Rail Vibrations A4

Higher Vibration Modes in Railway
Tracks at their Cutoff Frequencies

Markus R. Pfaffinger
Diss. ETH No. 13755

Accepted on the recommendation of
Prof. Dr. J. Dual
Prof. Dr. P. Hagedorn
Diss. ETH No. 13755
Higher Vibration Modes in Railway

Tracks at their Cutoff Frequencies
A thesis submitted to the

SWISS FEDERAL INSTITUTE OF TECHNOLOGY
for the degree of
Doctor of Technical Sciences
presented by
Dipl. Masch.-Ing. ETH
born July 29, 1970
citizen of Maur, ZH
and Rhode Island, USA
Accepted on the recommendation of
Prof. Dr. J. Dual, examiner
Prof. Dr. P. Hagedorn, coexaminer
Zrich, 2000
Acknowledgements
This work was carried out during my employment as an assistant at the Institute of Mechanical Systems, ETH Zrich. I would like to thank all people and
organizations who contributed in one form or another to this thesis, in particular:
Prof. Dr. J. Dual, my supervisor, for his kind support during the course of this
work. I enjoyed working together with Prof. Dual not only while undertaking
the work described in this thesis but also on various other occasions.
Prof. Dr. P. Hagedorn, my coexaminer, for accepting to review my thesis and
for making it possible to keep the ambitious schedule.
Prof. Dr. M. B. Sayir, the chairman of the institute, who managed to create a
liberal and friendly atmosphere among the institute members. The academic
freedom Prof. Sayir encourages provides an ideal environment for research.
Stephan Kaufmann for providing excellent computational services and for giving me the opportunity to participate in planning and administrating the computer network of the institute.
Tobias Leutenegger, Adrian Hof and Manuel Sprri, for writing their semester
or diploma thesis with great enthusiasm within the scope of this project. They
all made valuable contributions to this thesis.
Schweizer Electronic AG (Zofingen, Switzerland) for supporting the work
described in this thesis, especially Mr. P. Schweizer, Peter Maurer, Stefan Zimmermann and Daniel Habermacher.
The Commission for Technology and Innovation KTI (Switzerland) for
approval and funding of this project.
Angelika Schmidt for her help regarding the linguistics of this work.
Markus Pfaffinger
Zrich, July 2000
Table of contents
Table of contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
Zusammenfassung. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
List of symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1
Overview of research on dynamic railway track modeling . . . . . . . .5
1.2
Survey of dynamic track models . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6
1.3
Scope and outline of the present work. . . . . . . . . . . . . . . . . . . . . . . . .7
Theoretical considerations . . . . . . . . . . . . . . . . . . . . .11
2.1
Single mode behavior; beam on a viscoelastic foundation. . . . . . . .12

2.1.1 Forced vibration solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .12
2.1.2 Characterization of the near cutoff frequency behavior. . . . . . . .14
2.1.2.1 QBW-factor definition based on magnitude and phase behavior15
2.1.2.2 QBW-factor definition based on free vibration. . . . . . . . . . . . . .17
2.1.2.3 Study of the system energy . . . . . . . . . . . . . . . . . . . . . . . . . . . .19
2.1.2.3.1 Calculation of the energy components . . . . . . . . . . . . . . . . . .19
2.1.2.3.2 Discussion of the energy components . . . . . . . . . . . . . . . . . .21
a
2.2
Extension to multiple modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2.2.1 Modal solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.2 Multiple mode model on discrete supports . . . . . . . . . . . . . . . . 28
2.2.3 Discussion of the multiple mode model. . . . . . . . . . . . . . . . . . . 31
2.2.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2.3.2 Frequency spectra and receptances . . . . . . . . . . . . . . . . . . . . . 33
2.2.3.3 Beam displacements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.2.3.4 Damping and spatial decay . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.3
Wheel interaction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.3.1 Model extension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.3.2 Discussion of the wheel influence . . . . . . . . . . . . . . . . . . . . . . . 51
2.4
Generalization to the actual cross-section of the rail . . . . . . . . . . . 55

2.4.1 Wave propagation in elastic rods of arbitrary cross-section. . . . 55
2.4.2 Application to the rail. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.5
Conclusions from theoretical considerations. . . . . . . . . . . . . . . . . . 61
Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.1
Experimental setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.2
Frequency spectra and mode shapes of the rail . . . . . . . . . . . . . . . 65

3.2.1 Presentation of the results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.2.2 Discussion of the results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.3
Examination of the spatial decay . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.4
Modal damping of the rail. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

3.5
Influence of passing train wheels . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

ii
Table of contents
Conclusions and outlook . . . . . . . . . . . . . . . . . . . . . . 87
4.1
Final discussion of the results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .88
4.2
Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .91
Appendix A: Q-factor definitions of a single degree of freedom mechanical oscillator. . . . . . . . . . . . . . . . . . . . . 95

A.1 Q-factor definition based on magnitude and phase behavior . . . . .95
A.2 Q-factor definition based on free vibration . . . . . . . . . . . . . . . . . . . .96
A.3 Relation between Q-factor and system energy . . . . . . . . . . . . . . . . .97
Appendix B: Experimental modeshapes . . . . . . . . . . . . 100

Appendix C: Spatial decay . . . . . . . . . . . . . . . . . . . . . . . 103
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
Curriculum vitae . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .111
iii
iv
Abstract
This thesis examines forced vibrations of railway tracks excited at the cutoff
frequency of one of its wave modes. The system response of higher modes was
studied theoretically, numerically and experimentally in the frequency range
from 5 kHz to 50 kHz. The background of this work is the question, whether
the vibration field close to the excitation could be used as a detection principle
for passing train wheels. A feasibility study preceding this work showed a
strong interaction between passing wheels and forced rail vibrations but also
raised a number of questions regarding the reliability of a sensor based on this
idea.
The rail can be considered as a discretely supported, quasi-infinite waveguide.
In the first part of this work containing theoretical considerations, a general
study on the forced system response of a waveguide including higher modes
and modal damping was performed. This investigation was done with the help
of a model consisting of several elastically connected beams with modal
damping. It was shown, that the amplitude of the response reaches maxima at
the cutoff frequencies. The cutoff frequencies and mode shapes correspond to
the solutions of an eigenvalue problem associated with the cross-section of the
waveguide. This knowledge allowed the numerical calculation of the modes of
the actual rail, at which the experiments were carried out. The multiple mode
model revealed, that the cutoff frequencies and mode shapes of higher modes
are hardly influenced by support parameters. Although the experimentally
determined system response exhibits a high level of complexity, these findings
permitted a specific search for the modes of the test track at the numerically
predicted cutoff frequencies and made it possible to visualize the corresponding mode shapes of the real rail.
With the help of the multiple mode model actuated at one of its cutoff frequencies, it was shown that the system deformation is dominated by the corresponding mode shape close to the excitation. System damping and frequency
control the size of this so-called local vibration zone. At greater distances from
the excitation, lower wave modes, propagating with small attenuation rates,
become increasingly important to the system response. These characteristics
are also demonstrated within the experimental investigation.
Additional amplitude maxima originating from the discrete support are not
considered to be suitable for wheel detection because of their lack of a local
vibration zone and are therefore only dealt with in passing.
System damping was shown to be a key parameter regarding the spatial extent
of the local vibration zone, the amplitude and the time constant of the system
response. An experimental concept commonly used to characterize system
damping is the so-called quality factor or Q-factor. Using the model of a Ber-
a
noulli-Euler beam on a viscoelastic foundation, it was, however, pointed out,
that there are some important differences to be considered for a system which
both resonates and interacts with traveling waves. A modification of the Qfactor definition is proposed for this case. Experiments were carried out at the
test track in order to determine the quality factors of a number of modes. The
results obtained from different measurement methods showed a moderate correspondence, which was partially attributed to the complexity of the studied
system. For this reason, the achieved accuracy of the quality factor values
seemed reasonable.
The influence of passing train wheels was studied theoretically with the help
of the multiple mode model. It was shown, that the observed effect depends on
the position of the wheel contact point and of the receiving transducer within
the rail cross-section. It must be avoided, that neither is located at a node of the
cross-sectional vibration. Since the position of the wheel contact on the running surface is undefined, a vibration mode without nodes at the rail head must
be chosen for wheel detection.
A movable bogie was used to experimentally investigate the influence of passing train wheels on forced rail vibrations. Although these measurements in
general confirmed the previous statements, the overall reaction appeared to be
insufficient to be analyzed for single train wheels by an electronic device. The
moderate interaction between the vibration and the wheel was explained by the
low axle weight of the bogie used. Although the unloaded bogie can be considered as the worst case regarding axle load, even here an unconditional axle
detection must be demanded by a sensor operating within a safety relevant
system.
Based on the acquired knowledge, the studied detection principle cannot be
recommended for the use within a general-purpose wheel sensor. The required
safety standard cannot be achieved because of the insufficient reaction time of
the mechanical system and the uncertainty whether the excited vibration is
suitable for wheel detection. However, the concluding chapter proposes some
more specific applications of forced vibrations of waveguides and some ideas
for future research on this topic are outlined.
vi
Zusammenfassung
In dieser Arbeit werden erzwungene Schwingungen von Eisenbahnschienen
erforscht, welche bei ihren Cutoff Frequenzen angeregt werden. Die
Systemantwort von hheren Modes im Frequenzbereich von 5 kHz bis 50 kHz
wurde theoretisch, numerisch und experimentell untersucht. Insbesondere soll
die Frage beantwortet werden, ob das Schwingungsfeld nahe der Anregung
zur Detektion von Eisenbahnrdern gentzt werden knnte. Eine zuvor durchgefhrte Machbarkeitsstudie zeigte, dass erzwungene Schienenschwingungen
durch darberfahrende Zugsrder beeinflusst werden. Es wurden allerdings
auch Fragen bezglich der Zuverlssigkeit eines auf diesem Prinzip beruhenden Sensors aufgeworfen.
Die Schiene kann als ein diskret gelagerter, unendlich langer Wellenleiter
betrachtet werden. In einem ersten Teil dieser Arbeit, welcher theoretische
berlegungen enthlt, wurde eine grundstzliche Studie ber erzwungene
Schwingungen in Wellenleitern durchgefhrt. Diese Untersuchung wurde an
einem Modell bestehend aus mehreren elastisch verbundenen Balken vorgenommen, womit hhere Modes und modale Dmpfungen mit eingeschlossen
werden konnten. Es wurde gezeigt, dass die Amplitude der Systemantwort bei
den Cutoff Frequenzen Maxima erreicht. Die Cutoff Frequenzen und die dazugehrenden Schwingungsformen entsprechen den Lsungen eines Eigenwertproblems des Wellenleiterquerschnitts. Dieses Wissen erlaubte die numerische
Berechnung der Schwingungsformen der tatschlichen Schiene, an welcher
die Experimente durchgefhrt wurden. Untersuchungen am Multiple Mode
Model zeigten, dass die Cutoff Frequenzen und Schwingungsformen von
hheren Modes kaum durch Parameter der Lagerung beeinflusst werden.
Obwohl die experimentell bestimmte Systemantwort einen hohen Grad an
Komplexitt aufweist, ermglichte diese Erkenntnis die gezielte Suche nach
bestimmten Modes des Testgleises in der Nhe der numerisch vorhergesagten
Frequenzen und erlaubte damit die Visualisierung der entsprechenden Schwingungsformen der realen Schiene.
Anhand des Multiple Mode Models, welches bei einer seiner Cutoff Frequenzen angeregt wurde, konnte gezeigt werden, dass die Systemantwort nahe der
Anregung von der entsprechenden Schwingungsform dominiert wird. Die
Systemdmpfung und die Frequenz bestimmen die Grsse dieser sogenannten
lokalen Schwingungszone. Mit wachsendem Abstand von der Anregung
bekommen tiefere Wellenmodes, welche sich mit kleinen Abklingraten ausbreiten knnen, einen zunehmenden Einfluss auf die Systemantwort. Diese
Eigenschaften konnten auch bei den Experimenten beobachtet werden.
Zustzliche Amplitudenmaxima, welche aufgrund der diskreten Lagerung entstehen, werden wegen des Fehlens einer lokalen Schwingungszone bezglich
Achsdetektion als ungeeignet angesehen und werden deshalb nur am Rande
vii
a
behandelt.
Die Systemdmpfung wurde als einer der Hauptparameter identifiziert, welcher die rtliche Ausdehnung der lokalen Schwingungszone, die Amplitude
und die Zeitkonstante der Systemantwort beeinflusst. Eine gebruchliche
Messmethode zur Bestimmung der Dmpfungen ist der sogenannte Qualittsfaktor oder Q-Faktor. Anhand des Modells eines viskoelastisch gebetteten
Bernoulli-Euler Balkens konnte allerdings gezeigt werden, dass bei einem
resonanten System, welches Wellenausbreitung mit einschliesst, einige signifikante Unterschiede bercksichtigt werden mssen. Eine Anpassung der
Q-Faktor Definition wird fr diesen Fall vorgeschlagen. Die Qualittsfaktoren
von mehreren Schienenmodes wurden experimentell bestimmt. Die mit verschiedenen Messmethoden erhaltenen Resultate zeigten eine mssige bereinstimmung, was teilweise auf die hohe Komplexitt des untersuchten Systems
zurckgefhrt wurde. Aus diesem Grund erscheint die erreichte Genauigkeit
fr die Qualittsfaktoren vernnftig.
Der Einfluss eines Zugsrades wurde am Multiple Mode Model theoretisch diskutiert. Es wurde gezeigt, dass der beobachtete Effekt davon abhngt, an welcher Position auf dem Schienenprofil sich der Kontaktpunkt des Rades und der
Empfnger befinden. Keiner von beiden sollte auf einem Knoten der Querschnittsschwingung zu liegen kommen. Da die genaue Lage des Radkontaktes
auf der Laufflche undefiniert ist, mssen fr die Raddetektion Modes gewhlt
werden, welche keine Schwingungsknoten am Schienenkopf aufweisen.
Ein motorisiertes Drehgestell wurde verwendet, um den Einfluss von Zugsrdern auf erzwungene Schienenschwingungen experimentell zu untersuchen.
Diese Messungen besttigten im Wesentlichen die oben gemachten Aussagen.
Allerdings musste festgestellt werden, dass der beobachtete Effekt zu gering
war, um es einer Auswertungselektronik zu ermglichen, einzelne Zugsrder
zu erkennen. Die wenig ausgeprgte Wechselwirkung zwischen der Schwingung und den Rdern wurde durch die geringe Achslast des verwendeten
Drehgestells erklrt. Obschon ein einzelnes Drehgestell die ungnstigste Konfiguration bezglich Achslast darstellt, muss auch in diesem Fall eine einwandfreie Achsdetektion von einem Sensor gefordert werden, welcher
innerhalb eines sicherheitsrelevanten Systems eingesetzt werden soll.
Aufgrund der gewonnenen Erkenntnisse kann das untersuchte Verfahren nicht
zur Achsdetektion im Rahmen eines Allzwecksensors empfohlen werden. Die
geforderte Zuverlssigkeit kann nicht erreicht werden, da die Reaktionszeit
des mechanischen Systems ungengend ist und weil nicht garantiert werden
kann, dass die angeregte Schienenschwingung fr die Raddetektion geeignet
ist. Im abschliessenden Kapitel werden allerdings einige spezifischere Anwendungsmglichkeiten von erzwungenen Schwingungen in Wellenleitern vorgeschlagen und es wird ein Ausblick auf mgliche knftige Forschung auf
diesem Gebiet gegeben.
viii
List of symbols
A
B
C
D
M
S
T
V
W
cross-sectional area of beam

beam flexural stiffness coefficient matrix in multiple mode model
transfer matrix of unsupported part of the discretely supported model
transfer matrix of supported part of the discretely supported model
modal decomposition matrices without damping in multiple mode
model
matrix of modal displacement vectors of the multiple mode model
flexural beam stiffness
temporary matrix in order to calculate H
matrix, describing the radiation condition of the discretely supported
model
stiffness coefficient matrix of the elastic connections of the multiple
mode model
mass matrix of the multiple mode model
scaling matrix
transfer matrix of basic section of the discretely supported model
eigenvector of T
wheel influence matrix
a, b
cc, c
d
e
f
fc
g
i
k
constants
viscous damping constants
arbitrary length of a rigid body
Eulers constant
frequency
cutoff frequency
complex number
1
vector of wavenumbers of the multiple mode model, wavenumber of
Ew ,
Ek
Ek
EI
G
H
K
ix
a
rod of arbitrary cross-section
vector of wavenumbers without damping of the multiple mode model
k
m
mass of single degree of freedom mechanical oscillator
n
number of beams of the multiple mode model
p
amplitude of excitation force, vector of excitation forces at z = 0 of the
multiple mode model
q
eigenvalues of T
r
count of number of sections of the discretely supported model beginning at the excitation
s
spring constant of the single degree of freedom mechanical oscillator
support stiffness
sc
t
time variable
u
downward beam displacement, displacement vector of rod of arbitrary
cross-section, dimensionless downward displacement of single degree
of freedom mechanical oscillator in Appendix A
v
state vector at the section beginning of the discretely supported model
w
dimensionless beam displacement
x, y, z coordinates
a
b
g
d
z
h
k
l
x
dimensionless wavenumber
direct receptance
space scaling factor
dimensionless viscous damping constant
dimensionless space variable
hysteretic damping constant
dimensionless frequency
wavelength
vector of displacement amplitudes of the multiple mode model
x,
vectors of displacement amplitudes without damping of the multiple
xk
mode model
r
mass per unit volume
t
dimensionless time variable
c
wheel impedance
Yi(z) modal wave components matrices
y
modal coefficient vector
w
angular frequency
wi, wc angular natural frequency, angular cutoff frequency
(),i
derivative with respect to i
*
()
complex conjugate of complex quantity
( )
real part of a complex quantity
Numbers in [ ] brackets refer to the bibliography
Chapter 1
Introduction
A strong increase in railroad traffic has been observed during the last years,
regarding both passenger and cargo transportation. As the utilization of the
railway tracks rises, there is a strong demand in automated solutions to ensure
the safety of railway lines. A smooth and safe operation of railway tracks
becomes possible by dividing it into block sections. Only one train is permitted in a block section at one time. Signal boxes are used to control signals and
switches. In addition to this permanently installed infrastructure there is a need
for mobile warning systems to be employed at railway construction sites to
guarantee the safety of working staff and to ensure unhindered railway traffic.
Both permanent and mobile systems depend on reliable sensors, which indicate the presence of trains. The design concepts of these sensors differ depending on the field of application. This work focuses on the demands of mobile
systems. Devices for detecting the presence of railroad cars are called rail contacts.
Figure 1.1 shows a diagram of a railway construction site equipped with an
automatic crew warning system. Two rail contacts A and B are mounted in
some distance on both sides of the working area to detect approaching trains.
Commonly used rail contacts are mechanical switches, which are actuated by
the wheel rim. To reduce the wear, mechanical rail contacts are built in a
robust way and a damping element prevents the sensor from detecting single
wheels. This fact may lead to dangerous situations, as Figures 1.1. a) to d)
illustrate in the case, where train 1 passes the construction site and actuates
contact B, but comes to a stop above the rail contact. If a second train 2 enters
the construction area and if train 1 continues its journey in this constellation,
the warning system assumes that the track is free.
a)
train 1
wheel rim
mechanical
rail contact
train 1
b)
c)
d)
train 2
train 1
B
train 2
train 1
B
A
Figure 1.1: Diagram of a railway construction site equipped with a warning
system using the mechanical rail contacts A and B
The discussed danger situation could be avoided using rail contacts with the
ability to detect single train axles. An ideal rail contact for the use within a
mobile crew warning system should meet the following requirements:
- reliable detection of single axles independent of train speed
- insensitivity to environmental interferences including moisture, temperature variation and magnetic fields
- light weight construction
- easy mounting without any modifications of the rail and without
being an obstacle for special-purpose vehicles like snowblowers
- low energy consumption
- self-check during operation to detect malfunction
Besides the already mentioned mechanical contacts, todays most often used
rail contacts are inductive sensors or optical switches. None of these devices
meet all of the above requirements, thus leaving a strong industrial need for
improvements.
Introduction
Preceding this work, a feasibility study had been carried out to look at the suitability of the use of mechanical waves in the rail as a detection principle for
train wheels. Various tests have been performed with ultrasonic transducers
mounted on the rail operating in pulse/echo configuration. In these experiments, almost no influence of train wheels on the ultrasonic echo had been
observed. This can be explained by the fact that the contact surface between
the rail and the wheel is small as compared to the area from which the ultrasonic pulse is reflected. Since the exact location of the contact surface between
the wheel and the rail is undefined, classical pulse/echo ultrasonics did not
seem suitable for wheel detection.
In a different approach, a transducer mounted beneath the rail was used to harmonically excite a vibration in the rail, as illustrated in Figure 1.2. A second
transducer was used as a receiver to measure the magnitude and the phase of
the forced rail vibration. The examined frequency range was chosen beyond
the limit of human hearing at about 20 kHz. In this frequency range, the crosssection of the rail has a number of transverse resonance frequencies if subjected to plane strain. The experiments showed that the magnitude of the
vibration strongly increases at certain frequencies. If the rail is excited at one
of these frequencies, it had been observed, that passing train wheels influence
the magnitude and the phase of the vibration.
cross-section A-A
higher vibration mode
wheel
pad
rail
receiving transducer
ballast
sleeper
sending transducer
Figure 1.2: Diagram of the experimental setup
Amplitude of receiver signal (-)
Figure 1.3 shows the change of the receiver magnitude caused by a passing
wheel of a partially loaded freight car (two axles, approximate weight: 160 kN)
as measured within the feasibility study.
1
0.8
0.6
0.4
0.2
-2
-1
Distance of wheel from excitation (m)
-2
-1
Distance of wheel from excitation (m)
Figure 1.3: Wheel influence on magnitude of receiver signal measured within

