accomplish this replacement, the boundary conditions (3.5) and (3.6) that describe interaction between the beam and the layered halfspace should be employed. Application of the Fourier transform with respect to time
and the horizontal co-ordinates to (3.5) and (3.6) yields

zz ( k1 , k2 , 0, ) = F( ,2k)

sin k2 a
k2 a

xz ( k1 , k2 , 0, ) = yz ( k1 , k2 , 0, ) = 0
B

w , k1

1
=
2

w
1

, k1 , k2

( k1 , k2 , 0 ) dk2

(3.45)

(3.46)

with zz ( k1 , k2 , 0, ) = zzt 1 , xz ( k1 , k2 , 0, ) = xzt 1 , yz ( k1 , k2 , 0, ) = yzt 1 the Fourier images of the stresses on the surface of the layered half-space. Since

px , k1 ,k2 ( k1 , k2 ) = xz ( k1 , k2 , 0, ) ,
py , k1 ,k2 ( k1 , k2 ) = yz ( k1 , k2 , 0, ) ,
pz , k1 ,k2 ( k1 , k2 ) = zz ( k1 , k2 , 0, ) ,
72

(3.47)

substitution of the boundary condition (3.45) into Eq.(3.10), followed by employment of


Eq.(3.40), yields the following expression for the vertical displacement of the ground
surface:

w1 , k1 ,k2 ( k1 , k2 , 0 ) = 33 zz ( k1 , k2 , 0, ) = Q33 ( , k1 , k2 ) F( ,k)1


2

sin k2 a
k2 a

(3.48)

Substituting Eq.(3.48) into Eq.(3.46), we obtain


B

w , k1

1
=
2

sin k a
Q ( , k , k ) k a dk
33

( 2)
F , k1

(3.49)

or
( 2)

F , k1

1
=
2

sin k a
B
layer
B
Q33 ( , k1 , k2 ) k2 a2 dk2 w ,k1 = eq ( , k1 ) w ,k1

(3.50)

The multiplier of wB, k1 on the right-hand side of Eq.(3.50) is, by definition, the
equivalent vertical stiffness of the layered half-space. Substituting Eq.(3.50) into
Eq.(3.44), one obtains
E Jk4 m 2 + k i
( kP iP )
R
P
P
beam 1

( kP iP )
kP iP + kB iB ( mS + mB 3) 2

0
( kB iB + mB2 6)

beam
w,k1 2 P0 ( k1V )

2
( kB iB + mB 6) wS,k1 =
0

( kB iB mB2 3+ eqlayer ) w,k1


0

(3.51)
in which

layer
eq

1
( , k1 ) =
2

sin k a
Q33 ( , k1 , k2 ) k2 a2 dk2

(3.52)

is the equivalent stiffness of the layer.


In the next section, the equivalent stiffness is studied considering two ground
structures: 1) a layer of depth H fixed at its bottom and 2) a layer on a half-space.

Equivalent stiffness of a visco-elastic layer fixed at its bottom.


The simplest example of a layered solid is a single layer, the equivalent stiffness
of which is studied in this section. If the layer is fixed over its bottom, that it is capable
of being a model for a ground layer that overlies a bedrock.

73

To find an expression for the equivalent stiffness eqlayer ( , k1 ) , we have to calculate Q33 ( , k1 , k2 ) , see Eq. (3.52). A simple way to do so is to employ Eq. (3.42) and
calculate the product [ T]12 . This yields
Q33 ( , k1 , k2 ) =

2 RL Num ( , k1 , k2 )
,
cT2 Den ( , k1 , k2 )

(3.53)

with
Num ( , k1 , k2 ) = ( k12 + k22 ) sinh ( RT H ) cosh ( RL H ) RL RT sinh ( RL H ) cosh ( RT H ) ,
Den ( , k1 , k2 ) = 4 ( k12 + k22 ) RL RT + ( k12 + k22 )( 4 RL2 RT2 + 2 ) sinh ( RL H ) sinh ( RT H )

RL RT 4 ( k12 + k22 ) + 2 cosh ( RL H ) cosh ( RT H ) ,

= 2 ( k12 + k22 ) 2 cT2

(3.54)

To study eq ( , k1 ) , it is again customary (see Chapter 3) to introduce the following new variables:

Variable of integration = k2 k1 which represents the ratio between the wave numbers of waves propagating in the lateral and longitudinal directions;
Phase velocity of waves propagating along the beam, V ph = k1 ;

Ratio of V ph and complex velocities of longitudinal and shear waves in the layer,

L ,T = V ph cL ,T .
In terms of these variables, the equivalent stiffness reads

Num (V ph , k1 , ) sin k1a

* 2
layer
d , (3.55)
eq (V ph , k1 ) = 2 1 ik1V ph T RL ( )


Den (V ph , k1 , ) k1a

with

Num (V ph , k1 , ) = (1 + 2 ) sinh RT k1 H cosh RL k1 H RL RT sinh RL k1 H cosh RT k1 H ,

Den (V ph , k1 , ) = 4 (1 + 2 ) ( ) RL RT + (1 + 2 ) 4 RL2 RT2 + 2 ( ) sinh RL k1 H sinh RT k1 H

(
) (
, ( ) = 2 (1 + )

RL RT 4 (1 + 2 ) + 2 ( ) cosh RL k1 H cosh RT k1 H ,
RL ,T = 1 + 2 L2,T

2
T

(3.56)
Expression (3.55) coincides with that obtained by Metrikine and Popp [117].

74

Im ( eqlayer *108 ) , N m

Re ( eqlayer *108 ) , N m

V ph cT

k1

V ph cT

k1

a)

b)

Fig. 3.3. Equivalent stiffness of the layer versus the phase velocity and wave
number of waves in the beam: a) real part b) imaginary part.

eqlayer *108 , N m

k1 = 0.6 / a

Re ( eqlayer )

V ph cT
0
1
-1

-2

V ph*
Im ( eqlayer )

-3

Fig. 3.4. Equivalent stiffness of the layer versus the phase velocity of waves in the beam.

The integral in expression (3.55) can be evaluated numerically. Results of this


evaluation are depicted in Figure 3.3(a,b) and Figure 3.4. The former figure shows the
real and imaginary part of the equivalent stiffness as function of the phase velocity V ph
and the wave number k1 . In Figure 3.4, cross-sections of Figures 3.3(a,b) by the plane
k1 = 0.6 a m -1 are depicted.
Calculations were carried out using the following parameters of the system:
E = 108 N m 2 , * = 104 N*s m 2 , = 1000 kg m3 , = 0.3, H = 10m, a = 1.3m (3.57)

75

Figures 3.3(a,b) and Figure 3.4 demonstrate the following:


1. The imaginary part of the equivalent stiffness, which is related to the energy dissipation in the layer, becomes significant when the phase velocity exceeds the critical
value V ph* , which is pointed in Figure 3.4. This critical value depends upon the wave
number k1 as it is clearly seen from Figure 3.3(a,b). This dependence, shown in
Figure 3.5, represents variation of the phase velocity in the lowest dispersion branch
of the layer (see, for example, [117]). The cause of growth of the imaginary part is
generation of waves in the layer. These waves, transferring energy from the railway
track, provide strong attenuation of waves in the beam.
2. When the phase velocity tends to V ph* , both the real and imaginary parts of the
equivalent stiffness become quite small. This implies that a load moving on with velocity V = V ph* might cause resonance response of the structure ([30]).
3. Both the real and the imaginary part of the equivalent stiffness have a number of
minima as Vph > Vph* . These minima occur under the condition that, for a given value
of the wave number k1 , V ph is close to the group velocity of one of the higher modes
of the layer vibrations. If the layer had no damping, the equivalent stiffness in these
points would be zero.

Vph* cT

3 .0 0

2 .5 0

2 .0 0

1 .5 0

1 .0 0

Vph* = cR

0 .5 0
0 .0 0

0 .4 0

k
0 .8 0

1 .2 0

1 .6 0

2 .0 0

Fig. 3.5. Relative phase velocity V ph* cT versus wave number k = k1a
for the lowest dispersion branch of the layer.

Thus, the main difference between the equivalent stiffness of the layer and halfspace is caused by the multimode character of the layer vibration, which results in additional maxima and minima of the equivalent stiffness of the layer with respect to the
stiffness of the half-space. Such a difference implies that dynamic response of the railway track that lies on a layered ground could differ qualitatively from that of a track on
a homogeneous ground. This difference will be especially appreciable for small values
of material damping in the ground.

76

Im ( eqlayer *108 ) , N m

Re ( eqlayer *108 ) , N m

k1

V ph cT

k1

V ph cT

a)
2

b)
Re ( eqlayer )

eqlayer *108 , N m

k1 = 0.6 / a

V ph cT
0
1

-1

V ph*
-2

Im ( eqlayer )
-3

c)
Fig. 3.6. Equivalent stiffness of the shallow ( H = 2 m ) layer on the soft half-space versus the phase

velocity and wave number of waves in the beam: a) real part b) imaginary part
c) cross-section by the plane k1 = 0.6 a .

Equivalent stiffness of a layer on a half-space.


Consider the equivalent stiffness of a visco-elastic layer on a visco-elastic halfspace. In this case, analytical expression for the equivalent stiffness becomes rather
bulky and not informative. Therefore, studying the equivalent stiffness it is customary
to calculate Q33 ( , k1 , k2 ) in the expression for the equivalent stiffness numerically.
Calculations were carried out by using the following physical parameters of the
system:

77

Im ( eqlayer *108 ) , N m

Re ( eqlayer *108 ) , N m

k1

k1

V ph cT

V ph cT

a)

b)
2

Re ( eqlayer )

eqlayer *108 , N m

k1 = 0.6 / a

V ph cT
0
1

-1

V ph*
-2

Im ( eqlayer )
-3

c)
Fig. 3.7. Equivalent stiffness of the deep ( H = 10 m ) layer on the soft half-space versus the phase veloc-

ity and wave number of waves in the beam: a) real part b) imaginary part
c) cross-section by the plane k1 = 0.6 a .

Layer:

E = 108 N m2 , * = 104 N*s m2 , = 2000 kg m3 ,


= 0.3, a =1.3m, H = 2 or 10m.
Half-space:

78

(3.58)

Im ( eqlayer *108 ) , N m

Re ( eqlayer *108 ) , N m

k1

k1

V ph cT

V ph cT

a)

b)

Re ( eqlayer )

eqlayer *108 , N m

k1 = 0.6 / a

V ph cT
0
1

-1

-2

V ph*
Im ( eqlayer )

-3

c)
Fig. 3.8. Equivalent stiffness of the shallow ( H = 2 m ) layer on the stiff half-space versus the phase

velocity and wave number of waves in the beam: a) real part b) imaginary part. c) cross-section by the
plane k1 = 0.6 a .

h* = 104 N*s m 2 , h = 5000 kg m3 , h = 0.3,


10 108 N m 2
Eh =
8
2
0.5 10 N m

( stiff half-space ) , cR ,h =223km h


( soft half-space ) , cR,h =1000 km h

(3.59)

Variation of H in (3.58) is performed to study the effect of the layer depth. Two
magnitudes of Youngs modulus Eh of the half-space in (3.59) are chosen so that the
half-space is either softer or stiffer than the layer.

79

Im ( eqlayer *108 ) , N m

Re ( eqlayer *108 ) , N m

k1

V ph cT

k1

V ph cT

a)

b)
Re ( eqlayer )

eqlayer *108 , N m

k1 = 0.6 / a

V ph cT
0
1

-1

-2

V ph*

Im ( eqlayer )
-3

c)
Fig. 3.9. Equivalent stiffness of the deep ( H = 10 m ) layer on the stiff half-space versus the phase veloc-

ity and wave number of waves in the beam: a) real part b) imaginary part. c) cross-section by the plane
k1 = 0.6 a .

The real and imaginary parts of the equivalent stiffness are plotted in Figures
3.6-3.9. Figures 3.6-3.7 correspond to a soft half-space whereas Figures 3.8-3.9 correspond to a stiff half-space. Figures 3.6 and 3.8 are drawn for H = 2m , whilst Figures
3.7 and 3.9 are plotted for H = 10m .
Figures 3.6-3.9 demonstrate the following:

The equivalent stiffness of a layer on a half-space is similar to that of a layer


with fixed bottom. The similarity contains in existence of a critical phase velocity V ph* ( k1 ) that bounds a domain of phase velocities and wave numbers in
which the equivalent stiffness is almost real from that where it becomes significantly complex.
80

The minimum (with respect to k1 ) value of the critical velocity V ph* strongly depends upon the layers depth and the relative stiffness of the layer and the halfspace. The softer is the half-space, the smaller is the critical velocity. For the
soft half-space, the deeper is the layer, the larger is V ph* . For the stiff half-space,
on the contrary, the deeper is the layer, the smaller is the critical velocity.

Concluding this section we would like to underline its main result: the equivalent stiffness of a layered ground strongly depends on the physical parameters of the
ground and the depth of the layer(s). Thus, it might be expected that the dynamic response of a railway track is also strongly influenced by the properties of the ground.

3.3. Steady-state response of a beam on a layered half-space


to a moving load.
In this section, making use of the equivalent stiffness of the layered half-space,
we study the steady-state dynamic response of a beam on such a half-space to a moving
load in accordance with the problem statement (3.1)-(3.6). In correspondence with the
previous section, the beam is first assumed to overlay a layer with a fixed bottom. Then,
we proceed by replacing the layer by a layer on a half-space.
The Fourier image of the beam displacement wbeam
, k1 can be found from Eq.(3.51)
to give
wbeam
, k1 =

1 ( , k1 )
0 ( , k1 )

(3.60)

with
1 ( , k1 ) = 210 ( k1V ) ,
P0
12 ( k B i B ) ( k P i P ( mB + mS ) 2 ) +
12
+ mB 2 ( mB 2 4 ( k P i P ) + 4mS 2 ) +
10 ( , k1 ) =

+4 eqlayer ( 3 ( k P i P ) + 3 ( k B i B ) 3mS 2 mB 2 )

and

81

(3.61)

0 ( , k1 ) = eqlayer Ebeam Jk14 ( k P i P mS 2 ) + mS mR 4 ( k P i P ) 2 ( mS + mR )

(E

beam

Jk14 mR 2 )
12

(12 ( k

i B ) ( mB 2 eqlayer ) +

+ (12mS ( k B i B ) 4mS mB 2 mB2 2 + 4mB eqlayer ) +


2

( kP i P )

12

(12k (

( 4m m
R

layer
eq

(3.62)

mR 2 ) + 4 Ebeam Jk14 ( 3 ( k B i B ) mB 2 ) +

))

12 ( k B i B )( mB + mS ) + 4mB ( mS 2 eqlayer ) + mB2 2 .

Applying to (3.60) the inverse Fourier transforms over and k1 we obtain the following expression for the beam response to a uniformly moving load:
wbeam ( x, t ) =
=

1
2

1
=
2

1
4

1 ( , k1 )

( , k ) exp ( i t + ik x ) d dk
1

10 ( , k1 )
( k1V ) exp ( i t + ik1 x ) d dk1 =
0 ( , k1 )

(3.63)

10 ( k1V , k1 )
0 ( k1V , k1 ) exp ( ik1 ( x Vt ) ) dk1

The integral in Eq.(3.63) can be evaluated numerically. Results of this evaluation that
were obtained by employing the ground parameters (3.57), (3.58), (3.59) and the
following parameters of the superstructure:
mR = 60.34 kg m , Ebeam I = 1.22*107 N*m 2 ,
mS = 100 kg m , k P = 108 N m 2 , P = 1000 N*s m 2 ,

(3.64)

mB = 1000 kg m , k B = 108 N m 2 , B = 1000 N*s m 2 ,


are presented below. In all examples, the magnitude of the load is P0 = 20000 kg that
represents the axle loading of a train wagon.

Layer fixed at its bottom.


Consider the beam on a layer, the bottom of which is fixed.
Figures 3.14 (a,b) and 3.15 show the beam deflection at the loading point
beam
w (Vt , t ) as a function of the load velocity for three values of the material damping
in the layer. The bold dashed curves and the bold solid curves correspond to the shallow
( H = 2m ) and deep ( H = 10m ) layer, respectively. The vertical dashed line indicates
the Rayleigh wave speed.

82

0.0040

0.0040

H = 10 m

H = 10 m

0.0036

beam displacement, m

beam displacement, m

0.0036

0.0032

0.0028

0.0032

0.0028

0.0024

0.0024

H =2m

H =2m

0.0020
0

0.4

0.0020
0.8

1.2

1.6

0.4

0.8

1.2

1.6

V cT

V cT

a)

b)

Fig. 3.14. Beam displacement at the loading point versus the load velocity for (a) * = 103 N*s m 2

and (b) * = 104 N*s m 2 .


0.0040

beam displacement, m

0.0036

H = 10 m
0.0032

0.0028

0.0024

H =2m
0.0020
0

0.4

0.8

1.2

1.6

V cT

Fig. 3.15. Beam displacement at the loading point versus the load velocity for * = 105 N*s m 2 .

The following conclusions can be drawn from the figures:


1. The load moving on a supported by the layer beam has one critical velocity, likewise in the case of the beam on a half-space. If the material damping is small, this
velocity corresponds to the minimum phase velocity of waves in the beam interacting with the layer.
2. The critical velocity of the load depends in the depth of the layer. The smaller is the
depth, the higher is the critical velocity. As the depth of the layer increases, the critical velocity approaches the Rayleigh wave velocity likewise in the case of the halfspace. In general, it can be shown that the response of the beam on a visco-elastic

83

layer tends to that of the beam on a visco-elastic half-space (with the same material
properties) as the layer depth increases.
3. The depth of the layer influences the magnitude of the beam response. The thinner is
the layer, the smaller is the response.
4. The main effect of the material damping is that with increasing damping, the beam
response decreases.
Let us note that the beam displacement in the loading point is not the maximum
displacement of the beam. This can be seen from Figures 3.16-3.18, which show the
beam patterns that move along the beam together with the load. The vertical axis in
these figures is the beam deflection, whereas the horizontal axis shows the distance
= x Vt from the moving load. Figures 3.16-3.18 show that the maximum displacement of the beam takes place somewhat behind the load. Only in the sub-critical case,
this maximum would shift to the loading point under the condition that the material
damping vanished. The position at which the beam displacement is maximal depends on
the load velocity. Thus, to plot the dependence of the maximum beam displacement on
the load velocity, it is not enough to consider just the loading point. However, in the
sub-critical regime, which is the one that is practically relevant, the dependencies plotted in Figures 3.14 and 3.15 differ only slightly from those that are obtained by considering the maximum beam deflection instead of that of the loading point. Thus, for practical needs, Figures 3.14 and 3.15 are sufficient.
Let us return to Figures 3.16-3.18 that are plotted for H = 2 m , H = 5 m , and
H = 10 m , respectively. The material damping in these figures is * = 104 N*s m 2 .
Each figure presents the beam pattern for three velocities of the load: a) V = 250 km h ,
b) V = 470 km h and c) V = 600 km h . The Rayleigh-wave velocity for the parameters
under consideration is cR = 463km h . The critical velocity of the load, in accordance
with Figures 3.14 and 3.15, depends on the depth of the layer but is always larger than
the Rayleigh-wave velocity.
Figures 3.16-3.18 show that the moving load can cause two qualitatively different types of the beam response. First type of response can be seen in Figures 3.16(a),
3.17(a), 3.18(a) and 3.16(b). It corresponds to the sub-critical motion of the load and
occurs if the load velocity is smaller that the minimum phase velocity V ph* of waves in
the beam on the layer. This type of response is characterised by being localised in the
vicinity of the load and being almost symmetric with respect to the loading point. A
slight asymmetry is introduced by the material damping. Second type of response corresponds to the super-critical motion of the load and can be seen in Figures 3.16(b,c),
3.17(b,c), 3.18(c). This response is wavy and is not localised anymore around the
load. The higher is the load velocity, the wider is the region that this response occupies.
If the material damping were absent, the beam in this case would be disturbed everywhere, from minus to plus infinity. The super-critical response of the beam is fully
asymmetric with respect to the load. This is in accordance with the Doppler effect,
which causes the waves propagating in front of the load be shorter then those propagating behind the load. The larger is the load velocity, the higher is the degree of asymmetry.

84

0.001

0.001

0.001

-0.001

-0.001

-0.001

-0.002

-0.002

-0.002

-0.003

-0.003
-20

-10

10

20

-0.003
-20

-10

a)

10

20

-20

-10

b)

10

20

10

20

10

20

c)

Fig. 3.16. Steady-state beam deflections for a shallow layer, H = 2 m :


(a) V = 250 km h , (b) V = 470 km h and (c) V = 600 km h ( cR = 463km h ) .

0.001

0.001

0.001

-0.001

-0.001

-0.001

-0.002

-0.002

-0.002

-0.003

-0.003

-0.003

-0.004

-0.004
-20

-10

10

20

-0.004
-20

-10

a)

10

20

-20

-10

b)

c)

Fig. 3.17. Steady-state beam deflections for a layer with depth H = 5 m :


(a) V = 250 km h , (b) V = 470 km h and (c) V = 600 km h ( cR = 463km h ) .

0.001

0.001

0.001

-0.001

-0.001

-0.001

-0.002

-0.002

-0.002

-0.003

-0.003

-0.003

-0.004

-0.004
-20

-10

a)

10

20

-0.004
-20

-10

b)

10

20

-20

-10

c)

Fig. 3.18. Steady-state beam deflections for a deep layer H = 10 m :


(a) V = 250 km h , (b) V = 470 km h and (c) V = 600 km h ( cR = 463km h ) .

85

Let us study in more detail the influence of the layer depth on the beam deflection in the loading point. In Figure 3.19 this deflection is plotted versus velocity of the
load for five different values of the layer depth. In calculations, parameters (3.57) and
(3.64) were used.

0.0050

cR cT
H = 20 m

beam displacement, m

0.0045

H = 10 m
0.0040

H =3m

H =5 m

0.0035

0.0030

H =2m
0.0025
0

0.2

0.4

0.6

0.8

1.2

V cT

Fig. 3.19. Beam deflection in the loading point versus the load velocity
for different values of the layer depth.

Figure 3.19 shows that the critical velocity of the load steadily increases with
decreasing layer depth. At the same time, the beam deflection that corresponds to the
critical velocity decreases. This brings the following idea, which can be interesting for
practical applications. To make the dynamical amplification of the track vibrations at
the critical train velocities smaller, one can put a stiff strip under the track, providing an
artificial boundary. This boundary will slightly increase the critical velocity of the train
decrease the dynamic amplification associated with this velocity. Such a stiff strip could
be especially useful for soft subsoil, in which the critical velocities are quite small (in
the range of 200 250 km h ).

Layer on half-space.
Consider the beam on a layer that overlies a half-space. Let us start with the
situation that the half-space is softer than the layer. Results of calculations that were
performed by employing parameters (3.58), (3.59) and (3.64) are presented in Figures
3.20, 3.21, 3.22 and 3.23.
Figures 3.20 and 3.21 show the dependency of the beam displacement at the
loading point on the load velocity. The horizontal axis in these figures is normalised by
the shear wave velocity in the layer, cTlayer = 500 km h . Figures 3.20(a), 3.20(b) and 3.21
86

correspond to the material damping * = 103 N*s m 2 , * = 104 N*s m 2 and


* = 105 N*s m 2 , respectively. Bold dashed and solid lines correspond to the shallow,
H = 2m , and deep, H = 10m , layer, respectively. The vertical solid line indicates the
Rayleigh wave speed in the half-space, cRHS 0.414 cTlayer = 207 km h . The vertical
dashed line corresponds to the Rayleigh wave speed in the layer.

