Book - Rail Vehicle Dynamics

Klaus
Knothe · Sebastian Stichel
Rail Vehicle
Dynamics
Rail Vehicle Dynamics
Klaus Knothe Sebastian Stichel
•
Rail Vehicle Dynamics
123
Klaus Knothe Sebastian Stichel
Institute of Aeronautics and Astronautics KTH Royal Institute of Technology
Technical University of Berlin Stockholm
Berlin Sweden
Germany
ISBN 978-3-319-45374-3 ISBN 978-3-319-45376-7 (eBook)

DOI 10.1007/978-3-319-45376-7
Library of Congress Control Number: 2016949109
Original version published by SpringerVieweg 2003

© Springer International Publishing AG 2017
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made.
Printed on acid-free paper
This Springer imprint is published by Springer Nature

The registered company is Springer International Publishing AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
The book Rail Vehicle Dynamics was developed from a course book for the class
Rail Vehicle Dynamics at the Technical University (TU) Berlin. Klaus Knothe has
been responsible for the class for almost 25 years, 20 of these years together with
Prof. Robert Gasch. Therefore, Chaps. 4–7 are strongly influenced by Prof. Gasch.
Sebastian Stichel was a Ph.D. student for five years at the Department of Aeronautics
and Astronautics and during that time, substantially revised the course literature.
Today, he is a professor of rail vehicle dynamics at the Royal Institute of Technology
(KTH), in Stockholm. Further results from numerous research projects and Ph.D.
theses are included. A German version of the book was published in 2003.
The reason for writing a textbook on rail vehicle dynamics was that a com-
prehensive book on the topic did not exist, either in German or English. In the two
excellent books on Fahrzeugdynamik (Vehicle Dynamics) [1] and on
Systemdynamik und Regelung von Fahrzeugen (System Dynamics and Control of
Vehicles) [2], rail vehicle dynamics specifically could not be described extensively.
In English, there is the book of Garg and Dukkipati [3], but it is somewhat dated
now. The same applies to the book of Kovalev [4], as well as to the brochure of
Krugmann [5] or the still very valuable book from Hanneforth [6]. An anthology
with Klaus Knothe as one of the authors [7] deals with a number of aspects, but it
does not provide a general introduction. Since this book was published in German
in 2003, a number of very interesting books on the topic have appeared, e.g., the
book Fundamentals of Rail Vehicle Dynamics by Wickens [8] and Handbook on
Railway Vehicle Dynamics, edited by Iwnicki [9]. From Australia, the book Design
and Simulation of Rail Vehicles by Spyriagin and Cole et al. [10] has been pub-
lished. These books also partly fill gaps but still have a somewhat different focus
than our book. Books from the nineteenth century [11] or the first half of the
twentieth century [12, 13] cannot fill the gap either. Therefore, it was considered
worthwhile to fill the vacancy with this book. The situation is somewhat different in
track dynamics (Gleisdynamik) [14] and bridge dynamics (Brückendynamik) [15].
The text book has the character of an introduction. This means that the most
important aspects are described in detail, while further questions are touched
v
vi Preface
on only briefly. The approach is therefore almost inductive, i.e., basic theories are
given when needed. This has been successfully applied in our class on rail vehicle
dynamics, in which master's students in vehicle engineering and machine tech-
nology are in many cases confronted for the first time with equations of motion and
their solution.
With that said, we should mention the main audience for this book. It was
written for university students in engineering sciences who want to learn more
about rail vehicle dynamics and perhaps also to find an entrance into methods of
dealing with other dynamic problems in technical systems. It is also intended for
engineers in the rail vehicle industry, railway operators, infrastructure owners, and
authorities who want to deepen their knowledge of the topic.
Without help, we would not have been able to write this book. Firstly, we would
like to mention Prof. Gasch again, whose lecture notes of the first part of the class
Rail Vehicle Dynamics were the basis for Chaps. 4–7. Regarding the Ph.D. students
who participated in the lectures only, Dr.-Ing. Arnold Groß-Thebing and Dr.-Ing.
Burchard Ripke are mentioned here. Dr.-Ing. Walter Kik followed the development
of the lecture notes with great interest and offered many suggestions. He also helped
with both words and deeds in the final phase of the writing of the manuscript,
especially with calculation examples. The authors would like to thank Dr.-Ing.
André Theiler for help with Chap. 3. Those who contributed within their research
projects are mentioned in the respective sections in the book. We would also like to
thank Prof. Evert Andersson and Prof. Mats Berg. Sebastian Stichel is giving a class
on rail vehicle dynamics together with them at KTH. Parts from the lecture notes
are included in Chaps. 7 and 14. Special thanks also to Roger Enblom, who
proofread these two chapters and gave valuable input.
The drawings were made by Mrs. Christine Koll in her reliable manner. We owe
her enormous thanks. Special thanks also to the staff at Springer for their patience
with the authors.
Last but not least, the authors wish to thank their wives, who exhibited con-
siderable patience during the writing of the manuscript.
Berlin, Germany Klaus Knothe