the feasibility study at 42.70 kHz for both passing directions of the wheel
(Figure by courtesy of Schweizer Electronic AG)
The observed influence led to the idea to use forced vibrations in railway
tracks as a sensing technique in order to detect passing train wheels. A sensor
based on this principle would be likely to meet the above requirements, as the
interaction between the vibration and the train wheel is purely mechanical and
is expected to be insensitive to environmental and electrical interferences.
Since the proposed rail contact would consist of an active system with a sending and a receiving transducer, it would be possible to implement self-checking features, such as an automated testing of the sensor mounting.
However, the results of the feasibility study showed that the effect caused by
passing train wheels is more or less pronounced at different amplitude maxima
of the rail. It is the objective of this work to give an insight into effects of
forced rail vibrations and their interaction with train wheels in order to discuss
the suitability of these vibrations for axle detection. This study contains theoretical considerations and an experimental investigation. To give a background
to the solution concept pursued in the theoretical part of this work, the main
fields of research on dynamic track modeling are reviewed on the basis of
selected literature in the following section.
Introduction
1.1 Overview of research on dynamic railway track

modeling
Research in modeling the dynamic behavior of vehicle and track is mostly
stimulated by the need to understand the cause of practical problems arising
from dynamic loads and to develop solutions to those problems.
With increasing speed of trains the question of dynamic stability of a load
moving along a beam was risen. In this context, the word stability refers to the
situation where the amplitude of a steady state solution becomes infinite at a
certain critical speed. Frybas book [7] contains a survey of work on the moving load problem. However, the critical speed is found to be approximately
400 - 500 m/s with parameters of a railway track (see e.g. [16]). This is very
much greater than the speed of present trains. Thus, the moving load problem
is primarily of academic interest regarding track modeling.
Other problems of greater practical relevance have been studied using mathematical and numerical models of railway tracks. The main areas of concern are
introduced in the following sections. The frequency range of interest for the
different problems is also given.
The dynamics of the railway track becomes increasingly important at frequencies above approximately 20 Hz. Below this frequency limit, the track behaves
essentially as a relatively stiff spring, while the dynamic behavior of the vehicle itself is important with regard to curving, stability and passenger comfort.
Dynamic loads with frequencies up to about 500 Hz are connected to various
problems involving components of the bogie such as wheel bearings. Work in
this field has been published by Ahlbeck et al. [1] and by Lundn [18].
Studies concerned about the deterioration of the running surfaces of wheel and
rail have been a major stimulus for the development of reliable models of the
interaction between vehicle and track at frequencies up to about 1500 Hz. A lot
of theoretical and experimental work has been performed in this area, as
dynamic loads in the same frequency range are often held responsible for the
deterioration of track components including pads, sleepers and ballast as well
as of the rail itself. Newton and Clark [22] examined the effects of wheelflats
on the track. The corrugation of the railway track has been studied by multiple
authors including Hempelmann [15] and Tassilly [30]. The phenomenon of
rail corrugation, both of short wavelength (30 - 100 mm) and long wavelength
(300 - 1500 mm), is commonly explained using models in this frequency
range.
Much of the research in modeling high frequency vibration of the wheel and
rail up to about 5 kHz arises from concerns about noise. Irregularities of the
wheel and rail cause rolling and impact noise, while stick-slip, which primarily
occurs in tight curves, leads to wheel squeal. Research in the area of wheel-rail
a
noise generation has been performed by Remington e.g. [23], Thompson
[33,34,35,36,37] and Gry [11] amongst others.
No previously published work is found in which it is suggested to take advantage of rail resonances for a technical application as in the proposed wheel
detection concept.
In the following section the dynamic track models found in the literature are
briefly discussed.
1.2
Survey of dynamic track models
The history of dynamic modeling of railway tracks and of the interaction of

vehicle and track dates back to 1926, when Timoshenko [31] published a paper
in which he examined the effects of wheelflats. Timoshenko modeled the track
as a continuously supported Bernoulli-Euler beam excited by a harmonically
varying load or a moving irregularity. A separate layer of rigid bodies representing the sleepers was added to this model by Sato [26]. A model in which
the rail was represented as a Timoshenko beam on a continuous layer of rigid
sleepers was developed by Grassie et al. [8], thus taking the effects of shear
and rotary inertia into account. In the same publication, Grassie studies the
dynamic behavior of a Bernoulli-Euler beam supported on rigid bodies representing the sleepers. Grassie calculates the direct receptance for both a continuously and a discretely supported track model. The direct receptance is the
complex ratio of the displacement of the point of excitation to the exciting
force. A comparison shows, that the continuous support model is strictly valid
only for the calculation of the tracks dynamic response at frequencies lower
than about 500 Hz for vertical excitation and 400 Hz for lateral excitation.
However, the effects of discrete supports were found to affect the receptance
only at certain frequencies. This occurs if the semi-wavelength of the bending
wave corresponds to the sleeper spacing and is referred to as the pinnedpinned frequency. Research in various fields of dynamic track behavior using
discretely supported models is performed by Munjal and Heckl e.g. [21],
Bogacz et al. e.g. [3,4] and Heckl [14] amongst others.
At frequencies above about 1500 Hz the cross-section of the rail deforms significantly. Various concepts to take this into account for dynamic track modeling are found in the literature. Ripke and Knothe [24] used two Timoshenko
beams representing the head and the foot of the rail with an elastic coupling in
between. Scholl [28] calculated the dispersion relations of eleven wave modes
up to a frequency of 15 kHz by modeling the head, the web and the foot of the
rail by plates of constant thickness. Thompson [32] modeled a 10 mm long
section of the rail including sleepers, rail pads and ballast as equivalent continuous layers of mass and stiffness with finite elements, then applying periodic
Introduction
structure theory to calculate the response of the infinite system. Strzyzakowski

and Ziemanski [29] introduced an FEM cross-sectional solution together with
an analytical treatment in the longitudinal direction. Knothe et al. [17] compared models of rails as published by Grassie et al. [8,9,10], Ripke and Knothe
[24], Scholl [28] and Strzyzakowski [29] in the frequency range up to 15 kHz
and thus established the validity ranges of these models. The upper frequency
limit of 15 kHz is considered to be sufficient for the practical problems discussed involving rails, as it represents the limit of human hearing.
1.3
Scope and outline of the present work
A rail contact making use of forced vibrations of the rail would have to implement several capabilities in order to meet the requirements listed on page 2.
After the sensor is attached to the rail, it must find an appropriate rail vibration
by itself. Consequently a concept must be developed to reliably distinguish
vibrations showing a strong interaction with train wheels from other amplitude
maxima. During operation, the device must track the corresponding frequency
despite of changes in environmental conditions such as temperature. The stabilization of these frequencies can be achieved by using a phase-locked loop (PLL).
The scope of this work is to study forced rail vibrations and their interaction
with train wheels theoretically, numerically and experimentally. The suitability
of forced rail vibrations regarding safe wheel detection is analyzed with the
help of this investigation. The examination is performed based on the following questions:
- How do the vibration modes corresponding to the amplitude maxima
of the forced vibration response of the rail look like?
- What are the parameters influencing the forced vibration response of
the rail at its amplitude maxima?
- Which rail vibration modes are suited best for wheel detection?
- How can these vibration modes be found and stabilized by an industrial wheel sensor?
- How large is the spatial extent of the local vibration zone near the
excitation?
- How large is the spatial range within which the receiver signal is
influenced by wheels?
- What is the reaction time to be expected from a sensor of this type
and is this time sufficient to detect single wheels of fast traveling
trains?
- Is there a typical reaction of the receiver signal for which the analyzing electronic device must look for?
Signal power (dBm)
The present work is divided into Chapter 2 containing theoretical considerations and Chapter 3 in which the results of an experimental study are presented. A final assessment of the acquired information regarding the use of
forced rail vibrations as a wheel detection principle together with a more general view on the accomplished research is found in Chapter 4.
From the perspective to develop a rail contact based on forced rail vibrations,
the frequency range aimed at is above 20 kHz for various reasons. The cutoff
frequency of the first higher wave mode of a rail is typically found at approximately 1 kHz - 1.5 kHz. A cutoff frequency is the frequency limit below which
a wave mode cannot propagate and above which it can propagate. Often these
frequencies are also called transverse resonance frequencies. This work
focuses on cutoff frequencies.
There are resonances of the railway track below the first higher cutoff frequency originating from the discrete support on sleepers. However, these resonances are not considered to be suitable for wheel detection as will be
discussed within Chapter 2.
The signal-to-noise ratio of the sensor can be improved by using frequencies
above 15 kHz - 20 kHz, since the power spectrum of mechanical vibrations
caused by passing trains decreases towards higher frequencies. Figure 1.4
shows a typical power spectrum caused by a fast passenger train and justifies
this statement.
-10
-20
-30
-40
-50
-60
0
20000
40000
60000
Frequency (Hz)
80000
100000
Figure 1.4: Mean signal power measured during the passing of a

fast passenger train (Measurement performed by P. Maurer,
Figure by courtesy of Schweizer Electronic AG)
The use of frequencies in the audible range below 20 kHz would also cause an
unacceptable noise pollution.
The density of cutoff frequencies of the rail increases towards higher frequencies. As this fact makes it more and more difficult to find specific rail modes,
the upper limit of the frequency range covered within this investigation is
defined to be 50 kHz. Although the operating frequency of the sensor aimed at
Introduction
is above 20 kHz, some lower vibration modes of the rail are included and the
frequency range of this study is set to 5 kHz - 50 kHz.
Throughout this work, the rail is considered as an infinitely long waveguide,
i.e. a structure along which mechanical waves can travel without spreading in
all directions. As will be pointed out in Chapter 2, amplitude maxima of the
forced vibration in a waveguide occur at the cutoff frequencies. The concept to
use forced vibrations of the rail near its cutoff frequencies has a certain similarity to the idea of resonant sensors, which are used in various technical applications.
According to the literature, a high number of parameters are expected to influence the dynamic behavior of a railway track mounted on pads, sleepers and
ballast. The most important factors influencing the cutoff frequencies and
mode shapes of the rail are the geometry and material properties of the crosssection. Mainly three rail types are employed in Switzerland classified as
SBB I, SBB IV and SBB VI with weights per meter of 46 kg, 54 kg and 60 kg.
The most commonly used rail types across Europe are categorized by the
nomenclature UIC 541, UIC 60 and UIC 71. The shape of the rail cross-section additionally varies during its life span due to wear. This issue affects the
rail head in particular.
Besides the geometry and the material of the rail cross-section, parameters of
the supports are also expected to have an influence on rail vibrations. These
parameters include the type of support, e.g. wooden or concrete sleepers, the
geometry of the support, e.g. sleeper spacing and additional influences such as
the compacting of the ballast. System damping affects both the amplitude of
the response and the spatial extent of the local vibration zone near the excitation and is therefore regarded as an important issue to be covered within this
investigation.
Studies on the influence of support parameters on the dynamic track behavior
are found in the literature mainly in the frequency range up to 1 kHz which is
below the first cutoff frequency of the rail. Track models developed for frequencies up to 15 kHz are usually based on the assumption of an infinite continuous structure. The support is either taken into consideration as a
continuously distributed influence or ignored entirely. The effect of system
damping is usually not included in high frequency rail models. None of the
track models found in the literature appears to be suitable to describe the
forced vibration of the rail close to its cutoff frequencies considering all of the
parameters discussed above.
The solution concept pursued in this work is to first get an understanding of
the observed phenomenon by the use of models which can be treated analyti1. UIC = Union Internationale Chemin de Fer, 54 kg weight per meter
a
cally. The knowledge gained from these models in Sections 2.1 to 2.3 gives the
possibility to numerically address the realistic situation in an efficient manner
as will be elaborated in Section 2.4.
A commonly used value to characterize the damping of a system at certain resonance frequencies is the so-called quality factor or Q-factor. For resonant
sensors, the Q-factor is used as a measure regarding the measurement accuracy as well as the mechanical reaction time. Within this work the Q-factor
concept is discussed in the context of the forced vibration of a waveguide at its
cutoff frequencies.
10
Chapter 2
Theoretical considerations
As pointed out in Chapter 1, the frequency range of interest to be covered by
this investigation is chosen from 5 kHz to 50 kHz. This must be regarded as a
high frequency range in the context of railway track modeling. The upper frequency limit of track models found in the current literature is approximately
15 kHz.
The development of a detailed Finite Element model of the examined situation
would entail great expenses both regarding modeling and computational effort
while hiding the respective influences of the model parameters. As it would be
difficult to derive reliable statements about the detection capabilities from
such a model, a different approach is chosen in this work. The near cutoff frequency behavior of waveguides is studied using mechanical models which can
be treated analytically. The characteristics of a single wave mode representation is examined in Section 2.1 using the well known model of a BernoulliEuler beam on a continuous viscoelastic foundation excited by a harmonic
point force.
In Section 2.2 the model is extended to include higher transverse vibration
modes by studying a system of multiple elastically connected beams. This
model is later referred to as multiple mode model. A solution scheme is developed, which allows the implementation of modal damping. Both a continuously and a discretely supported configuration of the model are examined.
The interaction between a passing train wheel and the forced vibration of the
multiple mode model is studied in Section 2.3. Statements regarding the capability of certain modes to interact with passing wheels are derived from these
investigations.
In Section 2.4 the calculation of the cutoff frequencies and mode shapes of an
11
a
actual railway track is performed and discussed. The knowledge gained in
Sections 2.2 and 2.3 from the multiple mode model is then applied to choose
modes suitable for wheel detection.
2.1 Single mode behavior; beam on a viscoelastic foundation

2.1.1
Forced vibration solution
Figure 2.1 shows a purely elastic Bernoulli-Euler beam on a continuous visiwt

coelastic foundation excited by a harmonic point force p ( z = 0, t ) = p e
with angular frequency w.
EI, , A
sc, cc
~
p
Figure 2.1: Bernoulli-Euler beam on a continuous viscoelastic foundation

with harmonic excitation force at z = 0 m
The differential equation of this model in the case of harmonic motion is
2
EIu ,zzzz w r Au + iwc c u + s c ( 1 + i2h )u = 0 ,
(2.1)
where u=u(z,t) is the downward displacement and subscripts after the comma
denote partial differentiation with respect to the designated variable. In equation (2.1) EI, r and A are the flexural stiffness, the mass per unit volume and
the cross-sectional area of the beam, respectively. The support has the stiffness
sc and the viscous damping constant cc per unit length. In addition to viscous
damping, hysteretic damping h is introduced by adding a complex component
i2hsc to the support stiffness.
By defining dimensionless time and space variables t = w 1 t and z = gz ,
equation (2.1) may be put in the dimensionless form
2
w ,zzzz k w + 2ikdw + i2hw + w = 0 ,
(2.2)
3
where w 1 = s c rA , g = 4 s c EI , d = c c 4s c rA , w = ( EIg p )u
and k = w w 1 . Considering steady state solutions of the form
w ( z, t ) = b e
i ( a ( k )z + kt )
(2.3)
equation (2.2) is solved, if the dimensionless wavenumber a ( k ) satisfies the
12
a
condition
4
a ( k ) = k 1 i 2 ( dk + h )
(2.4)
giving four branches of the wavenumber. In the absence of damping, the wavenumber a is zero for k = 1, corresponding to a frequency w = w1. The frequency w1 is called cutoff frequency and corresponds to the resonance
frequency of a rigid body of mass m = rAd supported by a spring of stiffness
s = s c d , where A is the cross-sectional area and d is the arbitrary length of
the body.
In all subsequent work, the notation of the frequency dependency of the wavenumber will be omitted for the benefit of brevity. The complete solution for
the displacement w is
w ( z, t ) = ( b 1 e
i az
+ b2 e
i az
+ b3 e
az
az
+ b 4 e )e
4
i kt
(2.5)
iArg ( a ) 4
. Taking into
where bi are arbitrary constants and a = 4 a e
account the radiation condition for z and the boundary conditions at
z = 0 gives the forced vibration solution (2.6) for z 0 of an infinite beam on
a continuous viscoelastic support.
1 3 i az
az i kt
w ( z, t ) = a ( ie
+ e )e .
4
(2.6)
The complex ratio of the displacement at the point of excitation to the excitation force is called the direct receptance and gives
1 3
bc = a ( i + 1 )
4
(2.7)
in dimensionless notation. Equation (2.7) can be rewritten as

bc =
2
a
4
3p
i + 3Arg ( a )
4
(2.8)
using the absolute value a and the argument Arg ( a ) of the wavenumber. At
the minimum of a the direct receptance reaches its maximum and only system damping prevents the amplitude from increasing to infinity. In the case of
small damping parameters d and h the first approximation of the frequency,
where the direct receptance is maximal, is k = 1, corresponding to w = w1. In
the absence of damping the entire system performs a synchronous motion and
the amplitude tends towards infinity if excited at its cutoff frequency w1.
To illustrate these statements, Figure 2.2 shows the magnitude and phase of
the direct receptance as functions of the dimensionless frequency k for three
13
a
values of hysteretic damping and with d = 0.
20.
= 1/400
|c|
15.
= 1/200
10.
= 1/100
5.
0.
Arg(c)
0.5
= 1/400
1.
= 1/200
1.5
= 1/100
2.
2.5
3.
0.9
0.95
1.
1.05
Dimensionless frequency
1.1
Figure 2.2: Magnitude and phase of direct receptance bc ( d = 0 )

With increasing damping, the magnitude and the slope of the phase decrease at
the cutoff frequency k = 1, as we can see from Figure 2.2. In contrast to the
behavior of a single degree of freedom mechanical oscillator (see
Appendix A), the overall phase change of the system is less than p. The consequence of this characteristic difference regarding damping measurement is discussed in the following section.
2.1.2
Characterization of the near cutoff frequency behavior
The damping of a classical single degree of freedom mechanical oscillator is

commonly characterized by the quality factor or Q-factor. In Appendix A different definitions of the Q-factor are summarized. For small damping, one
method to determine the Q-factor is to calculate
Q =
1
kb ka
(2.9)
where k a and k b are the two dimensionless frequencies, above and below the
resonance frequency k = 1, for which the magnitude of the oscillation has
dropped to 1 2 of its resonance value, respectively. At k a and k b the
phase of the transfer function has shifted p 4 as compared to the phase at
14
the resonance frequency.

The overall phase change of a classical single degree of freedom mechanical
oscillator is p. In the case of the continuously supported beam, the total phase
shift of the receptance is 3p/4 as we can see from equations (2.8) and (2.4).
This obvious difference of a continuous beam as compared to a classical
mechanical oscillator indicates that the concept of the Q-factor cannot be
applied directly to characterize the damping behavior of a system which
includes wave propagation and leads to a more appropriate definition to be
used in the case considered here. This new measurement method is referred to
as QBW-factor, where the index stands for bending wave.
2.1.2.1
QBW-factor definition based on magnitude and phase behavior
At the cutoff frequency k = 1 the magnitude and the phase of the wavenumber
a (see equation (2.4)) are
a(1) =
2 ( d + h ) and Arg ( a ( 1 ) ) =
p
8
(2.10)
giving a magnitude and phase value of the receptance of

bc k = 1
8
4
=
( d + h ) and Arg ( b c )
8
4
k=1
3p 3p
.
+
8
4
(2.11)
Motivated by the single degree of freedom mechanical oscillator discussed in

Appendix A, the frequencies k a and k b are considered which satisfy the conditions
2
k a 1 + 2 ( dk a + h ) = 0 and
(2.12)
k b 1 2 ( dk b + h ) = 0 .
(2.13)
Taking into account that the frequencies must be positive, the solutions of
(2.12) and (2.13) are
2
k a = d + 1 + d 2 h and k b = d + 1 + d + 2h .
(2.14)
In the case of small damping parameters d and h the first order approximations
of the frequencies k a and k b are
k a = 1 d h and k b = 1 + d + h .
(2.15)
Adding and subtracting 2 ( dk a, b + h ) 2 ( dk a, b + h ) to equation (2.4) gives

a ( k a, b ) =
k a, b 1 2 ( dk a, b + h ) 2 ( dk a, b + h ) i2 ( dk a, b + h )
(2.16)
15
a
By using the relations (2.12) and (2.13) the wavenumber at the frequencies k a
and k b can be rewritten as
a ( k a, b ) =
2 ( dk a, b + h ) ( i 1 ) =
2 4 2 ( dk a, b + h ) e
p p
i --- ------
8 16
(2.17)
giving magnitude and phase values of the receptance function of

bc k
a, b
=
8
Arg ( b c )
3
---
8
4
( dk a, b + h ) and
8 8
(2.18)
3p 3p 3p
+
.
8
4
16
(2.19)
k a, b
=
2
By ignoring parts of the order d and dh , we can see from (2.18) and (2.19)
that the magnitude has dropped to 1 8 8 and the phase has shifted 3p 16
at the frequencies k a and k b as compared to the values at the cutoff frequency.
Let us define the ratio between the cutoff frequency and ( k b k a ) as the
QBW-factor of the continuously supported Bernoulli-Euler beam, which is
related to the system damping according to (2.20).
Q BW =
1
1
=
kb ka
2(d + h)
(2.20)
With dimensions the definition of the QBW-factor is

2
Q BW
w1
rAw 1
=
=
wb wa
c c w 1 + 2hs c
(2.21)
These relations between the QBW-factor and the damping parameters of the
continuously supported beam are analogous to the corresponding definitions
of a single degree of freedom mechanical oscillator.
The QBW-factor can also be described using the slope of the phase curve at the
cutoff frequency by taking the derivative of the receptance phase with respect
to k according to equations (2.22) and (2.23).
1d
2 ( dk + h )
d
d
Arg ( b c ) = 3 Arg ( a ) = 3
ArcTan 2
4dk
dk
dk
k 1
d Arg b
( c)
dk
16
=
k=1
3
4(d + h)
(2.22)
(2.23)
This leads to the relation between the slope of the phase curve at k = 1 and the
QBW-factor of
Q
BW
2 d
Arg ( b c )
=
3 dk
(2.24)
k=1
in the dimensionless case and

Q
BW
2w 1 d
Arg ( b c )
3 dw
(2.25)
w = w1
otherwise.
2.1.2.2
QBW-factor definition based on free vibration
It is standard in the field of acoustics, to determine the damping of a system by

measuring the reverberation time of specific vibration modes or in certain frequency bands (see i.e. [5] or [13]). This method is especially useful for systems with low damping, as the decaying envelope of the free vibration system
response following the excitation can be recorded with modest experimental
effort as compared to the high precision in phase measurement necessary in
these cases. The relation between the time constant of the free vibration
response of the continuously supported beam and the QBW-factor is derived in
this chapter.
Consider the steady state solution of a beam on a continuous viscoelastic support driven by an externally applied force at z = 0 as discussed above. If the
excitation force is turned off at the time t = 0, the system performs a damped,
free vibration. The corresponding boundary conditions in the case of small
damping parameters are
w ,z ( z = 0, t ) = 0 and
w ,zzz ( z = 0, t ) = 0 , 0 t.
(2.26)
Substituting the solution (2.5) in (2.26) and taking the radiation condition into
account leads to
i a a
3
ia a
b2
= 0.
(2.27)
b4
Nontrivial solutions for the constants b2 and b4 exist, if the determinant of the
matrix in (2.27) equals to zero, giving
2
k f = id + 1 d + i2h
(2.28)
17
a
for the corresponding frequency kf by taking into account that the real part of
the frequency must be positive. Linearization in the case of small damping
parameters leads to
kf = 1 + i ( d + h ) .
(2.29)
The real part of k f corresponds to the oscillatory part of the system response in
the free vibration case. The imaginary part causes an exponential decay of the
vibration amplitude. Defining the QBW-factor as the ratio between the real and
the imaginary part of k f in the same way as in the case of a single degree of
freedom oscillator gives
Q
BW
Re ( k f )
1
,
=
2Im ( k f )
2(d + h)
(2.30)
which is in agreement with the previous definitions.