0.010

beam displacement, m

H = 10 m

H =2m

0.008

0.006

H =2m

0.008

H = 10 m
0.006

0.004

0.004
0

0.4

0.8

1.2

1.6

0.4

0.8

1.2

V cTlayer

V cTlayer

a)

b)

1.6

Fig. 3.20. Beam displacement at the loading point versus the load velocity
for (a) * = 103 N*s m 2 and (b) * = 104 N*s m 2 . Layer on soft half-space.

0.010

beam displacement, m

beam displacement, m

0.010

H =2m

0.008

H = 10 m
0.006

0.004
0

0.4

0.8

1.2

1.6

V cTlayer

Fig. 3.21. Beam displacement at the loading point versus the load velocity for * = 105 N*s m 2 .
Layer on soft half-space.

87

The following conclusions can be drawn from Figures 3.20 3.21:


1. There exist two critical velocities of the load. The first critical velocity is very
close to the Rayleigh wave velocity of the half-space. The second one is related
to the Rayleigh wave velocity of the layer. The depth of the layer does not (almost) influence these critical velocities.
2. The thinner is the layer, the larger is the beam displacement at the first critical
velocity and the smaller is that at the second critical velocity. This implies that
for a thin layer laying on a soft half-space, the dynamic response is primarily determined by physical properties of the half-space.
3. The material damping influences mainly the beam displacement at the second
critical velocity and changes only slightly the displacement at the first critical
velocity.
4. Comparing the response of the beam on the layered half-space to that of the
beam on the layer with a fixed bottom (compare Figures 3.20-3.21 to Figures
3.14-3.15), it can be noticed that in the latter case the beam deflection is almost
twice as high. This is to be expected since the stiff half-space increases the compliance of the ground model.

0.002

0.002

-0.002

-0.002

-0.004

-0.004

-0.006

-0.006

-20

-10

10

-20

20

-10

a)
0.002

-0.002

-0.002

-0.004

-0.004

-0.006

-0.006

-10

10

20

10

20

b)

0.002

-20

10

-20

20

-10

c)

d)

Fig. 3.22. Steady-state beam deflections for a shallow layer, H = 2 m , on soft half-space:
(a) V = 150 km h (b) V = 250 km h , (c) V = 470 km h and (d) V = 600 km h

(c

layer
R

= 463km h , cRHS = 207 km h ) .

88

The beam patterns that are formed in the beam as it overlies the layer on stiff
half-space are depicted in Figures 3.22 and 3.23. The former and the latter figure correspond to the shallow layer, H = 2m , and the deep layer, H = 10m , respectively. Both
figures show the beam patterns for four velocities of the load: a) V = 150 km h , b)
V = 250 km h , c) V = 470 km h and d) V = 600 km h . Additional value of the load
velocity V = 150 km h has been considered to present the sub-critical motion since velocity V = 250 km h is larger than the Rayleigh wave velocity in the half-space.
Figures 3.22 and 3.23 show that the effect of the load velocity on the beam pattern remains the same as in the case of a fixed along the bottom layer: the sub-critical
response is localised and almost symmetric with respect to the load, whereas the supercritical response is wave, not localised and highly asymmetric.

0.002

0.002

-0.002

-0.002

-0.004

-0.004

-0.006

-0.006

-20

-10

10

-20

20

-10

a)
0.002

-0.002

-0.002

-0.004

-0.004

-0.006

-0.006

-10

10

20

b)

0.002

-20

10

-20

20

-10

c)

10

20

d)

Fig. 3.23. Steady-state beam deflections for a deep layer, H = 10 m , on soft half-space:
(a) V = 150 km h (b) V = 250 km h , (c) V = 470 km h and (d) V = 600 km h

(c

layer
R

= 463km h , cRHS = 207 km h ) .

Consider the situation that the half-space is stiffer than the layer. The velocitydependence of the beam deflection at the loading point, which is obtained by employing
parameters (3.58), (3.59) and (3.64), is presented in Figures 3.24-3.25. The horizontal
axis in these figures is normalised by the shear wave velocity in the layer,
cTlayer = 500 km h . Figures 3.24(a), 3.24(b) and 3.25 correspond to the material damping
89

* = 103 N*s m 2 , * = 104 N*s m 2 and * = 105 N*s m 2 , respectively. Bold dashed
and solid lines correspond to the shallow, H = 2m , and deep, H = 10m , layer, respectively. The vertical dashed line corresponds to the Rayleigh wave speed in the layer.
Figures 3.24 and 3.25 show that there is only one critical velocity of the load.
This is in correspondence with the fixed over the bottom layer (Figures 3.14 and 3.15)
and in contrast to the layer on soft half-space (Figures 3.20 and 3.21). This result is to
be expected, since a layer on a stiff half-space is similar to a layer that is just fixed over
its bottom. The fixation corresponds to an infinitely stiff half-space.

0.010

H = 10 m

beam displacement, m

beam displacement, m

0.010

0.008

H =2m
0.006

0.004

0.008

H = 10 m

H =2m

0.006

0.004
0

0.4

0.8

1.2

1.6

V cTlayer

0.4

0.8

1.2

1.6

V cTlayer

a)

b)

Fig. 3.24. Beam displacement at the loading point versus the load velocity for (a) * = 103 N*s m 2

and (b) * = 104 N*s m 2 . Layer on stiff half-space.

beam displacement, m

0.010

0.008

H = 10 m

H =2m

0.006

0.004
0

0.4

0.8

1.2

1.6

V cTlayer

Fig. 3.25. Beam displacement at the loading point versus the load velocity for * = 105 N*s m 2 .
Layer on stiff half-space.

90

The critical velocity in Figures 3.24 and 3.25 is close to the shear wave velocity
in the layer and, in contrast to the case of the layer on soft half-space, depends upon the
layers depth noticeably. The thinner is the layer, the larger is the critical velocity and
the smaller is the beam deflection that is associated with this critical velocity. This result is again similar to that we have obtained on the hand of a fixed over bottom layer.
The only difference between the fixed layer and the layer on stiff half-space is that the
beam displacement of the beam overlying the former is twice smaller, which has to do
with a much higher compliance of the half-space in comparison to the rigid foundation.
The beam patterns are depicted in Figures 3.26-3.27, which are similar to Figures 3.22- 3.23. The only difference between the former and the latter figures contains
in the magnitude of the beam deflection. As it is to be expected, the stiffer half-space
decreases this magnitude.

0.002

0.002

0.002

-0.002

-0.002

-0.002

-0.004

-0.004

-0.004

-0.006

-0.006
-20

-10

10

-0.006
-20

20

-10

a)

10

20

-20

-10

b)

10

20

c)

Fig. 3.26. Steady-state beam deflections for a shallow layer, H = 2 m , on stiff half-space:
(a) V = 250 km h , (b) V = 470 km h and (c) V = 600 km h ( cRlayer = 463km h , cRHS = 926 km h ) .

0.002

0.002

0.002

-0.002

-0.002

-0.002

-0.004

-0.004

-0.004

-0.006

-0.006

-0.006

-20

-10

a)

10

20

-20

-10

10

20

b)

-20

-10

10

20

c)

Fig. 3.27. Steady-state beam deflections for a deep layer, H = 10 m , on stiff half-space:

(a) V = 250 km h , (b) V = 470 km h and (c) V = 600 km h

(c

layer
R

= 463km h , cRHS = 926 km h ) .

Finalising the chapter, it is worth recalling its main result: the soil stratification
influences the dynamic response of the railway track strongly. Both the depth of the soil
layers and their physical properties are of importance. Their influence contains in

91

1. shifting and (sometimes) introducing new critical velocities of the train;


2. reducing or increasing the dynamic amplification of the track response;
3. scaling the spatial pattern of the rail response.

92

CHAPTER 4
Periodically Supported Beam on a Visco-elastic Layer
As a Model for Conventional Railway Track
In Chapters 1-3, the railway track was modelled as a structure whose properties
do not vary in the longitudinal direction. This kind of modelling is well verified if an
embedded track is considered, which is indeed homogeneous along its length. For a
conventional track, however, the assumption that the track is homogeneous is approximate because of the sleepers, which contact both the rails and the ballast in discrete positions. Thus, to model a conventional track accurately, discretely positioned sleepers
have to be taken into account. This is done in the present chapter by considering a
model for a railway track that consists of two beams on periodically spaced supports
that are mounted to a visco-elastic 3D layer. The beams, supports and layer model the
rails, sleepers and subsoil, respectively. The train axial loading is modelled with the
help of two moving loads of the same magnitude that are applied to the beams.
There are two main objectives of this chapter. First, a method will be developed
that allows for an accurate calculation of the steady-state response of a threedimensional, periodically inhomogeneous structure to a moving load. Second, employing this method, it will be clarified whether the effect of a three-dimensional ground in
modelling of a conventional track is as crucial as that in modelling of an embedded
track.
The chapter is structured as follows. In section 4.1, the equivalent dynamic stiffness of a visco-elastic layer at the interface with the sleepers is considered. It is shown
that the original three-dimensional (3D) model can be reduced to a one-dimensional
(1D) model by replacing the layer by a set of equivalent springs. The stiffness of these
springs is a function of the angular frequency and phase shift between vibrations of
neighbouring sleepers.
In section 4.2, the steady-state response of the beam to a set of vertical loads that
move uniformly on the beam is analysed. The main attention is paid to the vertical deflection of the rails and its dependence on the load velocity. Critical velocities of the
loads are determined and the effect of physical parameters of the model on the track response at these critical velocities is studied.
In section 4.3, the elastic drag experienced by a high-speed train due to excitation of track vibrations is analysed. First, the drag experienced by a single axle is calculated and compared to that obtained in Chapter 2 on the hand of the homogeneous
model. Second, to study the effect of the bogie wheelbase, the drag is studied for two
axles. Finally, the total elastic drag that is experienced by a TGV train is calculated and
compared to that obtained in Chapter 2 and to the rolling and aerodynamic drag.
Section 4.4 is devoted to a comparison of the steady-state response of 1D and
3D models for a conventional railway track. Advantages and disadvantages of both
models and restrictions on their application are discussed.
In the chapter, the idea of the equivalent stiffness of the ground, which was developed
in the paper of A.V. Metrikine and K. Popp [116], was used and extended. Results presented in sections 4.2-4.4 are original.
93

4.1. Equivalent dynamic stiffness of a visco-elastic layer at


the sleeper-layer interface.
In Chapters 2 and 3, it has been shown that introduction of the equivalent dynamic stiffness of a 3D foundation of a railway track is a powerful instrument in studying the dynamics of the track. In these chapters, the equivalent stiffness was used for
homogeneous models of the track. In this chapter, it is shown how the method of
equivalent stiffness can be extended to a periodically inhomogeneous model for a conventional track.
Consider the steady-state vibrations of the model presented in Figure 4.1. This
model is composed of two beams on periodically placed supports (sleepers) and a viscoelastic layer. The model for the sleepers and for their interface with the layer is described below. Vibrations of the structure are perturbed by four loads of constant amplitude P , uniformly moving with velocity V along the beams.

2a

2b

x, u

y, v
z, w

Fig. 4.1. Euler-Bernoulli beams mounted to a visco-elastic layer by periodically spaced sleepers as a
model for a conventional railway track.

The model is analysed under the following assumptions:


1. The beams are assumed to be of infinite length and to perform vertical vibrations
only.
2. The loading is symmetric with respect to the centreline y = 0 of the system.
This assumption allows for the following reduction of the model: two beams can
be replaced by one beam of width 2a and the loads corresponding to one axle
can be replaced by one load, which is uniformly distributed over the beam
width.
3. All supports are identical and consist of a mass, a spring and a dashpot.
4. The mass of the supports is concentrated at the middle point of the contact
sleeper-layer interface.
5. The contact area between each support and the layer is a rectangle of the size
2a 2b .
6. The stresses are uniformly distributed over the contact area.
7. The shear contact in the x -direction is modelled by springs of a constant stiffness K , which are uniformly distributed under each sleeper (see Figure 4.2,

94

where the vertical cross-section of the structure by the plane y = 0 is represented).


8. The shear contact in the y -direction is smooth.
9. The midpoint of each sleeper and the layer are in contact permanently.

d1

d1

d2

KP
z=0

z=0

w be am ( x , t )

u sn ( t )

w sn ( t )

u
w

Fig. 4.2. Vertical cross-section y = 0 with magnified sleepers layer interface.

With these assumptions the equations that govern the steady-state response of
the structure can be written as follows:

Equations of motion for an inner element of the layer:

u + ( + ) ( iu ) =

2u
,
t 2

(4.1)

where u ( x, y , z , t ) = ( u ( x, y , z , t ) , v ( x, y , z , t ) , w ( x, y , z , t ) ) [ m ] is the displacement


vector of the layer, kg m3 is its mass density, = + * / t and
= + * / t are operators that are used instead of Lams constants N m 2 and

N m 2 to describe the layer according to the Voigts phenomenological model.

Boundary conditions at the surface of the layer:

zz ( x, y, 0, t ) =

H (a y )

4ab

2 wsn

M
+ p ( wsn ( t ) wbeam ( nd , t ) ) + K p ( wsn ( t ) wbeam ( nd , t ) ) H ( b x nd )
2
t
t
n =

H (a y )
xz ( x, y, 0, t ) =
Ku ( nd , 0, 0, t ) H ( b x nd )
4ab
n =
(4.2)

95

Boundary conditions at the bottom of the layer:

yz ( x, y, 0, t ) = 0

(4.3)

u ( x, y , H , t ) = 0

Compatibility condition for the vertical motion of the sleepers and the layer:
wsn ( t ) = w ( nd , 0, 0, t )

(4.4)

Equation of motion for the beam:


2 wbeam
4 wbeam
+
E
I
=
beam
t 2
x 4
P ( ( x Vt ) + ( x Vt + d1 ) + ( x Vt + d1 + d 2 ) + ( x Vt + 2d1 + d 2 ) ) +

beam S

n =

(4.5)

ws ( t ) wbeam ( nd , t ) ) + K p ( wsn ( t ) wbeam ( nd , t ) ) ( x nd )


(
t

In equations (4.2)-(4.5), ( x, y, z , t ) N m 2 is the normal stress, xz ( x, y, z , t ) N m 2


and yz ( x, y, z , t ) N m 2 are the shear stresses, wbeam ( x, t ) [ m ] is the vertical beam
displacement, wn0 ( t ) [ m ] is the vertical deflection of the sleeper number n ,
S beam [ kg m ] and Ebeam I N m 2 are the mass per unit length and the bending stiff-

ness of the beam, respectively, K p [ N m ] and p [ N s m ] are the stiffness and damp-

ing coefficient of a pad, respectively, M [ kg ] is the mass of a sleeper, K N m 2 is the


stiffness per unit length of the shear springs at the sleeperlayer interface, d1 and d 2
are the distances between the loads as depicted in Figure 4.2, H (...) [ ] is the Heaviside
step function and (...) [1 m ] is the Dirac delta function.
As in Chapter 2, to solve the system of governing equations (4.1)-(4.5), it is
convenient to use the Helmholtz decomposition of the displacement vector u :

u = + ,

i = 0 .

(4.6)

In terms of the potentials and , the equations of motion for the layer read [3]:

cL2 2 =

with cL2 = + 2

2
,
t 2

cT2 2 =

2
,
t 2

i = 0

(4.7)

, cT2 = . Using these potentials, the components of the dis-

placement vector and essential for our study stresses in the layer take the following
from [3]:

96

u=

z y
z x
y x
, v=
, w=
,
+

+
+

x
y
z
y
x
z
z
x
y

2 y x
2

zz = + 2 2 +

z x
y
z
2 z y y x
+

xz = 2
+

z x x
y
xz z y

(4.8)
(4.9)
(4.10)

2 z x y x

yz = 2

+
z y x
y
yz z x

(4.11)

We will solve the problem at hand by using the following integral Fourier transforms over time and the horizontal co-ordinates:

{ f , g}( k1 , k2 , z, ) = { , }( x, y, z, t ) exp ( i t ik1 x ik2 y ) dxdydt


{u , v , w }( k1 , k2 , z, ) = {u, v, w}( x, y, z, t ) exp ( i t ) dt

{u

, k1 , k2 , v , k1 , k2 , w , k1 , k2 } ( k1 , k 2 , z , ) =

{u, v, w}( x, y, z, t ) exp ( it ik x ik y ) dxdydt (4.12)


1

beam

( x, ) = wbeam ( x, t ) exp ( i t ) dt ,

wbeam
, k1 ( k1 , ) =

w ( x, t ) exp ( i t ) dxdt
beam

Wn ( ) =

w ( t ) exp ( i t ) dt
n
s

Application of these transforms to equations (4.7), (4.2)-(4.5), with the use of Eqs.(4.8)(4.11), yields:

For the equations of motion for an inner element of the layer (from (4.7)):

2 f 2
2
2
+

k
k

f =0
1
2
z 2 cL2

2
2

g
+ 2 k12 k22 g = 0
2
z cT

ik1 g x + ik2 g y +

(4.13)

g z
=0
z

where c~L2 = c L2 i * + 2 * / , c~T2 = cT2 i * / are complex values, cL =

b + 2 g /

is the propagation velocity of the dilatational waves (P-waves), cT = / is the


propagation velocity of the shear waves (S-waves).

97

For the boundary conditions at the surface of the layer z = 0 (from(4.2), using (4.9)(4.11)):

g y
1 sin k1b sin k2 a
f
+ ik2 g z
Ku ( nd , 0, 0, ) exp ( ik1nd )
+ ik1 ( ik1 g y ik2 g x ) =
z z
z
k1b
k2 a n =
f
g
ik1 g z x + ik2 ( ik1 g y ik2 g x ) = 0
2ik2
z z
z
2ik1

2 f
sin k1b sin k2 a
k12 k22 f + 2 2 + ( ik1 g y ik2 g x ) =

2
z
k2 a
k1b
z

M 2Wn ( ) + ( K p i p ) ( w ( nd , 0, 0, ) wbeam ( nd , ) ) exp ( ik1nd )


n =

(4.14)
with = i * , = i * complex numbers.

For the boundary conditions at the bottom of the layer z = H (from (4.3), using
(4.8)):
ik1 f + ik2 g z

g y

=0
z
g
ik2 f ik1 g z + x = 0
z
f
+ ik1 g y ik2 g x = 0
z

For the compatibility condition (from (4.4)):


Wn ( ) = w ( nd , 0, 0, )

(4.15)

(4.16)

For the vertical motion for the beam (from (4.5), using (4.16))

beam

S 2 + Ebeam I k 4 ) wbeam
, k1 =

2 P ( kV ) 1 + exp id1 + exp i ( d1 + d 2 ) + exp i ( 2d1 + d 2 )


V
V
V

(4.17)

+ ( K p i p ) ( w ( nd , 0, 0, ) wbeam ( nd , ) ) exp ( ik1nd )


n =

Let us consider the system of equations (4.13)-(4.17). The general solution to the
first two equations of Eqs. (4.13) reads

98

f = A1e RL z + A2 e RL z
g x = A3e RT z + A4 e RT z

(4.18)

g y = A5e RT z + A6 e RT z
g z = A7 e RT z + A8e RT z

with RL ,T = k12 + k22 2 cL2,T . Substituting solution (4.18) into the boundary conditions (4.14) and(4.15), one obtains the following system of linear algebraic equations
with respect to Ai :
0

0
ik1a1

ik2 a1
RL a1

2ik1 RL

2ik2 RL

ik1

ik2

RT

ik1

ik2

ik1 / a1

RT b1

RT / b1

ik2b1

ik2 / a1

RT b1

RT / b1

ik1b1

RL / a1

ik2b1

ik2 / b1

ik1b1

ik1 / b1

2ik1 RL

k1k2

k1k2

2ik2 RL

R +k

2ik2 RT

2
T

2
2

R +k
2
T

2
2

2ik2 RT

( RT2 + k12 ) ( RT2 + k12 )

ik2 RT

k1k2

k1k2

ik1 RT

2ik1 RT

2ik1 RT

A1 0

RT A2 0

ik2 / b1 A3 0

ik1 / b1 A4 0
=
0 A5 0

ik2 RT A6 F1

ik1 RT A7 0

0 A8 F3
0

(4.19)
with

= 2 RT2 + 2 cT2 ,
F1 =

1 sin k1b sin k2 a


Ku ( nd , 0, 0, ) exp ( ik1nd ),
k1b
k2 a n =

F3 =

1 sin k1b sin k2 a

k1b
k2 a

(4.20)

M 2Wn ( ) + ( K p i p ) ( w ( nd , 0, 0, ) wbeam ( nd , ) ) exp ( ik1nd )


n =
The system of algebraic equations (4.19) can be solved readily to give analytical
expressions for Ai , i = 1,8 . Subsequently substituting these expressions into Eqs.(4.18)
and transformed Eqs.(4.8) the following representations for the Fourier displacements
of the layer surface in the x direction and z direction can be obtained:

u ,k1 ,k2 ( k1 , k2 , 0, ) = a11 F1 + a13 F3


w , k1 ,k2 ( k1 , k2 , 0, ) = a31 F1 + a33 F3
with

99

(4.21)

a11 =

1
k22 RL RT sinh RT H

(( 4 ( k

2
1

+ RT2 sinh RL H k22 ( + 4 RL2 ) sinh 2 RT H + k12


+ RL RT3 sinh RT H

( ( 2k

2
1

)
( 2 ( k

+ k22 ) + cosh RL H cosh RT H 2 ( + k12 + k22 )

+ k22 ) cosh 2 RT H +

2
1

) cosh RL H cosh RT H + 2k22 + k22 ( k12 + k22 ) sinh RL H sinh 2 RT H

a13 = k1a130 ,
a130 =

iRT
cosh RT H ( k12 + k22 ) sinh RL H sinh RT H +

)}

+ RL RT 2 RL RT sinh RL H sinh RT H + + 2 ( k12 + k22 ) (1 cosh RL H cosh RT H ) ,


a33 = cosh RT H

RL RT 2
RL RT sinh RL H cosh RT H ( k12 + k22 ) cosh RL H sinh RT H ,
cT2

a31 = a13 ,

{ (

= RT cosh RT H RL RT 4 ( k12 + k22 ) + 2 cosh RL H cosh RT H


2

( k12 + k22 ) ( 4 RL2 RT2 + 2 ) sinh RL H sinh RT H + 4 RL RT

)} = R
0

(4.22)

cosh RT H

Application of the inverse Fourier transforms over the wave numbers k1 and k2 to
equations(4.21), followed by substitution x = md and y = 0 yields
u ( md , 0, 0, ) =
w ( md , 0, 0, ) =

1
4

1
4 2

(a

F + a13 F3 ) eik1md dk1dk2

11 1

(a

F + a33 F3 ) e

31 1

ik1md

(4.23)

dk1dk2

Equations (4.23) establish the relationship between the displacements


u ( md , 0, 0, ) and w ( md , 0, 0, ) of the layer surface under the midpoints of the
sleepers and the vertical displacement wbeam ( nd , ) of the beam in the supported points
(the latter displacement is hidden in F3 , see(4.20)). To find the steady-state response of
the structure, we have to couple the relationship (4.23) with the equation of motion for
the beam (4.17).
Solving Eq.(4.17) with respect wbeam
, k1 , applying to the result the inverse Fourier
transforms over k1 to Eq. (4.17) and using the compatibility condition (4.16), we obtain

wbeam ( md , ) =
1
+
2

( K

n =

1
2

id1
i ( d1 + d 2 )
i ( 2 d1 + d 2 )
P ( k1V ) eik1md dk1
V
V
V
1
+
e
+
e
+
e
+
Ebeam Ik14 beam S 2

i p ) ( w ( nd , 0, 0, ) w

beam

eik1md e ik1nd dk1


( nd , ) )
E Ik 4 beam S 2
beam 1

By evaluating the first integral analytically, Eq.(4.24) can be simplified to

100

(4.24)

beam

w
+

1
2

i md
V

id1
i ( d1 + d 2 )
i ( 2 d1 + d 2 )

1
PV e
V
V
V
+e
( md , ) =
1 + e + e
+
2 Ebeam I 4 beam S 2V 4

( K

n =

i p ) ( w ( nd , 0, 0, ) wbeam ( nd , ) )

e
dk1
2
beam Ik beam S

(4.25)

ik1md ik1nd
4
1

The first term on the right-hand side of Eq.(4.25) is the loading term, which
determines the spatial variation of the steady-state response in the frequency domain.
This variation must be the same as that of the loading term, e.g. the response must be
proportional to exp ( i md V ) .
Thus, to find the steady-state response of the structure, we have to search for the
solution of equations (4.23) and (4.25) in the form:
wbeam ( md , ) = C0 ( ) exp ( i md V )
w ( md , 0, 0, ) = C ( ) exp ( i md V ) ,

(4.26)

u ( md , 0, 0, ) = A ( ) exp ( i md V )
where C0 ( ) , C ( ) and A ( ) are unknown functions of the radial frequency.
Substituting Eqs.(4.26) into Eqs. (4.23) and (4.25), the latter can be rewritten as

(
(C ( K

),
)C)

A = KJ11 A + J13 C0 ( K p i p ) + ( M 2 + K p i p ) C
C = KJ 31 A + J 33

2
p i p ) + ( M + K p i p

(4.27)

with
J ij =


4 2

aij

sin k1b sin k2 a


exp ( ik1d ( m n ) iq ( m n ) )dk1dk2 .
k1b
k2 a n =

(4.28)

In expression (4.28),
q=

(4.29)

and represents the phase shift between vibrations of neighbouring sleepers.