Stockholm, Sweden Sebastian Stichel
July 2016
References
1. K. Popp und W. Schiehlen. Fahrzeugdynamik. Eine Einführung in die Dynamik des Systems
Fahrzeug–Fahrweg (Vehicle Dynamics. An introduction into the dynamics of the system
vehicle–track). B.G. Teubner, Stuttgart, 1993
2. W. Kortüm und P. Lugner. Systemdynamik und Regelung von Fahrzeugen – Einführung und
Beispiele (System dynamics and vehicle control – Introduction and examples). Springer–
Verlag, Berlin e.a., 1994
Preface vii
3. V.K. Garg, R.V. Dukkipati, Dynamics of Railway Vehicle Systems (Academic Press, Toronto,
1984)
4. N.A. Kovalev, The Lateral Oscillation of Rolling Stock (Transheldhorizdat, Moskau, 1957)
5. H.-L. Krugmann, Lauf der Schienenfahrzeuge im Gleis (Running of railway vehicles on track)
(Oldenbourg, München, Wien, 1982)
6. W. Hanneforth, W. Fischer, Laufwerke (Bogies) (Transpress, Berlin, 1986)
7. J. Kisilowski, K. Knothe (eds.), Advanced Railway Vehicle System Dynamics (Wydawnictwa
Naukowo-Techniczne, Warszawa, 1991)
8. A.H. Wickens, in Advances in Engineering Series, Fundamentals of Rail Vehicle Dynamics
(CRC PressTaylor & Francis Group, Boca Raton, 2005)
9. S. Iwnicki (ed.), Handbook of Rail Vehicle Dynamics (CRC Press, Taylor & Francis Group,
Boca Raton, 2006)
10. M. Spyriagin, C. Cole, Y.Q. Sun, M. McClanachan, V. Spiryagin, T. McSweeney, Design and
Simulation of Rail Vehicles (CRC Press, Taylor & Francis Group, Braunschweig, Wiesbaden,
New York, 2014)
11. Chr. Boedecker. Die Wirkungen zwischen Rad und Schiene und ihre Einflüsse auf den Lauf
und den Bewegungswiderstand der Fahrzeuge in den Eisenbahnzügen (The effects between
wheel and rail and their inuences on the running behaviour and the resistance of vehicles in
railway trains). Hahn’sche Buchhandlung, Hannover, 1887
12. Y. Rocard. La Stabilité de Route des Locomotives, Premiére Partie, avec une note de
M. R. Lévy. Hermann & Cie., Paris, 1935. Teil II siehe [?]
13. M. Julien, Y. Rocard, La stabilité de route des locomotives, deuxième partie (Hermann &
Cie., Paris, 1935)
14. K. Knothe, Gleisdynamik (Track dynamics) (Ernst & Sohn, Berlin, 2001)
15. L. Frýba, Dynamics of railway bridges (Thomas Telford, London, 1996)
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 1
1.1 The Basic Function of the Wheel/rail System . . . . . . . . . . . .... 1
1.2 Significance of Dynamics on the Operation of Rail
Vehicles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 2
1.3 On the History of Research in the Field of Railway
Technology Since 1800 . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 4
1.3.1 1800–1945. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 4
1.3.2 A New Start After 1945: Japan and France . . . . . . .... 7
1.3.3 Research and Development Aimed at Overcoming
the “Boundaries of the Wheel/rail System” . . . . . . . . . . 10
1.4 Railway Industry in Europe . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.5 Overview of This Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.5.1 Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.5.2 Vertical and Lateral Vibrations . . . . . . . . . . . . . . . . . . . 13
1.5.3 Curving Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.5.4 Frequency- and Time-Domain Calculations . . . . . . . . . . 13
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2 Modeling of Vehicle, Track, and Excitation . . . . . . . . . . . . . . . .... 17
2.1 Prior Considerations and Coordinate Systems . . . . . . . . . . . .... 17
2.2 Vehicle Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 18
2.2.1 Bogie Frame Design, Primary Suspension,
and Bogie Guidance . . . . . . . . . . . . . . . . . . . . . . . .... 18
2.2.2 Mechanical Model of the Vehicle. Connecting
Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.3 Elastic Carbodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Modeling of Track and Excitation . . . . . . . . . . . . . . . . . . . . . . . 26
2.3.1 Track Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3.2 Modeling of the Excitation . . . . . . . . . . . . . . . . . . . . . . 28
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
ix
x Contents
3 Modeling of Wheel/Rail Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3.1 Profile Geometry. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Contact Kinematics Between Wheel and Rail . . . . . . . . . . . . . . . 36
3.2.1 Contact Kinematics with Conical and Circular
Profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2.2 Contact Kinematics for Arbitrary Profiles . . . . . . . . . . . 43
3.2.3 On the Determination of Equivalent Contact
Parameters by Means of Quasilinearization . . . . . . . . . . 45
3.2.4 Conversion into Equivalent Circular Profiles . . . . . . . . . 47
3.2.5 Linearized Contact Kinematics with Track
Irregularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.2.6 Creepage Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.3 Normal Contact Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.3.1 Overview of the Calculation of Contact
Point Stresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.3.2 Assumptions of the Normal Contact Problem . . . . . . . . 52
3.3.3 Nonelliptic Contact Patches . . . . . . . . . . . . . . . . . . . . . . 53
3.3.4 Hertzian Treatment of the Normal Contact Problem . . . 53
3.3.5 Spherical and Point Contact . . . . . . . . . . . . . . . . . . . . . 56
3.3.6 Ellipsoidal Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.3.7 Contact of Rollers, Line Contact . . . . . . . . . . . . . . . . . . 58
3.3.8 Linearized Replacement Model . . . . . . . . . . . . . . . . . . . 59
3.4 Tangential Contact Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.4.1 Introduction to the Tangential Contact Problem . . . . . . 60
3.4.2 Analytical Solution for Rolling Contact
(Line Contact) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.4.3 Kalker’s Theory of Rolling Contact for Ellipsoidal
Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.4.4 Approximate Solutions According
to Vermeulen–Johnson and Shen–Hedrick–Elkins . . . . . 67
3.4.5 Simplified Theory of Rolling Contact [54] . . . . . . . . . . 71
3.4.6 Adaptation of the Theory to Reality . . . . . . . . . . . . . . . 76
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4 Vertical Dynamics, Equations of Motion, and Free Vibrations .... 81
4.1 Notation and Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . .... 81
4.2 Equations of Motion with the Principle of Linear
and Angular Momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 82
4.2.1 Displacement Degrees of Freedom
for a Two-Axle Vehicle . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.2.2 Constraints. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.2.3 Forces in the Spring and Damper Elements . . . . . . . . . . 85
4.2.4 Free-Body Diagrams of the Masses . . . . . . . . . . . . . . . . 87
Contents xi
4.2.5 Principle of Linear and Angular Momentum