The time constant t 1 e , after which the amplitude of the vibration has
decayed to 1 e of its initial value, is
t 1 e = 2Q
BW
(2.31)
in the dimensionless case and

BW
t1 e
2Q
=
w1
(2.32)
otherwise.
As we can see from (2.31) and (2.32), the QBW-factor is related to the time constant t 1 e in exactly the same way as the Q-factor in the case of a single
degree of freedom oscillator and confirms the definitions derived in the previous section.
18
2.1.2.3
Study of the system energy
In the case of a single degree of freedom mechanical oscillator, the Q-factor

can also be related to the system energy (see Appendix A). If we look at the
system of a beam on a viscoelastic support, the total system energy is composed of five components: the kinetic energy, the potential energy stored in the
support, the potential energy stored in the deformed beam, the work, which is
dissipated by the damping and the work of the excitation force.
The goal of this section is to derive closed form solutions for these dimensionless energy components in order to study the energy behavior at frequencies
close to the cutoff frequencies.
2.1.2.3.1 Calculation of the energy components

The physical displacement w ( z, t ) of the beam corresponds to the real part of
the complex solution (2.6) found in Section 2.1.1. Throughout this work, all
real quantities which are a subset of the complex solutions are noted as ( ) .
By using the relation g = ( g + g* ) 2 , where g* is the complex conjugate
of the complex number g , the displacement of the beam on a viscoelastic
foundation can be written as
3
1
3
i az
a z ikt
ia* z
a* z i kt
w ( z, t ) = ( a ( ie
+ e ) e + a* ( i e
+e
)e
)
8
(2.33)
The total kinetic energy E k of the viscoelastic beam is

Ek ( t ) =
1 2
w dz =
2 ,t
w ,t dz .
(2.34)
The value of the integral is zero for z , giving the solution (2.35) for the
kinetic energy E k ( t ) .
k 3 ( 1 + i ) i2kt 3 ( 1 i ) i 2kt
Ek ( t ) =
e
e
+
7
7
128
a
a*
2
(2.35)
3
3
a a* ( a a* ) ( a i a* ) ( a + a* )
2
8 ( 1 i ) ( a i aa* a* )
19
a
The potential energy stored in the foundation stiffness E pf ( t ) and the potential energy stored in the beam deformation E pb ( t ) are calculated in a similar
way, giving the results
t
E pf ( t ) =
ww ,tdt dz =
w dz =
(2.36)
1 3 ( 1 + i ) i2kt 3 ( 1 i ) i 2kt
e
+
e
+
7
128 a 7
*
a
3
3
a a* ( a a* ) ( a i a* ) ( a + a* )
2
2
8 ( 1 i ) ( a i aa* a* )
and
E pb ( t ) =
2
1
w ,zz dz =
2
w ,zz dz =
(2.37)
1 ( 1 + i ) i2kt ( 1 i ) i 2kt
e
+
3
128 a 3
a*
8(1 + i)
.
*
*
*
( a a ) ( a i a ) ( a + a )
The rate, at which energy is dissipated in the foundation damping E d ( t ) is

calculated according to equation (2.38) and gives the result (2.39).
Ed ( t ) =
t
0
2 d w 2 + 2 h w 2 dt dz = 4 d + h
,t
k ,t
k
Ed ( t ) =
w ,t dt dz
(2.38)
0 0
2 ( dk + h ) 3 ( 1 i ) i2kt 3 ( 1 + i ) i 2kt
e
e
+
7
7
128
a
a*
(2.39)
2
2
16 ( 1 i ) ( a i aa* a* )kt
3
3
a a* ( a a* ) ( a i a* ) ( a + a* )
The energy, introduced into the system E e ( t ) by the excitation force gives
Ee ( t ) =
t
0
p ( t ) w ,t( z = 0, t ) dt =
1 ikt i kt
(e + e
) w ,t( z = 0, t ) dt = (2.40)
2
3
3
1 4 ( 1 + i ) i2kt 4 ( 1 i ) i 2kt 8 ( 1 + i ) ( a i a* )kt
e
+
e
.
3
3
3
128 a 3
*
*
a
a a
20
2.1.2.3.2 Discussion of the energy components

To illustrate the energy flow of the viscoelastically supported beam, Figure 2.3
shows the five calculated energy components over one vibration cycle at the
frequencies k = 0.9 , k = 1 and k = 1.1 . The damping constants are set to
d = 0 and h = 1/50. In Figure 2.3 the lines about which E d ( t ) and E e ( t )
oscillate are included in stroke-dotted style.
=0.9
2.
Energy
1.5
1.
0.5
Ek
Epf
Epb
Ed
Ee
0.
- 0.5
=1
40.
Energy
30.
20.
10.
Ek
Epf
Epb
Ed
Ee
0.
- 10.
8.
=1.1
Energy
6.
4.
2.
Ek
Epf
Epb
Ed
Ee
0.
- 2.
0.
0.2
0.4
0.6
(/(2))
0.8
1.
Figure 2.3: Energy components of the viscoelastically supported BernoulliEuler beam over one vibration cycle ( d = 0 , h = 1 50 )
21
a
Of course the law of energy conservation is met at any time, giving a constant
sum of all five energy components. The derived energy expressions illustrate
that all energy components have an oscillatory part with twice the excitation
frequency. The dissipated energy and the excitation energy consist of an additional time-linear part. By replacing a with its closed form solution (2.4), it
can be shown that these two time linear parts exactly cancel each other out,
showing the fact that the energy lost in the foundation damping has to be compensated by the excitation force.
Below the cutoff frequency, where the spatial decay of the system response is
high, the kinetic energy E k ( t ) , the foundations potential energy E pf ( t ) and
the beam deformation energy E pb ( t ) go back to approximately zero twice
during one vibration cycle. Towards higher frequencies the mean values about
which these three energy components oscillate are increasing as compared to
the magnitude of the oscillation. This implies that there is a constant amount of
these energy components stored throughout the beam in the propagating wave.
The magnitudes of the oscillatory energy components are
- for the kinetic energy:
Ek ( t ) =
3 2k
64 a
(2.41)
- for the potential energy stored in the foundation:

E pf ( t ) =
3 2
64 a
(2.42)
- for the beam deformation energy:

E pb ( t ) =
2
64 a
(2.43)
- for the dissipated energy:

Ed ( t ) =
3 2 2 ( dk + h )
64 a
(2.44)
- for the work of the excitation force:

Ee ( t ) =
4 2
64 a
(2.45)
At cutoff frequency k = 1 , the magnitude of the kinetic energy and the potential energy stored in the foundation match each other, implying that there is an
22
uncoupled energy flow between these two energy components on the one hand
and between E pb ( t ) , E d ( t ) and E e ( t ) on the other hand. At a general frequency there is an exchange between all five energy components.
Figure 2.4 shows the magnitudes of the oscillatory energy parts as a function
of frequency.
20.
Energy
15.
10.
888888
|Ek|
|Epf|
|Epb|
|Ed|
|Ee|
5.
0.
0.7
0.8
0.9
1.
1.1
1.2
1.3
Figure 2.4: Magnitudes of oscillatory energy parts

( d = 0 , h = 1 50 )
Figure 2.4 illustrates that the kinetic energy and the foundations potential
energy are the predominant energy components close to the cutoff frequency.
The observation that most of the potential energy is stored in the foundation
elasticity instead of the beam deformation at frequencies close to the cutoff
frequency is explained by the fact that the wavenumber a is small at these frequencies, implying a large wavelength l = 1 a and a small beam curvature.
Introducing the absolute value of the wavenumber a at the frequencies k a
and k b into expressions (2.41) to (2.45) shows that the oscillatory magnitudes
78
= 1 8 128 as
of E k , E pf and E d have dropped to approximately 1 2
compared to the cutoff frequency. The oscillatory magnitudes of E pb and E e
38
= 1 8 8 their cutoff frequency values.
have dropped to 1 2
Although the QBW-factor definitions derived in this chapter are consistent, the
physical interpretation is much less obvious as in the case of a classical single
degree of freedom oscillator, where the system energy is related quadratically
to the vibration amplitude und therefore has dropped to 1 2 at the corresponding frequencies k a and k b .
23
2.2
Extension to multiple modes
The next step is to extend the model studied in the previous section to include
higher vibration modes. Let us therefore consider a system of multiple Bernoulli-Euler beams interconnected by viscoelastic layers. Figure 2.5 shows a
model consisting of n = 3 beams.
z
A1, I1
s1
A2, I2
s2
A3, I3
beam 1
beam 2
beam 3
~
p
s3
Figure 2.5: Continuous multiple modes model consisting of n = 3 BernoulliEuler beams interconnected by elastic layers
Ripke and Knothe [24] used a model with two Timoshenko beams representing the head and the foot of the rail which are elastically connected by a plate
to simulate the tracks lateral response. Scholl [27] introduced a two dimensional model consisting of three strips representing head, web and foot of the
rail to model wave propagation along the rail. The elements of the model suggested in this work should not be viewed as a valid representation of physical
track components. The model serves merely as a vehicle to qualitatively study
effects of higher vibration modes at their cutoff frequency as well as effects of
discrete supports. Many features observed in the experimental investigations
presented and discussed in Chapter 3 can be explained using this simple
model. Since only the behavior close to the cutoff frequencies, which implies
large wavelengths, is of interest in this work, Bernoulli-Euler beams are used
instead of Timoshenko beams. This assumption is justified in greater detail in
Section 2.2.3.
2.2.1
Modal solution
The dynamic behavior of the model shown in Figure 2.5 is described by the set
of differential equations (2.46), where ui=ui(z,t) are the absolute downward
displacements of the three beams.
EI 1 0
u1
0 EI 2 0
u2
0 EI 3 u 3
rA 1 0
+
,zzzz
u1
0 rA 2 0
u2
0 rA 3 u 3
s1
s1
+ s1 s1 + s2
,tt
s2
0
s2
u1
u2 = 0
(2.46)
s2 + s3 u3
No damping is included in equation (2.46) yet in favor of modal damping,

which will be introduced at a later stage.
24
The set of equations (2.46) can be written in matrix notation as

Bu ,zzzz + Mu ,tt + Ku = 0 ,
(2.47)
where u is the displacement vector of the beams and K, M and B are the stiffness coefficient matrix of the elastic connections, the mass matrix and the flexural stiffness coefficient matrix of the beams, respectively.
The set of differential equations (2.47) is coupled by the stiffness matrix of the
elastic connections K. The equations could be directly solved by using appropriate functions similar to (2.3) in Section 2.1.1. However, in order to get a
better understanding of the vibration mechanisms, the final solution of the
damped system will be derived stepwise from several subproblems.
Let us first consider the subproblem
Mu ,tt + Ku = 0 ,
(2.48)
corresponding to the free vibration problem of the model cross-section.

Introducing
u = be
iwt
(2.49)
2
leads to the eigenvalue problem (2.50) with eigenvalues w i and eigenvectors

xi .
2
( w M + K )b = 0
(2.50)
Assembling the eigenvectors according to (2.51) gives the modal matrix E w

which is scaled to meet conditions (2.52).
E w = x 1 , ... , x n
(2.51)
E w KE w = diag w 2 , ... , w 2 = W
1
n
(2.52)
E w ME w = I
The modal matrix E w is now used to transform the set of differential equations
(2.47), giving
T
E w BE w u w ,zzzz + Iu w ,tt + W u w = 0 ,
(2.53)
which can be solved considering steady state solutions uw of the form

uw = bw e
i(k(w)z + wt)
(2.54)
where k ( w ) is the wavenumber of the undamped system. Introducing (2.54)
25
a
into (2.53) yields the eigenvalue problem
T
( k ( w ) E w BE w w I + W )b w = 0
(2.55)
with eigenvalues k i ( w ) and eigenvectors x ki ( w ) . The values k i ( w ) are the

wavenumbers of the wave propagation modes of the undamped system.
If the frequency is equal to one of the frequencies w i calculated above, the
2
2
rank of the matrix ( w I + W ) is reduced by one, implying that one of the
eigenvalues k i ( w ) is zero and the corresponding eigenvector x ki ( w ) equals x i .
This means that the frequencies w i calculated from the free vibration problem
of the cross-section (Equation (2.48)) are the cutoff frequencies of the system
and the vectors x i are the mode shapes of the respective modes at those frequencies.
Similar to above, the eigenvectors x ki ( w ) are assembled into the frequency
dependent modal matrix
E k ( w ) = x k1 ( w ) , ... , x kn ( w ) , where
T
E k ( w ) ( W w I )E k ( w ) = diag k ( w ) 4 , ... , k ( w ) 4
1
n
(2.56)
(2.57)
E k ( w )E w BE w E k ( w ) = I .
By applying the modal matrices E w and E k ( w ) , the original set of differential
equations can be frequency-wise decoupled, giving
T
E k ( w ) E w BE w E k ( w ) u k, zzzz + E k ( w ) E k ( w ) u k ,tt + E k ( w )W E k ( w ) u k = 0 , (2.58)
which can be solved by introducing solutions uk of the form

uk = bk e
i(k(w)z + wt)
(2.59)
Within the experimental investigation presented in Chapter 3 the damping values of several modes are measured at their cutoff frequencies. The resulting
quality factors are interpreted as damping measures assigned to these modes.
Motivated by this experimental consideration, both hysteretic and viscous
modal damping are introduced to the system at this stage as shown in (2.60).
T
4
2
2
k ( w ) I + E k ( w ) ( W w I )E k ( w ) +
ki ( 0 )
4
iwdiag 2
d i + idiag [ 2k i ( 0 ) h i ] b k = 0
wi
26
(2.60)
The undamped wavenumbers k i at w = 0 and the cutoff frequencies w i of

each mode are used to scale the damping matrices in order to use the same
order of magnitude for the damping parameters as in the previously studied
case of the continuously supported beam.
Solutions of the complex equation system (2.60) exist for the eigenvalues
4
k i ( w ) . By writing the corresponding complex eigenvectors x ki ( w ) as
E k ( w ) = x k1 ( w ) , ... , x kn ( w ) ,
(2.61)
the final solution of the original set of differential equations (2.47) can be
expressed by its modal components as
u = E w E k ( w ) E k ( w ) ( Y 1 ( z )y 1 + Y 2 ( z )y 2 + Y 3 ( z )y 3 + Y 4 ( z )y 4 ) e
where
Y 1 ( z ) = diag [ e
Y 3 ( z ) = diag [ e
ik i ( w ) z
], Y 2 ( z ) = diag [ e
ki ( w ) z
i ki ( w ) z
], Y 4 ( z ) = diag [ e
ki ( w ) z
iwt
],
(2.62)
(2.63)
and y1 to y4 are the modal coefficient vectors. In the case of a continuously

supported model as shown in Figure 2.5, which is harmonically excited by a
harmonic point force vector p acting on the beams at z=0, the modal components related to Y 1 and Y 4 must disappear to ensure the radiation condition
for z > 0 similarly to the beam on a viscoelastic foundation. The boundary
conditions at z=0 can be written in matrix notation as
3
BE w E k ( w ) E k ( w ) d Y 2 Y 2 ( 0 ) BE w E k ( w ) E k ( w ) d Y 3 Y 3 ( 0 ) y 2
Ew Ek ( w ) Ek ( w ) d Y2 Y2 ( 0 )
Ew Ek ( w ) Ek ( w ) d Y3 Y3 ( 0 )
y3
= p
0
(2.64)
where d Y 2 = diag [ i k i ( w ) ] and d Y 3 = diag [ k i ( w ) ] .

This shows that the system response of the model to the specified excitation is
composed of traveling wave shares related to Y 2 and of exponentially decaying shares related to Y 3 of each mode. Further investigation of this model
with help of a numerical example is presented in Section 2.2.3.
27
2.2.2
Multiple mode model on discrete supports
With the mounting of the rail on sleepers at discrete intervals in mind, the
influence of the discrete support on higher vibration modes of a waveguide
needs to be examined. It must be expected that the range of parameters regarding both damping and geometry can vary considerably especially for the supports of railway tracks. It is important to understand these influences to be
capable to make use of suitable vibration modes for a sensor based on forced
vibrations. For this reason, the solution of the multiple mode model on a discrete periodic support is derived in this section.
Lets consider the model shown in Figure 2.6, consisting of three elastically
connected beams, with equidistant elastically supported regions.
z
A1, I1
s1
A2, I2
s2
A3, I3
beam 1
beam 2
beam 3
~
p
s3
L
basic section
L1
L2
supported unsupported
part
part
Figure 2.6: Multiple modes model on discrete supports

Grassie et al [8] among other authors used the inter-element transfer matrix to
calculate the direct receptance of infinitely long, discretely supported systems.
The method applied here is based on this concept.
The system of Figure 2.6 can be divided into an infinite number of identical
sections of length L = L1 + L2 consisting of two unsupported parts of length
L1/2 and a supported part of length L2 as shown in Figure 2.7.
The dimensionless state vector at the beginning of the rth section to the right
hand side of the excitation is defined as
2
L
L
1
L
L
1
vr =
F 1, ... ,
F n, q 1, ... , q n,
M 1, ... ,
M n, y 1, ... , y n
EI 1
EI n
EI 1
L
EI n
L
(2.65)
= S [ F 1, ... , F n, q 1, ... , q n, M 1, ... , M n, y 1, ... , y n ]
where yi, qi, Fi, Mi are the displacement, the rotation, the shear force and
moment at beam i for a model consisting of n beams. Values, which are pre-set
at the excitation point z = 0 (applied force and beam rotations), stand first. The
state vector is scaled using the diagonal matrix S for better numerical behavior
28
a
of the problem.
+z
+y
+F
section (r-1)
section r
section (r+1)
+M
+
zl=0
L1/2
vr=vA
vB
vr
L2
L1/2
v(r+1)
v(r+1)
T
T
Figure 2.7: Basic section of discretely supported model

The first step is to find the transfer matrix T relating the state vector vr at the
beginning of the section r with the corresponding vector v(r+1) of section (r+1)
such that
v ( r + 1 ) = Tv r .
(2.66)
As illustrated in Figure 2.7, the matrix T is composed of the transfer matrices

C and D of the unsupported and the supported subsections of the model,
respectively. The displacement of the unsupported part corresponds to equation (2.62) with the foundation stiffness set to zero. By choosing a local variable zl running from -L1/4 to +L1/4, the boundary conditions at zl=L/4 can be
written as
L1
SC 1 y 1, y 2, y 3, y 4
4

3
BE w E k ( w )E k ( w )dY 1 Y 1 ( z l )

E w E k ( w )E k ( w )dY 1 Y 1 ( z l )
C1 ( zl ) =

2
B E w E k ( w )E k ( w )dY 1 Y 1 ( z l )

E w E k ( w )E k ( w )Y 1 ( z l )
= v B where

3
... BE w E k ( w )E k ( w )dY 4 Y 4 ( z l )

... E w E k ( w )E k ( w )dY 4 Y 4 ( z l )

2
... B E w E k ( w )E k ( w )dY 4 Y 4 ( z l )

...
E w E k ( w )E k ( w )Y 4 ( z l )
(2.67)
(2.68)
and d Y 1 = diag [ ik i ( w ) ] , d Y 2 = diag [ i k i ( w ) ] , d Y 3 = diag [ k i ( w ) ] and

d Y 4 = diag [ k i ( w ) ] . The boundary conditions at -L1/4 are solved for the modal
coefficient vector y 1, y 2, y 3, y 4 , giving
1 L 1
1
C 1 S v A = y 1, y 2, y 3, y 4
4
(2.69)
29
a
By eliminating the modal coefficient vector from equations (2.67) and (2.69),
the transfer matrix C can be expressed as1
L 1 1 L 1 1
v B = SC 1 C 1 S v A = C v A .
4
4
(2.70)
The transfer matrix D of the supported subsection is built up in a similar way

by using the solution (2.62) with corresponding model parameters, thus giving
the transfer matrix of the complete section T = CDC.
The state vector at the beginning of section (r+1) is computed from the state
vector at the excitation point v0 by repeated multiplication with T.
By using the eigenvectors Vi and the eigenvalues qi of T and thus writing the
state vector v0 as
a1
v 0 = V 1, ... , V 4n
(2.71)
a 4n
the state vector v(r+1) at the beginning of section (r+1) can be expressed as
r
v ( r + 1 ) = T v 0 = V 1, ... , V 4n diag [ q i ] a .
(2.72)
The eigenvalues of T appear in inverse pairs (qi,1/qi), corresponding to left and

right traveling waves, respectively. The case of qi = 0 does not occur in the
presence of damping. Terms in equation (2.72) which contain eigenvalues with
moduli greater than unity, say q2n+1 to q4n, must disappear to meet the requirement that vr tends towards zero with increasing distance r from the excitation
point. This implies that the state vector v0 can be written as
(1)
v0
(2)
v0
(1)
V1
(2)
V1
(1)
... V 2n
...
(2)
V 2n
a1
.
a 2n
( 1 ) a1
G
=
.
(2)
G
a 2n
(2.73)
from which [ a 1, ... , a 2n ] can be eliminated, giving the relation (2.74)