In Eqs.(4.27), A and C are the amplitudes of the horizontal (in the x direction) and vertical vibrations of the layer surface under the midpoint of a sleeper, respectively, whereas C0 is the amplitude of the vertical vibrations of the beam in the point
that is connected to this sleeper. These equations can be rewritten in the following matrix form:
KJ11 1
J13 ( M 2 + K p i p ) A
J

= C0 ( K p i p ) 13
KJ 31
J 33 ( M 2 + K p i p ) 1 C
J 33

101

(4.30)

Now, employing the idea of the equivalent dynamic stiffness, used in Chapters 2 and 3,
we introduce a complex stiffness matrix l s that describes the dynamic stiffness of the
layer in the contact point with a sleeper. The inverse to this matrix is defined as
J11
J 31

l1s =

J13
J 33

(4.31)

and depends on the angular frequency of vibrations of the sleepers and the phase
shift q between vibrations of the neighbouring sleepers. With the help of the equivalent
stiffness l s , Eq.(4.30) can be rewritten in the following, more comprehensible form

(K + K

ps

0
A
+ M ps + C ps + l s ) = ( K ps + C ps )
C
C0

(4.32)

with K the stiffness matrix of the interface between the sleepers and the layer surface,
K ps , M ps and C ps the stiffness, the mass and the damping matrices of the sleeper-pad
system. It is visible now that Eq.(4.32) relates the vertical vibrations of the beam in the
supported points to the vibrations of the layer surface beneath the supports.
To proceed with the analysis, we have to evaluate the compliance matrix l1s ,
that is the coefficients J ij . To do so, it is convenient to introduce a new index of summation l = m n in Eq.(4.28) and rewrite it as
J ij =

1
4 2

aij

sin k1b sin k2 a


exp ( i ( k1d q ) l )dk1dk2 .
k1b
k2 a l =

(4.33)

To evaluate Eq.(4.33), we present it in the following form

1
sin k2 a 0
J ij = 2
Sij ( k2 , ) dk2 ,

4 2iab k2

(4.34)

where

Sij0 ( k2 , ) =

n =

exp ( iqn ) aij

ik1 ( dn + b )

e
k1

ik1 ( dn b )

dk1

(4.35)

Let us first deal with the integral that enters Eq.(4.35). The integrand of this integral consists of two terms, one proportional to exp ( ik1 ( dn + b ) ) and the other one
proportional to exp ( ik1 ( dn b ) ) . Thus, considering the following integral, we can
evaluate both these integrals at ones:
Z ij ( r ) =

aij ( k1 , k2 , )

eik1r
dk1 ,
k1

102

r = nd b .

(4.36)

Integration of (4.36) will be performed using the method of contour integration.


To apply this method, singular points of the integrands should be determined. The integrands in Eq. (4.36) are single-valued functions despite of the presence of radicals RL
and RT . The radicals do not make the integrands multi-valued since the replacement
RL ,T RL ,T does not change the integrand. Thus, only poles of the integrands should
be considered. All these poles are simple and defined by the following equations:

for Z 33
0 = 0, k1 = 0

(4.37)

0 = 0

(4.38)

0 = 0 , cosh RT H = 0 and k1 = 0

(4.39)

for Z13

for Z11

( k1 )

Fig. 4.3. Integration contour for the case r > 0 .

Thus, in principle, we know positions of all the poles in the complex k1 plane.
Now, in accordance with the contour integration method, we have to choose a curve to
close the original integration path that runs from minus to plus infinity along the real
axis. This is normally done with the help of a semicircle that is positioned either in the
upper or the lower half-plane of the complex k1 plane. Which half-place is to be chosen is dictated by the Jordans lemma [1], which ensures that the integration along the
semi-circle vanishes. In application to the integral (4.36), the path has to be closed over
the upper half-plane if r > 0 (see Figure 4.3) and over the lover half-plane if r < 0 .
Now we can apply the residue theorem, according to which an integral along a
closed contour can be expressed through the sum of residues of the integrand. In our
case this theorem yields
103

for r > 0 :

a11 ( k1 , k2 , )

eik1r
eik1r
dk1 = Z11 ( r ) ia11 ( 0, k 2 , ) = 2 i res a11 ( k1 , k 2 , )

k1
k1 k = k m
m

ik1r

e
a ( k , k , ) k
33

ik1r

e
a ( k , k , ) k
13

e
,(4.40)
dk1 = Z 33 ( r ) ia33 ( 0, k 2 , ) = 2 i res a33 ( k1 , k 2 , )

k
m
m

1 k1 = k1
dk1 = Z13 ( r ) = 2 i res {a130 ( k1 , k 2 , ) eik1r }
m

ik1r

k1 = k1m

for r < 0 :

a11 ( k1 , k2 , )

eik1r
eik1r
dk1 = Z11 ( r ) + ia11 ( 0, k 2 , ) = 2 i res a11 ( k1 , k 2 , )

k1
k1 k = k l
l

ik1r

e
a ( k , k , ) k
33

ik1r

e
a ( k , k , ) k
13

e
.(4.41)
dk1 = Z 33 ( r ) + ia11 ( 0, k 2 , ) = 2 i res a33 ( k1 , k 2 , )

k
l
1 k1 = k1l

ik1r

dk1 = Z13 ( r ) = 2 i res {a130 ( k1 , k 2 , ) eik1r }


l

k1 = k1l

with k1m and k1l the poles located in the upper half-plane and the lower half-plane, re-

spectively Im ( k1m ) > 0, Im ( k1l ) < 0 .

Expressing Z ij ( r ) from equations (4.40) and (4.41), we obtain

for r > 0 :

eik1r
Z11 ( r ) = ia11 ( 0, k 2 , ) + 2 i res a11 ( k1 , k 2 , )

k1 k = k m
m

1
1

eik1r
,
Z 33 ( r ) = ia33 ( 0, k 2 , ) + 2 i res a33 ( k1 , k 2 , )

k1 k = k m
m

1
1
Z13 ( r ) = 2 i res {a130 ( k1 , k 2 , ) eik1r }
m

(4.42)

k1 = k1m

for r < 0 :

eik1r
Z11 ( r ) = ia11 ( 0, k 2 , ) 2 i res a11 ( k1 , k 2 , )

k1 k = k l
l

1
1

eik1r
Z 33 ( r ) = ia11 ( 0, k 2 , ) 2 i res a33 ( k1 , k 2 , )

k1 k = k l
l

1
1
Z13 ( r ) = 2 i res {a130 ( k1 , k 2 , ) eik1r }
l

104

k1 = k1l

(4.43)

Now that expressions for Z ij ( r ) have been found, we can return to Eq.(4.35) for
Sij0 ( k2 , ) , which can be expressed through Z ij ( r ) in the following manner
Sij0 ( k2 , ) =

exp ( iqn ) ( Z ( nd + b ) Z ( nd b ) ) =
ij

n =

ij

n =1

n =1

Z ij ( b ) Z ij ( b ) + e iqn ( Z ij ( nd + b ) Z ij ( nd b ) ) + eiqn ( Z ij ( nd + b ) Z ij ( nd b ) )

(4.44)
Since the distance between the neighbouring sleepers d is always larger than
the longitudinal size of the sleepers b , it is obvious that for n 1 , nd b > 0 and
nd b < 0 . This implies that calculating Z ij ( nd b ) we have to use expressions
(4.42), while evaluating Z ij ( nd b ) expressions (4.43) have to be used.
Equation (4.44) can be simplified by using a symmetry property of the functions
Z ij ( r ) . This property reads
Z11 ( r ) = Z11 ( r )
Z 33 ( r ) = Z 33 ( r )

(4.45)

Z13 ( r ) = Z13 ( r )
and follows from the following symmetry properties of the functions a11 , a33 and a13 :
a11 ( k1 , k2 , ) = a11 ( k1 , k2 , ) ,
a33 ( k1 , k2 , ) = a33 ( k1 , k2 , ) ,

(4.46)

a130 ( k1 , k2 , ) = a130 ( k1 , k2 , ) ,
which implies that

eik1r
eik1r
i
res
a
k
k
2 i res aii ( k1 , k2 , )
=
2

,
,

l ii ( 1 2 ) k l ,
k
m
m
1 k1 = k1
1 k1 = k1

2 i res {a130 ( k1 , k2 , ) e

ik1r

k1 = k1m

= 2 i res {a130 ( k1 , k2 , ) e
l

ik1r

k1 = k1l

i = 1,3
(4.47)

Employing relationships (4.45), Eq.(4.44) is reduced to

Sii0 ( k2 , ) = 2 Z ii ( b ) + ( einq + e inq ) Z ii ( nd + b ) Z ii ( nd b ) , i = 1,3


n =1

0
13

( k2 , ) = ( e
n =1

inq

inq

) ( Z ( nd + b ) Z ( nd b ) ) .
13

13

with

105

(4.48)


eik1r
Z ii ( r ) = 2 i res aij ( k1 , k2 , )
,

k1 k = k m
m

1
1

i = 1,3

(4.49)

To obtain formulae (4.48), we used the fact that

eik1r
+ iaii ( 0, k2 , )
Z ii ( nd + b ) Z ii ( nd b ) = 2 i res aii ( k1 , k2 , )

k1 k1 = k1m ,
m

r = nd + b

eik1r

2 i res aii ( k1 , k2 , )
+ iaii ( 0, k2 , ) =

m
k1 k1 = k1 ,
m

r = nd b

(4.50)

eik1r
eik1r
= 2 i res aii ( k1 , k2 , )
2 i res aii ( k1 , k2 , )
=

k1 k1 = k1m ,
k1 k1 = k1m ,
m
m

r = nd + b
r = nd b
= Z ii ( nd + b ) Z ii ( nd b )

Further simplification of Eqs.(4.48) can be done by analytically calculating the


infinite series with respect to n . Let us show the way to accomplish this calculation on
the hand of the following term (for all other terms, this can be done in exactly the same
manner):

Q = einq Z ii ( nd + b )

(4.51)

n =1

Substituting expression for Z ii given by (4.49) into this equation, we obtain

eik1r
Q = 2 i einq res aii ( k1 , k2 , )

k1 k = k m
n =1
m

1
1

(4.52)

Introducing the notation k1m = km + ikm with km and km real values and km > 0 (by definition, see (4.40)), and changing the order of summation, Eq.(4.52) can be rewritten as
ik nd + b

e 1 ( )
Q = 2 i einq res aij ( k1 , k2 , )
=

k1 k = k m
n =1
m

1
1

2 i e
m

ib ( km + ikm )

a ( k , k , )
res ij 1 2

k1

k1 = k1m

(4.53)

ind ( q + km ) ndk

n =1

The series over n in Eq.(4.53) is a sum of geometric progression with infinite number
ik nd k nd
of terms and the ratio of successive terms p = e inq e m e m , p < 1 .
Using the following formula [1] for the sum of the geometric progression:

106

n =1

p
,
1 p

(4.54)

expression (4.53) can be rewritten as


Q = 2 i e

ib ( km + ikm )

2 i e

ib ( km + ikm )

2 i e

ibk1m

aij ( k1 , k2 , )
res

k1

k1 = k1m

ind ( q + km ) ndkm

n =1

exp ( id ( q + km ) dkm )
aij ( k1 , k2 , )
res
=

k1

k1 = k1m 1 exp ( id ( q + km ) dkm )

(4.55)

exp ( ik1m d iq )
aij ( k1 , k2 , )
res

m
k1

k1 = k1m 1 exp ( ik1 d iq )

Carrying out the summation over n in the other terms in Eqs.(4.48), these equations can be reduced to the following form:
Sii0 ( k2 , ) = 2 iaii ( 0, k2 , ) +
a ( k , k , )
2 i res ii 1 2

k1
m =1

k1 = k1m

m
m
m
2eik1 b + eik1 b e ik1 b

S130 ( k2 , ) = 2 i res {a130 ( k1 , k2 , )}


m =1

k1 = k1m

(e

ik1m b

)
m

exp ( ik1m d + iq )
exp ( ik1m d iq )
+

m
m
1 exp ( ik1 d + iq ) 1 exp ( ik1 d iq )

e ik1 b

exp ( ik1m d iq )
exp ( ik1m d + iq )

m
m
1 exp ( ik1 d iq ) 1 exp ( ik1 d + iq )

(4.56)
Substituting Eqs.(4.56) into Eq.(4.34), and carrying out regular algebraic manipulations,
we finally obtain the following expressions for the coefficients J ij :
Jii =

m
aii ( k1, k2 , )
sin k2a
cos q eik1 d
ik1 b
m

+
+
a
k
res
e
i
k
b
0,
,
2
sin

( 1 ) cos k md cos q dk2

k2 ii ( 2 )

k1
4ab
m
m=1
(1 )

k1 =k1

sin ( k1mb) sin q


sin k2a
0
J13 =
res {a13 ( k1, k2 , )} m
dk2 ,
k2
k1 =k1
2ab
cos ( k1md ) cos q
m=1
i

i = 1,3

(4.57)
In order to analyse Eqs.(4.57) numerically, we should find the roots of the equations
cosh RT H = 0 and 0 = 0 (see (4.37)-(4.39)). The roots of the first equation

( cosh RT H = 0 ) can be easily found analytically to give:


2

1
2
2
k =i
m
k
,
+
+

2
H2
cT2
m
1

107

(4.58)

whereas the roots of the second equation ( 0 = 0 ) can be found only numerically (using
a program for finding complex roots of a transcendental equation). Having found all
these roots, the residues in Eqs.(4.57) can be calculated and summarised straightforwardly. Naturally, the summation should be carried out over a finite number of poles,
which do not have very large imaginary and real parts (the contribution of the other
terms into the sum in (4.57) is negligible). After the summation has been accomplished,
only a well convergent integration over k2 remains to be carried out to evaluate J ij .
This integration can be accomplished using a standard program for calculation of a single integral. Thus, we can consider J ij as known complex functions of the angular frequency and the phase shift q .
Now that J ij are known, Eq.(4.32) can be analysed further. Before doing so, let
us first study the equivalent stiffness of the layer in the contact points with the sleepers,
l s . The study will be limited by considering the vertical equivalent stiffness of the
ground, which is most influential in the vertical dynamics of the railway track. To find
an expression for this component of the equivalent stiffness, the stiffness of the shear
contact between the layer surface and the sleepers should be disregarded. In this case,
the matrix equation (4.32) can be reduced to the following algebraic equation that describes the vertical vibrations of the sleepers:
2
C ( lvert
s M + K p i p ) = C0 ( K p i p ) ,

(4.59)

with

lvert
s = 1 J 33 .

(4.60)

It is easy to see that Eq.(4.60) describes the vertical vibrations of a sleeper as it is shown
in Figure 4.4 (the sleeper is presented by a black rectangular).

KP

wbeam ( nd , t )

lvert
s

wsn ( t )

Fig. 4.4. Vertical vibrations of a sleeper.

The stiffness lvert


s in Eqs.(4.59), (4.60) and in Figure 4.4 is, obviously, the vertical equivalent stiffness of the layer in the contact points with the sleepers. It is important to underline that this stiffness does not depend on the number n of the sleeper, i.e.
the layer under each sleeper can be replaced by a set of identical springs with complex
stiffness lvert
s .

108

As follows from Eqs.(4.60) and (4.57), the equivalent vertical stiffness lvert
s is
given as
1

vert
l s

m
1 sin k a

a33 ( k1, k2 ,) ik1mb


cos q eik1 d
m
2

=
+
+
a
0,
k
,

2
res
e
i
sin
k
b
( 1 ) cos kmd cos q dk2

k2 33 ( 2 )

4ab
k1
m
m=1
(1 )

k1 =k1

(4.61)
and, therefore, is a complex-valued function of the angular frequency of the beam
vibrations and the phase shift q between vibrations of the neighbouring supports.
The results of the numerical evaluation of Eq.(4.61) are presented in Figure 4.5.
The equivalent stiffness is shown in this figure as a function of the angular frequency
for the fixed value of the phase shift q = d V equal to 0.5 . The solid and the dashed
lines in Figure 4.5 correspond to the real and imaginary parts of lvert
s , respectively. The
calculations were carried out using the following physical parameters of the layer and
geometrical parameters of the sleepers:

layer:

= 2.6 107 N m 2 , = 1960 kg m3 , = 0.3, * = 2.6 104 N*s m 2 , H = 5 m .

(4.62)

sleepers:
2a = 2.7 m, d = 0.6 m, 2b = 0.27 m.

(4.63)

The parameters of the layer are almost analogous to those used in Chapter 2, thus they
describe a realistic, quite stiff ground with the Rayleigh wave speed of cR = 384 km h ,
shear wave speed of cT = 414 km h , and dilatational wave speed of cL = 775 km h .
Geometric parameters of the sleepers are common for Western European railways.
Figure 4.5 shows that the equivalent stiffness is a complicated function of the
frequency. For frequencies smaller than = qcR d , the real part of the equivalent stiffness is positive and the imaginary part, although not zero, is very small. This implies
that at this band of frequencies, for the chosen q , vibrations of the supports generate no
waves in the layer. For frequencies close to = qcR d the equivalent stiffness becomes
very small. This happens because of the Rayleigh waves, which are generated in such a
way that they arrive to every sleeper in phase (resonance), providing a large displacement of the layer under the sleepers (see [116]). Figure 4.5 shows also that the equivalent stiffness can be quite large at some frequencies. This effect can be considered as
anti-resonance, which occurs when waves, as well as exponentially decaying displacement fields, generated by the sleepers, are almost in anti-phase under any support,
resulting in small displacements of the layer under the sleepers. The frequencies that
correspond to these large values of lvert
s can not be found analytically since all three
waves in the layer as well as the decaying displacement fields take part in the phenomenon of anti-resonance.

109

12

l s 108 ( N m )

Re ( l s )

0
100
-4

200

300

400

= qcR d
Im ( l s )

-8
Fig. 4.5. Equivalent stiffness of the layer as a function of the angular frequency for q = d / V = 0.5 .

The most important qualitative result of analysis of the equivalent stiffness is


that
can be very small at some frequencies, which are determined by the equation
= qcR d . Because of this, a uniform motion of a constant load with the velocity that
is equal to the Rayleigh wave speed can cause resonance in the structure. Indeed, the
phase shift between vibrations of the neighbouring sleepers q ( ) in the case of a convert
l s

stant load is given by expression (4.29). Correspondingly, the condition for the equivalent stiffness to be small takes the form

d
cR

d
V

(4.64)

Thus, if the load velocity becomes equal to the Rayleigh wave speed, that is
V = cR , the layer reaction becomes very small at all frequencies. As a result, vibrations
of the beam that are perturbed by such a load should have large amplitude. Thus, the
Rayleigh wave velocity is critical for both continuous and discrete models of the railway track.
Existence of the maxima of the equivalent stiffness should not be forgotten as
well. They also distinguish clearly the equivalent stiffness lvert
s from a constant stiffness, which is frequently used to replace the ground reaction.

110

4.2. Steady-state response of the beam to a set of moving


loads.
In this section, it will be demonstrated how the equivalent stiffness of the layer

l s should be used to analyse the beam response to a set of uniformly moving constant
loads. The demonstration will be carried out on the example of four loads, moving at the
fixed distances d1 and d 2 from each other, as shown in Fig. 4.2.
Transforming Eq.(4.5) into the frequency domain and employing Eqs.(4.26) and
(4.30) the governing equations for the steady-state response of the beam can be written
as

id
i( d + d )
i( 2 d + d )
4 beam
2 P ixV
2 beam
V
V
V
+e
Ebeam I 4 w ( x, ) beam S w ( x, ) =
e 1 + e + e
+
x
V

(4.65)
1

+ ( K p i p ) ( w ( nd ,0,0, ) wbeam ( nd , ) ) ( x nd )
n =

wbeam ( nd , ) = C0 ( ) exp ( i nd V )
w ( nd , 0, 0, ) = C ( ) exp ( i nd V )

(4.66)

u ( nd , 0, 0, ) = A ( ) exp ( i nd V ) ,

(
( C ( ) ( K

)
) C ( ) )

A ( ) = KJ11 A ( ) + J13 C0 ( ) ( K p i p ) + ( M 2 + K p i p ) C ( )
C ( ) = KJ 31 A ( ) + J 33

i p ) + ( M 2 + K p i p

(4.67)

The system of governing equations (4.65)-(4.67) can be simplified by eliminating A ( ) from Eqs.(4.67) and substituting w ( nd , 0, 0, ) and wbeam ( nd , ) into
Eq.(4.65). This yields
Ebeam I

id
i( d + d )
i( 2 d + d )
4 beam
2 P ixV
2 beam
V
V
V

+
+
+
w
x
,

w
x
,

e
1
e
e
e
(
)
(
)

beam

4
x
V

C0 ( ) ( K p i p ) B0 ( ) exp ( i nd V ) ( x nd ),

(4.68)

n =

wbeam ( nd , ) = C0 ( ) exp ( i nd V )

with
B0 = M 2 1/ J 33 + K p i p +

KJ13 J 31
J 33 ( J 33 (1 KJ11 ) + KJ13 J 31 )

(4.69)

For the analysis to follow, it is convenient to rewrite Eqs.(4.68) making no use


of the Dirac delta-functions. This can be accomplished employing the relationship between the classical and generalised derivatives [1] to give

111

Ebeam I

id
i( d + d )
i( 2 d + d )
4 beam
2 P ixV
2 beam
V
V
V (4.70)
w
x
,

w
x
,

e
1
e
e
e

+
+
+
( ) beam
( )

x4
V

2 beam
beam
wbeam
= w
= 2 w
=0
x = nd
x
x = nd x
x = nd
3

= C0 ( ) B0 ( ) exp ( i nd V )
Ebeam I 3 wbeam
x
x = nd
wbeam ( nd , ) = C0 ( ) exp ( i nd V )
where
the
square
brackets
denote
f ( x ) x = a = f ( x = a + 0 ) f ( x = a 0 ) .

the

following

(4.71)

(4.72)
difference:

Equation (4.70) describes vertical vibrations of the beam under the moving loads
in all points, which are not supported. In the supported points, the boundary conditions
given by Eqs.(4.71), (4.72) are to be employed.
To find the steady-state solution to the system of equations (4.70)-(4.72) we will
utilise the method presented in [158], [10]. This method is based on the fact that the
beam displacement in the frequency domain satisfies the following condition:

wbeam ( x + d , ) = wbeam ( x, ) exp id
V

(4.73)

This condition can be easily validated by direct substitution into Eqs.(4.70)-(4.72).