for the Formulation of the Equation System . . . . . . . . . 88
4.2.6 Elimination of the Constrained Forces.
Final System of Equations . . . . . . . . . . . . . . . . . . . . . . . 89
4.3 Principle of Virtual Displacements for Rigid-Body Systems . . . . 91
4.3.1 Initial Remark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.3.2 Formulation of the Principle of Virtual
Displacements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.3.3 Inclusion of Geometric Constraints
in the Principle of Virtual Displacement
Exemplified on a Two-Axle Vehicle . . . . . . . . . . . . . . . 94
4.4 Formalized Derivation of the Equations of Motion
and the Principle of Virtual Displacements . . . . . . . . . . . . . . . . . 95
4.4.1 Displacement Vector with the Degrees
of Freedom of the Free System . . . . . . . . . . . . . . . . . . . 95
4.4.2 Relation Between Spring Elongation and System
Displacement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.4.3 Spring Forces and the Formulation of the Virtual
Strain Energy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.4.4 Mass Matrix and Formulation of the Virtual Work
of the Inertia Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.4.5 External and Constraint Forces . . . . . . . . . . . . . . . . . . . 97
4.4.6 Equations of Motion for the Free System.
Introduction of Constraints . . . . . . . . . . . . . . . . . . . . . . 98
4.5 Equations of Motion for Elastic Car Bodies . . . . . . . . . . . . . . . . 99
4.6 Solution for Free Vibrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.7 Exercises for This Chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.7.1 Constraint Forces when Constraints are Met . . . . . . . . . 105
4.7.2 Validity of the Rolling Condition . . . . . . . . . . . . . . . . . 105
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5 Forced Vertical Vibrations for Excitation with Harmonic
and Periodic Track Irregularities (Frequency Domain
Solution) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 107
5.1 Complex Syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 108
5.2 Vertical Vibrations on a Track with Cosine-Shaped Track
Irregularity (Cosine Track) . . . . . . . . . . . . . . . . . . . . . . . . . .... 111
5.2.1 Track Irregularities and Base-Point Excitations . . . .... 111
5.2.2 Solution for the Vertical Motion . . . . . . . . . . . . . . .... 112
5.2.3 Interpretation of the Solution for Different
Suspensions Between Wheelset and Carbody . . . . .... 116
5.3 Vehicle on a General Periodic Track . . . . . . . . . . . . . . . . . .... 119
5.4 Solution for a Vehicle with Elastic Carbody Modes . . . . . . .... 123
xii Contents

5.5.1 Two-Axle Vehicle on Cosine Track . . . . . . . . . . . . . . . 124
5.5.2 Two-Axle Vehicle on Generally Periodic Track . . . . . . 125
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6 Random Vibrations due to Stochastic Track Irregularities . . . .... 127
6.1 Characterization of an Irregular Track with Help
of Root Mean Square Value and Power Spectrum . . . . . . . .... 127
6.2 Determination of Vehicle Response for Stochastic Track
Excitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 130
6.3 Power Spectra of Track Irregularities . . . . . . . . . . . . . . . . . .... 133
6.3.1 How to Derive Power Spectra of Track
Irregularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 133
6.3.2 Power Spectra for the DB Network . . . . . . . . . . . . .... 135
6.4 Supplementary Comments Regarding the Relationship
Between Measured Spectra of Spatial Angular Frequencies
and Spectra of Angular Frequencies as Function of Time. . .... 139
6.5 Interpretation of Response Spectrum. . . . . . . . . . . . . . . . . . .... 139
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 140
7 Human Perception of Vibrations - Ride Comfort . . . . . . . . . . . . . . . 141
7.1 Wertungsziffer According to Sperling . . . . . . . . . . . . . . . . . . . . . 142
7.1.1 Periodic Vibrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
7.1.2 Random Vibrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
7.2 ISO 2631 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
7.3 CEN Standard EN 12299 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
7.3.1 Simplified Criterion for Mean Comfort: NMV . . . . . . . . 151
7.3.2 Comfort on Curve Transitions: PCT . . . . . . . . . . . . . . . . 153
7.3.3 Comfort on Discrete Events: PDE . . . . . . . . . . . . . . . . . 153
7.4 Final Remarks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
7.4.1 Measure or Simulate . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
7.4.2 Ride Comfort as System Property . . . . . . . . . . . . . . . . . 156
7.4.3 Duration of Exposure of Vibration . . . . . . . . . . . . . . . . 156
7.5.1 Calculation of Wz-Values According to Sperling . . . . . 156
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
8 Introduction to Lateral Dynamics of Railway Vehicles . . . . . . . . . . 159
8.1 Preliminary Remark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
8.2 Hunting and Klingel Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 162
8.3 Assumptions for Derivation of the Klingel Equation . . . . . . . . . 165
8.4 Determination of Equivalent Conicity with Eq. (8.13) . . . . . . . . 167
8.4.1 Final Remark to This Chapter . . . . . . . . . . . . . . . . . . . . 168
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
Contents xiii
9 Derivation of Equations of Motion for Lateral Dynamics . . . . . . . . 169