(1)
2
2
between the given values v = [ L F 1 EI 1, ... , L F n EI n, q 1, ... , q n ] and the
(2)
unknown quantities v = [ LM 1 EI 1, ... , LM n EI n, y 1 L, ... , y n L ] at the
excitation point.
(2)
v0
1
= G
(2)
(1)
(1)
v0
(1)
= Hv 0
(2.74)
1. C 1 can be defined analytically. This improves the numerical stability of the solution.
30
2.2.3
Discussion of the multiple mode model
2.2.3.1
Introduction
In this section, the characteristics of the derived multiple mode model are studied using a numerical example consisting of n = 3 beams. The model is studied in three configurations as shown in Figure 2.8. The unsupported
configuration is used to study the behavior of wave propagation modes in the
waveguide on its own. Based on the assumption that the damping of the rail
itself is moderate as compared to the damping of the supports on sleepers and
ballast, the modal damping of the unsupported model is set considerably lower
than in the second configuration on a continuous support. The third model is a
combination of the first two configurations giving a discretely supported
model with alternating free and supported regions of the waveguide, thus taking into account the mounting of the rail on sleepers at equidistant locations.
The comparison of these three model configurations makes it possible to study
the influence of the support parameters on higher vibration modes.
Configuration A:
unsupported
Configuration B:
continuous support
Configuration C:
discrete support
Figure 2.8: Examined model configurations

The following model parameters are used:
EI2 = 1.7 106 [Nm2]
EI1 = 1.2 106 [Nm2]
rA1 = 15.0
[kgm-1]
EI3 = 0.5 106 [Nm2]
rA2 = 15.0
[kgm-1]
rA3 = 15.0
[kgm-1]
= 5.0 109 [Nm-2]
s2
= 5.0 109 [Nm-2]
s1
L
= 0.67
[m]
additional parameters of the unsupported model (Confg. A) and of the
unsupported model subsections (Confg. C):
= 0.0
[Nm-2]
L1 = 5.025 10-1 [m]
s3
hu,1 = 6.25 10-4 [-]
hu,2 = 6.25 10-4 [-]
hu,3 = 6.25 10-4 [-]
additional parameters of the continuously supported model (Confg. B)
and of the supported model subsections (Confg. C):
= 9.0 107 [Nm-2]
L2 = 1.675 10-1 [m]
s3
hs,2 = 5.0 10-3 [-]
hs,1 = 5.0 10-3 [-]
hs,3 = 5.0 10-3 [-]
31
a
It should be noted that hu,i and hs,i are the hysteretic modal damping parameters
according to the previous definitions. The viscous modal damping parameters
are set to du,i = 0 and ds,i = 0. Based on the assumption that the rail itself has a
moderate damping (see e.g. [9]), the modal damping of the model with support
is set eight times higher than in the unsupported case in the presented numerical example.
The following conditions together with some assumptions about the distribution of mass and stiffness where used to choose this set of parameters:
- The static deflection of a part of this model without support corresponds to the static deflection of a beam with the cross-section of a
rail of type SBB I. The experiments presented in Chapter 3 where
carried out at a rail with this cross-section.
- The stiffness of the elastic layers between the beams s1 and s2 give a
comparable cross-sectional deformation to static forces as in the case
of a rail of type SBB I.
- The frequency spectrum of the first mode is similar to the bending
mode of a beam with cross-section SBB I.
Using these approximate conditions, the value range of the results is in the
order of magnitude as for the real rail. However it must be pointed out once
again that the multiple mode model is merely used to qualitatively study and
understand higher mode behavior of a waveguide and is not intended to
exactly reproduce the system response of a rail with corresponding cross-section. A more specific investigation of the actual situation is found in
Section 2.4.
32
2.2.3.2
Frequency spectra and receptances
The frequency spectra of all three modes in both the unsupported and the continuously supported case are shown in Figure 2.9.
8000.
6000.
Frequency (Hz)
fc,3
4000.
fc,2
2000.
fc,1
+Im(k)
10.
+Re(k)
5.
5.
Wavenumber (rad/m)
continuous support
unsupported
10.
Figure 2.9: Frequency spectra of the unsupported and the continuously supported model
Figure 2.9 illustrates that the wavenumber ki has a large imaginary part below
the cutoff frequency fc,i of each mode. Due to damping, the absolute value of
the wavenumbers are not zero at the cutoff frequencies. With increasing frequencies above fc,i the imaginary parts of the complex wavenumbers tend
toward zero.
As we can see from Figure 2.9, the presence of the support shifts the cutoff
frequency of the first mode considerably, whereas the cutoff frequencies of the
higher modes are barely influenced. This is illustrated in Table 2.1, which
shows the cutoff frequencies and their relative deviations for all three model
configurations. The cutoff frequencies of the unsupported and the continuously supported models were calculated according to the solution of the free
vibration problem of the model cross-section (see equation (2.48) in
Section 2.2.1). The respective frequencies of the discretely supported model
were obtained from the receptance function (see later in this section).
33
a
Table 2.1 confirms that the cutoff frequencies of higher modes are barely
influenced by the presence of the support in the presented numerical example,
where the stiffness of the support is small as compared to the stiffness of the
connecting layers. With the rail mounted on sleepers and ballast in mind, this
assumption seems reasonable (see also [9]).
Mode
fc,A
[kHz]
fc,B
[kHz]
(fc,B-fc,A)/fc,B
[%]
fc,C
[kHz]
(fc,C-fc,A)/fc,C
[%]
0.000
0.224
100.000
0.112
100.000
2.906
2.919
0.449
2.909
0.110
5.033
5.035
0.050
5.033
0.005
Table 2.1: Cutoff frequencies of the unsupported (fc,A), the continuously

supported (fc,B) and the discretely supported model configuration (fc,C) and
their relative deviations
The first cutoff frequency is mainly influenced by the stiffness of the supports.
The discrete model is supported only at L2/L = 1/4 of its length, thus giving a
global stiffness of the support which is four times lower as compared to the
continuous model. The square root dependency between the foundation stiffness and the cutoff frequency (see Section 2.1) explains the ratio of approximately 1/2 between fc,C1 and fc,B1 in Table 2.1.
Although the frequency spectrum of each wave mode looks similar to the frequency spectrum of a single viscoelastically supported beam as discussed in
Section 2.1, it must be pointed out that they do not match exactly. This is due
to the fact that a frequency dependent transformation must be used to decouple
the differential equations of the multiple mode model (see Section 2.2.1).
Let us now look at the system responses of all three model configurations with
a harmonic excitation force at z=0 acting at beam 3. Figure 2.10 illustrates the
absolute value of the receptance y i ( z = 0 ) p 3 measured at beam 1 and at
beam 2. The receptance measured at beam 3 looks similar to the receptance of
beam 1 and is not shown for better clarity of the plot.
The three receptance plots show maxima at the cutoff frequencies of the
respective model configurations. At the cutoff frequency of mode 2, the amplitude maximum of the receptance of beam 2 is very weak for all three configurations.
Additional maxima and minima of the receptance functions are observed for
the discretely supported model configuration. The first maxima of this type is
found at fII,1 = 957 Hz. These maxima and minima originate in resonance
effects of the infinite series of identical sections. Such effects among other
characteristics of periodic systems have been studied by numerous research-
34
ers. Besides various authors working in the field of railway track modeling
mentioned in Section 1.1, the more general contributions of Mead [19,20] to
this research area deserve special attention.
The receptance of the studied system shows amplitude maxima caused by the
discrete support also at fII,2 = 3.02 kHz and at fII,6 = 5.15 kHz among other frequencies. The relative differences between these frequencies and the cutoff
frequencies fc,C2 and fc,C3 are less then 4%. This observation means that some
amplitude maxima of this type must be expected to be very close to the cutoff
frequencies of the rail. This presents an additional difficulty in experimentally
finding specific vibration modes of the rail.
|yi(z=0)/p3|
1. 107
fc,A2 fc,A3
1. 108
beam 1
1. 109
1. 1010
1. 1011
beam 2
unsupported
fc,B1
|yi(z=0)/p3|
1. 107
fc,B2 fc,B3
1. 108
beam 1
1. 109
1. 1010
1. 1011
beam 2
continuous support
fII,2
|yi(z=0)/p3|
1. 107
1. 108
1. 109
fII,1
fII,6
fc,C3
beam 1
fc,C1
1. 1010
1. 1011
fc,C2
beam 2
discrete support
100
200
500
1000 2000
5000
Frequency (Hz)
Figure 2.10: Receptance at beams 1 and 2 with harmonic excitation at beam 3
35
36
Displacement
Displacement
2.
0.
2.
4.
6.
2.5 107
1. 107
2.
0.
2.
4.
Figure 2.11: Beam displacements of model configuration B at fc,B1 and

of configuration C at fc,C1 (p3 = 1 N)
(b)
4.
(a)
6.
Distance from excitation (m)
7.5 107
3. 107
beam 3
beam 2
beam 1
0.
2.5 107
8888888888
5. 107
0.
1. 107
8888888888
2. 107
4.
2.5 107
1. 107
2. 10
6.
7.5 107
3. 107
5. 10
7.5 107
3. 107
2. 107
0.
2.5 107
8888888888
5. 107
0.
1. 107
2. 10
beam 3
7.5 107
3. 107
5. 10
7.5 107
3. 107
2. 107
0.
2.5 107
8888888888
5. 107
0.
1. 107
beam 2
2.5 107
2. 10
1. 107
7.5 107
5. 107
beam 1
3. 107
Configuration C:
Mode 1 at fc,C1
6.
2.2.3.3
Displacement
Configuration B:
Mode 1 at fc,B1
Beam displacements
Figures 2.11 to 2.13 show plots of the beam displacements of all three model
configurations at their respective cutoff frequencies. Hatched areas beneath the
displacement curve of beam 3 identify supported regions of the models. The
beam displacements of the discretely supported model at the frequencies fII,1,
fII,2 and fII,6 are shown in Figure 2.14.
Displacement
Displacement
Displacement
2.
4.
6.
1. 108
4. 108
0.
1. 108
4. 108
2.
5. 109
2. 108
0.
5. 109
1. 108
1. 108
0.
2. 108
4. 108
4. 108
5. 109
1. 108
2. 108
4. 108
0.
2.
4.
6.
2. 108
1. 108
0.
1. 108
2. 108
2. 108
1. 108
0.
1. 108
2. 108
2. 108
1. 108
0.
1. 108
2. 108
6.
2.
0.
2.
4.
(c)
4.
beam 3
beam 2
beam 1
Configuration C:
Mode 2 at fc,C2
Figure 2.12: Beam displacements of all three model configurations at cutoff frequency 2 (p3 = 1 N)
(b)
2.
(a)
4.
6.
beam 3
beam 2
beam 1
Configuration B:
Mode 2 at fc,B2
4.
0.
0.
6.
5. 109
2. 108
beam 3
5. 109
2. 108
beam 2
0.
5. 10
0.
1. 108
2. 108
beam 1
9
4. 108
Configuration A:
Mode 2 at fc,A2
6.
a
37
Displacement
Displacement
38
Displacement
0.
0.
6.
2.
0.
2.
4.
(a)
4.
beam 3
beam 2
6.
2. 109
0.
8888888888
1. 109
1. 109
2. 109
2. 109
6.
2.
0.
2.
4.
(b)
4.
6.
4. 109
0.
8888888888
2. 109
2. 109
4. 109
4. 109
2. 109
1. 109
2. 109
4. 109
0.
beam 3
beam 2
0.
1. 109
2. 109
4. 109
6.
2.
0.
2.
4.
(c)
4.
beam 3
beam 2
beam 1
Configuration C:
Mode 3 at fc,C3
Figure 2.13: Beam displacements of all three model configurations at cutoff frequency 3 (p3 = 1 N)
1. 108
5. 109
5. 10
1. 108
5. 109
5. 10
1. 108
2. 109
2. 109
1. 109
5. 109
0.
2. 109
4. 109
0.
beam 1
0.
1. 10
2. 109
9
beam 1
Configuration B:
Mode 3 at fc,B3
5. 109
1. 108
Configuration A:
Mode 3 at fc,A3
6.
Displacement
Displacement
4. 10
4.
2.
0.
2.
4.
6.
0.
2.
4.
(b)
2.
(a)
4.
6.
beam 3
beam 2
beam 1
4. 109
2. 109
2. 108
2. 109
0.
6.
beam 3
0.
2. 10
4. 10
4. 109
4. 109
2. 109
2. 108
0.
0.
4. 108
2. 109
4. 109
4. 109
2. 108
2. 109
2. 108
beam 2
0.
0.
2. 10
4. 109
9
beam 1
2. 108
4. 108
Configuration C:
at fII,2
6.
1. 109
1.
5.
5. 1010
0.
5.
5. 1010
1. 109
1. 109
5. 1010
0.
5. 1010
1. 109
1. 109
5. 1010
0.
5. 1010
1. 109
6.
2.
0.
2.
4.
(c)
4.
beam 3
beam 2
beam 1
Configuration C:
at fII,6
6.
Figure 2.14: Beam displacements of discretely supported model at amplitude maxima fII,1, fII,2 and fII,6 (p3 = 1 N)
Displacement
Configuration C:
at fII,1
a
39
a
The first mode of the investigated model is characterized by an in phase
motion of all three beams as illustrated in Figures 2.11 (a) and (b).
The characteristic feature of the mode shape at the second cutoff frequency is a
counter phase vibration of beams 1 and 3, whereas beam 2 shows almost no
movement. The circumstance that the system response is composed of a propagating wave part and of an exponentially decaying vibration component is
apparent not only from the unsupported and the continuously supported models, but also from the discretely supported model. The zone in which the vibration is predominant extends across 6 to 8 sleeper spacings. The spatial decay
of the local vibration zone is strongest for the continuously supported model
due to high damping throughout this model. Beyond this region of predominantly vibrating motion, only a propagating wave with comparably small
amplitude remains visible. The spatial decay of the wave component is moderate. These wave components are studied in greater detail later in this section.
A more quantitative definition and investigation of the local vibration zone can
be found within the discussion of system damping in Section 2.2.3.4.
The observed fact that the receptance at beam 2 at the second cutoff frequency
showed only a small peak is explained from the displacement plots of
Figure 2.12, as beam 2 is almost a node of the cross-sectional mode shape.
At the cutoff frequency of mode 3, the displacements of beams 1 and 3 are in
phase and beam 2 vibrates in counter phase, as is illustrated in Figure 2.13.
The spatial extent of the local vibration zone is slightly smaller than in the case
of cutoff frequency 2.
A graphic explanation of the additional amplitude maxima of the discretely
supported model as seen in Figure 2.10 is given in Figure 2.14 (a), which
shows the displacements of the discretely supported model at the frequency
fII,1 = 957 Hz. At this frequency, the wavelength of the first mode equals twice
the sleeper spacing. The resulting system response is a standing wave with
large displacements between the supports and vibration nodes at the supported
regions. In the literature, this phenomenon is often referred to as the pinnedpinned frequency of a beam on discrete supports (see e.g. [8]). At the frequency fII,2, which was found close to the cutoff frequency fc,C2, the beam displacements show the typical cross-sectional movement of mode 2. However,
the spatial behavior observed in Figure 2.14 (b) is completely different as
compared to Figure 2.12 (c). The amplitude of the system response decreases
at the supported areas of the model and builds up again in the unsupported
regions. The decay along the rail is moderate and no local vibration zone close
to the excitation is observed. The same statements can be made at the frequency fII,6. Although the cross-sectional movement corresponds closely to the
mode shape of mode 3, also here no local vibration zone is found.
40
Resonance effects originating from the discrete support are not considered to
be suitable for wheel detection because their lack of a local vibration zone
close to the excitation. A rail contact based on rail vibrations must avoid resonances of this type. The frequencies of these resonances can be very close to
the cutoff frequencies of the rail. This presents an additional difficulty as
already mentioned above. No further theoretical investigation of the effects
caused by the discrete support is performed within this work.
Although the chosen excitation vector doesnt allow the distinct excitation of a
specific mode, the model response at all cutoff frequencies is dominated by the
cross-sectional mode shape of the respective mode at least close to the excitation. This can be shown by plotting the absolute value of each component of
the modal coefficient vectors y2 and y3 with the given excitation for the continuously supported model (see Figure 2.15). Because of the symmetric situation, the relation between the modal coefficient vectors is y 2 = iy 3 for all
frequencies, as can be seen from equation (2.64).
Modal coefficients (-)
1.
0.1
0.01
888888
0.001
|21| = |31|
|22| = |32|
|23| = |33|
0.0001
0 1000 2000 3000 4000 5000 6000 7000
Frequency (Hz)
Figure 2.15: Absolute value of each component y2i of the modal coefficient
vector y2 for the continuously supported model with excitation at beam 3
Figure 2.15 illustrates that at each cutoff frequency the respective mode is
actuated primarily and with large amplitude.
A slightly different view is gained by looking at the actual contribution of each
mode to the beam displacement with increasing distance from the excitation as
in the example of beams 1 and 3 shown in Figure 2.16.
If the system is excited with the cutoff frequency of a certain mode, the displacement at the point of excitation is dominated by the mode shape of the
respective mode. At some distance from the excitation, the situation has
changed for higher modes due to the fact that modal damping has a strong
influence especially at the cutoff frequency of each mode. The modal damping
causes the mode which is excited at its cutoff frequency to decay at a much
higher rate as compared to modes with lower cutoff frequencies. This can be
seen in the frequency spectra shown in Figure 2.9 from the fact that the imagi-
41
a
nary part of the wavenumbers decrease towards higher frequencies. This
means that traveling waves of lower modes become more and more important
with increasing distance from the excitation. It should be noted that the modeshapes of these modes change toward higher frequencies, as can be seen from
the frequency dependent transformation used to obtain the solution.
Modal contributions (-)
1.
u1(z=0)
Mode 1
Mode 2
Mode 3
u3(z=0)
Mode 1
Mode 2
Mode 3
u1(z=4)
Mode 1
Mode 2
Mode 3
u3(z=4)
Mode 1
Mode 2
Mode 3
0.1
0.01
0.001
0.0001
Modal contributions (-)
1.
0.1
0.01
0.001
0.0001
1000 2000 3000 4000 5000 6000 7000
Frequency (Hz)
1000 2000 3000 4000 5000 6000 7000

Frequency (Hz)
Figure 2.16: Contributions of each mode to the displacements of beam 1 and 3

at z = 0 m and z = 4 m
With an excitation at a cutoff frequency of a certain mode, these observations
imply that the behavior of the respective mode dominates the system response
in the vicinity of the excitation. This justifies the use of Bernoulli-Euler beams
instead of Timoshenko beams in the model, as mainly the local vibration zone
close to the excitation is of importance for the examined application of train
wheel detection.
The study of the system energy in the case of the continuously supported beam
as performed in Section 2.1.2.3 showed that potential energy stored in foundation elasticity outweighs the potential energy stored in beam deformation at
the cutoff frequency. For the multiple mode model, the conclusion can be
drawn that the potential energy stored in the connecting springs is the predominant part of the potential energy at the cutoff frequency of each mode.
42
2.2.3.4
Damping and spatial decay
|y1|
As a next step, the effect of system damping is studied in greater detail. The
concept of the QBW-factor, developed in Section 2.1.2 is used to characterize
the damping of the overall system. This is illustrated in Figure 2.17 in which
the absolute value and the phase lag of the receptance at beam 1 close to the
cutoff frequency of mode 3 are shown for both the continuously and the discretely supported model configuration. The receptances are shown for three
values of the modal damping parameter hs,3 of the supported model and model
subsection, respectively. In the case of hs,3 = 1/200, the values used for the
QBW-factor determination according to Section 2.1.2 are shown with dashed
lines.
2.5 109
s,3 = 1/300
2. 109
s,3 = 1/200
1.5 109
s,3 = 1/130
1. 109
5. 1010
0.
s,3 = 1/300
Arg(y1)
0.5
s,3 = 1/200
1.
s,3 = 1/130
1.5
2.
2.5
continuous support
discrete support
3.
4950.
5000.
5050.
5100.
5150.
Frequency (Hz)
Figure 2.17: Determination of the QBW-factor of mode 3 from the magnitude

and the phase response (ds,i = 0)
Both the continuous and the discrete model show a phase shift close to 3p/4 at
the cutoff frequency of mode 3. The additional amplitude maximum of the discretely supported model is also visible at fII,6 = 5.15 kHz. The associated phase
shift is even less than p/4. Therefore neither the Q-factor nor the QBW-factor
are suitable to characterize the damping at this frequency.
Table 2.2 shows the QBW-factors of all modes and all model configurations
BW
BW
determined from the magnitude ( Q Abs ) and the phase curve ( Q Arg ) according
to equation (2.21). The relative deviations between the quality factor values is
also included.
43
a
Confg. A
BW
BW
Q Arg Q Abs Confg. B
----------------------------BW
BW
Q Abs
Q Arg
BW
Q Abs
BW
Q Arg
[-]
[-]
[%]
800.39
598.21
Mode
BW
Q Arg
BW
BW
Q Arg Q Abs Confg. C
----------------------------BW
BW
Q Abs
Q Arg
BW
BW
BW
Q Arg
Q Arg Q Abs
----------------------------BW
Q Arg
[-]
[-]
[%]
[-]
[-]
[%]
100.07
100.11
0.05
100.56
100.59
0.03
803.84
0.43
100.66
102.85
2.14
287.87
290.78
1.01
603.02
0.80
73.54
76.79
4.32
216.68
219.11
1.12
Table 2.2: QBW-factors of the unsupported (Confg. A), the continuously

supported (Confg. B) and the discretely supported model configuration
(Confg. C) as determined by the magnitude and the phase curve and their
relative deviations. Model parameters according to Section 2.2.3.1.
The QBW-factors determined by the magnitude and the phase curve correspond
well as could be expected from the similarity with the beam model. Table 2.2
also illustrates that the same damping parameters of the supported areas results
in a considerably higher QBW-factor of modes 2 and 3 in the case of the discretely supported model as compared to the continuously supported model.
The reason for this observation is the fact that the high modal damping at the
supported regions affects only L2/L = 1/4 of the total length of the discrete
model.
Applying the classical measurement method of the SDOF1 Q-factor (see
Appendix A) to the studied system gives the values summarized in Table 2.3
for the quality factor.
Confg. A
Q Arg Q Abs Confg. B