Condition (4.73) is commonly referred to as the periodicity condition. It gives a
relationship between the beam displacements in the points that are separated by the span
distance d . Physically this condition implies that in steady-state regime the beam displacement is repeated in time with the period T = d / V , but with translation in space
equal to d .
The periodicity condition (4.73) allows to obtain the solution to the problem
(4.70)-(4.72) performing the following three steps.
1. First, the general solution to Eq.(4.70) is to be written in the interval x [ 0, d ] . This
solution will contain four unknown constants Ci , i = 1..4 .
2. Second, using the periodicity condition, the solution in the interval x [d , 2d ] can
be obtained from that in the interval x [ 0, d ] by multiplying it by exp ( id V ) .
Obviously, the result of this multiplication will contain the same four constants
Ci , i = 1..4 .
3. Third, four unknown constants Ci , i = 1..4 and one extra constant that is related to
the Fourier-displacement of the sleeper number n = 1 are to be found by employing
five boundary conditions (4.71), (4.72) at the point x = d .
Let us accomplish this way of solution. The general solution to equation (4.70)
in the interval x [ 0, d ] can be written as

wbeam ( x ) = C1 exp ( x ) + C2 exp ( x ) + C3 exp ( i x ) + C4 exp ( i x ) F0 exp ix (4.74)
V

112

with

= ( beam Ebeam J ) ,
14

F0 =

id1
i ( d1 + d 2 )
i ( 2 d1 + d 2 )

2 PV 3
V
V
V
1
+
e
+
e
+
e

,
Ebeam I 4 2

(4.75)

and C j ( j = 1..4 ) unknown constants.


According to the periodicity condition (4.73), the solution to Eq.(4.70) in the interval x [ d , 2d ] can be expressed as

wbeam ( x ) = exp id C1 exp ( ( x d ) ) + C2 exp ( ( x d ) )
V

+ C3 exp ( i ( x d ) ) + C4 exp ( i ( x d ) ) F0 exp ix
V

(4.76)

Substitution of (4.74) and (4.76) into the boundary conditions (4.71), (4.72)
gives the following system of linear algebraic equations:
C1 exp ( d ) + C2 exp ( d ) + C3 exp ( i d ) + C4 exp ( i d ) = q0 ( C1 + C2 + C3 + C4 )
C1 exp ( d ) C2 exp ( d ) + iC3 exp ( i d ) iC4 exp ( i d ) = q0 ( C1 C2 + iC3 iC4 )
C1 exp ( d ) + C2 exp ( d ) C3 exp ( i d ) C4 exp ( i d ) = q0 ( C1 + C2 C3 C4 )

EI 3 q0 ( C1 C2 iC3 + iC4 ) ( C1 exp ( d ) C2 exp ( d ) iC3 exp ( i d ) + iC4 exp ( i d ) ) =


= C0 ( K p i p ) B0 ( ) q0
C1 + C2 + C3 + C4 F0 = C0

(4.77)
with q0 = exp ( i d V ) .
When solved for C j ( j = 1..4 ) this system gives:

113

C j = j ,

d
d

= 4iq0 4 EJ 3 B0 (T0 ( B0 T0 ) ) cos


cosh d cos
cos d +
V
V

+ cos
( sinh d sin d ) + cosh d sin d sinh d cos d
V

1 = iF0 q03 q02 ( exp ( d ) + 2 cos ( d ) ) + q0 (1 + 2 exp ( d ) cos ( d ) ) exp ( d ) ,

{
}
= F {q q ( exp ( i d ) + 2 cosh ( d ) ) + q (1 + 2 exp ( i d ) cosh ( d ) ) exp ( i d )} ,
= F {q q ( exp ( i d ) + 2 cosh ( d ) ) + q (1 + 2 exp ( i d ) cosh ( d ) ) exp ( i d )}

2 = iF0 q03 q02 ( exp ( d ) + 2 cos ( d ) ) + q0 (1 + 2 exp ( d ) cos ( d ) ) exp ( d ) ,


3
4

3
0

3
0

2
0

2
0

(4.78)
with T0 = K p i p .
Thus, the solution to system (4.70)-(4.72) is obtained in the frequency domain.
The Fourier-displacement of the beam in the interval x [ 0, d ] is determined by expressions (4.74) and (4.78). To get the solution for other values of the co-ordinate x , one
has to apply the periodicity condition (4.73). The beam displacement in time domain
can be found numerically as the inverse Fourier integral transform:
wbeam ( x, t ) =

1
2

ix
x
x
i x
i x
i t
V
+
+
+

C
e
C
e
C
e
C
e
F
e
e d
2
3
4
0
1

(4.79)

The further analysis will be carried out in the following way. First, the beam
displacement at the loading point will be analysed as a function of the load velocity.
Second, the influence of the damping in the supports (pads) on the amplitude of the
beam vibrations will be investigated. To comprehend the importance of the discrete
sleepers, this displacement will be compared to that of a correspondent homogeneous
model. Then, the beam profiles correspondent to a single load, two loads (bogie) and
four loads (wagon) will be considered.
The velocity dependence of the beam deflection of the loading point is calculated using the following parameters of the structure:

for the layer:

= 2.6 107 N m 2 , = 1960 kg m3 , = 0.3, * = 2.6 104 N*s m 2 , H = 5 m, (4.80)

for the beam and for the load:


Ebeam I = 1.222 107 N*m 2 , beam S = 2 60.34 kg m , P = 200000 N,

(4.81)

for the pads and the sleepers:


2a = 2.7 m, d = 0.6 m, 2b = 0.27 m, M discrete = 250kg,
K Pdiscrete = 108 N m , Pdiscrete = 105 N*s m ,

114

(4.82)

For comparison, the beam deflection is calculated on the basis of a correspondent homogeneous model, which can be retrieved from the model considered in Chapter
3. To have an exact correspondence between the models, the ballast should be removed
from the model of Chapter 3 and the remaining parameters should be taken as
M cont =

0.0045

M discrete
K discrete
discrete
, K Pcont = P
, Pcont = P
.
d
d
d

0.0035

wbeam ( m )

(4.83)

wbeam ( m )

0.0040

xload = 0

0.0035

xload = 0

0.0030

Homogeneous
model

Homogeneous
model

0.0030
0.0025

0.0025

0.0020
0.0015

0.0020

0.0010
0.0005

0.0015
0

100

200

300

V ( km h )

400

500

100

200

a)
0.0035

300

V ( km h )

400

500

b)
wbeam ( m )
xload = 0

0.0030

Homogeneous
model

0.0025

0.0020

0.0015
0

100

200

300

V ( km h )

400

500

c)
Fig. 4.6. Beam displacement versus load velocity for a) * = 2.6 104 N*s m 2 , b) * = 104 N*s m 2
and c) * = 103 N*s m 2 .

115

0.0040

0.0040

wbeam ( m )

wbeam ( m )

0.0035

0.0035

P = 105 N*s m

0.0030
0.0025

0.0025

0.0020

0.0020

0.0015

0.0015

0.0010

P = 106 N*s m

P = 105 N*s m

0.0030

0.0010

P = 5 106 N*s m

P = 106 N*s m

P = 5 106 N*s m

0.0005

0.0005
0

100

200

300

400

500

100

200

a)
0.0040

300

400

500

V ( km h )

V ( km h )

b)
wbeam ( m )

0.0035

P = 105 N*s m

0.0030
0.0025
0.0020
0.0015
0.0010

P = 106 N*s m

P = 5 106 N*s m

0.0005
0

100

200

300

400

500

V ( km h )

c)
Fig. 4.7. Beam displacement at the loading point versus load velocity for different values of viscosity in
the pads for a) * = 2.6 104 N*s m 2 , b) * = 104 N*s m 2 and c) * = 103 N*s m 2 .

Figure 4.6 (a,b,c) represents the beam displacement at the loading point as a
function of the load velocity for three values of the material damping in the layer. The
solid line corresponds to the periodically inhomogeneous model, whereas the dashed
line is related to the homogeneous one. For the inhomogeneous model, the displacement
is calculated at the time moment when the load passes the sleeper n = 0 ( xload = 0 ) 1.
1

Calculations show that the beam deflection at the loading point depends on the load position only
slightly.

116

Figure 4.6 shows that the velocity dependence of the beam deflection is almost
the same for both models. Slight difference takes place at small velocities and in the vicinity of the critical velocity (V = 390 km h ) . The latter is larger than the Rayleigh

(V = 384 km h )

wave velocity

and smaller than the shear wave velocity

(V = 414 km h ) .
Figure 4.7 (a,b,c) shows the effect of the damping in the pads on the beam displacement at the loading point. One can see from the figure that this effect is perceptible. The larger is the damping, the smaller is the amplitude of the beam vibrations. It
should be mentioned, however, that the smaller is the material damping in the layer, the
less sensitive is the beam response to the damping in the pads.
The characteristic beam profiles are presented in Figures 4.8-4.10 (a,b,c) for the
motion of a single load, two loads (bogie) and four loads (wagon), respectively. The
profiles are plotted for three values of the load velocity: 250 km h ( V < cR ), 400 km h
( cT < V < cL ) and 500 km h ( cT < V < cL ). These velocities are chosen to represent
three regimes of motion. The velocity of 250 km h corresponds to the subcritical regime and is a characteristic velocity of nowadays operating Western-European highspeed trains. The velocity 400 km h is representative for the near-critical response of
the beam. As for the last velocity, 500 km h , this is about the record for modern trains.
The displacements are calculated at the time moment when the front load
( x = Vt ) is in the point x = 0 , i.e. exactly above the sleeper n = 0 . For the distances between the loads, the dimensions of German ICE are adopted:
d1 = 3 m, d 2 = 8 m,

(4.84)

Here d1 is the bogie wheelbase, d 2 is the distance between two car bogies (see Figure
4.2).

0.003

0.003

wbeam ( m )

0.002

0.002

0.001

0.001

0.000

0.000

-0.001

-0.001

-0.002

-0.002

-0.003

-0.003

0.002

wbeam ( m )

wbeam ( m )

0.001

0.000

-0.001

-0.002

-0.004

-0.004
-30

-20

-10

x ( m)

a)

10

20

30

-0.003
-30

-20

-10

x ( m)

b)

10

20

30

-30

-20

-10

x ( m)

10

20

30

c)

Fig. 4.8. Characteristic beam deflections for single load, (a) 250 km h , (b) 400 km h and (c) 500 km h .

117

0.003

0.003

wbeam ( m )

0.002

0.002

0.001

0.001

0.000

0.000

-0.001

-0.001

-0.002

-0.002

-0.003

-0.003

0.002

wbeam ( m )

wbeam ( m )

0.001

0.000

-0.001

-0.002

-0.004

-0.004
-30

-20

-10

x ( m)

10

20

30

-0.003
-30

-20

-10

a)

x ( m)

10

20

30

-30

-20

-10

b)

x ( m)

10

20

30

c)

Fig. 4.9. Characteristic beam deflections for two loads (car bogie), (a) 250 km h , (b) 400 km h
and (c) 500 km h .

0.003

0.003

wbeam ( m )

0.003

wbeam ( m )

0.002

0.002

0.002

0.001

0.001

0.001

0.000

0.000

0.000

-0.001

-0.001

-0.001

-0.002

-0.002

-0.002

-0.003

-0.003

-0.003

-0.004

-0.004
-30

-20

-10

x ( m)

a)

10

20

30

wbeam ( m )

-0.004

-30

-20

-10

x ( m)

b)

10

20

30

-30

-20

-10

x (m)

10

20

30

c)

Fig. 4.10. Characteristic beam deflection for four loads (wagon), (a) 250 km h , (b) 400 km h
and (c) 500 km h .

Figures 4.8-4.10(a) show that in the subcritical motion the loads generate no
waves in the system. The response is almost symmetric with respect the loads. A small
asymmetry is related to the presence of the damping in the pads and the layer.
In the supercritical regime, the loads perturb waves in the beam. The length of
these waves can be in the order of the bogie wheelbase d1 and the car distance d 2 . Interfering between each other, these waves can result in a decrease (see Fig. 4.9(b) and
Fig. 4.10(b)) or increase (see Fig. 4.9(c) and Fig. 4.10(c)) of the maximum displacement
of the beam as compared to the case of a single load. This implies that for different velocities and for different values of d1 and d 2 the interference can lead to either decrease
or increase of the maximum beam displacement.
Before finishing this section, one should mention that the plotted curves are not
stationary in time. They are repeated with the period T = d V , but constantly changing,
breathing in time.

118

4.3. Elastic drag experienced by a high-speed train due to excitation of ground vibrations.
In this section, we proceed with investigation of the elastic drag experienced by
a train due to generation of waves in the railway track. First, we will analyse the elastic
drag for a single load. Then, the influence of the bogie wheelbase on the visco-elastic
drag will be studied for two loads. And, at last, on the basis of the obtained results, the
elastic drag experienced by a high-speed train (French TGV) will be calculated and
compared to that obtained in Chapter 2 on the hand of the homogeneous model.
The formula (3.38) given in Chapter 2 for elastic drag De experienced by constant loads moving over a beam remains valid for periodically inhomogeneous system.
For the case of a single load (wheel) this formula takes the form:
Dewheel =

P wbeam
V t

(4.85)
x =Vt

Results of numerical integration of equation (4.85) are presented in Figure 4.11


that shows the dependence of the elastic drag Dewheel on the load velocity V at the moment t = 0 , when the load passes the sleeper n = 0 . The figure presents three curves
that are plotted for three values of the damping coefficient * , namely for
1* = 103 Ns/m 2 , 2* = 104 Ns/m 2 , 3* = 2.6*104 Ns/m 2 . The other parameters are given
by (4.80)-(4.82).

0.7

Dewheel ( kN )

0.6

*1
*2

0.5

*3

0.4
0.3
0.2
0.1
0
0

100

200

300

400

500

V ( km h )

Fig. 4.11. Elastic drag for a single load versus the load velocity.

Figure 4.11 shows that the elastic drag grow rapidly as the load velocity approaches the critical one. This result is in perfect correspondence with that obtained in
Chapter 2 for homogeneous model. The specific feature of the drag in the periodically
119

inhomogeneous model is that the drag is a periodic (T = d V ) function of time (not a


constant as for homogeneous models). Therefore, to obtain a physically relevant value
for the drag its value should be averaged with respect to sleeper period to give
d

Dewheel =

P wbeam
Vd 0 t

dx

(4.86)

t=x V

For a single load, however, Dewheel and Dewheel appear to be almost the same.
Therefore, Figure 4.11 is representative for Dewheel as well.
Consider now the elastic drag experienced by two loads moving at a fixed distance d1 from each other (bogie). This drag is given as
P wbeam

V t

Debogie =

+
x =Vt

wbeam
t

x =Vt d1

(4.87)

To study the effect of the bogie wheelbase d1 , Debogie is shown in Figure 4.12 as
a function of d1 for two velocities of the loads. Figure 4.12(a) presents the drag at the
time moment t = 0 , whereas Figure 4.12(b) shows the drag at t = d 2V .

1.000

1.000

Debogie ( kN )

Debogie ( kN )

V = 250 km h

V = 250 km h

V = 400 km h

V = 400 km h

0.100

0.100

0.010

0.010

0.001

0.001
0

d1 ( m )

12

16

20

a)

d1 ( m )

12

16

b)

Fig. 4.12. Elastic drag experienced by car bogie versus the wheelbase for (a) x = 0 and (b) x = d 2 .

120

20

The figure demonstrates the following:


1) Elastic drag in the subcritical case ( 250 km h ) is less than that in the supercritical one ( 400 km h ) .
2) Dependence of the elastic drag on the wheelbase has a minimum located approximately at d = 2 m .
3) Elastic drag experienced by a bogie depends upon the wheelbase and is quite
sensitive to variations of this distance, especially in the subcritical regime of
motion. This is a consequence of the periodicity of the structure.
4) Elastic drag depends on the time moment, at which it is calculated.
Let us study the average value of the drag, which is given by the expression
Debogie =

d
P wbeam

Vd 0 t

+
x =Vt

wbeam
t

dx

x =Vt d1

(4.88)

The result of the averaging is shown in Figure 4.14.

1.00

Debogie ( kN )

V = 250 km h
V = 400 km h

0.10

0.01
0

12

16

20

d1 ( m )

Fig. 4.14. Elastic drag averaged over the structure period versus the wheelbase for V = 250 km h and
V = 400 km h .

The figure shows that the averaging removes the oscillatory dependence of the
drag on the wheelbase d1 . Additionally, as the distance d1 exceeds 10m , the drag becomes almost independent of this distance. This implies that the drag experienced by
two bogies of one wagon can be found by simple doubling of the drag acting on one bogie.

121

Consider now the drag experienced by a high-speed train. To be able to compare


results to those obtained in Chapter 2, we again analyse a TGV train, which consists of
two power cars and eight passenger cars. The weight of the power car equals 80000 kg ,
the weight of the passenger car is 40000 kg . Thus, each wheelset of power cars undergoes loading of 200000 N whereas each wheelset of a passenger car has 100000 N of
constant loading. Besides this, the TGV train has the following geometric parameters:
d1 = 3 m, d 2 = 14 m, d3 = 3.27 m,

(4.89)

where d1 is the wheelbase, d 2 is the distance between the mid points of the bogies, and
d3 is the distance between the last wheelset of the first wagon and the first wheelset of
the second wagon.
The calculation result obtained by employing parameters (4.80)-(4.82) is shown
in Figure 4.15, which presents the drag as a function of the train speed. The drag is
evaluated in accordance with assumption that there is an aerodynamic resistance to the
train motion, which presses the train against the rails (see Chapter 2). This pressure is
accounted for by replacing the weight of the locomotive P by P (1 + V 2 ) , where is
a coefficient, which reflects the fact that the aerodynamic drag influences the contact
force between the train wheels and the rails. The coefficient was taken as = 0.001 ,
likewise in Chapter 2.

1000

Da
Dr

D ( kN )

Deperiod
Decont

100
10
1
0.1

Rayleigh wave
speed

0.01
0.001
0

100

200

300

V ( km h )

400

500

Fig. 4.15. Elastic drag for inhomogeneous model ( Deperiod ) , elastic drag for homogeneous model ( Decont ) ,
rolling drag ( Dr ) and aerodynamic drag ( Da ) versus the load velocity.

122

To compare the periodically inhomogeneous model to the correspondent homogeneous model, the velocity dependence of the elastic drag for the latter model is also
shown in Figure 4.15. The aerodynamic drag and the rolling drag are plotted in accordance with the results of paper [61] and are the same as in Figure 2.14.
Figure 4.15 demonstrates that the aerodynamic drag prevails among the other
drags as the train moves subcritically. However, as soon as the train velocity approaches
the Rayleigh wave speed, the elastic drag becomes comparable with the aerodynamic
drag and then exceeds it.
Comparing the homogeneous and inhomogeneous models, one can conclude that
the difference between them becomes apparent in the supercritical regime only. Figure
4.15 shows that in the supercritical regime the elastic drag for the homogeneous model
is larger. This implies that for the embedded track the energy consumption at high velocities should be expected to be higher that for the corresponding conventional track.
Thus, we have shown that the elastic drag for the conventional track, as well as
for the embedded track, is an important design criterion for high-speed trains, which
should be accounted for as a source of considerable energy loss.

4.4. Comparison of responses of 3D and 1D models of the


conventional track.
In the previous sections of this chapter it was shown that if the load velocity is
close to the Rayleigh wave speed, the waves in the ground play an important role in the
dynamic response of the conventional railway track. Therefore, modelling the train motion with velocities that are close or larger than the Rayleigh wave speed one has to account for three-dimensionality of the ground.

0.002

wbeam ( m )

3D model
1D model

0.001

0.002

3D model
1D model

0.001

-0.001

-0.001

-0.002

-0.002

-0.003

-0.003

-0.004

wbeam ( m )

-0.004
-30

-20

-10

10

20

30

x d

-30

-20

-10

0
x d

a)

b)

10

20

30

Fig. 4.16. Beam deflection profiles for 3D and 1D models for (a) V = 50 km h and (b) V = 100 km h .

123

0.002

wbeam ( m )

3D model
1D model

0.001

0.002

-0.001

-0.001

-0.002

-0.002

-0.003

-0.003

-0.004

-0.004

-20

-10

10

20

3D model
1D model

0.001

-30

wbeam ( m )

30

-30

-20

-10

x d

x d

a)

b)

10

20

30

Fig. 4.17. Beam deflection profiles for 3D and 1D models for (a) V = 200 km h and (b) V = 400 km h .

On the other hand, one could guess that if a train moves slowly in comparison
to the ground waves, a 1D model might be used instead of the 3D model described in
this chapter. To check, whether this is possible indeed, in this section, the dynamic response of the 3D model is compared to that of a related 1D model, in which the equivalent stiffness of the layer l s is replaced by a constant W . In other words, the sleepers
are assumed to be supported by classical (vertical) springs with the stiffness W . The
magnitude of this stiffness is taken as W = 0.75 108 N m , which is closed to the statically measured one and also corresponds to the magnitude of l s at low frequencies.
The beam deflections for both models are presented in Figures 4.16-4.17(a,b).
Four load velocities are considered, namely V = 50 km h , V = 100 km h ,
V = 200 km h , and V = 400 km h . The calculations were performed by using parameters (4.80)-(4.81).
Figures 4.16 and 4.17 show that if the load velocity is smaller than the Rayleigh
wave speed in the ground, the 1D model is in a good agreement with the 3D. With increasing velocity, this agreement becomes worse.
Thus, modelling the dynamic response of a railway track, it is possible to use a
1D model as long as the train speed is small in comparison to the wave speeds in the
ground.
Let us note, however, that this statement holds under the condition that the magnitude of the train axle loading is considered as constant. If the oscillatory part of this
loading were accounted for, then the range of applicability of the 1D model would be
reduced.

124

CHAPTER 5
Locomotive Detection of Derailment of a Wagon of a
Freight Train: Theory and Experiment
The previous chapters of this development dealt with three-dimensional models
of a railway track, promoting the idea that modelling the track response to high-speed
trains, it is ultimately necessary to employ three-dimensional models of the track subsoil. It should be realized, however, that the part of train traffic that falls on high-speed
trains is relatively small since vast majority of trains does not operate at high speeds.
Especially, the freight trains move slowly and, most likely, will always do so. These
slow trains, although operated for more than a century, still bring problems to engineers.
This chapter is devoted to one of these problems, namely to possible detection of derailment of a freight trains wagon by a set up mounted on the trains locomotive.
Wagons of a long freight train can derail when the train moves downhill and
slightly decelerates. Such a derailment can cause other wagons to derail and, as a consequence, the train to crash. This is the worse scenario. More often, the derailed wagon
returns back to the track or the trains driver manages to stop the train before the catastrophe happens. In both these cases, however, the derailed wagon leaves behind a considerably demolished piece of track.
To minimize the destructive effect that a derailed wagon can produce, the moment of derailment must become known to the trains driver as soon as possible. The
easiest way to do so is to mount to each bogie of the train an accelerometer that would
register a high acceleration that inevitably occurs due to the impact of a wheel against a
sleeper. This solution is, however, quite expensive, especially for companies that exploit
a large number of wagons and only several locomotives. For this kind of companies it
would be advantageous to develop a technique that would allow for detection of the derailment with the help of accelerometers that are mounted to the locomotives bogies
only. In principle, this seems to be possible, since the derailed wagon should perturb
anomalous vibrations of the rails, which could be then detected under the locomotive.
To explore whether such detection is indeed possible and to find criteria that suit this
detection the best, a theoretical as well as an experimental investigation is needed.
In this chapter, vibrations of the rails that are perturbed by a derailed wheelset
and then measured under the locomotive are studied both theoretically and experimentally. In the theoretical analysis, a Timoshenko beam on discrete, periodically spaced
supports is employed as a model for the railway track. The supports are identical and are
composed of a mass that is connected to the beam via a visco elastic element and is
supported by a spring. The mass, visco-elastic element and the spring are employed to
model the sleepers, pads and ground reaction, respectively. A slow train motion justifies
modelling the ground reaction by a spring with a constant stiffness.
To model the axle loading of the train a constant load is used that uniformly
moves along the track. If there is no derailment, e.g. the normal motion of the train
takes place, the load is applied to the beam (rails). In the case of derailment, which is
referred to as an emergency motion, the load is applied to the supports (sleepers).