9.1 Principle of Virtual Displacements for a Suspended
Wheelset with All Degrees of Freedom . . . . . . . . . . . . . . . . . . . 169
9.1.1 Evaluated System and Acting Forces . . . . . . . . . . . . . . . 169
9.1.2 Formulation of Principle of Virtual Displacements . . . . 171
9.1.3 Determination of the Virtual Displacements . . . . . . . . . 172
9.1.4 Equilibrium Conditions in the x-Direction
and Around the y-Axis . . . . . . . . . . . . . . . . . . . . . . . . . 174
9.1.5 Equilibrium Conditions in the y-Direction
and Around the z-Axis . . . . . . . . . . . . . . . . . . . . . . . . . 175
9.2.1 Interpretation of Creep Force Terms in Eq. (9.13) . . . . . 176
9.2.2 Rolling Resistance Due to Spin Creepage . . . . . . . . . . . 177
9.2.3 Equations of Motions for Forced Lateral Vibrations . . . 177
9.2.4 Rolling Resistance in the Simplified Theory . . . . . . . . . 177
9.2.5 Equations of Motion of a Suspended Wheelset
with Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
9.2.6 Creep Forces Under the Assumption of a Reference
State with Creep Forces . . . . . . . . . . . . . . . . . . . . . . . . . 177
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
10 Lateral Eigenbehavior and Stability of a Wheelset on Straight
Track . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
10.1 Determination of Eigenvalues and Eigenvectors . . . . . . . . . . . . . 179
10.2 Root Loci Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
10.3 Approximate Solution for Low Speeds . . . . . . . . . . . . . . . . . . . . 183
10.4 Stability Investigation with Hurwitz Criterion . . . . . . . . . . . . . . . 186
10.5 Critical Speed of a Wheelset. . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
10.6 Interpretation of Stability Criteria of a Single Wheelset . . . . . . . 189
10.7.1 Characteristic Equation . . . . . . . . . . . . . . . . . . . . . . . . . 193
10.7.2 Transformation of Wheelset Equations of Motion . . . . . 194
10.7.3 Graphical Illustration of the Root Loci Curves
of a Suspended Wheelset and Determination
of the Critical Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
10.7.4 Independently Rotating Wheels . . . . . . . . . . . . . . . . . . . 194
Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
11 Lateral Eigenbehaviour and Stability of Bogies . . . . . . . . . . . . .... 195
11.1 Numerical Calculation of Eigenvalues and Critical Speed . .... 195
11.2 Analytic Approximations for Bogies . . . . . . . . . . . . . . . . . . .... 201
11.2.1 Coordinate Transformation to Introduce
Generalized Degrees of Freedom . . . . . . . . . . . . . . .... 203
11.2.2 Bogie with Infinitely High Bending and Shear
Stiffness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 211
xiv Contents
11.2.3 Realization of Infinitely High Bending Stiffness

or Shear Stiffness. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
11.2.4 Bogies with Infinite Shear Stiffness . . . . . . . . . . . . . . . . 216
11.2.5 Bogies with Infinite Bending Stiffness . . . . . . . . . . . . . . 219
11.2.6 Bogies with Finite Bending and Shear Stiffness . . . . . . 220
11.3.1 Equations of Motion of a Bogie . . . . . . . . . . . . . . . . . . 222
11.3.2 Equations of Motion of a Freely Rolling Wheelset
at Low Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
11.3.3 Equations for Bending and Shear Stiffness . . . . . . . . . . 223
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
12 Lateral Eigenbehavior and Stability of Bogie Vehicles . . . . . . . . . . . 225
12.1 Stability of a Train with Two Vehicles . . . . . . . . . . . . . . . . . . . . 225
12.2 General Conclusions Regarding the Stability of a Bogie
Vehicle According to Matsudaira [4] . . . . . . . . . . . . . . . . . . . . . 228
12.2.1 Theory and Simulation of Bogie Vehicle Hunting . . . . . 229
12.2.2 Bogie Hunting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
12.2.3 Influence of Friction Yaw Damping on Bogie
Hunting (Nonlinear Stability Investigation) . . . . . . . . . . 233
12.3 Suggestions for Further Work for This Chapter . . . . . . . . . . . . . 235
12.3.1 Influence of Bending and Shear Stiffness
on the Stability of Bogie Vehicles . . . . . . . . . . . . . . . . . 235
12.3.2 Stability of a Vehicle with Independently Rotating
Wheels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
12.3.3 Friction and Viscous Yaw Damping . . . . . . . . . . . . . . . 235
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
13 Introduction to Non-linear Stability Investigations . . . . . . . . . . .... 237
13.1 Preamble . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 237
13.2 Nonlinear Critical Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 238
13.3 Fourier Decomposition of Nonlinear Limit Cycle Motions:
The Method of Urabe and Reiter . . . . . . . . . . . . . . . . . . . . .... 240
13.4 Simplified Investigation of Nonlinear Stability
with the Method of Quasilinearization . . . . . . . . . . . . . . . . . . . . 243
13.5 Limits of Fourier Decomposition . . . . . . . . . . . . . . . . . . . . . . . . 245
13.6 Nonlinear Stability Investigations in the Time Domain . . . . . . . . 245
13.7 Ideas for Further Study in this Chapter . . . . . . . . . . . . . . . . . . . . 246
13.7.1 Stability Investigation for the Boedecker Vehicle . . . . . 246
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
14 Quasistatic Curving Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
14.1 Historical Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
14.2 General Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
14.3 A Single Wheelset in a Curve . . . . . . . . . . . . . . . . . . . . . . . . . . 251
14.3.1 Free Wheelset in a Curve (Kinematic Curving) . . . . . . . 251
Contents xv
14.4 Wheelset Guided in Track Following Frame . . . . . . . . . . . . . . . 252

14.5 Curving of Bogies and Entire Vehicles . . . . . . . . . . . . . . . . . . . . 259
14.5.1 Curving According to Uebelacker and Heumann . . . . . . 260
14.5.2 Curving of Bogies with Suspension. . . . . . . . . . . . . . . . 264
14.6 Wear Calculation in the Wheel–Rail Contact . . . . . . . . . . . . . . . 266
14.7.1 Directions of Creep Forces for Different Wheelset
Positions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 269
14.7.2 Deviation of Angle of Attack or Lateral
Displacement of Wheelsets . . . . . . . . . . . . . . . . . . .... 270
14.7.3 Curving of Single Wheelset . . . . . . . . . . . . . . . . . . .... 270
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 270
15 Determination of Load Collectives for Vehicle Components . . . . . . 273
15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
15.2 General Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
15.3 Stress Calculation in Components . . . . . . . . . . . . . . . . . . . . . . . . 275
15.3.1 Finite Element Calculation in Each Time Step . . . . . . . 276
15.3.2 Stress Calculation with the Help of Transformation
Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
15.4 Determination of Load Collectives . . . . . . . . . . . . . . . . . . . . . . . 279
15.4.1 Determination of Sustainable Stresses . . . . . . . . . . . . . . 279
15.4.2 Cycle Counting Methods to Determine Load
Collectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
15.4.3 Conversion from a Two-Parameter to a One
Parameter Collective . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
15.4.4 Superposition to Total Load Collective . . . . . . . . . . . . . 289
15.5 Damage Accumulation: Proof of Strength. . . . . . . . . . . . . . . . . . 290
15.5.1 Damage Accumulation Hypotheses . . . . . . . . . . . . . . . . 290
15.5.2 Concepts for Variable-Amplitude Fatigue Strength
Proof for Rail Vehicles . . . . . . . . . . . . . . . . . . . . . . . . . 292
15.6.1 Transformation Matrix Between Degrees
of Freedom in a Multibody Simulation Model
and Stresses in the Bogie Frame . . . . . . . . . . . . . . . . . . 293
15.6.2 Determination of Collective of Spring Forces
with Spectral Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 293
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
16 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
16.1 List of Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
16.2 Coordinate Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
xvi Contents
16.3 Fundamentals of Contact Mechanics . . . . . . . . . . . . . . . . . . .... 304