----------------------------Q Abs
Q
Q Abs
Q Arg
[-]
[-]
[%]
649.29
485.27
Mode
Q Arg
Q Arg Q Abs Confg. C

----------------------------Q Abs
Q Arg
Q Arg
Q Arg Q Abs
----------------------------Q Arg
[-]
[-]
[%]
[-]
[-]
[%]
81.17
57.83
33.58
81.57
58.09
33.62
466.10
32.85
81.67
60.58
29.65
233.67
169.37
31.91
350.83
32.16
59.48
45.73
26.15
175.76
127.74
31.65
Table 2.3: Q-factors of the unsupported (Confg. A), the continuously

supported (Confg. B) and the discretely supported model configuration
(Confg. C) as determined by the magnitude and the phase curve and their
relative deviations. Model parameters according to Section 2.2.3.1.
As we can see from Table 2.3, the Q-factors determined by the phase response
Q Arg are approximately 30% lower as compared to Q Abs . This discrepancy
1. Single degree of freedom mechanical oscillator
44
indicates that the SDOF measurement method is not suitable for the studied
system. The Q-factor values are considerably smaller than the QBW-factor presented in Table 2.2. This can be explained by the fact that the SDOF Q-factor
method doesnt include wave propagation effects and therefore attributes
energy lost by wave propagation to system damping. The consequence is that
an adequate measurement method such as the QBW-factor must be applied to
allow a determination of the quality factor with sufficient accuracy. A good
agreement between the quality factor values determined from the magnitude
and the phase measurement can be rated as an indicator that the used measurement method is appropriate within experiments.
The influence of the inverse damping parameter value on the QBW-factor of
both the continuously and the discretely supported model is plotted in
Figure 2.18 for all three modes. In the case of the discretely supported model
only the modal damping parameters hs,i of the supported areas are modified,
whereas the damping of the unsupported parts is left unchanged.
800
mode 1
QBW-factor (-)
mode 2
600
400
mode 3
200
continuous support
discrete support
0
0
250
500
750
1/s,i
1000
1250
1500
Figure 2.18: Relation between modal damping parameter of the support hs,i
and QBW-factor
As in the model of a single beam on a viscoelastic support (see equation
(2.20)), the relation between the inverse damping parameter value 1/hs,i and
the QBW-factor is linear for all three modes in the case of the continuously supported model. The slope of the relating line is smaller in the case of mode 3 as
compared to the other two modes, which can be attributed to the way the
modal damping parameters were defined in equation (2.60).
Mode 1 of the model on discrete supports also shows a close to linear relation
between the inverse damping parameter and the QBW-factor in the examined
parameter range.
The QBW-factors of modes 2 and 3 of the discrete model are larger as compared
to the continuously supported model for large damping parameters hs,i (small
values of the inverse 1/hs,i) as would be expected from the fact that only 25%
45
a
of the model is subjected to this high damping. With decreasing support damping, the damping of the unsupported model regions become increasingly
important, which can be observed by the decreasing slope of the QBW-factor
curves.
Measurements carried out at a railway track on wooden sleepers (see
Chapter 3) gave QBW-factor values in the range between 100 and 500 for various modes. The results of model calculations presented in Figure 2.18 show
that in this parameter range the modal damping of the supported regions hs,i of
the discrete model must be set substantially higher than the overall damping of
the continuously supported model. This is illustrated by the following table,
where the damping parameters are summarized for three selected values of
QBW-factors for both higher modes.
Mode 2
Confg. B
QBW-factor
s,B2
[-]
Confg. C ________
s,C2 - s,B2
s,C2
s,C2
[-]
[%]
Mode 3
Confg. B
QBW-factor
s,B3
[-]
Confg. C ________
s,C3 - s,B3
s,C3
s,C3
[-]
[%]
180
2.80 10-3 9.19 10-3
69.6
180
2.06 10-3 6.40 10-3
67.7
340
1.48 10-3 3.92 10-3
26.6
340
1.09 10-3 2.51 10-3
56.4
500
1.01 10-3 2.02 10-3
50.2
500
7.44 10-4 1.11 10-3
33.0
____________
s,2(180) - s,2(500)
s,2(180)
[%]
63.9
78.0
____________
s,3(180) - s,3(500)
s,3(180)
[%]
63.9
82.7
Table 2.4: Relation between the modal damping parameters hs,i and the
QBW-factors of the continuously supported (Confg. B) and the discretely
supported model configuration (Confg. C)
To what extent this concentration of damping to the supports has an influence
on the global behavior of the system is to be studied next with special attention
paid to the local vibration zone close to the excitation.
In Figure 2.19 the absolute value of the displacement of beam 1 at the cutoff
frequency of mode 3 is plotted as a function of the distance from the excitation. The results for both the continuously and the discretely supported model
are shown for different values of the QBW-factor. The QBW-factor of the discrete model was adjusted by setting a high modal damping to the supported
areas in the same way as mentioned above and by leaving the damping of the
free regions unchanged.
46
a
4 109
|y1|
3 109
continuous support
discrete support
increasing QBW-factor
QBW = 480
QBW = 360
QBW = 240
QBW = 120
2 109
1 109
Figure 2.19: Absolute value of system response of mode 3 at beam 1 for

different values of the QBW-factor
Despite substantial differences in the distribution of system damping, the differences between the continuous and the discrete model are moderate, as can
be seen from Figure 2.19. Both models show the characteristic that the system
displacement is large only close to the excitation, for the assumed QBW-factor
values. This region of large vibration amplitude was called local vibration
zone earlier in this section. The extent of the local vibration zone gets smaller
with decreasing QBW-factor and the amplitude of the system response is
reduced. Lets define the measure Dz as
Dz
1
yi z = = yi ( z = 0 ) e
(2.75)
to quantitatively describe the size of the local vibration zone.

Figure 2.20 illustrates the relation between the QBW-factor and the size of the
local vibration zone Dz for modes 2 and 3 of the continuously and the discretely supported model.
2.4
z (m)
2.2
mode 2
2
1.8
mode 3
1.6
continuous support
discrete support
1.4
100
200
300
400
QBW-factor (-)
500
600
Figure 2.20: Relation between QBW-factor and the size of the

local vibration zone Dz
47
a
In the realistic range of QBW-factors the size of the local vibration zone Dz varies between about 1.7 to 2.3 m for mode 2 and between 1.4 to 1.9 m for mode
3. The relative change of the size of the local vibration zones is about 30%
although the damping parameter of the support hs,i is modified by about 65%
for the continuous model and by about 80% for the discrete model. This moderate influence of support damping on the spatial extent of the local vibration
zone is of benefit for the reliable application of these types of vibrations as a
detection principle. An open question is, if the expected size of the local vibration zone is small enough to detect individual train axles. The observation that
the local vibration zone with the same QBW-factor is smaller for mode 3 means
that higher vibration modes are more localized at the excitation.
In summary Figure 2.20 illustrates, that the spatial extent of the local vibration
zone decreases towards higher cutoff frequencies on the one hand and with
increasing damping on the other hand. As a small spatial extent is rated as an
advantage regarding the interaction with single train wheels, it seems to be
favorable to use vibration modes at high frequencies.
48
2.3
Wheel interaction
2.3.1
Model extension
Now lets consider a discontinuity in the transfer matrices of the discretely supported multiple mode model, representing a wheel in the distance zw of the
excitation as shown in Figure 2.21. Only the steady-state response of the system with the wheel standing in a certain distance from the excitation is considered here.
infinitely long segment
to the left
infinitely long segment

to the right
segment with wheel

zw
+M
+
z(R) +y
+F
+y z(L)
+F
+M
+
zwl
wheel
~
p
v0L
v0R
L1/2
L2
L1/2
0
zwl
L1/2:
Cw
L1/2 < zwl

L1/2+L2:
Dw
L1/2+L2 < zwl

L:
Cw
Tw
vR
Figure 2.21 Multiple modes model including the influence of a train wheel
In the configuration shown in Figure 2.21, the model can be divided into three
global segments: an infinitely long segment to the left of the excitation, a segment right of the excitation including the span with the wheel, which is again
followed by an infinitely long segment to the right.
The state vectors at the beginning of both infinitely long segments must fulfill
equation (2.74), in order to ensure that the solution stays bounded for
z ( L ) and z ( R ) , giving the relations (2.76) and (2.77) between the
(1) (1)
(2) (2)
forces and rotations v 0L , v R and the moments and displacements v 0L , v R of
the beams.
(2)
(1)
v 0L = Hv 0L
(2)
vR
(1)
= Hv R
(2.76)
(2.77)
It should be noted that the coordinate systems used to the right and to the left
49
a
of the excitation in Figure 2.21 are defined differently. This difference allows
the direct use of equation (2.76) for the model segment left of the excitation.
The transfer matrix of the section with the wheel Tw is depending on the wheel
position zwl. Lets consider the case shown in Figure 2.21, where the wheel is
standing on the first unsupported part of the section.
The transfer matrix of the part with the wheel Cw = Cw(zwl) can be written as
(cf. explanation of equation (2.70))
L1 z
L1 z
wl
wl
z wl 1 z wl 1
2
1 2
C w ( z wl ) = SC 1
C1
WC 1 C 1 S . (2.78)
2
2
2
2
The matrix W describes the wheel influence. Lets assume for the following
investigation
W =
diag ( c w, 0, 0 )
0
0
0
I
0
0
0
I
0
0
0
I
, (2.79)
where cw is the wheel impedance, which relates the displacement of beam 1 to

the force as a result of the wheel. Considering the small contact region
between the wheel and the rail, an influence on the beam rotation or the
applied moment is not assumed in equation (2.79).
The transfer matrix of the section with the wheel on the first unsupported part
becomes Cw(zwl)DC. The other cases with the wheel on the supported part or
on the second unsupported part are dealt with in a similar way.
With a number rw of basic sections before the section with the wheel, the transrw
fer matrix of the middle model segment becomes T w T , thus giving the relation (2.80) between vR and v0R.
(1)
v 0R
(2)
v 0R
rw 1 1
( T ) Tw
(1)
vR
(2)
vR
(1)
R 11 R 12 v R
R 21 R 22 v ( 2 )
R
(2.80)
As already mentioned in Section 2.2.2, all inverse transfer matrices can be

expressed analytically. Because of the inverse pairs of eigenvalues of T, the
numerical calculation of T-1 would cause numerical problems.
Keeping in mind that the beam rotations to the left and to the right of the excitation point have opposite signs and that the sum of the forces at each beam
equals the respective excitation force, the relation between the state vectors v0L
50
and v0R at the excitation point can be written in the form

2
(1)
v 0L
(2)
v 0L
(1)
v 0R
=
+ I 0
2
(
)
0 I v
0R
L
diag p i
EI i
,
diag ( 0 )
diag ( 0 )
diag ( 0 )
(2.81)
where pi is the excitation force at beam i.

By using the relations (2.76), (2.77) and (2.80), equation (2.81) can be rewritten in the form
2
R 11 + R 12 H
L
diag p i
EI i
(1)
v 0L
H ( R 21 + R 22 H ) v ( 1 )
R
diag ( 0 )
diag ( 0 )
diag ( 0 )
(2.82)
(1)
from which v R can be eliminated, giving

2
(1)
v 0L = ( I + ( R 11 + R 12 H ) ( R 21 + R 22 H ) H )
L
diag EI p i .
i
(2.83)
diag ( 0 )
This result is substituted into (2.76) and (2.80) to find the other components of
the state vectors v0L and v0R at the excitation point.
2.3.2
Discussion of the wheel influence
The model parameters introduced in Section 2.2.3 are used to numerically

study the changes of the steady-state system response to a discontinuity in the
periodic system representing a wheel. The QBW-factor using these values is
290 for mode 2 and 220 for mode 3. The wheel impedance is set to
2
c w = m w w i corresponding to a discontinuity caused by inertial effects of the
mass mw.
In Figure 2.22 the magnitude as well as the phase of the displacement
y i ( z = 0 ) in comparison with the corresponding values without wheel
y i ( z = 0 ) are plotted against the distance of the wheel from the excitation zw
for all three modes. Both the system response observed at beam 1 and at beam
2 are shown. The positions, where the wheel is standing at the middle of the
supported areas are marked by thin vertical lines.
51
52
|yi(z=0)|/|~yi(z=0)| (%)
Arg(yi(z=0))-Arg(~yi(z=0))
7.5
5.
2.5
0.
2.5
0.
0.05
0.1
0.15
0.2
7.5
0.
0.2
0.4
0.6
0.8
1.
20.
40.
60.
80.
100.
3.
beam 2
beam 1
beam 2
beam 1
(b)
2.
1.
0.
1.
2.
Distance from excitation zw (m)
Mode 2:
3.
0.
0.2
0.4
0.6
0.8
40.
60.
80.
100.
3.
beam 2
beam 1
beam 2
beam 1
(c)
2.
1.
0.
1.
2.
Mode 3:
Figure 2.22: Influence of passing train wheel on the magnitude and the phase of the
displacement yi at z = 0 m measured at beams 1 and 2 for all three modes.
2
(wheel impedance c w = m w w i , mw = 1 kg)
(a)
5.
beam 2
beam 1
0.3
0.25
beam 2
beam 1
85.
90.
95.
100.
105.
Mode 1:
3.
Figure 2.22 shows that the system response is the same at beam 1 and beam 2
for mode 1. The spatial range, at which the wheel causes a clear change of the
signal is about 5 m, which is about 7 sleeper spacings. The magnitude of the
signal has decreased to about 83% of y i ( z = 0 ), with the wheel exactly above
the excitation. With the wheel in a distance of about 3 m from the excitation,
the magnitude of the signal at both beams reaches a maximum and has
increased to about 104% of y i ( z = 0 ). The phase shift as a result of the passing wheel takes place synchronous with the change of the magnitude.
At mode 2, there is a discrepancy between the responses at beam 1 and
beam 2. The response at beam 1 shows a strong decrease in magnitude with
the wheel at the excitation and a maximum with the wheel standing at a distance of about 1 m. At beam 2, several oscillations of the signal as a function
of the wheel distance are observed. As stated in Section 2.2.3, the amplitude of
the vibration at beam 2 is small at mode 2, because beam 2 is almost a node of
the cross-sectional mode shape. The consequence of this observation regarding wheel detection is that the mounting location of the transducers is of great
importance in order to achieve a high signal quality. It can also be expected
that the position of the contact point between the wheel and the rail has a large
influence on the characteristics of the system response. By choosing a suitable
vibration mode it should be avoided that the wheel can run along a node of the
cross-sectional mode shape.
The system response at both beams for mode 3 shows a strong decrease in
amplitude while the wheel is standing above the excitation point and an
increase to a slight local maxima with a wheel distance of about 0.8 m. The
phase of the signal responds to the passing train wheel in a similar, but slightly
more complicated way.
The range at which the wheel shows a clear influence on the system response
is slightly smaller for mode 3. For both modes, the size of this range matches
approximately the size of the local vibration zone determined in Section 2.2.3
for the corresponding QBW-factors. With a typical distance between the axles
of a bogie of about 2 to 3 m the influence of the train wheel on the vibration
could be just local enough to enable a distinction of separate train wheels.
Both the magnitude or the phase of the receiver signal can be used to measure
the influence of passing wheels.
2
The wheel impedance was set to c w = m w w i with mw = 1 kg in Figure 2.22.
As already mentioned, this corresponds to a discontinuity caused by inertial
effects. Of course the wheel does not behave as a rigid body in the studied high
frequency range. This is taken into account by setting the mass mw considerably lower than the total mass of a train wheel. The calculated influences illustrated in Figure 2.22 show a similarity with the measured wheel influences
53
a
presented in Section 3.5 and therefore the simplistic model of the wheel interaction is regarded to be sufficient for the present purpose.
Figure 2.23 shows the influence of the mass parameter mw on the magnitude of
the displacement measured at beam 1 with the wheel standing at zw = 0 m.
|y1(z=0)|/|~y1(z=0)| (%)
100.
mode 2
mode 3
90.
80.
70.
60.
50.
40.
0.
0.2
0.4
0.6
mw (kg)
0.8
1.
Figure 2.23: Influence of the mass parameter mw on the magnitude of the

displacement y1(z=0) of beam 1 with the wheel standing at zw = 0 m.
In the studied value range, the mass parameter mw has a close to linear influence at mode 2. At mode 3 the influence on the magnitude of yi is more pronounced and not linear. The former can be attributed to the quadratic
2
frequency dependency of the chosen wheel impedance c w = m w w i .
An important parameter which must be expected to influence the wheel
impedance of the realistic situation is the axle load. According to Hertzian
contact theory, the contact area of the wheel and the rail increases with
increasing axle load, thus improving the dynamic interaction between the
wheel and the rail. This corresponds to increasing values of the wheel impedance function with rising axle load. It must be experimentally investigated,
whether the influence on forced rail vibrations caused by railroad vehicles
with small axle loads (i.e. load per wheel of approximately 5 kN) is sufficiently strong to allow wheel detection.
54
2.4
Generalization to the actual cross-section of the rail
Up until now, simplistic models have been developed which allowed a comprehensive study of the behavior of higher vibration modes in a waveguide
near their cutoff frequency. Using these models, it was shown that the cutoff
frequencies correspond to the vibration modes of the cross-section. The system response to an excitation at a certain cutoff frequency is dominated by the
vibration of the respective mode. The comparison between a continuously and
a discretely supported version of the model showed that the cutoff frequencies
and mode shapes of higher vibration modes are barely influenced by the type
of the support. It was further illustrated that the interaction between the vibration and a passing train wheel depends largely on the position, at which the
transducers are mounted and on the location of the contact point between the
wheel and the rail. Nodes of the cross-sectional vibration must be avoided.
This makes it necessary to choose a suitable vibration mode for the application
as a wheel sensor. The calculation of mode shapes and cutoff frequencies of
the actual rail are therefore the scope of this section.
2.4.1
Wave propagation in elastic rods of arbitrary cross-section
Consider an infinitely long waveguide with constant cross-section. Assuming

that a linear theory of wave motion is applicable and taking the z direction as
the direction of propagation, plane waves with angular frequency w and wave
number k of the form
u ( x, y, z, t ) = e
i ( kz + w t )
u ( x, y )
(2.84)
are sought, where u(x,y,z,t) is the displacement vector in the rod. Rosenfeld
and Keller [25] studied waves of this type in isotropic rods of arbitrary crosssection without damping. They obtained asymptotic expansions of the exact
solutions for both small and large values of the dimensionless wavenumber
(ka), where a is a typical dimension of the cross-section. In addition to four
modes propagating at low frequencies, referred to as the lowest longitudinal,
torsional and flexural modes, there is an infinite number of other modes. These
modes are referred to as higher modes throughout this work. Each of them has
a cutoff frequency w c, i below which it cannot propagate. At the cutoff frequency, the wavenumber is zero and the wavelength tends to infinity in the
absence of damping and supports. To analyze these higher modes near their
cutoff frequencies, Rosenfeld and Keller expanded the displacement vector
u ( x, y ) and the phase velocity in powers of (ka). By using these expansions in
the equations of motion and in the boundary condition, they obtained equations for the coefficients in the expansions. The modes were found to be of two
55
a
types. The first type shows a mainly longitudinal displacement. For the second
type the displacement is primarily transverse. There are infinitely many modes
of each type. Their cutoff frequencies w c, i are the eigenvalues of two different eigenvalue problems associated with the cross-section. Because of the fact
that a large in plane movement of the cross-section is desired in expectation of
a good interaction with passing train wheels and because of experimental considerations both regarding excitation and measurement, only transverse modes
are considered here. This corresponds to the eigenvalue problem of the crosssection subjected to plane strain.
2.4.2
Application to the rail
All experiments presented in Chapter 3 were carried out at a track with rails of
the swiss type SBB I. The geometry of this rail type is shown in Figure 2.24.
75.53
1:4
R6
R1
45
R13
R300
R80
14
65
145
R30
center of inertia
25
1:4
R1
R2
R6
R30
69.47
14
125
Figure 2.24: Cross-section of studied rail type SBB I, dimensions in mm