125

The chapter is structured as follows. In section 5.1, the normal motion of the
train is studied assuming that the load is applied to the beam. The steady-state response
of the beam is analysed at a point that moves at a fixed distance from the load. This
point is associated with a peace of the rail under the locomotive and is hereinafter referred to as an observation point. The main attention is paid to calculation of the spectrum of vibrations in this point.
In section 5.2, the emergency motion is studied assuming that the load
jumps from one sleeper to another one with a constant speed. The goal of the study is
to find the spectrum of the rail vibrations at the observation point and then compare the
spectra in the normal and emergency regimes.
In section 5.3, effect of the loads (wagon wheels) that are located between the
observation point and the loading point on the spectra of the rail vibrations in the observation point is studied. A frequency band, at which the derailment of a wagon can be
detected, is determined.
In section 5.4, theoretical results obtained in sections 5.1-5.3 are compared to
experimental data, collected during measurements carried out by Administration of
Gorky Railway (Russia) next to the station Ingenernaya in August and December of
the year 2000 [91]-[92]. A method of spectral analysis of the data that allows to increase
sensitivity of the derailment detection is proposed. Perspectives of using the locomotive
detection for the wagon derailment are discussed.
Results presented in this chapter are original and served as a basis for paper
[166] that has been submitted recently.

5.1. Steady-state response of a railway track to axle loading.


Normal regime.
In this section we consider a so-called normal type of the axle loading of a
uniformly moving train. This type of loading is depicted in Figure 5.1. It occurs when
there is no derailment and the desired train motion is achieved. In this case, the load is
applied to the rail, normal to its undisturbed surface (vertically), has amplitude P and
moves along the rail with a constant velocity V so that the horizontal co-ordinate of the
loading point is defined as x = Vt .

w beam ( x , t )

V
x0

pad

0
n

Observation
point

(t )

sleeper

rail

ground
d

Fig. 5.1. Model for the normal motion of the load.

126

The governing equations that define the steady-state dynamic response of the
track can be written as a system of equations that consists of

equations of motion for the beam:

2 wbeam
2 wbeam
wbeam

GS
+

= P0 ( x Vt )
1
2
t 2
t
x
x

2
wbeam
beam S 2 + 2
Ebeam I 2 + GS
=0
t
t
x
x

beam S

(5.1)

boundary conditions at the contact points of the beam and the sleepers ( x = nd ) :

= wbeam
=0

x = nd
x x = nd

[ ]x =nd =

wbeam

= K + ( wbeam ( nd , t ) wn0 ( t ) )
GS

t
x x = nd

(5.2)

equation of motion for the sleeper number n


M

d 2 wn0
d

t + K + ( wn0 ( t ) wbeam ( nd , t ) ) + wn0 ( t ) = 0


2 ( )
dt
dt

(5.3)

In equations (5.1)-(5.3), wbeam ( x, t ) is the displacement of the neutral axis of the beam,

( x, t ) is the angular rotation of the cross section, wn0 ( t ) is the vertical deflection of
the sleeper number n , beam is the beam mass density, S is the beam cross-sectional
area, I is the moment of inertia of the beam cross-section, Ebeam and G are the
Youngs modulus and shear modulus of the beam material, respectively, is the shear
correction factor (Timoshenko coefficient), K and are the stiffness and damping coefficient of a pad, respectively, is the stiffness of the ground, M is the mass of a
sleeper, (...) is the Dirac delta function, and the square brackets denote the following
difference: f ( x ) x = a = f ( x = a + 0 ) f ( x = a 0 ) .
Let us note that equations (5.1) that describe the vertical motion of the beam,
contain terms 1 w t and 2 t representing a distributed damping in the beam.
These terms are introduced to model the internal friction in the beam material. Normally, this friction is modelled with the help of operators of the form 3 t x 2 . In this
paper the simpler operator t is chosen, for in slightly damped systems both operators ( t and 3 t x 2 ) have almost the same effect on the solution.
The central role in the solution procedure that is presented below plays a socalled periodicity condition [10,158]. This condition is based on the fact that the system
of governing equations (5.1)-(5.3) remains unchanged if the following transformation of
the co-ordinate system and time is performed: { x, t} { x + nd , t + nd V } . In other

127

words, the system of governing equations is invariant with respect to the following replacement:
wbeam ( x, t ) = wbeam ( x + nd , t + nd V )

(5.4)

( x, t ) = ( x + nd , t + nd V ) .

Correctness of this statement can be easily checked by substitution. The periodicity


condition implies that the displacement pattern in the beam repeats itself in time with
the period T = d V experiencing, however, the spatial translation equal to d . This condition, although formulated in the frequency-domain, has been already used in Chapter
4.
To solve the governing equations (5.1)-(5.3), the following integral Fourier
transforms over time are applied:

{w

beam

, ( x, ) =

{w

beam

} ( x, t ) exp ( i t ) dt

W ( ) =
0
n

(5.5)

w ( t ) exp ( i t ) dt
0
n

Application of these transforms yields:

2 wbeam
P i V x
+ ( beam S 2 + i 1 ) wbeam GS
= e
2
x
x V
wbeam
2
2
Ebeam I 2 + ( beam I + i 2 ) GS + GS
=0
x
x

[ ]x =nd = = wbeam x =nd = 0


x x = nd

GS

wbeam
= ( K i ) ( wbeam ( nd , ) Wn0 ( ) )
GS
x x = nd

M 2Wn0 ( ) + ( K i ) (Wn0 ( ) wbeam ( nd , ) ) + Wn0 ( ) = 0

(5.6)

(5.7)

(5.8)

To solve the system of equations (5.6)-(5.8), we will make use of the periodicity
condition, which, after application of the Fourier transforms (5.5) to Eq.(5.4), takes the
form

wbeam ( x + nd , ) = wbeam ( x, ) exp ind


V

( x + nd , ) = ( x, ) exp ind
V

(5.9)

Equations (5.9) relate the Fourier-displacements of the beam in the interval x [0, d ] to
those in any other interval x [nd , ( n + 1) d ] . As already described in Chapter 4, the pe-

128

riodicity condition allows to obtain the solution to system (5.6)-(5.8) performing the
following three steps.
1. First, the general solution to the system of equations (5.6) is to be written in the interval x [ 0, d ] . This solution will contain four unknown constants Ai , i = 1..4 .
2. Second, using the periodicity condition, the solution in the interval x [d , 2d ] can
be obtained from that in the interval x [ 0, d ] by multiplying it by exp ( id V ) .
Obviously, the result of this multiplication will contain the same four constants
Ai , i = 1..4 .
3. Third, four unknown constants Ai , i = 1..4 and one extra constant that is related to
the Fourier-displacement of the sleeper number n = 1 are to be found by employing
four boundary conditions (5.7) at the point x = d and the equation of motion for the
sleepers (5.8).
Let us accomplish this way of solution. The general solution to equations (5.6)
in the interval x [ 0, d ] can be written as:
beam

( x, ) = A1e

i1 x

( x, ) = 1 A1e

+ A2 e

i1 x

i1 x

A2 e

i1 x

+ A3e

i2 x

+ A4 e

)+ (A e
2

i2 x

i2 x

F1e

+ A4 e

i2 x

) F e
2

(5.10)

where Ai , i = 1..4 are unknown constants, and


1,2 =

beam SI ( Ebeam + G ) 2 + i ( 1 Ebeam I + 2 GS ) D


2 GEbeam I

2
D = beam
S 2 I 2 ( Ebeam G ) 4 + 4 2G 2 S 3 Ebeam I beam 2
2

2i beam SI ( Ebeam G )( 2 GS 1 Ebeam I ) 3 ( 2 GS 1 Ebeam I ) 2 + 4i 2G 2 S 2 Ebeam I 1 ,


2

i =

i GS i
,
Ebeam I beam I 2 + GS i 2

2
i

PV V 2 beam Ebeam I 2 GSV 2 + i V 2


2

,
F1 =

F
iPV 2 GS
F2 =
.

F
F = IS G beamV 2 Ebeam beamV 2 3

( (

)(

))

iV 2 1 I Ebeam beamV 2 + 2 S G beamV 2 2 V 4 GS 2 beam + 1 2 i 1 SV 4

(5.11)
Multiplying (5.10) by exp ( id V ) , the following expressions for wbeam ( x, )
and ( x, ) are obtained that are valid in the interval x [ d , 2d ] :

129

(
( x, ) = ( ( A e
wbeam ( x, ) = A1e
1

i1 ( x d )

i1 ( x d )

+ A2 e

i1 ( x d )

A2 e

i1 ( x d )

+ A3e

i2 ( x d )

)+ (A e
2

+ A4 e

i2 ( x d )

i2 ( x d )

A4 e

i d

e V F1e V

i2 ( x d )

)) e

i d
V

F2 e V

(5.12)
x

From expressions (5.12) it follows that the Fourier-displacement of the beam in the
point x = d is given as


wbeam ( d , ) = A5 exp i d , where A5 = A1 + A2 + A3 + A4 F1 .
V

(5.13)

Obviously, the Fourier-displacement of the sleeper n = 1 has to be of the same form,


e.g.

W10 ( ) = A6 exp i d
V

(5.14)

Substitution of (5.10), (5.12)-(5.14) in the boundary conditions gives the following system of linear algebraic equations:

1
1
( q a1 ) A1 + q A2 + ( q b1 ) A3 + q A4 = 0
a1
b1

1 ( q a1 ) A1 1 q 1 A2 + i 2 ( q b1 ) A3 i 2 q 1 A4 = 0
a1
b1

1
1
1 1 ( q a1 ) A1 + 1 1 q A2 + 2 2 ( q b1 ) A3 + 2 2 q A4 = 0
a1
b1

i ( q a ) A i q 1 A + i ( q b ) A i q 1 A + T qA T qA = 0
1
1
1
2
2
1
3
2
4
0
5
0
6
1
a1
b1

A1 + A2 + A3 + A4 A5 = F1
T A BA = 0
6
0 5

with


q = exp i d ,
V

a1 = exp ( 1d ) ,

b1 = exp ( i2 d ) ,

(5.15)

T0 = k i ,

B = M 2 + k + i .
Solving system (5.15) with respect to Aj ( j = 1..6 ) , the following expressions
are found (only first four coefficients that are necessary to describe the motion of the
beam are presented):

130

Ai =

i
, i = 1..4,

1 = i 2 F1 q 3 q 2 ( 2 cos 2 d + e i d ) + q (1 + 2e i d cos 2 d ) e i d

{
= i F {q q ( 2 cos d + e ) + q (1 + 2e
= i F {q q ( 2 cos d + e
) + q (1 + 2e
= i F {q q ( 2 cos d + e ) + q (1 + 2e
1

2
3
4

i1d

2 1

1 1

i1d

i2 d

1 1

cos 2 d ei1d

i2 d

i2 d

i2 d

)
}
cos d ) e
}

cos 1d e i2 d
1

i2 d

(5.16)

2 B GS
d
d
= 4q
( 1 2 2 1 ) cos 1d cos
cos 2 d cos

V
V

T0 ( B T0 )
d
d

1 sin 2 d cos 1d cos


+ 2 sin 1d cos 2 d cos

V
V

Thus, the problem has been solved in the frequency domain. Inverting solution
(5.10), the deflection of the beam can be found in the interval x [0, d ] :
wbeam ( x, t ) =

1
2

1
( x, t ) =
2

(Ae
1

i1 x

+ A2 e i1x + A3ei2 x + A4 e i2 x F1ei x V ) exp ( i t ) d

( ( Ae
1

i1 x

A2 e i1x ) + 2 ( A3ei2 x + A4 e i2 x ) F2 ei x V exp ( i t ) d

(5.17)

The deflection in the other spans can be easily obtained by employing the periodicity
condition (5.9).

The amplitude spectrum of vibrations in the observation point.


If the locomotive registration of derailment is utilized, then the rail vibrations
perturbed by the derailed bogie are measured with the help of an accelerometer mounted
to a bogie of the locomotive. In mathematical terms this means that the vibrations are
observed in a point that moves with the same velocity as the load but at a certain distance from it.
One of the most sensitive techniques by means of which an information on vibrations can be gathered is the spectral analysis. Performing in-situ measurements, this
analysis is carried out on the basis of registration of vibrations during a finite time interval. To be consistent with the measurements technique, let us calculate the spectrum
w ( x, ) of the beam vertical displacement wbeam ( x, t ) over a finite time interval
0 < t < t * . This spectrum in the point x = x0 + Vt reads
t*

w ( ) = wbeam ( x0 + Vt , t ) exp ( it ) dt
0

131

(5.18)

Assume that the time interval t * , over which the measurements are carried out, is divisible by the sleeper-passing period T = d V , e.g. t * = ( N + 1) T = ( N + 1) d / V , where
N is an integer. With this assumption, expression (5.18) can be simplified as
N

( n +1)T

n =0

nT

w ( ) =

wbeam ( x0 + Vt , t ) exp ( it ) dt

(5.19)

Using the periodicity condition (5.4), it can be shown that w ( x0 + Vt , t ) is a periodic


function of time with the period T = d V :
w ( x0 + V ( t + nT ) , ( t + nT ) ) = w ( x0 + Vt + nd , t + nd V ) = w ( x0 + Vt , t ) .

(5.20)

Thus, introducing in expression (5.19) a new variable of integration = t nT , this expression can be reduced to
N

( n+1)T

n=0

nT

w ( ) =

( n+1)T

n=0

nT

w( x +Vt, t ) exp ( it ) dt = w( x +V ( + nT ) , ( + nT ) ) exp ( i( + nT ) ) d


0

n=0

= exp ( inT ) w( x0 +V , ) exp ( i ) d =

1 exp ( iNT ) T

w( x +V , ) exp ( i ) d

1 exp ( iT )

(5.21)
Expression (5.21) gives the amplitude spectrum of the vertical vibrations of the
beam at the point x = Vt + d 0 + md . These vibrations are generated in the steady-state
regime by the moving load that is applied to the structure at the point x = Vt .
The integral in expression (5.21) can be evaluated analytically by making use of
solution (5.17) and the periodicity condition (5.9). If the distance x0 is represented as
d 0 + md , where 0 d 0 d and m is an integer, then expression (5.21) takes the form:
w ( ) =

d
i( + 1V ) dV0
i
1 exp ( iNT )
A1
V
e
1
e

i ( + 1V )
1 exp ( iT )

d
i ( + 1V ) 0
i( + 1V )Vd
V
e
e

d
i( 1V ) dV0
i
A2
1+ e V
e
i ( 1V )

d
d
d
i( + 2V ) dV0
i ( + 2V ) 0
i i ( + 2V )
A3
V
V
e
1+ e V e
e
i ( + 2V )

d
d
d
i( 2V ) dV0
i ( 2V ) 0
i i ( 2V )
A4
V
V
1+ e V e
e
e
i ( 2V )

d
d
d
d
i 0
i i
F1 i V0
1 + e V e V e V
e
i

d
i ( 1V ) 0
i( 1V )Vd
V
e
e

i d ( m +1)
d
e V

(5.22)

132

where Ai , i = 1..4 are defined by expressions (5.16).

In the next section, we will calculate the spectrum w ( ) in the emergency

case and compare it to that given by Eq.(5.22).

5.2. Steady-state response of a railway track to axle loading


in the emergency case. Comparison with normal regime.
In order to describe the emergency regime of the train motion we consider the
load jump over sleepers as depicted in Figure 5.2.

w beam ( x , t )

x0
V

0
n

(t )

Observation
point

x
d

Fig. 5.2. Model for emergency regime of the load motion.

Assuming that the train moves uniformly despite of derailment of a wagon, we assume
that the load jumps over sleepers with a constant speed. With this assumption, the
governing equations for the vertical vibrations of the beam and the sleepers take the following form:
2 wbeam
2 wbeam
wbeam

GS
+

=0
1
2
t 2
t
x
x

2
wbeam
beam S 2 + 2
Ebeam I 2 + GS
=0
x
t
t
x

beam S

(5.23)

[ ]x =nd = [ x ]x=nd = wbeam x=nd = 0

= K + ( wbeam ( nd , t ) wn0 ( t ) )
GS w,beam
x
x = nd
t
M

(5.24)

d 2 wn0 ( t )
d
Vt

+ K + ( wn0 ( t ) wbeam ( nd , t ) ) + wn0 ( t ) = P n


2
dx
dt
d

Applying to Eqs.(5.23) and (5.24) the integral Fourier transforms defined by Eq.(5.5),
one obtains:

133

2 wbeam

+ ( beam S 2 + i 1 ) wbeam GS
=0
2
x
x
2
wbeam
Ebeam I 2 + ( beam I 2 + i 2 ) GS + GS = 0
x
x

[ ]x =nd = = wbeam x =nd = 0


x x = nd

GS

wbeam
= ( K i ) ( wbeam ( nd , ) Wn0 ( ) )
GS

x = nd
Pd i V nd
M 2Wn0 ( ) + ( K i ) (Wn0 ( ) wbeam ( nd , ) ) + Wn0 ( ) =
e
V

(5.25)

(5.26)

(5.27)

The general solution to the system of equations (5.25) has the form
wbeam ( x, ) = A1ei1x + A2 e i1 x + A3ei2 x + A4 e i2 x

( x, ) = 1 A1ei1 x A2 e i1 x + 2 A3ei2 x A4 e i2 x

(5.28)

where expressions for i are the same as in (5.11), and constants Aj should be found by
using the boundary conditions (5.26) and (5.27). Since Eqs.(5.25)-(5.27) satisfy the periodicity condition (5.9), the procedure of finding the system of algebraic equations with
respect to the constants Aj is exactly the same as described in the previous section. This
procedure results in

1
1
( q a1 ) A1 + q A2 + ( q b1 ) A3 + q A4 = 0
a1
b1

1 ( q a1 ) A1 1 q 1 A2 + i 2 ( q b1 ) A3 i 2 q 1 A4 = 0

a1
b1

1
1

1 1 ( q a1 ) A1 + 1 1 q A2 + 2 2 ( q b1 ) A3 + 2 2 q A4 = 0

a1
b1

1
1
i1 ( q a1 ) A1 i1 q a A2 + i2 ( q b1 ) A3 i2 q b A4 + T0 qA5 T0 qA6 = 0
1
1

A1 + A2 + A3 + A4 A5 = 0

T A BA = Pd
6
0 5
V

Solving system (5.29) with respect to Aj , we obtain:

134

(5.29)

Ai =

i
, i = 1..4,

}
}

Pd 3
q q 2 2 cos 2 d + e i1d + q 1 + 2e i1d cos 2 d e i1d
V
Pd 3
2 = 2
q q 2 2 cos 2 d + ei1d + q 1 + 2ei1d cos 2 d ei1d
V
Pd 3
3 = 1
q q 2 2 cos 1d + e i2 d + q 1 + 2e i2 d cos 1d e i2 d
V
(5.30)
Pd 3
i2 d
i2 d
i2 d
2
4 = 1
+ q 1 + 2e
cos 1d e
q q 2 cos 1d + e
V
2 B GS
d
d
= 4q ( B T0 )
( 1 2 2 1 ) cos 1d cos cos 2 d cos
V
V

T0 ( B T0 )
d
d

1 sin 2 d cos 1d cos


+ 2 sin 1d cos 2 d cos

V
V

1 = 2

{
{

) (

) (

) (

) (

Thus, we have found the beam response in the frequency domain. Applying now
the inverse Fourier transform to Eq.(5.28), we can obtain the following expression for
the steady-state response of the beam in the space-time domain:
beam

1
( x, t ) =
2

1
( x, t ) =
2

(Ae
1

i1 x

+ A2 e i1x + A3ei2 x + A4 e i2 x ) exp ( i t ) d

( ( Ae
1

i1 x

A2 e

i1 x

)+ (A e
2

i2 x

+ A4 e

i2 x

) ) exp ( i t ) d

(5.31)

with Aj defined by Eq.(5.30). Expressions (5.31) describe the beam response in the interval x [0, d ] and can be extended to the other intervals with the help of the periodicity condition (5.4).
The amplitude spectrum of vibrations at the observation point can be obtained
employing expression (5.21), since it was obtained without specifying whether the load
moves over the beam or jumps over the sleepers (only uniformity of motion was required). Thus, substituting Eq.(5.28) into Eq. (5.21), after simple evaluations, we obtain
the following expression for the spectrum of vibrations in the observation point that is
valid in the emergency case:
d
d
d
i( + 1V ) dV0
i ( + 1V ) 0
i i ( + 1V )
1 exp ( iNT )
A1
V
V
V
1+ e
e
w ( ) =
e
i ( + 1V ) e
1 exp ( iT )

d
d
d
i( 1V ) dV0
i i ( 1V )
i ( 1V ) 0
A2
V
V
V
+
1 + e
e
e
e
i ( 1V )

d
i( + 2V ) dV0
i
A3
V
e
1
e

i ( + 2V )

i( + 2V )Vd
e

d
i( 2V ) dV0
i
A4
V
1
e

+
e

i ( 2V )

i( 2V )Vd
e

d
i ( + 2V ) 0
V
e
+

i ( 2V ) 0 i d ( m +1)
V
e
e
d
V

(5.32)
135

Comparison of the rail vibrations in the normal and emergency


cases.
Let us study the spectrum of the rail vibrations in the observation points. Using
expressions (5.22) and (5.32), this spectrum can be easily calculated numerically for
both the normal and emergency regimes of the load motion. To perform calculations, we used the following physical parameters of the system:
S = 2 7.687 103 m 2 , beam S = 2 60.34 kg/m, E = 2 1011 N/m 2 , = 0.28

G = E / ( 2 (1 + ) ) , I = 2 3.055 10 5 m 4 , =0.82, M = 250 kg, K =108 N/m

(5.33)

=108 N/m, =103 N*s/m, P = 50000 N, d = 0.5 m, V = 20 km/h


These parameters are valid for the railway track that was used for performing an experiment that will be described in section 5.4.
First, to show the effect of the time window t * , we choose the time interval t * ,
over which the spectrum is calculated, as t * = T . With the sleeper distance d = 0.5 m
and the train velocity V = 20 km/h , this equals 0.09 s . Then, to be consistent with the
experimental data we employ more realistic time window t * = 11T = 0.99s , which is
almost the same that was used in the experiments (1s ) (See 5.4).
The spectrum of vibrations for t * = T in the normal case is compared to that in
the emergency case in Figure 5.3. The comparison is carried out for four magnitudes of
the distance x0 between the load and the observation point, namely for x0 = 10d = 5 m ,
x0 = 50d = 25 m , x0 = 100d = 50 m , x0 = 200d = 100 m .
Figure 5.3 shows that the spectrum in the normal case has a quasi-repetitive
character as a function of frequency. The minima in the spectrum are a consequence of
the averaging over a finite time window t * . In the case of the periodic function
wbeam ( x0 + Vt , t ) this averaging always gives a number of zeroes of the spectrum. If t *
is not very large with respect to period T , these minima are always to be expected. The
minima in Figure 5.3 do not correspond exactly to the sleeper passing frequency
f SP = V d = 11.1 Hz and its super-harmonics nf SP ( n is integer) although, from the figure, it could seem to be so. The closeness of the minima to the above-mentioned frequencies is due to small value of the load velocity that was used in calculations. The
point is that in the case of small velocities, function wbeam ( x0 + Vt , t ) in the interval

t [ 0, T ] has almost a constant average around which it quickly oscillates. Therefore,


2t
integration of the function wbeam ( x0 + Vt , t ) exp i
over period T gives a close to
T
zero result.
It should be noticed that with increase of the distance x0 the amplitude of the
normal spectrum first decreases but then, when this distance is large enough (more
than 25 m ), almost does not change. It implies that only the far wave field determines
the spectrum in the observation point if the distance x0 becomes large.