16.3.1 Hertz Contact Mechanics . . . . . . . . . . . . . . . . . . . . .... 304
16.3.2 Contact Equation . . . . . . . . . . . . . . . . . . . . . . . . . . .... 306
16.3.3 Basic Equations for the Tangential Contact
Problem According to Carter . . . . . . . . . . . . . . . . . .... 308
16.4 Function Φ for the Vermeulen–Johnson Solution . . . . . . . . .... 311
16.5 Basic Equations of the Simplified Theory of Rolling
Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 312
16.6 Stability Criteria of Characteristic Equations with Hurwitz
Criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 313
16.7 Critical Speed of Single Wheelset Taking Nondiagonal
Elements of Creepage Damping Matrix into Account . . . . . .... 314
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 315
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
Chapter 1
Introduction
1.1 The Basic Function of the Wheel/rail System
Wheel and rail are the fundamental elements of every rail vehicle. The interaction
of both constitutes the wheel/rail system, which has multiple functions:
• The wheel/rail system has a load-bearing function. It ensures that the weight of
the vehicle as well as the payload (freight, passengers) is transmitted to the ground.
• The wheel/rail system must ensure that the lateral deviation of the vehicle is kept
to a minimum while running on straight track. In curves, the deviations should be
kept within tight boundaries (guiding function).
• The wheel/rail system must ensure acceleration and deceleration (traction func-
tion and braking function).
The transfer of all forces required for these functions occurs in the contact patch
between wheel and rail, which has the approximate size of a circle with a 10 mm
diameter. Naturally, the loads that occur in this contact are extremely high. The
contact between wheel and rail also has great influence on the overall behavior of
the railway vehicle on straight track as well as in curves.
Neither on straight track nor in curves can ideal rail conditions be expected. Track
irregularities act as disturbances that can excite vibrations in the vehicle, which might
also be felt by passengers. This can lead to a decrease in the level of comfort.
A feature of the wheel/rail system is its guiding function. The simplest wheel
profile has the shape of a cone. Both wheels on one axle form a double cone, if only the
running surface is taken into consideration (Figure 1.1). This profile setting stipulates
that an axle that is subjected to a small disturbance on straight track and deviates
normally will return to the center position. Only in the case of big disturbances or
when negotiating a tight curve will the wheel flange take over the guiding function.
As a consequence of higher velocities, an axle may, even if subjected only to ever
so small disturbances, not return to a center position. Instead, disturbances can add
up: the system becomes unstable, and in the most unfavorable case, only the wheel
flange prevents the vehicle from derailing.
© Springer International Publishing AG 2017 1

K. Knothe and S. Stichel, Rail Vehicle Dynamics,
DOI 10.1007/978-3-319-45376-7_1
2 1 Introduction
These introductory words should suffice to define the scope for the dynamics of
rail vehicles.
• The load-bearing function is linked to loss of comfort and wear of compo-
nents due to track irregularities. Models are needed that allow simulation of these
processes.
• The guiding function can also be linked to loss of comfort due to track irregular-
ities. More important, however, is that the guiding function can induce instability
of the whole system. These processes also need to be modeled in order to allow
their simulation.
• Curve negotiation of rail vehicles is in most cases determined by the quasistatic
curving properties, which itself also are a part of the guiding behavior. This does
not apply to curve entrance or exit nor in short curves.
• Acceleration and braking are not completely uncoupled from the processes of
load-bearing and guiding. It is, however, possible to consider them independently.
Therefore, we have decided not to consider them in this context.
The investigation of load-bearing behavior is essentially the field of vertical

dynamics, while investigation of guiding behavior falls within the field of lateral
dynamics. This separation is possible owing to the fact that most rail vehicles pos-
sess a plane of symmetry, which is mounted through the axle in the longitudinal and
vertical directions. On straight track, therefore, the vertical and lateral dynamics are
only slightly coupled. Acceleration and braking are in turn viewed as parts of the
vertical and longitudinal dynamics.
1.2 Significance of Dynamics on the Operation of Rail

Vehicles
Many aspects that are important to the operating company relate to the dynamic
properties of the vehicle. Most important are the safety aspects. For example, a
vehicle must not derail due to poor running characteristics. In order to be acceptable
to passengers, a vehicle has to offer a high level of ride comfort. This leads us to
the next important topic, profitability. A vehicle with a poor level of comfort will be
avoided by passengers, which leads to a loss of revenue. An overview of the relevant
aspects for the dynamics of a rail vehicle can be obtained by reviewing the existing
assessment criteria [1–4]. Assessment criteria are the basis for admission of new
rail vehicles. Also, when a contract is signed between an operating company and a
manufacturer of rail vehicles, permitted limit values, regarding, e.g., comfort are of
decisive importance. At this point, the list of assessment criteria helps to shed light
on why a simulation is being conducted. Taken a step further, these criteria form
the starting point for the calculation of life cycle costs. Analysis of life cycle costs
(LCC), however, is beyond the scope of this book.
1.2 Significance of Dynamics on the Operation of Rail Vehicles 3
Table 1.1 Assessment criteria for short-term dynamics

1. The rail vehicle must not derail
2. The rail vehicle must not turn over
3. Collision with other parts of the track system must be avoided
4. The vehicle must not shift the rail in lateral direction
5. The payload (freight or passengers) must not be exposed to unacceptable conditions
• It has to be ensured that there is a sufficient level of comfort for passengers
• Also for freight transport the threshold values for acceleration must be kept
• The level of noise must not exceed the allowed threshold value
6. Also, in the vicinity of a railway line threshold values must not be exceeded, especially
regarding noise and ground vibration
7. In order to reach and maintain the desired velocity or to successfully negotiate ascents, the
traction system must provide enough traction force and the drive line must facilitate the
transfer of this force from wheel to rail
Table 1.2 Assessment criteria for long-term behavior