The higher vibration modes of this cross-section, if subjected to plane strain,
were calculated using a standard finite element code1.
Counting the torsional and the two flexural modes, which exist at low frequencies as discussed above, as the first three modes with transverse motion, the
first higher transverse vibration mode is called mode 4. Although not written
explicitly throughout the rest of this work, all discussed vibration modes are
transverse modes.
The results are presented in Figures 2.25 to 2.27.
1. SDRC I-DEAS
56
Mode 4:
1.46 kHz
Mode 5:
4.26 kHz
Mode 6:
6.24 kHz
Mode 7:
11.57 kHz
Mode 8:
11.62 kHz
Mode 9:
17.12 kHz
displacement
Figure 2.25: Calculated modeshapes of modes 4 to 9
57
a
Mode 11:
25.19 kHz
Mode 12:
26.69 kHz
Mode 13:
32.52 kHz
Mode 14:
35.70 kHz
Mode 15:
36.12 kHz
58
displacement
Mode 10:
24.19 kHz
Mode 16:
39.75 kHz
Mode 17:
41.35 kHz
Mode 18:
42.61 kHz
Mode 19:
44.67 kHz
Mode 20:
49.82 kHz
displacement
59
a
A total of 17 higher vibration modes were found in the frequency range up to
50 kHz for the studied rail.
The mode shapes shown in Figures 2.25 to 2.27, can be divided into modes
with asymmetric movement (e.g. modes 4 and 5) and modes with symmetric
movement (e.g. modes 6 and 7). The complexity of the mode shapes increases
toward higher frequencies. This can be observed at the railfoot in particular. At
modes 4 and 5 the railfoot basically acts as a rigid body. At mode 7 with a frequency of about 12 kHz, the deformation of the railfoot corresponds to half the
period of a sinus, which increases to 5 1/2 periods at mode 20 with a frequency
of 50 kHz. The same applies to the rail head, which approximately behaves as
a rigid body up to mode 12 with a frequency of 27 kHz.
From the knowledge obtained throughout the previous sections, modes with
large displacements and no vibration nodes at the rail head are expected to be
the most promising regarding their interaction with train wheels. From the calculated mode shapes of Figures 2.25 to 2.27 up to 50 kHz, modes 7, 17 and 18
seem to meet these requirements most obviously. Also, mode 6 shows a synchronous movement of the railhead, which could be suitable for wheel detection, although the displacement of the railfoot predominates the overall
vibration at this mode. Of course, these statements must be experimentally
verified.
As mentioned above, all experiments were carried out at a test track with a
cross-section of type SBB I, which is why the numerical calculations were performed at this rail type. A widely used rail type throughout Europe is the socalled UIC 60 (see [6]), which closely corresponds to the Swiss rail type
SBB VI. With an overall height of 172 mm and a width of the rail head and the
railfoot of 72 mm and 150 mm, respectively, the geometric dimensions of this
rail type are considerably larger than for the studied type SBB I, whereas the
overall shape of the cross-section is similar. Therefore it can be expected that
the results obtained throughout this work in general also apply to the situation
found at the UIC 60 rail, although the cutoff frequencies are shifted toward
lower values and the sequence of the mode shapes and their detailed displacement behavior are slightly different.
60
2.5
Conclusions from theoretical considerations
Using the multiple mode model, it was shown that the amplitude maxima of
the system response of the complete rail system including the support on
sleepers and ballast are to be found close to the cutoff frequencies of the rail as
a waveguide itself. The influence of support parameters is expected to be moderate with regard to the cutoff frequencies and mode shapes of higher modes.
The cutoff frequencies and mode shapes at these frequencies correspond to the
eigenvalue problem of the cross-section of the rail if subjected to plane strain.
This allowed the numerical calculation of the cutoff frequencies and mode
shapes of the studied rail type SBB I.
Along with amplitude maxima at the cutoff frequencies, a second type of resonance mechanism originating from the discrete support on sleepers is found at
railway tracks. The study of the multiple mode model revealed that these additional amplitude maxima can occur at frequencies close to the cutoff frequencies of the waveguide. In these cases the cross-sectional movement looks
similar to the mode shape of the adjacent cutoff frequency. However, the spatial behavior of the system response at these frequencies is completely different. The amplitude decreases at the sleepers and builds up again in the
following unsupported section. The overall decay along the waveguide is
moderate. Because of their lack of a local vibration zone these amplitude maxima are not considered to be suitable for the application as a rail contact. At
these frequencies one wheel influences the vibration several times. Furthermore the exact location of these frequencies depends strongly on support
parameters, such as sleeper spacing, which can vary in practice. This fact
would make it difficult for the sensor to reliably find specific vibration modes
of this kind suitable for wheel detection.
With an excitation at a cutoff frequency of a specific mode, the amplitude of
the system response is only limited by system damping. The multiple mode
model allowed a straightforward implementation of modal damping. As the
behavior of individual modes close to their cutoff frequencies is of interest
within this investigation, the concept of modal damping is particularly suitable. A commonly used experimental technique to characterize the damping
behavior of a resonating system is the quality factor or Q-factor. However, it
was shown with the help of a Bernoulli-Euler beam on a viscoelastic foundation that this measurement method must be adopted in order to give good
results for a system which includes wave propagation. This leads to the definition of the QBW-factor for the studied beam. As the multiple mode model consists of several elastically connected Bernoulli-Euler beams, the QBW-factor
proved to be a method suitable to characterize its damping at each mode. At
the same QBW-factor, the continuously and the discretely supported model
61
a
show a similar behavior both regarding the amplitude of the system response
and the spatial extent of the local vibration zone, despite of considerable differences in the distribution of damping of these models. This independence
from the physical distribution of the track damping, which is expected to be
concentrated at the sleepers in reality, is rated as an advantage for using the
local vibration zone as a wheel detection principle. The actual value of the
QBW-factor is an important measure regarding the spatial extent of the local
vibration zone and the reaction time of the sensor. An important question in
this context to be investigated by experiments is, which measurement method,
the Q-factor or the QBW-factor, is suited best to characterize the damping of
the real track.
Calculations using the multiple mode model gave a spatial extent of the local
vibration zone of about 1.4 - 2.3 m. Within this zone, the beam vibration is
dominated by the mode shape of the mode, which is actuated at its cutoff frequency. With increasing distance from the excitation, the contributions of
lower modes to the system displacement become increasingly relevant, as
these modes can travel with moderate attenuation according to their frequency
spectra. This effect and the actual size of the local vibration zone will be investigated by experiments within Chapter 3. If the calculated spatial extent of
about 1.4 - 2.3 m would prove to be realistic and with a distance between the
wheel sets of a bogie of approximately 2 - 3 m it is to be expected that the
vibration is not localized enough to allow a reliable detection of separate axles.
Of course it must be experimentally confirmed, that the general statements
derived from the multiple mode model are valid for a real railway track. The
verification of the theoretical observations by comparison with experimental
results is performed in the following Chapter 3.
62
Chapter 3
Experiments
The forced vibration response of a waveguide including higher modes was
studied using mechanical models in Chapter 2. In Section 2.2 a model was
chosen which could be treated analytically and thus allowed an examination of
the overall scheme of higher mode behavior in a waveguide. The general statements derived from this model need to be verified with the help of an experimental investigation as pointed out in the concluding Section 2.5 of the
theoretical considerations. In order to pursue this task, a number of experiments where carried out at a test range with a track on wooden sleepers. The
experimental setup used is described in Section 3.1.
The first objective of the experimental investigation is the verification that the
numerically predicted modes can be excited and measured. This is done in
Section 3.2 by performing a modal analysis at the test track and by comparison
with the numerical results.
The spatial extent of the local vibration zone and the system response at some
distance from the excitation transducer are examined in Section 3.3.
The theoretical investigations performed in Chapter 2 showed that damping
has an important influence on the system response of a waveguide which is
excited at one of its cutoff frequencies. Damping measurements at a number of
cutoff frequencies of the rail were performed using various methods. The
results are presented and discussed in Section 3.4.
The actual influence of passing wheels on the forced rail vibration was experimentally studied using a movable bogie. Section 3.5 contains the corresponding results together with a preliminary discussion.
63
3.1
Experimental setup
All measurements presented throughout this chapter were performed at a test

range with rails of type SBB I (see Figure 2.24 for geometry details) laid on
wooden sleepers and ballast. The sleeper spacing was between 0.65 m and
0.70 m.
Piezoelectric transducers1 were used for vibration excitation and measurement. A circular inertial mass (stainless steel, 10 mm length, 20 mm diameter)
was glued to each piezoelectric disc in order to achieve an amplitude amplification, giving a first transducer resonance of approximately 130 kHz2, which
is well above the studied frequency range. The transducers were glued3 to the
rail. A sending transducer mounted at the center of the railfoot and a receiver
at a distance of 70 mm were used to study symmetric modes. To examine
asymmetric modes, the sending transducer was attached to the side of the rail
head and the receiver was fixed to the web of rail according to Figure 3.1.
wheel 2
2.7
wheel 1
position
measurement
receiving transducer 1
head excitation
receiving transducer 2
foot excitation
-0.33
0.685
0
0.66
0.33
0.675
0.675
Figure 3.1: Schematics of the experimental setup

Transducers and appropriate fixation devices have been designed for the excitation and measurement of rail vibrations in the course of this project. However, to eliminate the influence of the inertia of these transducers and
mountings, all experiments presented in this chapter were carried out using the
configuration described above.
The influence of train wheels was studied using a bogie equipped with an electrical axle drive. The axle base of the bogie was 2.7 m. The weight of the bogie
was approximately 25 kN giving a load per wheel of 6.25 kN. The weight per
wheel of the partially loaded freight car used during the feasibility study is
estimated to 40 kN.
1. Ferroperm Piezoceramics, ceramic type PZ 26, 20x1 mm circular disc
2. First longitudinal mode of the inertial mass in the clamped-free configuration
3. Permabond double bubble fast cure epoxy
64
Experiments
3.2
Frequency spectra and mode shapes of the rail
3.2.1
Presentation of the results
The frequency spectra and the mode shapes of the rail at their cutoff frequencies were experimentally examined up to the frequency limit of 50 kHz. A
scanning vibrometer1 based on a heterodyne laser interferometer was used to
measure the velocity at the surface of the rail. The laser beam is deflected
using two built-in mirrors, thus allowing the measurement at each point of a
defined grid. The recorded data is processed using the software of the scanning vibrometer. The mode shapes are calculated at user defined bands by performing a standard modal analysis based on the Fast Fourier Transform
(FFT). The results are graphically presented.
Measurements of the mode shapes on top of the rail head, at the side of the rail
head, at the web of the rail and on top of the railfoot were performed. The
experimental setup showing both positions of the scanning vibrometer and the
used measurement equipment is illustrated in Figure 3.2.
2
3
1
3
2
(c)
(b)
Other components:
1: Rail type SBB I
8: Krohn-Hite KH 7500 amplifier
2: Movable bogie
9: LeCroy 9350AL oscilloscope
Scanning Vibrometer:
10: SR850 DSP lockin amplifier
3: Laser head
11: PC running LabView
4: Video control box
5: HP 33120A function generator
4
6: OFV3001S laser demodulator
7: Scanning Vibrometer PC
5
11
7
9
8
10
(a)
Figure 3.2: Experimental setup: equipment (a),

measurement on top of the rail head and on top of the railfoot (b) and
measurement at the side of the rail head and at the web of the rail (c)
1. Polytec Scanning Vibrometer PSV-200
65
a
A periodic chirp signal was used to excite the rail. The total frequency range
was divided into intervals of 0 - 20 kHz, 20 - 30 kHz, 30 - 40 kHz and
40 - 50 kHz in order to have sufficient signal energy at each interval. The periodic chirp signal was downloaded to the function generator and repeated continuously. A sampling frequency of 128 kHz was used to record the velocity
output of the laser demodulator and the voltage of the excitation signal. The
time signals were sampled continuously in arrays of 8192 data points. The
duration of the periodic chirp signal exactly matched the sampling period
(0.064 s) and the data acquisition was synchronized with the excitation using
the trigger output of the function generator. The frequency resolution of the
measurement is 15.6 Hz and the highest detectable frequency is 64 kHz.
The transducers mounted beneath the railfoot and at the side of the rail head
were used to measure symmetric and asymmetric modes, respectively. The
used measurement grid for both configurations is shown in Figure 3.3.
(a)
(b)
Figure 3.3: Measurement grid used with excitation at the railfoot (a) and with
excitation at the rail head (b). The following views are shown:
top view of rail head (top picture), side view of rail head and web of the rail
(middle picture) and top view of railfoot (bottom picture)
The rail looks asymmetric in the top views of Figure 3.3. This optical effect is
caused by the fact, that the laser head of the scanning vibrometer is not positioned exactly above the running surface (see Figure 3.2 (b)).
To give an idea of the complexity of the system response, the absolute value
and the argument of a measured transfer function is shown in Figure 3.4. The
transfer function has been recorded between the voltage of the excitation signal and the velocity of a point on the running surface. All numerically calculated cutoff frequencies (see Section 2.4.2) are marked as stroke-dotted lines
in Figure 3.4. At each frequency sub range, a total of 500 measurements were
recorded and averaged in the frequency domain. For better clarity of the phase
66
Experiments
a
plot, phase jumps of 2p have been eliminated.
|G0| dB(m/s/V)
0.
20.
40.
60.
80.
0.
Arg(G0)
25.
50.
75.
100.
125.
150.
0.
10000.
20000.
30000.
Frequency (Hz)
40000.
50000.
Figure 3.4: Transfer function between voltage of excitation signal and velocity
of point on top of rail head above excitation with foot excitation.
Calculated cutoff frequencies are marked as stroke-dotted lines
_
|G 0| dB(m/s/V)
10.
foot excitation
20.
30.
40.
50.
_
|G 0| dB(m/s/V)
60.
10.
head excitation
20.
30.
40.
50.
60.
0.
10000.
20000.
30000.
Frequency (Hz)
40000.
50000.
Figure 3.5: Average spectrum over all points of measurement grid with foot
excitation (890 pts) and head excitation (933 pts).
Calculated cutoff frequencies are marked as stroke-dotted lines
67
a
Acting as high pass filters, the used transducers produce a sufficient excitation
amplitude only above approximately 4 kHz. This explains the high noise of the
transfer function below this frequency limit in Figure 3.4. Although some distinct amplitude peaks can be observed starting with a prominent peak at the
cutoff frequency of mode 6, it appears to be impossible to determine the cutoff
frequencies of the other modes by a single point measurement because of the
high density of amplitude maxima. The situation can be largely improved by
calculating the average spectrum of all measurement points. In Figure 3.5 the
average spectrum above 4 kHz is plotted for both excitation types.
Although the amplitude maxima still outnumber the calculated cutoff frequencies, the number of amplitude maxima visible in both average spectra of
Figure 3.5 has decreased as compared to the single point measurement presented in Figure 3.4.
By looking at the mode shapes at the peaks in the averaged spectra close to the
numerically predicted cutoff frequencies, the modes could be matched with
the calculated ones. Except for mode 16 all calculated higher modes in the frequency range from 5 kHz to 50 kHz could be located. The measured mode
shapes of six modes are presented in Figures 3.6 and 3.7. These modes have
been selected because they cover the complete range of observed effects. The
mode shapes of the remaining modes can be found in Appendix B.
The calculated and experimentally determined cutoff frequencies and their relative deviations are summarized in the following table.
Mode
Calculated Measured
[kHz]
[kHz]
relative
deviation
[%]
Mode
Calculated Measured
[kHz]
[kHz]
relative
deviation
[%]
6.24
6.20
0.6
11.62
11.42
1.7
11.57
11.58
0.1
17.12
16.98
0.8
10
24.19
24.11
0.3
11
25.19
24.94
1.0
12
26.69
27.64
3.6
14
35.70
35.77
0.2
13
32.52
31.69
2.6
16
39.75
15
36.12
36.17
0.1
17
41.35
41.47
0.3
18
42.61
42.69
0.2
19
44.67
45.06
0.9
20
49.82
49.20
1.2
(a)
(b)
Table 3.1: Experimentally determined cutoff frequencies,
(a): symmetric modes with foot excitation,
(b): asymmetric modes with head excitation
68
Experiments
Mode 7, foot excitation, 11.58 kHz:
Mode 8, head excitation, 11.42 kHz:
Figure 3.6: Calculated (boxed pictures) and measured mode shapes of modes 6, 7 and 8 (scaling on the right hand side)
69
70
Figure 3.7: Calculated (boxed pictures) and measured mode shapes of modes 10, 17 and 18 (scaling on the right hand side)
Experiments
3.2.2
Discussion of the results
Except for modes 4, 5 and 16, all expected modes up to 50 kHz were recognized by comparison of their modeshapes with the numerically calculated
ones. All modes show a synchronous motion according to their respective
mode shape up to a distance of about 0.1 m to 0.5 m from the excitation. The
vibration amplitudes mostly decreases at the sleepers. A more detailed study
of the spatial behavior of the modes in greater distance from the excitation is
found in Section 3.3.
The calculated and the experimentally determined cutoff frequencies agree
within 4% and mostly even better than 1%. This confirms the statement
derived in Section 2.2.3 that the type of support has a marginal influence on
the cutoff frequencies of higher vibration modes.
The cutoff frequencies of modes 7 and 8 are very close to each other. The
experiments showed that for the studied rail, the cutoff frequency of mode 8 is
slightly lower than of mode 7, while the numerical calculation gave the reverse
order.
The missing of modes 4 and 5 is attributed to the high pass characteristics of
the transducers used, which did not produce sufficient excitation amplitude
below 5 kHz. The missing of the asymmetric mode 16 can be explained by the
unfavorable position of the exciting transducer, as the motion of the rail head
has a small amplitude in mainly vertical direction for this mode.
The following practical consequences are derived from discussed mode shape
measurements regarding a sensor which relies on the excitation of a specific
vibration mode:
It appeared that the high density of amplitude peaks observed at a single measurement point makes it necessary to have multiple receiver signals in order to
be able to distinguish between different vibration modes for an unknown rail
geometry. These transducers should be mounted advantageously at several
positions along the rail contour in order to find the characteristic motion of a
certain vibration mode.
71
3.3
3.3.1
Examination of the spatial decay

Measurements of the vibration on top of the rail head of three adjoining railsleeper sections were performed using the scanning vibrometer in order to
study the spatial behavior of the vibration. The results at these three measurement subsections were joined to give a continuous picture of the vibration of
each mode. The following figure shows the measurement grids used.
Figure 3.8: Measurement grids on the running surface over a distance of three
adjoining rail-sleeper sections, excitation at (z = 0 m)
The actual measured areas had an overlapping of about 0.1 m - 0.2 m and the
comparison of the vibration in this region allowed the determination of the
correct phase relation between the three sections. For mode 6 and at the transition from the excited span to the second span of mode 7 the phase relation
could not be determined for sure because of the small vibration amplitudes at
the sleepers.
The rail was excited in the same way as described in the previous section,
using a broad band excitation in four frequency sub-ranges. The excitation is
located at z = 0 m.
The results of the selected modes 6, 7, 8, 10, 17 and 18 are presented in
Figures 3.9 and 3.10. The plots of the remaining modes can be found in
Appendix C. It should be noted in the figures that the value ranges of each
measurement subsection are different in favour of resolution.
72
Mode 8, head excitation, 11.42 kHz
Mode 7, foot excitation, 11.63 kHz
Figure 3.9: Spatial decay of modes 6, 7 and 8, excitation at z = 0 m
a
Experiments
73
74
Figure 3.10: Spatial decay of modes 10, 17 and 18, excitation at z = 0 m
Experiments
3.3.2
Two basic phenomenons can be observed by looking at the vibration over a