136

1E-005

Modulus of amplitude spectrum w (m s)

Modulus of amplitude spectrum w (m s)

normal case
emergency case

0.0001

x0 = 10d

1E-006
1E-007
1E-008
1E-009
1E-010
1E-011
1E-012
1E-013
0

20

40

60

0.0001

normal case
emergency case

1E-005
1E-006
1E-007
1E-008
1E-009
1E-010
1E-011
1E-012
1E-013
1E-014
0

80 100 120 140 160 180 200

20

40

Frequency f (Hz)

60

Modulus of amplitude spectrum w (m s)

Modulus of amplitude spectrum w (m s)

b)

normal case
emergency case

1E-005

x0 = 100d

1E-006
1E-007
1E-008
1E-009
1E-010
1E-011
1E-012
1E-013
1E-014
0

20

40

60

80 100 120 140 160 180 200

Frequency f (Hz)

a)
0.0001

x0 = 50d

0.0001

normal case
emergency case

1E-005

x0 = 200d

1E-006
1E-007
1E-008
1E-009
1E-010
1E-011
1E-012
1E-013
1E-014

80 100 120 140 160 180 200

Frequency f (Hz)

c)

20

40

60

80 100 120 140 160 180 200

Frequency f (Hz)

d)

Fig. 5.3. Amplitude spectrum for the normal (solid line) and the emergency (dashed line) cases at the
point (a) x = 10d + Vt , (b) x = 50d + Vt , (c) x = 100d + Vt , (d) x = 200d + Vt .

On the contrary to the normal case, the spectrum in the emergency case is visibly amplified in the frequency band 100 150 Hz . This amplification takes place around
the natural frequency of the sleepers that is given as
fs =

1
2

K+
140 Hz .
M

(5.34)

The reason for the amplification is that the natural vibrations of the sleepers are
perturbed more effectively in the case of the direct loading of the sleepers, e.g. in the
emergency case.
Thus, the spectrum in the emergency case differs from that in the normal case in
the frequency band that surrounds the natural frequency of the sleepers. Unfortunately,
this difference decreases dramatically with the increase of the distance between the load

137

0.0001

Modulus of amplitude spectrum w (m s)

Modulus of amplitude spectrum w (m s )

and the observation point. Thus, this difference can be used as a criterion for detecting
the emergency situation only if the train is not long. To gain an idea of how long the
train should be to make this criterion working, an extra investigation is needed that
takes into account vibrations perturbed in the rails by the other (moving in the normal
regime) bogies of the train. This investigation is carried out in the following section.

x0 = 10d

1E-005
1E-006
1E-007

normal case
emergency case

1E-008
1E-009
1E-010
1E-011
1E-012
1E-013
0

20

40

60

0.0001

normal case
emergency case

1E-005
1E-006
1E-007
1E-008
1E-009
1E-010
1E-011
1E-012
1E-013
0

80 100 120 140 160 180 200

20

40

Frequency f (Hz)

60

1E-005

b)
Modulus of amplitude spectrum w (m s)

Modulus of amplitude spectrum w (m s)

normal case
emergency case

x0 = 100d

1E-006
1E-007
1E-008
1E-009
1E-010
1E-011
1E-012
1E-013
0

20

40

60

80 100 120 140 160 180 200

Frequency f (Hz)

a)
0.0001

x0 = 50d

80 100 120 140 160 180 200

Frequency f (Hz)

c)

0.0001

normal case
emergency case

1E-005

x0 = 200d

1E-006
1E-007
1E-008
1E-009
1E-010
1E-011
1E-012
1E-013
0

20

40

60

80 100 120 140 160 180 200

Frequency f (Hz)

d)

Fig. 5.4. Amplitude spectrum for the normal (solid line) and the emergency (dashed line) cases at the
point (a) x = 10d + Vt , (b) x = 50d + Vt , (c) x = 100d + Vt , (d) x = 200d + Vt .

Figure 5.4 represents results obtained for more realistic time window
t = 11T = 0.99s , which was employed in the experimental data processing. As follows
from the figure, the main change of the spectra with respect to the case of t * = T is related only to the positions and number of minima and maxima.
*

138

5.3. Effect of multi-axle loading on the track response.


If one of the train bogies should derail, vibrations that are perturbed by this bogie in the track would interfere with vibrations that are induced by the other bogies. The
latter vibrations, especially those that are induced by the bogies of the first wagon,
could be powerful enough to make the signal coming from the derailed bogie not distinguishable. In this section, it is checked, whether this is indeed the case. To simplify
analysis of this question, the time window t * is considered equal to T = 0.09s .
First, it is customary to determine the number of the closest to the locomotive
bogies (moving in the normal regime) that provide a significant contribution to the
measured spectrum. To this end, the spectrum is studied of vibrations that are induced
by the bogies of the first, closest to the locomotive, wagon of the train. The situation at
hand is presented schematically in Figure 5.5.

w beam ( x , t )
w

0
n

d2

d1

d1

Observation
point

D0

(t )

x
d

Fig. 5.5. Normal motion of the first wagon.

To obtain the spectrum generated by four loads depicted in Figure 5.5 (each load
represents one axis of the train that moves in the normal regime), it is necessary to calculate the integral (5.22) four times, using four different magnitudes of x0 = d 0 + md .
The result of this calculation is presented in Figure 5.6. For calculations, the set of parameters (5.33) was used. The distances d1 and d 2 (see Figure 5.5) were considered
equal to 1.85 m and 8.65 m , respectively. The distance D0 between the first wheelset
and the observation point was given four different values, e.g. 4 m , 4.05 m , 4.25 m
and 4.375 m , yielding the graphs (a), (b), (c) and (d), respectively.
Figure 5.6 demonstrates that the contribution of the second bogie to the spectrum is negligible in comparison with the contribution of the first bogie (compare the
spectrum for 2 and 4 loads). This result is independent of the distance D0 , despite a
high sensitivity of the spectrum to small variations of this distance. Thus, should no bogie of the train be derailed, vibrations of the rail under the locomotive could be well
modelled as the rail response to the first bogie.

139

1 load
2 loads
4 loads

Modulus of amplitude spectrum w (m s )

Modulus of amplitude spectrum w (m s )

1E-006

(a)

1E-007

1E-008

1E-009

1E-010

1E-011
0

20

40

60

1E-006

1 load
2 loads
4 loads

1E-007

1E-008

1E-009

1E-010

1E-011
0

80 100 120 140 160 180 200

Frequency f (Hz)

20

40

60

Modulus of amplitude spectrum w (m s )

Modulus of amplitude spectrum w (m s )

b)
1 load
2 loads
4 loads

(c)

1E-007

1E-008

1E-009

1E-010

1E-011
0

20

40

60

80 100 120 140 160 180 200

Frequency f (Hz)

a)
1E-006

(b)

80 100 120 140 160 180 200

Frequency f (Hz)

c)

1E-006

1 load
2 loads
4 loads

(d)

1E-007

1E-008

1E-009

1E-010

1E-011
0

20

40

60

80 100 120 140 160 180 200

Frequency f (Hz)

d)

Fig. 5.6. Amplitude spectrum of vibrations generated by the first wheel, the first bogie and the first wagon
of the train for (a) D0 = 4 m , (b) D0 = 4.05 m , (c) D0 = 4.25 m , (d) D0 = 4.375 m .

Consider a situation when one wheelset of the train is derailed. This situation is
depicted in Figure 5.7, which schematically shows the first wagon (moving in the normal regime) and one axle that is derailed.
The spectrum of the response to both the first wagon and the emergency load
is presented in Figure 5.8. Four distances between the observation point and the emergency load are considered, namely 25 m , 50 m (left figure), 75 m and 100 m (right
figure).
The
other
parameters
are
defined
by
(5.33)
and
d1 = 1.85 m, d 2 = 8.65 m, D0 = 4 m . The spectrum of vibrations perturbed by both the
first wagon and the emergency load is compared to the spectrum produced by the first
wagon only (see the bold line).

140

V
d1

w ( x, t )

d2

d1 P

0
n

(t )

D0

Observation
point

x0

1E-005

normal motion

1E-006

x0 = 50 d
x0 = 100 d

1E-007
1E-008
1E-009
1E-010
1E-011
0

40

80

120

160

200

Frequency f (Hz)

Modulus of amplitude spectrum | w | (m s)

Modulus of amplitude spectrum | w | (m s)

Fig. 5.7. Emergency load and the first wagon.

1E-005

normal motion

1E-006

x0 = 150 d
x0 = 200 d

1E-007
1E-008
1E-009
1E-010
1E-011
0

40

80

120

160

Frequency f (Hz)

Fig. 5.8. Effect of the distance between the emergency load and the observation point on the amplitude
spectrum.

Figure 5.8 shows that the effect of the emergency load on the spectrum decreases dramatically as the distance x0 between this load and the observation point increases. Already for x0 = 150 m (less than 10 wagons) this effect becomes almost invisible. A slight difference remains at the frequency band 100 140 Hz . As it will be
shown in the next section, in experiment, this frequency band indeed shows the highest
sensitivity to the derailment.

5.4. Comparison of experimental results to theoretical predictions.


In this section, the results of the theory developed for the rail vibrations perturbed by a derailed bogie in sections 5.1-5.3 are compared to the experimental data.

141

200

First, results of measurements carried out by Gorky Railway in August and December
of the year 2000 are presented. Then, on the basis of the carried out comparative analysis of the theory and the experiment, a new spectral method of data processing that allows for increasing sensitivity of the derailment detection is proposed. Discussion of
perspectives of the locomotive detection method finalizes this section.

Experimental set-up.
In the experiment, the train was formed in the way that is schematically depicted
in Figure 5.9. The locomotive (L) moved the train consisting of ten wagons with the
constant velocity V of 20 km h .

10

d1

d2

d1

D0 = 4m

x0 = 152.5 m
Fig. 5.9. Schematics of the experiment with L the locomotive, R the position of the accelerometer, E the
emergency wheelset, 1, 2,...,10 the wagons.

The accelerometer-receiver (R) was mounted on the axlebox of the locomotive


as shown in Figure 5.10. Mounted in this way, the receiver was registering only the
translational motions of the wheelbase.

receiver

axlebox (immovable part).

receiver
Fig. 5.10. Position of the receiver.

142

12 BOLT M24 170


SHIM 12 GOST 13463-77-12

CYLINDRICAL HEADER

Fig. 5.11. Scheme of the emergency wheel.

260 mm
240 mm

Sleepers

2m

Obstacle
Header

rails

Fig. 5.12. Scheme of the derailment device.

143

The construction of the last wheelbase (of the last tenth wagon) slightly differed
from that of the other ones. Figure 5.11 shows that the one of the wheels had a special
cylindrical header.
During the experiments the emergency wheelset (E) was extruded from the
track by the special obstacle mounted to sleepers in a certain place, as it is shown in
Figure 5.12. To make the derailment process gradual, the extruding obstacle had length
of 2 meters. At the moment of the derailment a radio signal was generated to give the
reference for the data processing. The distance between the receiver and the derailed
wheelset was 152.5 m .

a)

b)

Fig. 5.13. Experiment: a) during derailment, b) final scene.

To minimise the damage to the railway, after the derailment the train moved
with the constant velocity only 10 seconds (see Fig. 5.13). To repeat the experiment, an
additional train with crane was used after each testing to put the derailed wheelset on
the rails. Four measurements, two in August and two in December of the year 2000,
were carried out. Totally, 47 people took part in the experiments.

Measurement set-up.
The block-scheme of the measurement set-up is shown in Fig. 5.14.

Notebook

Charge amplifier -2
Signal

16-bit DAQCard
Transducer
-57
LPF

Memory

HPF

Fig. 5.14. Measurement chain.

144

Digital spectral analysis

During the experiment the signal is measured by the accelerometer-receiver and


then sent to the charge amplifier. The receiver is a piezoelectric transducer possessing
the following parameters:
transducer -57 (made in Russia, JSC GLOBALTEST, Sarov):

Axial sensitivity
Relative transversal sensitivity
Maximum impact
Working temperature range
Frequency band
Amplitude range
Eigen frequency
Capacity
Isolation resistance
Material of hull
Mass without cable
European analog
mark.

80 pC g 1;
< 5% ;
4000 g ;

60 +150 C0 ;
0.5 8000 Hz ;
2000 g ;
> 20 kHz ;
700 pF ;
> 100 G ;
Stainless steel;
0.04 kg .
transducer 4366 of company Bruel & Kjaer, Den-

Fig. 5.15. Universal transducer -57.

The general view of the transducer is shown in Fig. 5.15.


Charge amplifier 3-2 (made in Russia, JSC GLOBALTEST, Sarov):

The amplifier includes a low-pass (LP) and a high-pass (HP) Butterworth filter,
where the cut-off frequencies may be chosen as:
1) for LPF: 0.3; 1; 3; 10; 30 kHz;
2) for HPF: 1; 3; 10; 30; 100 Hz.
The work frequency band of the amplifier is 0.4 20000 Hz . The amplitudes
drop with a slope of 40 dB decade away from the passband. The amplification factor
may be discretely varied from 0.1 to 1000. Working temperature range is 10 +50 C0 .
The European analog of 3-2 is charge amplifier 2635 of company Bruel & Kjaer,
Denmark.
1

g = 9.8 m s 2 is the gravity acceleration.

145

After amplification, the signal is digitised by means of DAQCard-AI-16XE-50,


which is incorporated in the notebook (see Fig. 5.16). DAQCard-AI-16XE-50 analog
input circuitry is calibrated for bipolar polarity, i.e. the input voltage range is between
Vref and +Vref . Having been digitised the data are saved onto the hard disk.

Fig. 5.16. A typical configuration for DAQCard.

Data processing.
All measurements were carried out in the frequency band from 5 to 1000 Hz at
a sampling frequency of 2000 Hz . The sensitivity of the transducer was set at the level
0.002 g . The voltage range of DAQCard-AI-16XE-50 analog input circuitry was chosen as Vref = 0.8 V . To process the data the time window of 1 s was used.
The characteristic shapes of the time dependency prior to the derailment and after the derailment are shown in Figure 5.17.
The digital spectral analysis method, or method of digital spectral estimation,
used during the data processing is based on a finite sampling { xk ,k = 0 ,...,N S } taken
from time series (see, for instance, [51], [107] or [109])

xk = x ( k S ) =

x ( t ) h ( t k S ) dt =

X () e
S

i 2 k S

d , k ( , + ) ,2

(5.35)

which originates from analog signal x ( t ) , t ( , + ) digitised with the sampling frequency f S = 1 S (index S denotes the sampling):
2

It is assumed that the impulse response h ( t ) = ( t ) .

146

0,8

0,8

0,6

0,6

0,4

0,4

0,2

0,2

-0,2

-0,2

-0,4

-0,4

-0,6

-0,6

-0,8

-0,8

-1

-1

a)

b)

Fig. 5.17. Characteristic shape of the time dependency (a) prior to the derailment, (b) after the derailment.

xS ( t ) = S

x ( k ) (t k ) = X ( ) e
S

k =

i 2 t

d .

(5.36)

Here

XS ( f ) =

xS ( t ) ei 2 t d =

= S

x [k ] e

k =

i 2 f S k

x ( k ) ( t k )e
S

k =

i 2 t

d =

(5.37)

X ( f + kf )
S

k =

and X ( f ) are the spectral components of the analysed analog signal x ( t ) .


The number of samples N S taken for estimation of the real spectrum is determined by the size of time window t * and by the sampling frequency f S :

NS = fS t*

(5.38)

The characteristic shapes of the spectrum prior to the derailment and after the
derailment are shown in Figure 5.18. It is seen that although the spectrum before the
derailment slightly differs from that after the derailment, this difference is not apparent
in any regular sense. There are few reasons for this to be so. First, the receiver was
tuned on the maximal sensibility. That is why there are a lot of parasitic noises in the
spectrum, such as knocks of wheels over rails joints or natural noises of the locomotive. Second, as it was found out in the previous section, the vibrations of the rail under
the locomotive could be well modelled as the rail response to the first bogie. Because of
that, the larger is the distance between the receiver and emergency wheelset, the smaller
is the contribution of the emergency load into the spectrum.

147

(a)

1E-007

1E-009

1E-011
0

100

200

300

Modulus of amplitude spectrum | w | (m s)

Modulus of amplitude spectrum | w | (m s)

before derailment

1E-005

400

after derailment

1E-005

(b)

1E-007

1E-009

1E-011
0

100

Frequency f (Hz)

200

300

400

Frequency f (Hz)

a)

b)

Fig. 5.18. Characteristic shape of the spectrum (a) prior to the derailment, (b) after the derailment.

Thus, to detect the derailment, the data should be processed further. The following method was proposed. The obtained spectrum was first subdivided into frequency
bands of the width of f = 50 Hz , e.g. 0 50 Hz , 50 100 Hz , 100 150 Hz . Then,
a summation of the spectral components was calculated at each frequency band:

j 1
S k = w 2 k f + f
, k = 0,1, 2...

N f
j =1

Nf

(5.39)

In expression (5.39), f = 50 Hz is the bandwidth, N f is the number of measured frequencies in the band (for the sampling frequency of 2000 Hz , N f = 50 ).

Processed in this way, experimental data showed the following. There exist only
two frequency bands, namely the band 100 150 Hz and the band 300 350 Hz , in
which the sum Sk significantly changes at the moment of derailment. In Figure 5.19,
this sum is plotted versus time for the frequency band 100 150 Hz . The moment of the
derailment is depicted with the help of the vertical dashed line.
Figure 5.19 shows that after the derailment, a significant amplification takes
place of the sum S 2 . This is in good agreement with the theoretical prediction of the
previous sections, in accordance with which the emergency load should amplify the
spectrum around the natural frequency of the sleepers ( 140 Hz ). This fact is reflected
in Figure 5.19 by two dashed horizontal lines. These lines correspond to Figure 5.8 and
show the result of summation of the spectral components (with the sampling frequency
of 2000 Hz ) in the frequency band 100 150 Hz . In the time interval before the derailment, the response to the normal motion of the train is used, i.e. the solid line in Figure 5.8. The line after the derailment is obtained by using the response that takes into
account the emergency load that moves at the distance 150 m (the dashed line in the
right part of Figure 5.8).

148

Sum of spectral components S2 (m s)

1E-005

moment of derailment
measurements
theory

1E-006

1E-007
0

10

15

20

25

30

Time t (s)

Fig. 5.19. Summation of the spectral components versus time for the frequency band 100-150 Hz.

Thus, both the experiment and the theoretical investigation show that at the moment of derailment, vibrations of the rails under the locomotive are amplified in the frequency band that surrounds the natural frequency of vibrations of the sleepers. For the
railway track at hand this frequency is about 140 Hz . It is important to note that the
amplification measured in the experiment is much higher than that predicted by the
theoretical modelling. This gives a hope that the locomotive detection of the derailment
is possible at much larger distances than that predicted by the theory (not more than
150 m ). To prove it an extra experimental study is needed. The theoretical model should
be also improved by taking into account the transient character of the derailment and a
more accurate modelling of the impact of the derailed bogie against the sleepers.
The spectral method is also to be improved for increasing the distance at which
the locomotive detection of derailment is possible. One of possibilities to do this is to
calculate not the sum (5.39) but the following ratio
Nf

j =1
Nf

Sk =

f j w(2) ( f j ) w(1) ( f j )
fj f

( w ( f ) w ( f ))
f w (f )
f f
j =1

(2)

Nf

j =1
Nf

(1)

(1)

f j = k f + f

j 1
,
Nf

k = 0,1, 2... ,

(5.40)

w ( f )
j =1

(1)

*
where f is a frequency that is close to the natural frequency of vibrations of the sleepfj
is, actually, a spectral window that allows deers f s . In this method, factor
fj f *

creasing contribution of the frequencies, which do not surround frequency f s . Addi-

149

tionally, the procedure of normalization on the summation of spectral components in


(5.40) removes the contribution of the loads closest to the observation point and allows
to avoid difficulties associated with sensibility of the spectrum to variations of the dis/
tance between the first wagon and the locomotive. If the ratio S k is larger than 1, it implies that there exists a derailment.
/
Numerical calculations show that the ratio S k depends strongly upon the fre/

quency f * . Below, the table of values of S k is presented for different values of distance D between the emergency load and the receiver. Five values of the frequency f *
are used.

f * [ Hz ]

D [m]

110

25
50
100
150
25
50
100
150
25
50
100
150
25
50
100
150
25
50
100
150

122

122.5

123

130

Sk
0.9380
0.9212
0.9639
1.0358
3.1503
3.8007
4.7180
1.9903
5.6097
6.8605
8.5468
2.9933
2.4715
2.9664
3.5690
1.5396
0.9016
0.9568
0.6041
0.9146

The table shows that using the sum (5.40), the detection distance can be increased by factor two, i.e. the derailment can be identified at distances much larger than
/
150 m . As it is also seen from the table, S k is larger than 1 only in the case that the
frequency f * is located in a quite narrow frequency band: from 120 to 125 Hz . Other/

wise, S k is about 0.9 1 . In practice, properties of the ground can vary considerably.
Consequently, f * can be found only from the experimental data. Different measures
show that the stiffnesses K and , used in our model, change in the range from 2 107
to 5 108 N/m [33]. Because of that, the frequency f * can vary for the concrete sleepers with mass of 250 kg in the frequency range 50 350 Hz .

150

Finalizing the chapter, it has to be underlined that the model employed for theoretical predictions of the railway track vibrations should undergo a significant improvement. This improvement is concerned with modelling the train bogies. They
should be modelled as mass-spring-dashpot systems, performing its own vibrations and
reflecting waves that propagate in the track. The latter factor is especially important,
since, presumably, it will lead to a decrease of the distance at which the locomotive detection of derailment is possible. Thus, one can suppose that this distance cannot exceed
200 m .

151

CONCLUSIONS.
One of the most important requirements to a high-speed railway track is that it
has to be sufficiently straight. For peat and organic clay soil shear wave velocity can be
even lower than 150 km/h (for instance, in The Netherlands and in Sweden). Sites with
such low shear wave velocities are particularly susceptible to significant amplification
of train-track vibrations at high train speeds. This amplification, occurring when the
train speed is close to the Rayleigh wave velocity, is not only worrisome for its environmental impacts (discomfort to people and damage to buildings), deterioration of the
railway track and elevation of power supply to the trains but, being a threatening phenomenon that might cause derailment, raises concerns about the running safety of the
trains. To prevent this amplification from happening, a theoretical study of train-track
dynamics at high speeds is needed that would enable to design the high-speed railway
track properly. In order to accomplish this, dynamic three-dimensional models that account for the track-subsoil interaction are to be developed.
In this thesis, the steady-state response of various 3D models for a railway track
to a set of uniformly moving constant loads (train) have been investigated to reveal the
factors, which are of primary importance in dynamics of a high-speed railway track and
to compare different track models. Concerning the parametric analysis of the models,
the main attention has been paid to the configuration of the ground subsoil; damping in
the ground, in the pads and in the ballast; and to the effect of the spatially discontinuous
rail-ground interaction that is caused by the sleepers.