1. Components in a vehicle must be reliable throughout their expected life time
• Safety relevant components must not fail
•The damage or wear of components (e.g. due to fatigue) should be as small as possible
2. Analogously the functionality of track components must be ensured
• Unacceptable track irregularities (especially due to ballast settlement) must be avoided
• Rails and other track components (sleepers, rail fastenings) must not break
• The general wear of rails should be kept to a minimum
The previously mentioned assessment criteria are not a final compilation. There
can be additional criteria. In reviewing them, it becomes clear that some of them do
not solely concern the vehicle but the whole vehicle/track system. Such a systematic
approach is necessary for the field of rail vehicle technology.
Items 1–4 in Table 1.1 concern safety aspects, while items 5 and 6 are important
from a comfort and environmental point of view. Finally, item 7 is a precondition for
the operation of high-speed and heavy haul traffic.
The assessment criteria can be divided into two groups. One the one hand, dynamic
or (quasi) static processes that occur while the vehicle is running are evaluated. These
problems are called short-term dynamics. On the other hand, processes are evaluated
that take effect after hundreds, thousands, or even millions of cycles. In those cases,
we talk of long-term processes. This includes most of the defects (see Table 1.2).
Problems of the short-term dynamics and the long-term behavior were dealt with in
a DFG-programme [5].
The scope of rail vehicle dynamics is primarily to investigate the problems of
short-term dynamics.
4 1 Introduction
1.3 On the History of Research in the Field of Railway

Technology Since 1800
1.3.1 1800–1945
Research on the dynamic behavior of rail vehicles began about 200 years ago. A
starting point occurred when cylindrical wheel profiles were replaced with conical
wheel profiles. Wheel sets with cylindrical wheel profiles have a tendency to drift
sideways with only the wheel flange in place to stop the drift. Wheel sets with con-
ical wheel profiles, however, have a tendency to center themselves. This kinematic
behavior was described very precisely long ago by Stephenson1 ([6], cited in [7]):
It must be understood that the form of edge railway wheels is conical, that is the outer is
rather less than the inner diameter about 3/16 of an inch. Then from a small irregularity of
the railway the wheels may be thrown a little to the right or a little to the left, when the
former happens the right wheel will expose a larger and the left one a smaller diameter to
the bearing surface of the rail which will cause the latter to loose ground of the former but at
the same time in moving forward it gradually exposes a greater diameter to the rail while the
right one on the contrary is gradually exposing a lesser which will cause it to loose ground
of the left one but will regain it on its progress as has been described alternately gaining and
loosing ground of each other which will cause the wheels to proceed in an oscillatory but
easy motion on the rails.
In Germany, Ferdinand Redtenbacher,2 from Karlsruhe, was the first to try to

explain the behavior of wheel sets in track in his still readable book “Die Gesetze
des Lokomotivbaus” (The laws of locomotive design) [8]. Redtenbacher’s primary
concern was the behavior of locomotives in negotiating curves. Figure 1.1 is taken
from his book. It is recognizable that Redtenbacher was familiar with the kinematics
of the “double cone,” first formulated by Stephenson. Redtenbacher applied this
knowledge in order to examine the position of a rail vehicle while negotiating a
curve. He was probably the first scientist in Germany who dealt with resonance
phenomena of rail vehicles.
Regarding the second half of the nineteenth century, the first name that comes to
mind is that of Johann Klingel. In 1883, Klingel3 published a work entitled “Über den
1 George Stephenson, 1781–1848, was the son to a fireman on a steam engine, from a village close to
Newcastle. When Stephenson died, he was one of the wealthiest and most distinguished engineers
in England. He was the inventor of the locomotive (1814 for coal hauling); in 1822, he founded a
factory for locomotives where among others, The Rocket for the Liverpool–Manchester Railway
was built. The citation is probably taken from a patent application.
2 Ferdinand Redtenbacher (1809–1863) was a student at the Polytechnical Institute in Vienna. From
1833 to 1841, he was a teacher of mathematics at the Polytechnicum in Zurich, and from 1841 to
1863, he was a professor of applied mechanics and machine design at the Polytechnicum Karlsruhe.
In the years 1862–1865, his most important work was published, Der Maschinenbau (Machine
construction).
3 Johann Klingel was born in Heidelberg in 1819. From 1833/34 to 1837/38 he was a student at
the Polytechnikum in Karlsruhe, where he studied engineering science. In 1840, he successfully

completed his state examination. From 1840 to 1882/83 he was employed at the Baden Railways in
1.3 On the History of Research in the Field of Railway Technology Since 1800 5
r
r
C a A 2 2
nf
B C
D1
C1
C
F1 m 1 H 1 G1
A n1 p1 B
F n G
m E p E1
Fig. 1.1 Position of bogie with two wheel sets for stationary curving according to Redtenbacher
[8]. Principle of double cone (top), Redtenbacher’s ideas for the normal contact problem (bottom)
Lauf von Eisenbahnfahrzeugen auf gerader Bahn” [9] (On the running of rail vehicles
on straight track). Stephenson’s problem was picked up again and solved analytically.
Klingel calculated the wavelength of the motion that today is called sinusoidal motion
or hunting motion; cf. Fig. 1.2. The question whether this movement is stable or
unstable, that is, whether the movement decreases or increases over time only to be
limited by the wheel flanges, remained unsolved by Klingel.
The first one who is known to have asked the question of movement stability of
rail vehicles, was Christoph Boedecker4 [10]. For this purpose he introduced not
(Footnote 3 continued)
Karlsruhe. Starting as a trainee, he finished his career with the title of chief building officer at the
general management of the Baden state railways. The publication [9] is probably the only one he
ever wrote. Supposedly, Klingel knew the works of Redtenbacher. Klingel passed away in January
1888.
4 Christoph Boedecker was born about 1845 and was a student at the Bauakademie (academy of
civil engineering) in Berlin. From 1876 on, Boedecker was Regierungsbaumeister (governmental
master builder) with the Prussian railways. In the winter term of 1880/81, Boedecker became a
privatdozent (adjunct professor) at the University of Technology in Berlin-Charlottenburg. From
the winter term 1880/81 until the summer term 1885, he read “Theoretische Kapitel aus dem
Eisenbahnbau” (theoretical chapters for railway construction). Subsequently, until his retirement
in 1910, Boedecker was employed by the Royal Prussian Railways. By the time he retired, he held
6 1 Introduction
L
2
e
g 0 01
A1
B A2
A
f g
d b
b1
a c
h
B1
B B2
Fig. 1.2 Hunting motion of single wheelset according to Klingel [9]
Fig. 1.3 Nonlinear unstable Y