range of three sleeper spacings. At the first group of modes a local vibration
zone with an extent of one sleeper spacing or less is observed. Beyond this
region a wave is traveling away from the excitation. Traveling waves can be
distinguished from vibrations by looking at animations of the measured mode
shapes. The observation of a local vibration zone and a traveling wave beyond
that region corresponds basically to the predictions made for modes at their
cutoff frequencies in Chapter 2. The measured mode shapes of modes 8 and 17
(and modes 9, 11 and 20 shown in Appendix C) show this characteristic especially pronounced.
The second group of modes shows a vibration corresponding to the respective
mode shape along the whole examined area. The amplitude of the vibration
usually decreases at the sleepers but builds up again in the following unsupported section of the rail. This type of system response was observed at the
additional resonances of the multiple mode model caused by the discrete support. Modes 6 and 7 (and mode 13 shown in Appendix C) show this behavior.
At mode 10 a combination of both behaviors is observed, as it shows a vibration corresponding to the mode shape in the excited and its adjoined span. A
traveling wave overlays this vibration.
The motion of the rail head caused by traveling waves in Figures 3.9, 3.10 and
C.1 to C.3 often does not correspond to the mode shape of the respective
modes. This characteristic was also predicted by the multiple mode model. It
originates from the fact that the mode which is excited at its cutoff frequency
predominates the system response only close to the excitation whereas other
wave modes, which can travel with comparably low attenuation according to
their frequency spectra, become increasingly important at greater distances
from the excitation.
The head movement at the wave propagation zone of modes 9, 10, 11 and 17
show a certain similarity. This leads to the assumption that the observed waves
belong to the same lower mode. The wavelength of this wave mode decreases
with larger frequency from approximately 0.15 m to 0.10 m. This decrease in
wavelength is in agreement with wave propagation theory. From the observed
asymmetric motion at the rail head it could be attributed to mode 4 or mode 5.
However, at this stage this assignment is purely speculative and additional
research would be necessary for verification.
75
3.4
Modal damping of the rail
3.4.1
The Q-factors of several modes at their cutoff frequencies have been experimentally determined by the building-up and decay characteristics and by the
magnitude and phase response. Of special interest in this section is the question which measurement method, the classical Q-factor or the QBW-factor as
discussed in Section 2.1.2, is more appropriate for the examined system. This
issue was investigated by measuring the quality factors according to different
experimental methods and by comparing the results.
As already mentioned, the building-up and decay characteristics of the forced
vibration response at the cutoff frequency of a specific mode have been
recorded. Figure 3.11 shows the amplitude maxima of the system response at
each vibration cycle in the example of mode 6. A least-square fit of an exponential function to this set of experimental data was performed. The Q-factor
could be calculated according to (2.32) from the time constants t of the building-up and the decaying envelopes giving the values Qb and Qd in Table 3.3.
As stated in Section 2.1.2.2, the time constant t is the reaction time after
which the magnitude of the free vibration response has dropped to 1 e its initial value if the excitation is switched off. The exponential building-up of the
forced vibration amplitude takes place in a similar way. The relation between
the time constant t and the Q-factor is t = 2Q/w = 2Q/(2pf). It should be
noted that the definition of the Q-factor and the QBW-factor based on the time
constant of the system are exactly the same.
amplitude envelope
0.8
decay
envelope
building-up
envelope
0.6
0.4
0.2
0.
0.01
0.02
0.03
Time (s)
0.04
0.16
0.17
0.18
0.19
Time (s)
Figure 3.11: Building-up and decay envelope of mode 6
76
0.2
0.21
Experiments
On the other hand, the transfer function be(w) of several modes between the
sending and the receiving transducer has been measured by stepwise changing
the excitation frequency close to the cutoff frequencies. The magnitude and
phase response has been recorded using a lock-in amplifier.
The following Figures 3.12 to 3.14 show the measured normalized magnitude
and phase responses of several modes.
A least-square fit of the experimental data has been performed in order to estimate the parameters wc and a1 to a6 of the approximation functions (3.1) and
(3.2), which are also shown in Figures 3.12 to 3.14.
b e ( w ) Fit =
a1
2
( wc
2 2
w ) + a2
(3.1)
+ a3
a5
Arg ( b e ( w ) ) Fit = a 4 Arctan 2
+ a6
w c w 2
(3.2)
Using these approximations, the Q-factor and the QBW-factor have been determined by finding the frequencies, where the magnitude and phase have
changed by 1 2 and p 4 or by 1 8 8 and 3p 16 as compared to
the values at the cutoff frequency.
Mode 6
1.
Mode 7
0.8
|e|
0.6
0.4
0.2
3.
2.
Arg(e)
1.
0.
1.
2.
3.
6100
6150
6200
6250
Frequency (Hz)
6300
11540
11560
11580
11600
Frequency (Hz)
Figure 3.12: Normalized magnitude and phase response of modes 6 and 7
77
Mode 8
1.
Mode 9
0.8
|e|
0.6
0.4
0.2
3.
2.
Arg(e)
1.
0.
1.
2.
3.
11250 11275 11300 11325 11350 11375
16825 16850 16875 16900 16925 16950 16975
Frequency (Hz)
Frequency (Hz)
Mode 10
1.
Mode 18
0.8
|e|
0.6
0.4
0.2
3.
2.
Arg(e)
1.
0.
1.
2.
3.
6100
24050
24100
24150
Frequency (Hz)
24200
42500
42600
42700
42800
Frequency (Hz)
78
Experiments
A summary of Q-factors and QBW-factors as determined according to the previous descriptions and their relative differences can be found in Table 3.2.
Additionally, the table contains the total phase change Dj as estimated by the
least square fit of the approximation function (3.2).
Mode
f
[kHz]
Dj
Q Abs
Q Arg
[]
[-]
[-]
Q Abs Q Arg
-------------------------------Q
[%]
BW
BW
Q Abs
BW
Q Arg
[-]
[-]
BW
Q Abs Q Arg
-------------------------------BW
Q
[%]
6.20
189
155.1
156.3
0.8
185.1
233.3
23.0
11.57
136
325.2
233.4
32.9
421.5
403.0
4.5
11.31
171
256.6
247.3
3.7
306.4
376.3
20.5
16.89
186
377.9
376.5
0.4
456.2
558.9
20.2
10
24.13
153
418.3
333.6
22.5
505.5
531.8
5.1
18
42.58
142
156.0
106.3
37.9
193.5
192.4
0.6
Table 3.2: Quality factors from the magnitude and phase response of several
modes and their relative difference
The quality factors Q b and Q d determined by the building-up and decay characteristics of four modes are presented in Table 3.3.
Mode
f
[kHz]
Qb
Qd
[-]
[-]
BW
Qb Qd
--------------------Q b ,d
2 Q Q b ,d
--------------------------Q + Q b ,d
2 Q Q b ,d
---------------------------------BW
Q + Q b ,d
[%]
[%]
[%]
6.20
209.3
205.2
2.0
28.0
0.9
11.57
475.8
493.4
3.6
53.8
16.1
10
24.13
374.9
365.6
2.5
1.5
33.4
18
42.58
188.3
182.6
3.1
34.3
4.0
Table 3.3: Quality factors from building-up and decay characteristics and
comparison with values from Table 3.2
Tables 3.2 and 3.3 contain all measurements performed within this experimental study.
79
3.4.2
Especially the transfer functions of modes 6, 7 and 18 differ from an ideal resonance behavior both regarding symmetry and smoothness of the signal.
These differences cannot be attributed to measurement errors, as the standard
deviation of several measurements was always below one percent. Also a nonlinearity of the amplitude response could be excluded by repeated measurements using different amplifications of the excitation voltage. Concluding
from the mode shape measurements, it is assumed for modes 6 and 7 that the
cutoff frequency and an additional resonance frequency originating from the
discrete support are very close to each other. This circumstance could cause
the supplemental peaks of modes 6 and 18 and is suspected to be the reason for
the atypical shape of these transfer functions.
The values of the total phase change Dj cover the whole range between 180
and 135. A phase change of 180 corresponds to the phase change of a single
degree of freedom mechanical oscillator whereas 135 = 3p/4 was found in
the case of a Bernoulli-Euler beam on a viscoelastic foundation. The reason
why these two values were found to limit the observed range of phase changes
is not understood and further investigation is needed. In Section 2.2.3 a phase
change of the receptance of less than p/4 was observed for a resonance of the
discretely supported model. None of the experimental determined transfer
functions showed an equally small phase change. This is astonishing especially for modes 6 and 7 which are predominated by effects from the discrete
support according to Figure 3.9. Additional research on the phase behavior of
periodic systems would be necessary to give a conclusive explanation of this
issue. All experimentally observed phase changes lie in the range where the
measurement methods of the Q-factor and the QBW-factor can be applied. The
results obtained for the Q-factor and QBW-factor are discussed in the following
sections.
The Q-factors Q Abs and Q Arg from the magnitude and phase response give
similar values within 4% in the cases of modes 6, 8 and 9. These are the modes
with total phase changes close to 180. At modes 7, 10 and 18 with total phase
BW
BW
changes closer to 135, the QBW-factors Q Abs and Q Arg match within approximately 5%.
A good correspondence between the values from the magnitude and the phase
response is rated as an indicator which value, the Q-factor or the QBW-factor,
should be used to characterize system damping. The Q-factors Q b and Q d
determined by the building-up and decay characteristics are used to verify this
statement. The relative difference between Q b and Q d is smaller than 4% for
all modes according to Table 3.3.
BW
A comparison of the mean value Q b ,d with the mean values Q and Q
gives
80
Experiments
a contradictory view on which quality factors are the most dependable. At
modes 7 and 18, where a good agreement between the values from the magnitude and phase response consider the QBW-factor to be more reliable, the Q b ,d
BW
value is indeed closer to Q . On the other hand, the Q b ,d value of mode 6 is
BW
closer to Q
and the Q b ,d value of mode 10 is closer to Q , although the
agreement in magnitude and phase response supposes the respective other values to be more reliable. The reason of this unexpected observation is unclear
and means that no definitive decision can be made which of the measured
quality factor values is the most trustworthy for each mode. Considering the
complexity of the studied system especially in the examined high frequency
range, an uncertainty for the damping values of about 30% - 40% seems to be
reasonable at the present time.
The time constant t, which is directly related to the quality factor as mentioned above, characterizes the reaction time of the vibrating system at a certain cutoff frequency. For the experimentally studied modes presented in the
4
previous section, the time constants are located between 8 10 s for
2
Q = 106.3 of mode 18 and 1.4 10 s in the case of Q = 493.4 for mode 7.
2
A train wheel traveling at the velocity of 140 km/h needs 1.5 10 s to pass the
distance of 0.6 m. This distance corresponds to the typical sleeper spacing of
Swiss railway tracks and equals approximately the extent of the local vibration
zone of a number of modes as observed in Section 3.3. The passing time of
2
2
1.5 10 s is close to the time constant of 1.4 10 s . This means that the reaction time of the sensor to passing train wheels could be insufficient, depending
on the Q-factor of the used vibration mode. As discussed in Section 2.2.3, the
main source of damping is attributed to the support on sleepers and ballast.
The parameters of the support must be expected to diverge in a broad bandwidth due to variances of the track-laying process and due to environmental
influences like humidity or temperature. Therefore, the lack of a reserve
regarding the observed time constants must be rated as an argument against the
use of forced vibrations as a reliable detection principle. It must also be considered that the electronic device necessary to obtain the magnitude and phase
of the receiver signal has a time lag itself. This further complicates the situation regarding reliable detection of changes of the receiver signal at high rates.
81
3.5
Influence of passing train wheels
3.5.1
In this section, the reaction of the receiver signal is studied at several modes
while the bogie passes the transducer location. A lock-in amplifier was used to
obtain the transfer function be (see Section 3.4.1). The magnitude and the
phase of the transfer function be are plotted as a function of the distance of
wheel 1 from the excitation in Figures 3.15 to 3.20. The time constant of the
lock-in amplifier was set to 100 ms. A good reproducibility of the measurements was achieved using this high value of the time constant but also a filtering effect of fast signal changes must be accepted. The measured values were
transferred through a GPIB interface (IEEE 488) from the lock-in amplifier to
the evaluating computer1, where they were combined with the readout from
the position measurement. The positions, where wheel 1 and 2 pass the measurement location are marked as stroke-dotted lines in the figures.
The bogie was moved back and forth five times for each measurement. The
curves of the mean value and the standard deviation are shown in the plots
additionally to the actual measurement points. Due to a small ascent of the
track, the velocity traveling to the right hand side was 0.65 m/s (2.3 km/h)
compared with 0.67 m/s (2.4 km/h) in the other direction.
1.
|e|
0.8
0.6
0.4
Arg(e)
0.2
0.8
0.6
0.4
0.2
0.
-0.2
-0.4
0.
1.
2.
3.
Distance of wheel 1 from excitation (m)
0.
1.
2.
3.
Figure 3.15: Wheel influence on the transfer function be of mode 6 (6.19 kHz)
1. Personal Computer running National Instruments LabView
82
Experiments
1.
|e|
0.8
0.6
0.4
0.2
Arg(e)
0.2
0.1
0.
0.1
0.2
0.
1.
2.
3.
0.
1.
2.
3.
Figure 3.16: Wheel influence on the transfer function be of mode 7

(11.64 kHz)
1.
|e|
0.8
0.6
0.4
0.2
Arg(e)
0.2
0.1
0.
-0.1
-0.2
0.
1.
2.
3.
0.
1.
2.
3.

(11.30 kHz)
83
1.
|e|
0.8
0.6
0.4
0.2
Arg(e)
0.4
0.3
0.2
0.1
0.
0.
1.
2.
3.
0.
1.
2.
3.

(24.12 kHz)
1.
|e|
0.8
0.6
0.4
0.2
Arg(e)
-2.5
-2.6
-2.7
-2.8
-2.9
0.
1.
2.
3.
0.
1.
2.
3.

(41.90 kHz)
84
Experiments
1.
|e|
0.8
0.6
0.4
0.2
Arg(e)
0.2
0.1
0.
-0.1
-0.2
0.
1.
2.
3.
0.
1.
2.
3.

(42.28 kHz)
3.5.2
The most distinct reaction of the receiver signal to both wheels can be
observed at mode 6. The magnitude is reduced by at least 30% by the influence
of wheel 1 and by approximately 50% in the case of wheel 2. The phase of the
receiver signal shifts essentially synchronous to the magnitude change. At
mode 6 additional smaller reactions of the signal can be observed with the
wheels in a distance of approximately 0.8 m to 1 m from the excitation. It is
assumed that this reaction is caused by the wheel standing on the vibration of
neighboring span which was observed for this mode in Section 3.2. Although
the axle base of the bogie is equivalent to about 4 sleeper spacings, it seems
that there is an interdependence between the influences caused by both wheels.
This statement more or less applies to all studied modes as can be seen from
the fact that the receiver signal is not constant between the two wheels.
Except for a delay as compared to the wheel position, the direction of the
bogie only has a marginal influence on the signal of mode 6. This delay can be
observed at all acquired signals in Figures 3.15 to 3.20. The cause is not the
slow reaction of the mechanical rail vibration, but the time lag introduced by
the time constant of the lock-in amplifier and by the slow communication rate
85
a
through the GPIB interface. The time lag within the lock-in amplifier of 0.1 s
together with the time increment between two data points of about 0.11 s allow
the bogie to move approximately 0.14 m, before a clear reaction of the receiver
signal can be expected. This corresponds roughly to the delay seen in the figures.
As already stated above, the influence of the wheels at mode 6 differs mainly
in the intensity of the reaction. At higher modes, also the shape of the measured signals varies considerably depending on which wheel passes the sensor.
Additionally, the direction of the bogie appears to have an important influence
on the measured signal with an increasing tendency towards higher modes.
These observations can be explained by differences in the contact point of
each wheel depending on the direction. In agreement with the theoretical considerations of Chapter 2, the location of the contact point gets increasingly
important towards higher frequencies, since the complexity of the mode
shapes rises. Furthermore it is assumed that the axle load of the bogie is
unequally distributed due to geometry faults of the chassis, causing the differences between the two wheels at modes 6 and 7.
The mode shapes of modes 6 and 7 (see Section 2.4.2 and Section 3.2) have a
certain similarity. Both show a synchronous movement of the rail head and
therefore should be suited well for wheel detection. Still the signal change is
clearer in the case of mode 6. The reason for this difference is not yet understood. It could originate in a frequency dependence in the wheel/rail contact
which causes a smaller interaction at higher frequencies. Regarding the application as a sensor, both of these modes are unsuited because of their frequencies in the audible range, which cause a considerable noise pollution.
Overall it must be stated that all the measured influences were significantly
smaller as compared to the results obtained during the feasibility study mentioned in Chapter 1. At that time, a complete freight car was used for the
experiments. The approximately six times lower axle load of the bogie is held
responsible for the decrease in signal quality. Also the axle base of approximately 10 m of the freight car prevented a simultaneous influence of both
wheels on the receiver signal. Although the unloaded bogie can be viewed as
worst case regarding axle load, even here an unconditional axle detection must
be requested from a sensor operating within a safety relevant system. The signal quality in Figures 3.15 to 3.20 is insufficient to reliably recognize single
wheel influences.
Final conclusions from both the theoretical considerations and the experiments
with respect to the application of the studied phenomenon as a resonating sensor and a more general view on the accomplished study are found in
Chapter 4.
86
Chapter 4
Conclusions and outlook

The behavior of higher vibration modes in railway tracks close to their cutoff
frequencies has been examined theoretically, numerically and experimentally.
The background of this work was the idea to make use of forced rail vibrations
which are influenced by passing train wheels as a detection principle for a rail
contact. A feasibility study demonstrated the suitability of this concept but
also revealed that the observed influence of train wheels is more or less pronounced at different vibration modes. One major objective of this work was to
examine the reasons for these differences and to provide criteria to select adequate vibration modes. Additional questions to be covered by this investigation include:
- What are the parameters which influence the forced vibration
response at the cutoff frequencies of the rail?
- Which rail vibration modes are suited best for wheel detection?
- How can these vibration modes be found and stabilized by an industrial wheel sensor?
- How large is the spatial extent of the local vibration zone near the
excitation and how large is the spatial range within which the
receiver signal is influenced by wheels?
- What is the reaction time to be expected from a sensor of this type
and is this time sufficient to detect single wheels of fast traveling
trains?
- Is there a typical reaction of the receiver signal for which the analyzing electronic device must look for?
The acquired theoretical and experimental results are discussed in the next sec-
87
a
tion. A final assessment regarding the suitability of forced rail vibrations as a
wheel detection principle is made followed by a more general view on the
accomplished research including some ideas for future work.
4.1
Final discussion of the results
A general study of the forced vibration of higher modes in a waveguide was

performed using a model of multiple, elastically connected Bernoulli-Euler
beams with modal damping. With a model of this type, the number of crosssectional modes equals the number of beams. As the multiple mode model can
be solved analytically and thus allows the separation of the solution into its
modal components, this model gave a good insight into the characteristics of
the forced vibration response of a waveguide.
The eigenvalue problem of the cross-section, if subjected to plane strain, gives
the cutoff frequencies of the waveguide and the corresponding mode shapes.
Rosenfeld and Keller [25] showed this statement to be true in the general case
of a waveguide with arbitrary cross-section. This knowledge allowed the
numerical calculation of the cutoff frequencies and mode shapes of the rail, at
which the experiments were carried out within this work. A total number of 17
higher modes were calculated up to the defined frequency limit of 50 kHz.
Using the multiple mode model, it was shown that amplitude maxima of the
forced vibration response occur at the cutoff frequencies of the waveguide and
only system damping prevents the amplitude from increasing to infinity. The
model was studied on different types of supports, including a configuration
with supported areas alternating with unsupported model sections, thus
describing the discrete support of the rail on sleepers. In contrary to modes at
low frequencies, it was shown that the type of the support has hardly any influence on both the cutoff frequency and the mode shape of higher modes. This
knowledge allowed the selective search of vibration modes within the high
number of amplitude peaks observed during experiments.
The multiple mode model showed that in the presence of damping the characteristic movement of the mode which is excited at its cutoff frequency predominates the system response close to the excitation. This region was called
local vibration zone. At a greater distance from the excitation, lower modes,
which can propagate with small attenuation according to their frequency spectra, become increasingly relevant for the system response.
The study of the discretely supported model revealed that the characteristic
motion of the cross-section is also observed at additional amplitude peaks
close to the cutoff frequencies. No local vibration zone can be found for these
amplitude maxima caused by the discrete support.
It was possible to actually measure 14 of 15 modes which had been predicted
88

by the numerical calculations in the frequency range from 5 kHz to 50 kHz.
The mode shapes were visualized using a scanning vibrometer. The calculated
and the experimentally determined cutoff frequencies correspond within 4%
and mostly even better than 1%. A local vibration zone could be observed for
most of the experimentally determined modes. The size of these local vibration zones was one sleeper spacing or less. Beyond this region, traveling
waves could be visualized. The observed spatial extent of the local vibration
zone of these modes seems to be reasonable regarding the use within a rail
contact.
Three modes showed the spatial behavior which was theoretically predicted
for the additional resonances originating from the discrete support. For these
modes the vibration amplitude decreases at the sleepers and builds up again at
the following unsupported part of the rail. The frequencies at which this type
of system response occurs cannot be reliably predicted because they depend
largely on parameters of the support. These parameters such as sleeper spacing
or type of fixation vary within a great bandwidth in practice. Because of their
lack of a local vibration zone, modes showing this characteristic seem to be
less suitable for wheel detection.
In summary, a vibration mode is suitable for wheel detection, if the following
conditions are met:
- A local vibration zone must exist, i.e. the mode is excited at its cutoff
frequency.
- The mode has a large vibration amplitude at the contact point
between the rail and the wheel.
- The mode shape has no vibration node at the running surface.
The following consequences regarding the technical realization of a resonating
sensor for wheel detection can be drawn from the mode shape measurements.
To be sure that a suitable mode is excited at its cutoff frequency, it is necessary
to check for the characteristic motion of this mode. In order to do so, a sensor
required to operate at a specific mode would need several receiving transducers along the rail contour. Additional receivers mounted at some distance from
the excitation are necessary to exclude modes which are dominated by the type
of response originating from the discrete support.
A method widely used to characterize the damping of a system, especially in
the context of sensors, is the quality factor or Q-factor. It was shown in
Section 2.1.2 that the Q-factor concept is not directly applicable to systems
which include wave propagation. Derived from the model of a Bernoulli-Euler
beam on a viscoelastic foundation, the measurement technique of the QBW-factor was introduced and appeared to be suitable also in the case of the multiple
mode model. Unlike a single-degree of freedom mechanical oscillator with a
89
a
total phase change of p/2 in the region of its resonance frequency, the phase
change at the cutoff frequencies of these models was found to be 3p/4 = 135.
The damping of a number of modes of the rail were experimentally measured
using both, the Q-factor and the QBW-factor method. The total phase changes
were found to be between 3p/4 and p/2. The quality factor, determined by
the magnitude and phase behavior and by the free vibration response, gave
values differing about 30% - 40%. Always keeping the complexity of the system in mind, especially in the studied high frequency range, this accuracy
seems to be adequate.
The measured quality factors cover the range from approximately 110 to 560.
The quality factor is related to the system response time giving values between
4
2
8 10 s and 1.4 10 s . Considering the time which the wheel of an express
train needs to pass the local vibration zone of one sleeper spacing in extent and
taking into account that the evaluating electronic device itself has a certain
time constant to determine the magnitude and phase of the receiver signal, this
reaction time is regarded to be insufficient for the aimed at application.
The influence of a passing train wheel was modeled as a discontinuity of the
discretely supported multiple mode model. It was shown theoretically that
both the magnitude and the phase of the system response are affected by the
wheel within a range which corresponds approximately to the spatial extent of
the local vibration zone. Unfortunately, the response varies significantly
depending on the mode used and on the position of the wheel contact point and
of the receiver. In particular it must be avoided that neither the contact point of
the wheel nor the receiving transducer happen to be at a node of the cross-sectional vibration. Experiments using a movable bogie showed a wide range of
measured system responses at different modes. The clearest reaction to passing wheels was observed at modes with frequencies at the lower limit of the
studied frequency range. This can be explained by the fact that these modes
show a synchronous movement of the rail head without any vibration nodes.
The system response gets increasingly complex towards higher frequencies
with the response differing for both wheels of the bogie. The measured system
response also looks differently depending on the driving direction (left to right
or right to left) of the bogie. These effects are attributed to the circumstance,
that the contact points of the wheels change depending on the direction and it
is suspected that the axle weight of the used bogie is irregularly distributed.
The signals measured during the feasibility study mentioned above appeared
to be more distinct, which is explained by the fact that the freight car used at
that time had a six times higher axle load. Although the unloaded bogie can be
considered as the worst case regarding axle load, even in this case an unconditional axle detection must be demanded by a sensor operating within a safety
90
relevant system. This requirement is not met by the investigated detection