Main results.
It has been shown that the critical velocity of the train that is conditioned by its
dead weight (constant load) is approximately equal to the Rayleigh wave speed in the
ground surrounding the railway track. This result is valid as for a homogeneous (embedded) track as for an inhomogeneous (conventional) track. The magnitude of the
critical velocity depends mainly upon the physical properties and saturation of the subsoil. If the ground is soft, the value of the critical velocity, varying from 150 to 300
km/h, is attainable by nowadays operating high-speed trains. This implies that a careful
study is necessary on what would happen if such a regime of motion were to occur.
The investigation of influence of the ground stratification on the dynamic response of the railway track to a constant moving load has demonstrated that the soil
stratification strongly affects the dynamic response of the railway track. Variations in
depth of the soil layers and in their physical properties can result in 1) shifting and
(sometimes) introducing new critical velocities of the train; 2) reducing or increasing
the dynamic amplification of the track response; 3) scaling the spatial pattern of the rail
response.
Elastic drag experienced by a high-speed train due to excitation of ground vibrations has been analysed. It has been shown that the elastic drag can be comparable and,
starting from the velocities close to the Rayleigh wave velocity, can exceed the aerodynamic drag. Thus, the elastic drag is an important design criterion for high-speed trains
to be accounted for as a source of perceptible energy loss.

152

Comparative analysis of the dynamic responses of models for the embedded


track and for the conventional track carried out in Chapter 4 has shown qualitative and
quantitative resemblance of the obtained results for both models. This allows to conclude that for constant moving load one can use simpler homogeneous 3D models for
the railway track. The latter circumstance can be very useful for railway engineers, for
numerical calculations for the conventional track are much slower than those for the
embedded track. However, we ought to underline that this conclusion is right only for
constant loading.
The study of the effect of the damping in the considered models (Chapters 1-4)
has shown that, although the increase of the damping (in the subsoil, in the pads and in
the ballast) generally reduces the amplification of the track vibrations at the critical velocity, it leads also to a substantial increase of the elastic drag in the subcritical case
(Chapter 2). That is why the increase of the damping can cost a perceptible energy loss,
which is unfavourable for the high-speed transportation.
In Chapter 5, a possibility to detect derailment of a wagon of a freight train by
means of measurements of a railway track response under a locomotive has been investigated. It has been demonstrated that the spectrum of vibrations in the emergency case
(with one derailed wheelbase) differs from that in the normal case in the frequency
band, which surrounds the natural frequency of the sleepers and that this difference can
be detected at distances not exceeding 200 meters.

Discussion of the results.


The main goal of this thesis was to point it out that there exist two important
problems in high-speed transportation. The first one is the problem of the critical velocity, which is close to the Rayleigh wave speed and can be reached by modern highspeed trains. The second problem is the elastic drag, which, from the authors point of
view, must be an important design criterion to be taken into consideration as a source of
considerable energy loss. Both problems have been extensively elaborated in Chapters
1-4. However, the possible ways of solution to these problems have not been discussed
in the thesis. This is the aim of this paragraph.
Let us first discuss the problem of the critical velocity and raise the question on
how to avoid the amplification of the track vibrations at this velocity. In our perception,
there are three ways to solve the problem. The first way, which is now being realized in
the south of the Netherlands, is to decouple vibrations of the ground and those of the
rails by means of piles on which the track is to be mounted. This method, however, has
two drawbacks: it is very expensive and it is not clear whether the structure is robust
against the dynamic loading by high-speed trains.
The second solution to the problem is to make the track stiffer, likewise in the
case of the embedded track. Such a track has two advantages: a low maintenance cost
and a high critical velocity. It is not clear, however, what will be the long-term behaviour of the slab of the embedded track after it has been laid on a soft soil. Will it sink
into the soil? Will it keep its straight (in the vertical direction) shape? This will be
known quite soon.
The third method to get rid of the critical velocity by significantly increasing it
is to move the train traffic into tunnels, which should be laid in stiff layers of the
ground. This way also does not seem to be perspective to the author, for it is the most

153

expensive one. However, from the environmental point of view, the idea of replacing
the trains from the earth surface is very attractive.
The problem of the elastic drag raises another important question: how to decrease this drag? From the authors point of view, there are four crucial points, which
should be accounted for while dealing with this problem. Firstly, as has been shown in
Chapters 2 and 4, the elastic drag is conditioned by generation of elastic waves in the
track and in the track subsoil by a moving train. These waves consume energy of the
locomotive engine. The energy consumption can become perceptible if the train velocity
is close to the critical velocity. To avoid extra energy losses at high speeds, one should
create conditions for reduction of the energy of these waves. This purpose can be
achieved by increasing the critical velocity, for instance, by means of employment of
the embedded track.
The second basic point is the effect of the inhomogeneity of the conventional
railway track. Although it has been shown in Chapter 4 that the elastic drag for the conventional track almost does not differ from that for the embedded track, we should bear
in mind that: 1) this is valid for a constant loading; 2) the conventional track, being inhomogeneous over its length, always holds a possibility to cause the transient radiation
of elastic waves that takes place at any train velocity. The latter circumstance leads to an
increase of the energy loss on wave radiation. Thus, in our opinion, replacement of the
conventional track by the homogeneous embedded track could also improve the situation in this respect.
The third factor that seems to be important is a distribution of the trains weight.
In Chapter 2, it has been demonstrated that the elastic drag is proportional to the trains
weight square. This implies that the contribution of heavy wagons will be considerably
larger than that of more lightweight wagons. Solution to this problem is a uniform distribution of the weight over trains length. This, although because of other reasons
(mounting of engines to each wagon to increase power), is being realized in modern
high-speed trains (for example, in last generation of German ICE).
The last criterion that should be accounted for in design of high-speed trains is
the bogie wheelbase, the proper choice of which can also reduce the elastic drag. As has
been shown in Chapters 2 and 4, the elastic drag has a minimum if the bogie wheelbase
is approximately equal to 2 m . The minimal value of the elastic drag in this case is almost twice smaller than that for distances exceeding 4 m .

Future perspectives in modelling dynamics of the railway


track.
The railway track is a very complicated system. Modelling dynamic behaviour
of the track, it is impossible to take into consideration all its properties. However, moving step-by-step in development of the track models, one can account for more and
more peculiarities of the railway track and of the trains.
The study presented in this thesis was carried out under certain assumptions,
which introduce severe restrictions to the range of applicability of the considered models. These assumptions, namely the constant loads, uniform distribution of the stresses
over the track-ground contact, symmetric loading with respect to the centreline of the
track, etc., make the developed models incapable of giving an entire representation of
the train-track dynamics. Thus, the thesis leaves a vast field for future developments and
investigations.
154

Possible improvements of the dynamic models developed in this work are listed
below:
1.
2.
3.
4.
5.
6.
7.

Accounting for internal degrees of freedom of the train.


Introduction of asymmetric loading (with respect to the centreline of the track).
Accounting for proper boundary conditions at the track-ground interface.
Taking into consideration the lateral vibrations of the rails.
Accounting for possibility of break of the wheel-rail contact.
Accounting for inhomogeneity of the ground over the track length.
Considering the effect of the aerodynamic drag on the contact force between the
trains wheels and the rails.

All these factors should be objectives of future investigations and the author believes
that the models developed in this monograph could be considered as a reference point
for the studies to come.

155

References.
[1] Abramowitz, M., Stegun, I.A., Handbook of mathematical functions; with formulas,
graphs, and mathematical tables, New York, Dover, 1970.
[2] Achenbach J.D., Sun C.T. Moving Load on a Flexible Supported Timoshenko
Beam, International Journal of Solid and Structures V.1: 353-370, 1965.
[3] Achenbach J.D., Wave propagation in elastic solids, Amsterdam-London, NorthHolland Publishing Company, 1973.
[4] Andersen L, Nielsen SRK, Kirkegaard PH, Finite element modelling of infinite
Euler beam on Kelvin foundations exposed to moving loads in convected co-ordinates,
Journal of Sound and Vibration V.241 (4): 587-604, April 5, 2001.
[5] Auersch L., Schmid G., A simple boundary element formulation and its application
to wavefield excited soil-structure interaction, Earthquake Engineering and Structural
Dynamics V.19: 931-947, 1990.
[6] Auersch L., Wave propagation in layered soils: theoretical solution in wavenumber
domain and experimental results of hammer and railway traffic excitation, Journal of
Sound and Vibration V.173 (2): 233-264, June 2, 1994.
[7] Auersch L., Vehicle-track-interaction and soil dynamics, Vehicle System Dynamics V.29: 553-558, Suppl. S, 1998.
[8] Bakker M.C.M., Verweij M.D., Kooij B.J., Dieterman H.A., The travelling point
load revisited, Wave Motion V.29: 119-135, 1999.
[9] Belotserkovskiy P.M. High-frequency vertical vibrations of a rail under the action
of a mobile harmonic force, Mechanics of Solids V.30(3): 177-185, 1995.
[10] Belotserkovskiy P.M., On the oscillations of infinite periodic beams subjected to a
moving concentrated force, Journal of Sound and Vibration V.193(3): 706-712, 1996.
[11] Belotserkovskiy P.M. Periodic spring response to an impact and a suddenly applied concentrated stationary force, Journal of Sound and Vibration V.228(1): 51-68,
November 18, 1999.
[12] Blohine E.P., Manaskine L.A., Dynamics of train (in Russian), Moscow, Transport, 1982.
[13] Bogacz R., On self-excitation of moving oscillator interacting at two points with a
continuous system, Nonlinear vibration system V.19: 239-250, 1979.
[14] Bogacz R., Krzyzynski T. and Popp K., Influence of Beam Models on the Solution of the Generalized Jobs Problem, Z. Angew. Math. Mech. V.69: 320-321, 1989.

156

[15] Bogacz R., Krzyzynski T. and Popp K., On the Generalization of Mathews Problem of the Vibrations of a Beam on Elastic Foundation, Z. Angew. Math. Mech. V.69:
243-252, 1989.
[16] Bogacz R., Krzyzynski T. and Popp K., On Dynamics of Systems Modelling Continuous and Periodic Guideways, Arch. Mech. V.45: 575-593, 1993.
[17] Bychenkov V.A., Krysov S.V., Resistance to motion of a wheel over 1D elastic
system (in Russian), Dynamics of systems, GSU, Gorky: 79-82, 1986.
[18] Bychenkov V.A., Krysov S.V., Wave resistance to rolling of a rigid wheel over
1D visco-elastic support (in Russian), Mashinovedenie 3: 60-66, 1988.
[19] Cheng Y.S., Au F.T.K., Cheung Y.K., Zheng D.Y. On the separation between
moving vehicle and bridge, Journal of Sound and Vibration V.222(5): 781-801 May
20, 1999.
[20] Choros H. and Adams G.G., A Steadily Moving Load on an elastic Beam Resting
on a Tensionless Winkler Foundation, J.of Appl. Mech. V.46: 175-180, 1979.
[21] Clouteau D, Elhabre ML, Aubry D, Periodic BEM and FEM-BEM coupling
Application to seismic behaviour of very long structures, Comput. Mech. V.25(6): 567577, June, 2000.
[22] Cole J. and Huth J., Stress produced in a half-plane by moving loads, Journal of
Applied Mechanics V.25: 433-436, 1958.
[23] Cornejo Crdova C.J., Elastodynamics with hysteretic damping. PhD thesis, Delft,
Delft University Press , 2002.
[24] Criner H.E. and Mc Cann G.D. Rails on Elastic Foundation Under the Influence
of High-Speed Traveling Loads, J. Appl. Mech. V.20: 13-22, 1953.
[25] Degrande G, Lombaert G, An efficient formulation of Krylovs prediction model
for train induced vibrations based on the dynamic reciprocity theorem, J. Acoust. Soc.
AM V.110(3): 1379-1390, Part 1, September, 2001.
[26] Degrande G, Schillemans L, Free field vibrations during the passage of a Thalys
high-speed train at variable speed, Journal if Sound and Vibration V.247 (1): 131-144,
October 11, 2001.
[27] Denisov G.G., Novikov V.V., Kugusheva E.K., To the problem of stability of 1D
infinite systems (in Russian), Applied Mathematics and Mechanics V.49: 691-696,
1985.
[28] Dieterman H.A. Critical velocities of moving loads on soft soil, Report of Delft
University N 03.21.0.22.23, April, 1995.

157

[29] Dieterman H.A., Metrikine A.V., The equivalent stiffness of a half-space interacting with a beam. Critical velocities of a moving load along the beam, European Journal of Mechanics A/Solids V.15(1): 67-90, 1996.
[30] Dieterman H.A., Metrikine A.V., Critical velocities of a harmonic load moving
uniformly along an elastic layer, Trans. ASME J. of Applied Mechanics V.64: 596-600,
1997.
[31] Dieterman H.A. and Metrikine A.V., Steady-state displacements of a beam on an
elastic half-space due to a uniformly moving constant load, European Journal of Mechanics A/Solids 16(2): 295-306, 1997.
[32] Dominguez J., Dynamic stiffness of rectangular foundations, Publ. No. R78-20,
Department of Civil Engineering M.I.T., August, 1978.
[33] Esveld C., Modern railway track, MRT-Productions, W. Germany, 1989.
[34] Esveld C. (Ed.), Kritische treinsnelhelheden. Technical Report 7-95-110-8, Technische Universiteit Delft, Faculteit der Civiele Techniek, Railbouwkunde (Dutch), 1995.
[35] Felszeghy S.F. The Timoshenko beam on an elastic foundation and subject to a
moving step load, Journal Vib. Acoust. V.118(3): 277-284, July, 1996.
[36] Filippov A.P., Steady-State Vibrations of a Infinite Beam on an elastic half-space
Under Moving Load (in Russian), Izvestia of AN USSR OTN Mechanica and Mashinostroenie V.6: 97-105, 1961.
[37] Filippov A.P., Vibrations of Distorted Systems (in Russian), Moscow,
Mashinostroenie, 1970.
[38] Frischgesell T, Krzyzynski T, Bogacz R, Popp K, On the dynamics and control of
a guideway under a moving mass, Heavy Vehicle Systems V.6 (1-4): 176-189, 1999.
[39] Frba L., Vibrations of Solids and Structures Under Moving Loads, Groningen,
Noordhoff International Publishing, 1972.
[40] Frba L., Vibrations of Solids and Structures Under Moving Loads (third edition),
Prague, Academy of Sciences of the Czech Republic, 1999.
[41] Frba L, Nakagiri S, Yoshikawa N, Stochastic finite-elements for a beam on a
random foundation with uncertain damping under a moving force, Journal of Sound
and Vibration V.163(1): 31-45, May 8, 1993.
[42] Frba L, History of Winkler foundation, Vehicle Syst. Dyn. V.24: 7-12, Suppl. S,
1995.
[43] Frba L, Yoshikawa N, Bounds analysis of a beam based on the convex model of
uncertain foundation, Journal of Sound and Vibration V.212 (3): 547-557, May 7,
1998.

158

[44] Frba L, Pirner M, Load tests and modal analysis of bridges, Eng. Struct.
V.23(1): 102-109, January, 2001.
[45] Frba L, A rough assessment of railway bridges for high-speed trains, Eng.
Struct. V.23(5): 548-556, May, 2001.
[46] Fuchs B.A., Shabat B.V., Berry J., Functions of complex variables and some of
their applications, Oxford, Pergamon, 1964.
[47] Gaitanaros A.P., Karabalis D.L., Dynamic analysis of 3D flexible embedded
foundations by a frequency domain BEM-FEM, Earthquake Engineering and Structural Dynamics V. 16: 653-674 1988.
[48] Gakenheimer D.C., Miklowitz J., Transient excitation of an elastic half-space by a
point load traveling on the surface, Journal of Applied Mechanics V.36: 505-515,
January, 1969.
[49] Gakenheimer D.C., Response of an elastic half space to expanding surface loads,
Transactions of the ASME. Journal of Applied Mechanics V.38(1): 99-110, March,
1971.
[50] Georgiadis H.G., Barber J.R., Steady-state transonic motion of a line load over an
elastic half-space the corrected Cole/Huth solution, Journal of Applied Mechanics
V.60(3): 772-774, September, 1993.
[51] Gonorovsky I.S., Radiotechnical circuits and signals (in Russian), Moscow, Radio i Svyaz, 1986.
[52] Gracheva L.O., Hamoev A.D., Martynuke A.V., Derailments of wagons: causes
and means of their prevention (in Russian), Zheleznodorozhniy transport V.1: 40-43,
1996.
[53] Gradstein I.S., Ryzhik I.M., Tables of integrals, sums, series and products (in Russian), Moscow, Nauka, 1971.
[54] Grassie S.L., Gregory R.W., Harrison D. and Jonhson K.L., The dynamic response of railway track to high frequency vertical/lateral/longitudinal excitation, Journal of Mechanical Engineering Science V.24: 77-90/91-96/97-102, 1982.
[55] Grassie SL, Saxon MJ, Smith JD, Measurement of longitudinal rail irregularities
and criteria for acceptable grinding, Journal of Sound and Vibration V.227 (5): 949964, November 11, 1999.
[56] Grassie SL, Elkins JA, Rail corrugation on North American Transit systems, Vehicle System Dynamics V. 29: 5-17, Suppl. S, 1998.
[57] Grundmann H, Hartmann C, Waubke H, Structures subjected to stationary stochastic loadings. Preliminary assessment by statistical linearization combined with an
evolutionary algorithm, Computers and Structures V.67(1-3): 53-64, April-May, 1998.

159

[58] Grundmann, H., Lieb, M., Trommer, E., The response of a layered half-space to
traffic loads moving along its surface, Archive of Applied Mechanics V.69(1): 55-67,
February, 1999.
[59] Grundmann H, Trommer E, Transform methods what can they contribute to
(computational) dynamics? Computers and Structures V. 79 (22-25): 2091-2102, September, 2001.
[60] Haskel N.A., The dispersion of surface waves on multiplayer media, Bulletin of
the Seismological Society of America V.43: 17-34, 1953.
[61] Hopkins, T., Silva, J.P., Marder, B., Turban, B., Kelley, B., Maglift Monorail: A
high performance, low cost, and low risk solution for high-speed transportation, Proceedings of High Speed Ground Transportation Association Annual Conference. Seatle,
June 6-9, 1999.
[62] Hunaidi M.O., P.A. Chen, J.H. Rainer, Tremblay M. Shear moduli and damping
in frozen clay by resonant column tests, Canadian Geotech. Journal V.33: 510-514,
1996.
[63] Jezequel L., Analysis of critical speeds of a moving load on an infinite periodically supported beam, Journal of Sound and Vibration V.73(4): 606-610, 1980.
[64] Jezequel L., Response of periodic systems to a moving load, ASME Journal of
Applied Mechanics V.48(3): 603-618, 1981.
[65] Karabalis D.L. and Beskos D.E., Dynamic response of 3D rigid surface foundation by time domain boundary element method, Earthquake Engineering and Structural Dynamics V.12: 73-93, 1984.
[66] Karamanlidis D., Prakash V., Buckling and vibration analysis of flexible beams
resting on an elastic half-space, Earthquake Engineering and Structural Dynamics V.
16(8): 1103-1114, November, 1988.
[67] Kausel E. and Rosset J.M., Stiffness matrices for layered soils, Bulletin of the
Seismological Society of America V.71: 1743-1761, 1981.
[68] Kaynia A.M., Madhus C., Zackrisson P., Ground vibration from high-speed
trains: prediction and countermeasure, J. Geotech. Geoenviron. V.126(6): 531-537,
June, 2000.
[69] Kenney J.T., Steady-State Vibrations of Beam on Elastic Foundation for Moving
Load, Journal of Applied Mechanics V.21: 359-364, 1954.
[70] Kerr A.D., The Continuously Supported Rail Subjected to an axial Force and a
Moving Load, Int. J. Mech. Sci. V.14: 7178, 1972.
[71] Kerr A.D On the determination of the rail support modulus K, International
Journal of Solid and Structures V.37(32): 4335-4351, 2000.

160

[72] Kim, D.S., Lee J.S., Source and attenuation characteristics of various ground vibrations, Geotechnical Special Publication (2), ASCE, Reston, VA, USA: 1507-1517,
1998.
[73] Knothe K, Ripke B, The effect of the parameters of wheelset, track and running
conditions on the growth-rate of rail corrugations, Vehicle System Dynamics V.18:
345-356, Suppl. S, 1989.
[74] Knothe K. and Grassie S.L., Modelling of Railway and Vehicle/Track Interaction
at High Frequencies, Vehicle System Dynamics V.22: 209-262, 1993.
[75] Knothe K. and Wu Y., Receptance Behaviour of Railway Track and Subgrade,
Archive of Applied Mechanics V.68: 457-470, 1998.
[76] Knothe K, Bohm F, History of stability of railway and road vehicles, Vehicle
System Dynamics V.31(5-6): 283-323, June, 1999.
[77] Knothe K, Grassie SL, Workshop on Rail Corrugations and Out-of-Round
Wheels, Journal of Sound and Vibration V.227 (5): 895-897, November 11, 1999.
[78] Knothe K. Gleisdynamik und Wechselwirkung zwischen Fahrzeug and Fahrweg,
Z. Angew Math. Mech. V.79(11): 723-737, 1999.
[79] Kononov A.V., Dieterman H.A., The elastic field generated by two loads moving
along two springs on an elastically supported membrane, Journal of Sound and Vibration V.214(4): 725-746, 1998.
[80] Kononov A.V., Wolfert A.R.M, Load motion along a beam on a visco-elastic
half-space, European Journal of Mechanics and Solid V.19(2): 361-371, March, 2000.
[81] Kruse H., Metrikine A.V. and Popp K. Eigenfrequencies of a two-mass oscillator
uniformly moving along a spring on a visco-elastic foundation, Journal of Sound and
Vibration V.218(1): 103-116, 1998.
[82] Kruse H, Popp K. and Krzyzynski T., On Steady State Dynamics of Railway
Track Modelled as Continuous Periodic Structures, Machine Dynamics Problems
V.20: 149-166, 1998.
[83] Kruse H, Popp K, A modular algorithm for linear, periodic train-track models,
Archive of Applied Mechanics V.71(6-7): 473-486, July, 2001.
[84] Krzyzynski T. and Popp K., A Mathematical Treatise on Periodic Structures under
Traveling Loads with an Application to Railway Track, Proc. 9th Conf. Mathematics in
Industry. Teubner: 93-100, 1997.
[85] Krzyzynski T., On dynamics of a railway track modelled as a two-dimensional
periodic structure, Heavy Vehicle Systems V.6(1-4): 330-344, 1999.
[86] Krylov V.V., Generation of ground vibrations by superfast trains, Applied Acoustics V.44(2): 149-164, 1995.

161

[87] Krylov V.V., Effect of Track Properties on Ground Vibrations Generated by


High-Speed Trains, Acustica Acta Acustica V.84: 78-90, 1998.
[88] Krysov S.V., Forced vibrations and resonance in elastic systems with moving
loads (in Russian), Gorkiy, GSU, Educational guidance, 1985.
[89] Krysov S.V., Holuev V.V., Forces of resistance to motion of constant loads along
elastic ways (in Russian), Dynamics of systems, Gorky: 142-149, 1985.
[90] Krysov S.V., Holuev V.V., Forces of resistance to motion of variables loads over
railway (in Russian), Mashinovedenie V.1: 91-94, 1988.
[91] Kulemin V.N., Merkulov V.I., Misevich V.N., Misevich P.V., Vantsev S.S.,
Churazov A.G., Locomotive detection of derailment of a wheelset of a freight wagon.
Results of the experiments carried out in August 11, 2000 on station Inzhenernaya (in
Russian), Report GD-A-129-42, August, 2000.
[92] Kulemin V.N., Merkulov V.I., Misevich V.N., Misevich P.V., Vantsev S.S.,
Churazov A.G., Locomotive detection of derailment of a wheelset of a freight wagon.
Results of the experiments carried out in December 23, 2000 on station Inzhenernaya
(in Russian), Report GD-A-129-71, December, 2000.
[93] Labra J.J., An axially stressed railroad track on an elastic continuum subjected to
a moving load, Acta Mechanica V.22: 113-129, 1975.
[94] Lamb H., On the Propagation of Tremors Over the Surface of an Elastic Solid,
Phil. Trans. Roy. Soc. London, Ser. A, V.CCIII(1): 1-42, 1904.
[95] Landau L.D., Lifshits E.M., Short course of theoretical physics. Electrodynamics
of continuous mediums (in Russian), Moscow, Nauka, 1982.
[96] Landau L.D., Lifshits E.M., Short course of theoretical physics. Mechanics (in
Russian), Moscow, Nauka, 1988.
[97] Lansing, D.L., The displacements in an elastic half-space due to a moving concentrated normal load, NASA technical report, TR R-238, 1966.
[98] Laturelle F.G., The stresses produced in an elastic half-space by a normal step
loading over a circular area. Analytical and numerical results, Wave Motion V.12(2):
107-27, March, 1990
[99] Lee H.P., Dynamic response of a Timoshenko Beam on a Winkler foundation
subjected to a moving mass, Appl. Acoust. V.55(3): 203-215, November, 1998.
[100] Lependine L.F., Acoustics (in Russian), Moscow, High school, 1978.
[101] Lieb M., Sudret B., A fast algorithm for soil dynamics calculations by wavelet
decomposition, Archive of Applied Mechanics V.68(3-4): 147-157, 1990.