wheelset motion according
to [10] M
M
2
X
2
M1
L
2
only kinematic relations in the wheel–rail contact but also physical relations in the
form of Coulomb’s law in a local formulation. Boedecker’s conclusion was rather
surprising: rail vehicles always run in an unstable state. It is only the wheel flanges
that ensure that no derailment occurs; cf. Fig. 1.3.
A correct treatment of the so-called sinusoidal motion was first successfully
accomplished by Frederick William Carter5 [13]. A precondition for this was that the
(Footnote 4 continued)
the title of Geheimer Baurat (privy commissioner of public works) and was an executive of the
Betriebsinspektion (company inspection) Berlin. No other publications by Boedecker than [10] are
known. Boedecker passed away in 1937 or 1938 in Berlin.
5 Carter was born in 1870. He was an electrical engineer. In 1915, he held his pioneering lecture on
the subject “The electric locomotive,” in which the running stability of locomotives was mentioned
in the marginalia. In this presentation without deduction, a linear adhesion slip law was introduced.
In 1926, Carter delivered a deduction of this law [11]. Already in 1922, Carter was occupied with
a book dealing with the problems of electric traction of locomotives. In 1928, [12] he once again
went into detail on the stability problem. Carter passed away in 1952.
processes occurring during tangential contact, which were described by Boedecker

with a local version of Coulomb’s law, were understood. During his investigation in
1915, Carter applied this law intuitively (correctly). Not until 1926, however, did he
justify this application in another publication [11]. At the same time, the principles
of the tangential rolling contact problem were derived in Berlin in the dissertation
of Hans Fromm.6 They were, however, not published until one year later (1927)
[14]. In 1928, once again Carter engaged with the fundamentals of the stability of
locomotives. Ives Rocard,7 who published two volumes on the running stability
of locomotives in 1935 [18, 19], does not cite the work of Carter. The agreement
between the two, however, is in certain passages astonishing.
During World War II, research on the stability problems of rail vehicles was
stagnant, whereas the stability of vehicles with pneumatic tires was investigated
intensively in order to come to grips with the instability of aircraft on takeoff and
landing; see, e.g., [20].
1.3.2 A New Start After 1945: Japan and France
Right after the war, Germany had other problems than to advance high-speed rail
transport. In Japan, things were different. Arms research was, as in Germany, prohib-
ited after 1945. In 1946, the Japanese government assigned 120 maritime scientists
to advance fundamental research on high-speed rail transport. At the outset, nothing
was known of this work in Europe, probably due to the fact that all publications were
written solely in Japanese [21]. That situation did not change until the end of the
1950s.
In Europe, the SNCF (French National Railway Company) began research at the
beginning of the 1950s with the aim to develop a high-speed train.
Test runs on March 29, 1955, resulted in a disaster; see Fig. 1.4. On the custom-
built and restored line Bordeaux–Hendaye, a train with a locomotive type BB 9104
6 Hans Fromm was born in 1892 in Kreuznach Germany. From 1912 to 1920—interrupted by four
years of military service–he was a student of mechanical engineering at the University of Technology
in Berlin. In 1922, Fromm was employed as Oberingenieur (chief engineer) at the laboratory of
solid mechanics of the University of Technology in Berlin. In this position, he received his PhD in
1926. From 1931 on, Fromm, meanwhile promoted to professor, was the director of the laboratory
of solid mechanics. During his time in Berlin, Fromm was engaged in tire dynamics and the stability
of airplane running gears, among other things. In 1934, Fromm was called to the chair of mechanics
at the University of Technology in Danzig. Here he published works in the field of material science.
For a short time after the war, Fromm was assistant professor in mechanics at the university in
Mainz. He passed away in 1952.
7 Yves Rocard was born in 1903 in Vannes, France. In 1927, Rocard passed his doctoral examination
at the École Normale Supérieure in the field of mathematics. There he also accepted the chair of
physics from 1945 to 1973. In the field of mechanics, Rocard became well known through his work
on dynamics [15, 16], which in part was translated into English [17]. Furthermore, Rocard was
active in many other fields of physics. He passed away in Paris, in 1992. His son, Michel Rocard,
was the prime minister of France from 1988 to 1991.
8 1 Introduction
Fig. 1.4 Track displacement

after high-speed test of
SNCF between Bordeaux
and Hendaye on March 29,
1955 [22]
reached a top speed of 331 km/h. The track was displaced sideways by the loco-
motive’s hunting movements; the pantograph and the overhead traction line were
torn down. From today’s point of view, all that can be said for certain is that the
locomotive had begun to run unstably. The resulting near catastrophe probably had
one more cause. Not long before the trials, maintenance measures were performed
on the track. It has been suspected [1] that the resistance to lateral displacement was
decreased (see item 4 in Table 1.1), leading to a situation in which the high lateral
forces between the locomotive’s wheels and the rails during the sinusoidal motion
displaced the track. Whether the development of the high-speed train included inves-
tigation into stability in accordance with the recommendations put forth in Rocard’s
books [18, 19] twenty years earlier is not known.
In principle, the problems with high-speed rail traffic were known in Europe,
only it was not called unstable sinusoidal motion but rather the “hunting problem.”
In May 1955, the C9 committee (Interaction between vehicle and rail) of the ORE
(Office de Recherche et d’Essais de l’Union International des Chemins de Fer, the