principle.
Based on the acquired knowledge discussed above, the use of the forced vibration of higher modes in a railway track as a general-purpose wheel detection
principle cannot be recommended. The reasons can be summarized as follows:
- The reaction time of the mechanical system is insufficient.
- The existence of suitable vibration modes for wheel detection cannot
be guaranteed because of additional, non local resonance effects.
- The influence of wheels with small axle loads is insufficient.
However, both the theoretical and experimental results acquired throughout
this work revealed some interesting features of forced vibrations of a
waveguide close to its cutoff frequencies and could be the starting point of
future research as will be outlined in the following section.
4.2
Outlook
Although the use of forced vibrations of the rail is not applicable as a general
purpose rail contact, the studied effect could be employed in more specific
configurations, such as to detect wheels of wagons moving at low velocities at
a marshalling yard. One main advantage to make use of forced rail vibrations
is the circumstance that no modification of the track is necessary in order to
install the sensor.
Because a dependence on the axle load of the receiver signal can be assumed,
a calibrated sensor of this type could also be used to determine the weight of
cargo trailers. To eliminate the influence of the contact point between wheel
and rail in such an application, rail modes at relatively low frequencies should
be applied.
As the system response shows amplitude maxima at the cutoff frequencies,
these frequencies are considered to be especially important for practical problems such as noise or the interaction between wheel and rail. The measurement
of modal damping, which was shown to be a key parameter regarding the system response, seems to be an important issue when dealing with these problems. There is little work found in the literature heading in this direction.
Some concepts for modal damping measurements of a railway track have been
developed and applied within this investigation, but also a number of questions have come up which should be investigated in greater detail. Among
other issues, a discrepancy in phase behavior between the theoretical calculations and the measurements was discovered for resonances caused by the discrete support. Further research in this direction could lead to a more
91
a
appropriate definition of the quality factor for these modes.
With an excitation at a cutoff frequency, the corresponding mode predominates the system response only within an area close to the excitation. At
greater distances, lower modes, which can travel with comparably small attenuation, become increasingly important. Also modes which are predominated
by the type of response originating from the discrete support show a moderate
attenuation along the rail. The examination of modes which can travel over a
long distance in the rail could be an interesting topic of future work. Such
waves may be the effect, which was reportedly used by Indians to hear
approaching trains from a distance of several miles.
Some investigations regarding the frequency spectra of wave propagation
modes in railway tracks is found in the literature, e.g. Scholl [28]. However,
these studies are mostly theoretical and in the purely elastic case. An implementation of modal damping to more general numerical models of the rail as a
waveguide could prove to be a promising approach to this subject. The experimental determination of the frequency spectra of rail modes certainly poses an
interesting challenge. The suggested approach to this topic is to use a broadband excitation and to record the system response at each point of a two
dimensional measurement grid along the contour and in longitudinal direction
of the rail. The acquired data would then be analyzed using Fast-FourierTransform or a spectrum estimation method such as Linear-Prediction or
Matrix-Pencil in order to calculate the frequency-wavenumber relations to be
found in the waveguide, thus giving the frequency spectrum. These techniques
have already been successfully applied to cylindrical structures by Vollmann et
al. [38] and Gsell et al. [12]. The extension of these methods to wave guides
with arbitrary cross-section is certainly interesting not only from an academic
point of view. The rail represents an ideal quasi-infinite testing object, therefore making it unnecessary to deal with reflections.
92
93
94
Appendix A
Appendix A: Q-factor definitions of a single degree

of freedom mechanical oscillator
Consider a single degree of freedom mechanical oscillator which is driven by
an externally applied force. The differential equation of such a system in the
case of harmonic motion, written in dimensionless form and including viscous
and hysteretic damping, is
u ,tt + 2du ,t + i2hu + u = e
i kt
(A.1)
where w 1 = s m , k = w w 1 and t = w 1 t , are the natural frequency,

the dimensionless frequency and the dimensionless time, respectively.
The solution of (A.1) is
u(t) = b e
b =
i kt
, with
(A.2)
(A.3)
1 k + i2 ( dk + h )
being the receptance of the system.

In the following sections, some Q-factor definitions of a single degree of freedom mechanical oscillator are summarized. These definitions can be found in
standard literature, such as [2].
A.1 Q-factor definition based on magnitude and phase

behavior
At the resonance frequency k = 1 the magnitude and the phase of the receptance are
bk=1 =
1
and Arg ( b )
2(d + h)
k=1
p
= .
2
(A.4)
Now consider the two frequencies ka and kb, which satisfy

2
1 k a 2 ( dk a + h ) = 0 and 1 k b + 2 ( dk b + h ) = 0 .
(A.5)
The solutions with positive real values of these equations are

2
k a = d + 1 + d 2h and k b = d + 1 + d + 2h ,
(A.6)
which can be approximated by the first order of their Taylor series in the case
95
a
of small damping parameters d and h, thus giving
k a = 1 d h and k b = 1 + d + h .
(A.7)
By expanding the denominator of b with 2 ( dk a, b + h ) 2 ( dk a, b + h ) and by

using the relations (A.5), the receptance function at the frequencies ka and kb
can be written as
b
k a, b
1
.
2 ( dk a, b + h ) ( i 1 )
(A.8)
Ignoring parts of order d2 and dh, the magnitude and phase of the receptance
function are
1
and Arg ( b )
2 2(d + h)
b ka, b =
k a, b
p p
.
2 4
(A.9)
At the frequencies ka and kb the magnitude has dropped to 1 2 and the

phase has shifted p 4 as compared to the values at the resonance frequency. The Q-factor can be defined as the ratio between the resonance frequency and (kb - ka) and is related to the system damping according to (A.10).
1
1
=
2(d + h)
kb ka
Q =
(A.10)
The Q-factor can also be related to the slope of the phase curve of the receptance at the resonance frequency as shown in (A.11).
Q =
1d
Arg ( b )
2 dk
=
k=1
1d
2 ( dk + h )
ArcTan
2
2 dk
1k
=
k=1
1
(A.11)
2(d + h)
Please note, that no approximations are necessary in this Q-factor definition.
A.2 Q-factor definition based on free vibration

The solution of the homogeneous differential equation of harmonic motion
u h, tt + 2du h, t + i2hu h + u h = 0
(A.12)
gives
uh ( t ) = b e
i ( i d + 1 d + i2 d )t
= be
i kf t
(A.13)
by taking into consideration, that the real part of the frequency is positive. The
96
Appendix A
linearization of the complex frequency k f in the case of small damping parameters is

kf = 1 + i ( d + h ) .
(A.14)
Using the real and imaginary part of k f , the Q-factor can be defined as
Q =
Re ( k f )
1
.
=
2Im ( k f )
2(d + h)
(A.15)
The time constant t 1 e , after which the amplitude of the vibration has
decayed to 1 e its initial value is directly related to the Q-factor according to
(A.16).
t 1 e = 2Q
(A.16)
A.3 Relation between Q-factor and system energy

If we look at the system of a single degree of freedom mechanical oscillator,
which is harmonically excited, the total energy can be divided into four energy
components: the kinetic energy, the potential energy, the work dissipated by
the damping and the work of the excitation force. The real part u ( t ) of the
complex displacement solution function (A.2) must be taken in order to calculate these energy components in the same way as performed in Section 2.1.2.3.
Closed form solutions for the kinetic energy E k , the potential energy E p , the
dissipated work E d and for the work of the exciting force E e are found
according to (A.17) - (A.20).
2
2 i2kt
1 2
k
2 i2kt
E k ( t ) = u ,t = ( b e
b* e
+ 2 bb* )
8
2
Ep ( t ) =
uu ,tdt =
2 i2kt
1 2
1 2 i2kt
u = (b e
+ b* e
+ 2 bb* )
2
8
(A.17)
(A.18)
Ed ( t ) =
t
0
2 i2kt
1
2 i2kt
h 2
2 d + u ,tdt = ( dk + h ) ( i b e
i b* e
+ 4 bb* kt )
4
k
Ee ( t ) =
i2kt
i2kt
1
pu ,tdt = ( b e
+ b* e
+ i2 ( b b* )kt )
8
(A.19)
(A.20)
Figure A.1 shows all four energy components of the single degree of freedom
mechanical oscillator over one vibration cycle at the frequencies k = 0.9 ,
k = 1 and k = 1.1 . The damping constants are set to d = 0 and h = 1/50. In
Figure A.1 the lines about which E d ( t ) and E e ( t ) oscillate are included in
97
a
stroke-dotted style.
=0.9
12.5
10.
Energy
7.5
5.
2.5
0.
Ek
Ep
Ed
Ee
- 2.5
- 5.
=1
300.
Energy
200.
100.
Ek
Ep
Ed
Ee
0.
=1.1
12.5
10.
Energy
7.5
5.
2.5
0.
Ek
Ep
Ed
Ee
- 2.5
- 5.
0.
0.2
0.4
0.6
(/(2))
0.8
1.
Figure A.1: Energy components of single degree of freedom mechanical

oscillator over one vibration cycle ( d = 0 , h = 1 50 )
The magnitudes of the oscillatory energy parts as a function of frequency are
shown in Figure A.2.
98
Appendix A
150.
Energy
125.
100.
75.
50.
|Ek|
|Ep|
|Ed|
|Ee|
25.
0.
0.7
0.8
0.9
1.
1.1
1.2
1.3
Figure A.2: Magnitudes of oscillatory energy parts

( d = 0 , h = 1 50 )
For the present investigation, the following statements are important:
- Introducing the solution (A.3) for b into (A.19) and (A.20) shows,
that the components linear in time of the dissipated work and of the
work of the exciting force exactly compensate each other.
- At the resonance frequency k = 1 the kinetic energy and the potential
energy match each other, as can be seen from (A.17) and (A.18).
This implies, that there is a decoupled energy flow between E k and
E p on the one hand and between E d and E e on the other hand at this
frequency.
- At the frequencies ka and kb used to determine the Q-factor, the
kinetic energy and the potential energy have decayed to half their
resonant values, because of the quadratic relation with the vibration
amplitude b. The Q-factor can also be defined as the relation
between the maximum kinetic energy stored and the energy dissipated per cycle, thus giving
1 *
bb
2
Q = 2p
2p
Ed Ed ( 0 )
k
1
2
k=1
8(d + h)
1
= 2p
=
p
2(d + h)
2(d + h)
(A.21)
99
Appendix B: Experimental modeshapes
100
Figure B.1: Measured mode shapes of modes 9 and 11
Figure B.2: Calculated (boxed pictures) and measured mode shapes of modes 12, 13 and 14 (scaling on the right hand side)
Appendix B
101
102
Figure B.3: Calculated (boxed pictures) and measured mode shapes of modes 15, 19 and 20 (scaling on the right hand side)
Figure C.1: Spatial decay of modes 9 and 11, excitation at z = 0 m
a
Appendix C
Appendix C: Spatial decay
103
104
Figure C.2: Spatial decay of modes 12, 13 and 14, excitation at z = 0 m
Figure C.3: Spatial decay of modes 15, 19 and 20, excitation at z = 0 m
a
Appendix C
105
106
Bibliography
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
Ahlbeck, D.R., A study of dynamic impact load effects due to railroad

wheel profile roughness. 10th IAVSD Symposium on Dynamics of Vehicles on Roads and Tracks, Prague, 13 - 16, 1987
Bishop, R.E.D., Johnson, D.C., The Mechanics of Vibration. Cambridge
University Press, London, 1979
Bogacz, R., Krzyzynski, T., Popp, K., Influence of shear deformation
and rotatory inertia on the solutions of the generalized Mathews problem. Z. angew. Math. Mech., 73(1), 5 - 13, 1993
Bogacz, R., Krzyzynski, T., Popp, K., On dynamics of system modelling
continuous and periodic guideways. Arch. Mech., 45(5), 575 - 593, 1993
Cremer, L., Heckl, M., Krperschall. Springer-Verlag, Berlin, 1996
Fastenrath, F. (Publisher), Die Eisenbahnschiene. Verlag von Wilhelm
Ernst und Sohn, Berlin, 1977
Fryba, L., Vibrations of solids and structures under moving loads.
Thomas Telford, London, Third ed. 1999.
Grassie, S.L., Gregory, R.W., Harrison, D., Johnson, K.L., The dynamic
response of railway track to high frequency vertical excitation. J. Mech.
Engng. Sci. 24(2), 77 - 90, 1982
Grassie, S.L., Gregory, R.W., Johnson, K.L., The dynamic response of
railway track to high frequency lateral excitation. J. Mech. Engng. Sci.
24(2), 91 - 96, 1982
Grassie, S.L., Gregory, R.W., Johnson, K.L., The dynamic response of
railway track to high frequency longitudinal excitation. J. Mech. Engng.
Sci. 24(2), 97 - 102, 1982
Gry, L., Gontier, C., Dynamic modelling of railway track: a periodic
model based on a generalized beam formulation. J. Sound Vib. 199(4),
531 - 558, 1997
Gsell, D., Profunser, D., Dual, J., Measurement of the dispersion relation
of guided nonaxisymmetric waves in filament-wound cylindrical structures. Proceedings of Ultrasonics International 1999, 29 June - 1 July
1997, Copenhagen (to appear)
Heckl, M., Mller, H.A., Taschenbuch der Technischen Akustik.
Springer-Verlag, Berlin, 1995
Heckl, M.A., Railway noise - can random sleeper spacings help?. Acustica 81, 559 - 564, 1995
Hempelmann, K., Hiss, F., Knothe, K., Ripke, B., The formation of wear
patterns on rail tread. Wear 144, 179 - 195, 1991
107
a
[16] Ilias, H., Knothe, K., Ein diskret-kontinuierliches Gleismodell unter dem
Einfluss schnell bewegter, harmonisch schwankender Wanderlasten.
VDI Fortschritt-Bericht, Reihe 12, Nr. 171, Dsseldorf, 1992
[17] Knothe, K., Strzyzakowski, Z., Willner, K., Rail vibrations in the high
frequency range. J. Sound Vib. 169(1), 111 - 123, 1994
[18] Lundn, R., Fatigue durability of tread-braked railway wheels - on
admissible combinations of axle load, train speed and signaling distance. J. Rail and Rapid Transit, Proc. Instn. Mech. Engrs., 205F, 21 33, 1991
[19] Mead, D.J., Wave propagation and natural modes in periodic systems: I.
Mono-coupled systems. J. Sound Vib. 40(1), 1 - 18, 1975
[20] Mead, D.J., Wave propagation and natural modes in periodic systems:
II. Multi-coupled systems, with and without damping. J. Sound Vib.
40(1), 19 - 39, 1975
[21] Munjal, M.L., Heckl, M., Vibrations of a periodic rail-sleeper system
excited by an oscillatory, stationary, transverse point force. J. Sound
Vib. 81(4), 491 - 500, 1982
[22] Newton, S.G., Clark, R.A., An investigation into the dynamic effects on
the track of wheelflats on railway vehicles. J. Mech. Engng. Sci. 21, 287
- 297, 1979
[23] Remington, P.J., Wheel/rail squeal and impact noise: What do we know?
What dont we know? Where do we go from here? J. Sound Vib. 116(2),
339 - 353, 1985
[24] Ripke, B., Knothe, K., Die unendlich lange Schiene auf diskreten
Schwellen bei harmonischer Einzellasterregung. VDI Fortschritt-Bericht, Reihe 11, Nr. 155, Dsseldorf, 1991
[25] Rosenfeld, G., Keller, J.B., Wave propagation in elastic rods of arbitrary
cross section. J. Acoust. Soc. Am. 55, 555 - 561, 1974
[26] Sato, Y., Study on high-frequency vibrations in track operated with highspeed trains. Japanese National Railways, Railway Technical Research
Institute, Quarterly Reports 18(3), 109 - 114, 1977
[27] Scholl, W., Schwingungsuntersuchungen an Eisenbahnschienen. Acustica 52, 10 - 15, 1982
[28] Scholl, W., Darstellung des Krperschalls in Platten durch bertragungsmatrizen und Anwendung auf die Berechnung der Schwingungsformen von Eisenbahnschienen. VDI Fortschritt-Bericht, Reihe 11, Nr.
93, Dsseldorf, 1987
[29] Strzyzakowski, Z., Ziemanski, L., On the application of the finite strips
method to dynamical analysis of vehicle-track system. Z. angew. Math.
Mech., 71(4), 221 - 224, 1991
108
Bibliography
[30] Tassilly, E., Vincent, N., A linear model for the corrugation of rails. J.
Sound Vib. 150, 25 - 45, 1991
[31] Timoshenko, S., Method of analysis of statical and dynamical stresses in
rail. Proc. Second Int. Congress for Appl. Mech., Zrich, 407 - 418,
1926
[32] Thompson, D.J., Theoretical modelling of wheel-rail noise generation.
Journal of Rail and Rapid Transit, Proc. Instn. Mech. Engrs., 205F, 137 149, 1991
[33] Thompson, D.J., Wheel-rail noise generation, part I: introduction and
interaction model. J. Sound Vib. 161(3), 387 - 400, 1993
[34] Thompson, D.J., Wheel-rail noise generation, part II: wheel vibration. J.
Sound Vib. 161(3), 401 - 419, 1993
[35] Thompson, D.J., Wheel-rail noise generation, part III: rail vibration. J.
Sound Vib. 161(3), 421 - 446, 1993
[36] Thompson, D.J., Wheel-rail noise generation, part IV: contact zone and
results. J. Sound Vib. 161(3), 447 - 466, 1993
[37] Thompson, D.J., Wheel-rail noise generation, part V: inclusion of wheel
rotation. J. Sound Vib. 161(3), 467 - 482, 1993
[38] Vollmann, J., Breu, R., Dual, J., High-resolution analysis of the complex
wave spectrum in a cylindrical shell containing a viscoelastic medium.
2. Experimental results versus theory. J. Acoust. Soc. Am. 102 (2), 909 920, 1997
109
110
Curriculum vitae
Born on July 29, 1970 in Warwick, Rhode Island, USA
Citizen of Switzerland and USA
1977 - 1983
Primary school in Aesch, ZH, Switzerland.
1983 - 1989
Kantonsschule Hohe Promenade, Zrich, Switzerland. Graduation with a matura, type B.
1990
Practical training at the MacNeal Schwendler Corp. in Los

Angeles, USA.
1990 - 1995
Studies at the Mechanical Engineering Department of the Swiss

Federal Institute of Technology ETH, Zrich, Switzerland.
Graduation with honors with the degree Dipl. Masch.-Ing. ETH
1995 - 2000
Doctoral student, research and teaching assistant at the Institute

of Mechanical Systems (formerly Institute of Mechanics) of the
the ETH Zrich.
111
112

Rail Vibrations A4

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Rail Vibrations A4

Hochgeladen von

Copyright:

Verfügbare Formate

Higher Vibration Modes in Railway

Tracks at their Cutoff Frequencies

Diss. ETH No. 13755

Diss. ETH No. 13755

Higher Vibration Modes in Railway

A thesis submitted to the

Overview of research on dynamic railway track modeling . . . . . . . .5

Survey of dynamic track models . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6

Scope and outline of the present work. . . . . . . . . . . . . . . . . . . . . . . . .7

Theoretical considerations . . . . . . . . . . . . . . . . . . . . .11

Single mode behavior; beam on a viscoelastic foundation. . . . . . . .12

Extension to multiple modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

Generalization to the actual cross-section of the rail . . . . . . . . . . . 55

Conclusions from theoretical considerations. . . . . . . . . . . . . . . . . . 61

Frequency spectra and mode shapes of the rail . . . . . . . . . . . . . . . 65

Examination of the spatial decay . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

Modal damping of the rail. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

Influence of passing train wheels . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

Conclusions and outlook . . . . . . . . . . . . . . . . . . . . . . 87

Final discussion of the results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .88

Appendix A: Q-factor definitions of a single degree of freedom mechanical oscillator. . . . . . . . . . . . . . . . . . . . . 95

Appendix B: Experimental modeshapes . . . . . . . . . . . . 100

cross-sectional area of beam

Figure 1.2: Diagram of the experimental setup

Amplitude of receiver signal (-)

Distance of wheel from excitation (m)

Distance of wheel from excitation (m)

Figure 1.3: Wheel influence on magnitude of receiver signal measured within

1.1 Overview of research on dynamic railway track

Survey of dynamic track models

The history of dynamic modeling of railway tracks and of the interaction of

structure theory to calculate the response of the infinite system. Strzyzakowski

Scope and outline of the present work

Signal power (dBm)

Figure 1.4: Mean signal power measured during the passing of a

2.1 Single mode behavior; beam on a viscoelastic foundation

Forced vibration solution

Figure 2.1 shows a purely elastic Bernoulli-Euler beam on a continuous visiwt

Figure 2.1: Bernoulli-Euler beam on a continuous viscoelastic foundation

EIu ,zzzz w r Au + iwc c u + s c ( 1 + i2h )u = 0 ,

w ,zzzz k w + 2ikdw + i2hw + w = 0 ,

equation (2.2) is solved, if the dimensionless wavenumber a ( k ) satisfies the

in dimensionless notation. Equation (2.7) can be rewritten as

Figure 2.2: Magnitude and phase of direct receptance bc ( d = 0 )

Characterization of the near cutoff frequency behavior

The damping of a classical single degree of freedom mechanical oscillator is

the resonance frequency.

QBW-factor definition based on magnitude and phase behavior

giving a magnitude and phase value of the receptance of

Motivated by the single degree of freedom mechanical oscillator discussed in

Adding and subtracting 2 ( dk a, b + h ) 2 ( dk a, b + h ) to equation (2.4) gives

giving magnitude and phase values of the receptance function of

With dimensions the definition of the QBW-factor is

in the dimensionless case and

QBW-factor definition based on free vibration

It is standard in the field of acoustics, to determine the damping of a system by

which is in agreement with the previous definitions.

in the dimensionless case and

Study of the system energy

In the case of a single degree of freedom mechanical oscillator, the Q-factor

2.1.2.3.1 Calculation of the energy components

The total kinetic energy E k of the viscoelastic beam is

The rate, at which energy is dissipated in the foundation damping E d ( t ) is