162

[102] Lombaert G, Degrande G, Clouteau D, Numerical modelling of free field traffic


vibrations, Soil Dyn. Earthq. Eng. V.19(7): 473-488, October, 2000.
[103] Lombaert G, Degrande G, Experimental validation of a numerical prediction
model for free field traffic induced vibrations by in situ experiments, Soil Dyn. Earthq.
Eng. V.21(6): 485-497, August, 2001.
[104] Lombaert G, Degrande G, Clouteau D, The influence of the soil stratification on
free field traffic-induced vibrations, Archive of Applied Mechanics V.71(10): 661-678,
October, 2001.
[105] Luco J.E., Apsel R.J., On the Greens functions for a layered half-space. Part 1,
Bulletin of the Seismological Society of America V.73(4): 909-929, August, 1983.
[106] Lysuke V.S., Causes and mechanism of wheel derailment. Problem of wear of
wheels and rails (in Russian), Moscow, Transport, 1997.
[107] Marple, S.Lawrence (Jr.), Digital spectral analysis: with applications, Englewood
Cliffs Prentice-Hall, 1987.
[108] Mathews P.M. Vibrations of a Beam on Elastic Foundation, Z. Angew. Math. U.
Mech. V.38: 105-115, 1958.
[109] Max G., Methods and technique of signal processing at physical measurements
(in Russian), Moscow, Mir, V.1, 1983.
[110] Mead, D. J., Vibration response and wave propagation in periodic structures,
ASME Journal of Engineering for Industry Series B 93 V. 81: 783-792, 1971.
[111] Metrikine, A.V., Dieterman H.A., Three-dimensional vibrations of a beam on an
elastic half-space: resonance interaction of vertical-longitudinal and lateral beam
waves. Trans. ASME J. of Applied Mechanics V.64: 951-956, 1997.
[112] Metrikine A.V., Dieterman H.A., Lateral vibrations of an axially compressed
beam on an elastic half-space due to a moving lateral load, European Journal of Mechanics A/Solids V.18: 147-158, 1999.
[113] Metrikine A.V. and Dieterman H.A., The equivalent vertical stiffness of an elastic half-space interacting with a beam, including the shear stresses at the beam - halfspace interface, European Journal of Mechanics A/Solids V.16(3): 515-527, 1997.
[114] Metrikine A.V. and Dieterman H.A., Instability of vibrations of a mass moving
uniformly along an axially compressed beam on a visco-elastic foundation, Journal of
Sound and Vibration V.201(5): 567-576, 1997.
[115] Metrikine A.V. and Popp K., "Instability of vibrations of an oscillator moving
along a beam on an elastic half-space", European Journal of Mechanics A/Solids
V.18(2): 331-349, 1999.

163

[116] Metrikine A.V. and Popp K., "Vibration of a periodically supported beam on an
elastic half-space ", European Journal of Mechanics A/Solids V.18(4): 679-701, July,
1999.
[117] Metrikine AV, Popp K, Steady-state vibrations of an elastic beam on a viscoelastic layer under moving load, Archive of Applied Mechanics V.70 (6): 399-408,
July, 2000.
[118] Metrikine A.V., Vostrukhov A.V., How much energy does a high-speed train
loose to excite a ground vibration? Abstracts of EUROMECH Colloquium 409 Dynamics and Long-term Behaviour of Railway Vehicles, Track and Subgrade, Hannover, March 6-9: 10, 2000.
[119] Metrikine A.V., Vostroukhov A.V., Vrouwenvelder A.C.W.M., Drag experienced by a high-speed train due to excitation of ground vibrations, International Journal of Solid and Structures V.38: 8851-8868, 2001.
[120] Metrikine A.V., Wolfert A.R.M. and Dieterman H.A., "Transitional radiation in
an elastically supported spring. Abrupt and smooth variation of the support stiffness",
Wave motion V.27: 291-305, 1998.
[121] Metrikine A.V., Verichev S.N., Instability of a moving two-mass oscillator on a
flexibly supported Timoshenko beam, Archive of Applied Mechanics V.71: 613-624,
2001.
[122] Metrikine A.V., Wolfert A.R.M., Vrouwenvelder A.C.W.M., Steady-state response of periodically supported structures to a moving load, Heron V.4(2): 91-107,
1999.
[123] Metrikine A.V., Vrouwenvelder A.C.W.M., Surface ground vibration due to a
moving train in a tunnel: two-dimensional model, Journal of Sound and Vibration
V.234(1): 43-66, 2000.
[124] Miklowitz J., The Theory of Elastic Waves and Waveguides, Amsterdam, NHPC,
1978.
[125] Mohammadi M., Karabalis D.L., Dynamic 3D soil-railway track interaction by
BEM-FEM, Earthquake Engineering and Structural Dynamics V.24 (9): 1177-1193,
September, 1995.
[126] Muller S, Knothe K, Stability of wheel-track dynamics for high frequencies,
Archive of Applied Mechanics V.67(6): 353-363, August, 1997.
[127] Naprstek J, Frba L, Stochastic modeling of track and its substructure, Vehicle
System Dynamics V.24: 297-310, Suppl. S, 1995.
[128] Nordborg A., Rail/wheel parametric excitation: Laboratory and field measurements, Acustica V.85(3), 355-365, May-June, 1999.

164

[129] Norwood F.R., Exact transient response of an elastic half space loaded over a
rectangular region of its surface, Transactions of the ASME. Journal of Applied Mechanics V.36(3): 516-22, September, 1969.
[130] Popp, K., Kruse, H., Kaiser, I., Vehicle-track dynamics in the mid-frequency
range. Vehicle System Dynamics V.31(5-6): 423-464, 1999.
[131] Popp K., Muller P.C., Ein Beitrag zur Gleisdynamil, Z. Angew. Math und Mech.
V.62: 65-67, 1982.
[132] Rabotnov Y.N., Mechanics of Solid, Moscow, Nauka, 1988.
[133] Rasmussen KM, Nielsen SRK, Kirkegaard PH, Boundary element method solution in the time domain for a moving time-dependent force, Comput. Struct. V.79(7):
691-701, March, 2001.
[134] Ripke B, Knothe K, Simulation of high-frequency vehicle track interactions,
Vehicle System Dynamics V.24: 72-85, Suppl. S., 1995.
[135] Rosset J.M., Chang D.-W. and Stokoe K.H., Comparison of 2D and 3D models
for analysis of surface wave tests, Proceedings of the 5th Conference on Soil Dynamics
and Earthquake Engineering. University of Karlsruhe: 111-126, 1991.
[136] Sheng, X., Jones, C.J.C., Petyt, M., Ground vibration generated by a harmonic
load acting on a railway track, Journal of Sound and Vibration V.225(1): 2-28, 1999.
[137] Sheng, X., Jones, C.J.C., Petyt, M., Ground vibration generated by a load moving along a railway track. Journal of Sound and Vibration V.228(1): 129-156, 1999.
[138] Stichel S, Knothe K, Fatigue life prediction for an S-train bogie, Vehicle System
Dynamics V.29: 390-403, Suppl. S., 1998.
[139] Suiker ASJ, de Borst R, Esveld C, Critical behaviour of a Timoshenko beam half
plane system under a moving load, Archive of Applied Mechanics V.68 (3-4): 158-168,
April, 1998.
[140] Suiker ASJ, Chang CS, de Borst R, et al., Surface waves in a stratified half-space
with enhanced continuum properties. Part 1: Formulation of the boundary value problem, European Journal of Mechanics A/Solids V.18(5): 749-768, September-October,
1999.
[141] Suiker ASJ, Chang CS, de Borst R, et al., Surface waves in a stratified half-space
with enhanced continuum properties. Part 2: Analysis of the wave characteristics in regard to high-speed railway tracks, European Journal of Mechanics A/Solids V.18(5):
769-784, September-October, 1999.
[142] Suiker A.S.J., deBorst R., Esveld C., Critical behaviour of a Timoshenko beamhalf plane system under a moving load, Archive of Applied Mechanics V.68(3-4): 158168, April, 1998.

165

[143] Suiker ASJ, de Borst R, Chang CS, Micro-mechanical modelling of granular material. Part 1: Derivation of a second-gradient micro-polar constitutive theory, Acta
Mechanica V.149(1-4): 161-180, 2001.
[144] Suiker ASJ, de Borst R, Chang CS, Micro-mechanical modelling of granular material. Part 2: Plane wave propagation in infinite media, Acta Mechanica V.149(1-4):
181-200, 2001.
[145] Suiker ASJ, Metrikine AV, De Borst R, Dynamic behaviour of a layer of discrete
particles, part 1: Analysis of body waves and eigenmodes, Journal of Sound and Vibration V.240(1): 1-18 , February 8, 2001.
[146] Suiker ASJ, Metrikine AV, De Borst R, Dynamic behaviour of a layer of discrete
particles, part 2: Response to a uniformly moving harmonically vibrating load, Journal
of Sound and Vibration V.240(1): 19-39, February 8, 2001.
[147] Suiker ASJ, Metrikine AV, De Borst R, Comparison of wave propagation characteristics of the Cosserat continuum model and corresponding discrete lattice models,
International Journal of Solid and Structures V.38(9): 1563-1583, February, 2001.
[148] Sun L. A closed-form solution of beam on visco-elastic subgrade subjected to
moving load, Computers and Structures V.80(1): 1-8, 2002.
[149] Szolc T., Medium frequency dynamic investigation of the railway wheelset-track
system using a discrete-continuous model, Archive of Applied Mechanics V.68(1): 3045, 1998.
[150] Thompson W.T., Transmission of elastic waves through a stratified soil medium, Journal of Applied Physics V.21: 89-93, 1950.
[151] Timoshenko S., Stresses in rails (in Russian), Transactions of the Institute of
Ways of Communications, 1915.
[152] Timoshenko S., Strength of Materials, Part 2. New York, Van Nostrand, 1942.
[153] Triantafyllidis T., 3D time domain BEM using half-space Greens functions,
Engineering Analysis with Boundary Elements V.8(3): 115-124, June, 1991.
[154] Vershinskiy S.V., Longitudinal dynamics of wagons in freight trains (in Russian),
Moscow, Transzheldorizdat, 1957.
[155] Vershinskiy S.V., Danikin V.N., Husidov V.D., Dynamics of wagon (in Russian),
Moscow, Transport, 1991.
[156] Vinogradova M.B., Roudenko O.V., Soukhoroukov A.P., Theory of waves (in
Russian), Moscow, Nauka, 1979.
[157] Verruijt A., Cornejo Crdova C.J., Moving loads on an elastic half-plane with
hysteretic damping, Journal of Apllied Mechanics, ASME V.68: 915-922, November,
2001.

166

[158] Vestnitskii A.I. and Metrikine A.V., Transient radiation in a periodically nonuniform guide, Mechanics of Solids V.28: 158-162, 1993.
[159] Vestnitskii A.I. and Metrikine A.V., Transition radiation in Mechanics, Physics
Uspehi V.39: 983-1007, 1996.
[160] Vestnitskii A.I., Kononov A.V., Metrikine A.V., Transition radiation in twodimensional elastic systmes, Journal of Applied Mechanics and Technical Physics
V.36(3), 1995.
[161] Volkov E.A., Numerical methods (in Russian), Moscow, Nauka, 1987.
[162] Vostrukhov A.V., Metrikine A.V., To a question about work of force providing
uniform motion of an object along a two-dimensional elastic system (in Russian),
Validations of materials and constructers: Scientific treatises of Nf IMAS RAN, Nizhniy
Novgorod, Russia: 241-247, 2000.
[163] Vostrukhov A.V., Metrikine A.V., Energy loss of a source maintaining uniform
motion of a load along elastically supported membrane, Acoustic Journal V.46(2):
177-181, 2000.
[164] Vostrukhov A.V., Work provided by a source maintaining a uniform motion of a
load over a surface of a half-space and an elastically supported membrane, Proceedings of Third Scientific Conference on Radiophysics, Nizhny Novgorod, Russia, May 47: 273-274, 1999.
[165] Vostrukhov A.V., Metrikine A.V., Vrouwenvelder A.C.W.M., Energy Loss of a
High-speed Train Due to Elastic Wave Radiation, Proceedings of conference Actual
Problems in Mechanics, S. Petersburg, June 1-10: 1-14, 2000.
[166] A.V. Vostroukhov, A.V. Metrikine, A.C.W.M. Vrouwenvelder, Kulemin V.N.,
V.I. Merkulov, V.N. Misevich, G.A. Utkin, Locomotive detection of derailment of a
wagon of a fright train: theory and experiment, Archive of Applied Mechanics (accepted for publication).
[167] Vrettos Ch., In-plane vibrations of soil deposits with variable shear modulus: 1.
Surface waves, International Journal for Numerical and Analytical Methods in Geomechanics V.14: 209-222, 1990.
[168] Winkler E., Die Lehre von der Elastizitt und Festigkeit, Prague, Verlag H. Dominikus, 1867.
[169] Wolfert A.R.M., Metrikine A.V. and Dieterman H.A., Wave radiation in a onedimensional system due to a non-uniformly moving constant load, Wave Motion
V.24(2): 185-196, 1996.
[170] Wolfert A.R.M., Metrikine A.V. and Dieterman H.A., Passing through the 'Elastic Wave Barrier' by a Load Moving along a Waveguide, Journal of Sound and Vibration V.203(4): 597-606, 1997.

167

[171] Wu TX, Thompson DJ, The effect of local preload on the foundation stiffness
and vertical vibration of railway track, Journal of Sound and Vibration V.219 (5): 881904, February 4, 1999.
[172] Wu TX, Thompson DJ, A double Timoshenko beam model for vertical vibration
analysis of railway track at high frequencies, Journal of Sound and Vibration
V.224(2): 329-348, July 8, 1999.
[173] Wu TX, Thompson DJ, Analysis of lateral vibration behaviour of railway track
at high frequencies using a continuously supported multiple beam model, J. Acoust.
Soc. AM V.106(3): 1369-1376, Part 1, September, 1999.
[174] Wu TX, Thompson DJ, The influence of random sleeper spacing and ballast
stiffness on the vibration behaviour of railway track, Acustica V.86(2): 313-321,
March-April, 2000.
[175] Wu TX, Thompson DJ, The vibration behavior of railway track at high frequencies under multiple preloads and wheel interactions, J. Acoust. Soc. AM V.108(3):
1046-1053, Part 1, September, 2000.
[176] Wu TX, Thompson DJ, Application of multiple-beam model for lateral vibration
analysis of a discretely supported rail at high frequencies, J. Acoust. Soc. AM V.108(3):
1341-1344, Part 1, September, 2000.
[177] Wu TX, Thompson DJ, Vibration analysis of railway track with multiple wheels
on the rail, Journal of Sound and Vibration V.239(1): 69-97, January 4, 2001.
[178] Wu YS, Yang YB, Yau JD, Three-dimensional analysis of train-rail-bridge interaction problems, Vehicle System Dynamics V.36(1): 1-35, July, 2001.
[179] Zhai WM, Cai CB, Wang QC, Wu TX, Dynamic effects of vehicles on track in
the case of raising train speeds, Proceedings of the Institution of Mechanical Engineers. Part F-Journal of Rail and Rapid Transit V.215 (2): 125-135, 2001.
[180] Zhang JJ, Knothe K, Statistical linearization of wheel rail contact nonlinearities
for investigation of curving behaviour with random track irregularities, Vehicle System
Dynamics V.25: 731-745, Suppl. S., 1996.
[181] Zheng D.Y., Au F.T.K., Y.K. Cheung, Vibration of vehicle on compressed rail
on viscoelastic foundation, J. Eng. Mech. - ASCE V.126(11): 1141-1147, November
2000.

168

SUMMARY
Three-dimensional Dynamic Models of a Railway Track for
High-speed Trains

Modern means of railway transportation in Western Europe are being currently


developed to further reduce the travelling time for passengers. Cruise velocities of such
high-speed trains as French TGV, German ICE, Swedish X-2000, etc. are nowadays in
the range of 200-300 km/h and increase continuously. This velocity increase brought a
new problem to railway engineering, namely the problem of significant amplification of
the train-track vibrations at high train speeds. This amplification occurs when a train
moves with a velocity that is close to the Rayleigh wave velocity in the subsoil of a
railway track, which varies from 150 to 800 km/h, depending on the soil type. Obviously, modern high-speed trains can easily reach the lower threshold.
The amplification of the train and track vibrations at high-speed is a threatening
phenomenon, which leads to a rapid deterioration of the track structure and might cause
derailment of the train. To prevent this amplification from happening, a theoretical
study of train-track dynamics at high speeds is needed that would enable to design the
high-speed railway track properly. In order to accomplish this, dynamic threedimensional models that account for the track-subsoil interaction are to be developed.
The main objective of this thesis is to study the steady-state dynamic response of
various three-dimensional models for railway track to uniformly moving loads of a constant magnitude. The practical goal behind this objective is investigation of the effect of
physical parameters of the track on its dynamics that would allow for choosing these
parameters so that this will preclude from large increase of the track vibrations at high
trains velocities. In particular, this thesis aims to:

Study the effect of material damping in the ground, in the pads and in the ballast;
Analyse the influence of physical properties and stratification of the track subsoil;
Study elastic drag that is a measure of energy loss of a high-speed train on excitation of waves in the track;
Develop a 3D model for railway track, which would account for inhomogeneity
of the track along its length (due to sleepers), and study dynamic response of
such a model to a set of constant, uniformly moving loads;
Study possibility to detect derailment of a wagon of a long freight train by
means of measurement of the spectrum of rail vibration under locomotives
wheels.

The thesis is structured using the principle: to move from simple to more sophisticated. First, the dynamic response of an elastically supported membrane and a
visco-elastic half-space to a constant, uniformly moving load is considered to demonstrate specific features of the responses of basic two- and three-dimensional systems and
to study the effect of the material damping on the response and on the elastic drag. It is
shown that for both models, under a supercritically moving load, a so-called Mach an-

169

gle arises on the surface, at the edges of which the displacement field varies rapidly
likewise at the Mach cone that forms behind a supersonically moving plane. The elastic
drag increases rapidly as the load velocity approaches the critical velocity. The larger is
the material damping, the higher is the elastic drag in the subcritical case and the
smaller it is in the supercritical one.
As a second step, a simplest 3D model for the railway track, namely an EulerBernoulli beam overlying a visco-elastic half-space, is considered. This model, although
disregarding many properties of real railway tracks, allows a) to expose a crucial role of
the Rayleigh waves in the ground on the train-track dynamics; b) to study the effect of
shear stresses in the interface between the ground and the overlying track structure, and
to estimate the elastic drag experienced by a high-speed train. On the hand of this model
it is demonstrated that the critical velocity of the train conditioned by its dead weight
(constant load) is approximately equal to the Rayleigh wave speed in the ground surrounding the railway track. If the track subsoil is soft, the value of the critical velocity
can be attainable by nowadays-operating high-speed trains. Further, it is shown that the
elastic drag experienced by French TGV can be comparable and, starting from the velocities close to the critical velocity, can exceed the aerodynamic drag. This implies that
the elastic drag is an important source of the energy loss for high-speed trains.
The third step towards development of 3D models of a railway track is to account for stratification of the ground and for visco-elastic and inertial properties of the
rail-supporting structure that overlies the ground. Implementation of these improvements makes the model capable of describing the low-frequency dynamics of a conventional railway track and to be employed for studying dynamics of modern embedded
tracks. It is shown that the soil stratification influences the dynamic response of the
railway track strongly. Both the depth of the soil layers and their physical properties are
of importance. Their influence contains in 1) shifting and (sometimes) introducing new
critical velocities of the train; 2) reducing or increasing the dynamic amplification of the
track response; 3) scaling the spatial pattern of the rail response.
To model a conventional track accurately, one has to account for discretely positioned sleepers, i.e. the inhomogeneity of the railway track over its length is to be considered. This is the next step of this development. To accomplish it, a model consisting
of two beams on periodically spaced supports that are mounted to a visco-elastic 3D
layer is considered. The Euler-Bernoulli beams, supports and layer model the rails,
sleepers and subsoil, respectively. It is shown that for constant loading all results obtained for the homogeneous model of the railway track are valid for the inhomogeneous
one. There exists both qualitative and quantitative resemblance of the results. This allows to conclude that in engineering practice one can use the homogeneous model,
which can be studied much quicker than the inhomogeneous one.
The last aim of the thesis is concerned with theoretical and experimental investigation of locomotive detection of a derailment of a wagon of a long freight train. Although solution of this problem cannot be included in the general conception of development of 3D models (since quite slow motion of freight trains allows to consider 1D
model for the railway track), the locomotive detection of a derailment is a very important problem for companies that exploit a large number of wagons. For theoretical
modelling of the railway track, a 1D Timoshenko beam on discrete, equidistantly
spaced supports is used. The steady-state response of the beam to a load that either
uniformly moves along the beam (normal motion) or uniformly jumps over the
supports (the action of a derailed wheelset) is investigated in a point, which moves at a
fixed distance from the load. It is shown that the effect of the derailed wheelset contains
in amplification of the rails response in the frequency band that surrounds the natural
frequency of the sleepers vibrations. This theoretical prediction is confirmed
170

the sleepers vibrations. This theoretical prediction is confirmed experimentally. A


method is proposed of processing the measured data that allows for stable detection of
the derailment. A maximum distance between the locomotive and the derailed wagon is
estimated, at which the detection is possible.

Alexei Vostroukhov
Delft, July 2002

171

SAMENVATTING

172

173

CURRICULUM VITAE
April 23, 1975

Born in Pavlovo, Gorky region, USSR.

September 1982 June 1992

Secondary school 1, Pavlovo, Gorky region,


USSR. Distinction: silver medal.

September 1992 June 1998

Student at Radiophysical Faculty, State University


of Nizhny Novgorod, Russia. Degree: Master of
Science (Physics of Radiowaves).

January 1995 June 1998

Specialization at Chair of Electrodynamics, Radiophysical Faculty, State University of Nizhny


Novgorod, Russia.

July 1998 June 2001

Ph.D. student at Radiophysical Faculty, Chair of


Applied Mathematics, State University of Nizhny
Novgorod, Russia.

March 2001 March 2002

Construction Mechanics Group, Faculty of Civil


Engineering and Geosciences, Delft University of
Technology, The Netherlands.

December 13, 2001

Defence of Candidate Thesis in Russia.


Specialization dynamics, durability of
machinery, devices and apparatus. Degree:
Candidate of Science (Physics and Mathematics).

April 2002 present

Junior Researcher of Road and Railway Engineering Group, Faculty of Civil Engineering and Geosciences, Delft University of Technology, The
Netherlands.

174

S-ar putea să vă placă și