research and testing bureau of the International Railway Union), announced a contest
for solving the hunting problem. It is tempting to suspect a direct connection between
the near catastrophe at SNCF and the contest. This was not the case, however, since
the preparations for the contest date to the year 1952. Of the three winners of the
contest, one was the Japanese naval engineer Tadashi Matsudaira. Essential parts of
the work were probably already published in 1952 in Japanese [21]. This work was
unknown in Europe, however. Unfortunately, the contest work was published only as
a report of the ORE C9 committee [23] and never in a journal. In the second part of
Chap. 11, we will return to Matsudaira’s work in more detail. In 1952, resulting from
Matsudaira’s work, a new bogie was designed for the Tokaido high-speed train [24].
At the same time, intensive theoretical investigations into the “hunting problem”
were conducted in England, mainly in Alan Wickens’s group [25, 26]. Wickens had
a basic understanding of aircraft design. He immediately concluded that instability
problems such as wing vibration due to aeroelasticity could be described by the
same type of equations as the instability problem of sinusoidal motion. It appears
that interdisciplinary ideas were needed to advance knowledge of basic mechanical
problems of rail vehicle dynamics. The Englishman Carter was an electrical engineer,
the French Rocard was a physicist, the Japanese Matsudaira had a naval background,
and Wickens worked in aircraft design.
At the time, the leading expert in Germany for all problems related to rail vehicle
dynamics was Carl Theodor Müller.8 In 1969, he authored a scientific paper giving
a very convincing presentation of the hunting problem [27]. He clearly differenti-
ated between forced vibrations and the free, self-excited vibrations that occur when
velocity is increased. Since Müller was an honorary professor at the University of
Technology in Munich, in addition to his duties as a director at the federal railways, it
can be assumed that he treated the problem with equal clarity in his lectures. Surpris-
ing, therefore, are the accounts in a volume by Hans-Ludwig Krugmann [28], who
considered himself a student of Müller. Although the volume, published in 1982,
aimed at “closing a gap,” the publications by the ORE C9 committee and the pub-
lications by Wickens are not mentioned. “Hunting” was explained as a problem of
forced vibrations caused by stochastic track irregularities.9 Although instability is
mentioned, the physical cause of instability, in contrast to Müller, is not recognized.
It may be assumed that outside of Müller’s lectures in Munich, such questions related
to high-speed traffic were not discussed at any German university.
8 Carl Theodor Müller was born in 1903 in Kherson, in Ukraine. From 1924 to 1929 he studied
mechanical engineering at the University of Technology in Berlin. In 1934, he received his PhD at
the University of Technology in Aachen under the direction of Hermann Heumann. In 1934, he was
employed in the experimental department of the Reichsbahn repair center in Grunewald, Berlin. In
1952, he became an assistant professor in the field of steam locomotive design and vehicle dynamics
at the University of Technology in Munich, and in 1961, he became an honorary professor. In 1962,
Prof. Dr.-Ing. Theodor Müller became the director of the research center in Minden. He passed
away in 1970 in Boppard am Rhein.
9 “If there were no more track irregularities, the sinusoidal motion would cease after a few half
waves.” ([28], p. 82).

10 1 Introduction
1.3.3 Research and Development Aimed at Overcoming

the “Boundaries of the Wheel/rail System”
Support Programs in the Federal Republic of Germany.

At the beginning of the 1970s, a massive amount of public money was put into
research on rail transport, which lead to the construction and testing of the Trans-
rapid, a magnetic levitation train, or maglev for short. As a consequence, an initiative,
mainly driven by Krupp,10 was initiated and financed by the minister of research and
technology. The initiative was called “Research on the boundaries of the wheel/rail
system,” which began in 1972 [29]. The main concern of the enterprise consisted
of three research areas: the area track, the area vehicle, and the area interaction
of vehicle and track. While high-speed trains were already in service in Japan and
France, research in Germany concerned itself with the fundamentals of running
stability and contact mechanics [30]. Polemically formulated, the heavy industries of
federal Germany learned, financed with public money, how to conduct investigations
into running stability. The result of the BMFT-backed research was, besides the fact
that the industry gained the tools to design rail vehicles, a test rig for whole rail
vehicles (located in München-Freimann), a test vehicle Intercity-Experimental (ICE),
which served as a prototype for the first series of ICE-trains, and, and this cannot
be underestimated, an understanding of the concepts of basic contact mechanics and
linear and nonlinear stability investigations (keyword: limit cycle calculation). From
1976 on, it was required that software be developed within the framework of the
undertaking. This development was mostly done at DLR,11 MAN Neue Technologie
(New Technology), in Munich, and the University of Technology in Berlin. This
resulted in the software Medyna and later in the software Simpack.
With these newly developed tools, the industry was able to conduct investigations
into riding comfort, stability, and curving behavior.
New System Boundaries at the Beginning of the 1990s.
One effect of the project to research the boundaries of the wheel/rail system was that
the boundaries were pushed into the daily praxis of rail traffic. The consequences of
higher velocities, higher axle loads, increased performance, increased traction, and
design changes on rails, vehicles, and the system as a whole were hardly investi-
gated. The results of these research and development undertakings must therefore be
described as subsystem optimization. It was foreseeable that the wheel/rail system
would hit new boundaries. These boundaries were primarily those governed by long-
term behavior. Initiatives from research institutions, universities, and industry that
requested from the ministry a new research project remained unsuccessful, mainly
because the available resources were bound up in maglev technology. In the latter
10 Krupp, since 1999 ThyssenKrupp, is a German industrial conglomerate that sprung from the Krupp
family business. In 1852/53, Alfred Krupp patented the seamless wheel rim, which significantly
contributed to the company’s economic success at the time.
11 National Aeronautics and Space Research Center of the Federal Republic of Germany.
half of the 1980s, several projects were carried out, primarily at universities. The
focal point of these projects was the track and the loads that result from the interac-
tion of wheel and rail. This resulted in a program at the DFG12 with the title “system
dynamics and long-term behavior of vehicles, rail, and subsoil.” For the classic rail
vehicle dynamics, the topics that were treated in this program are special problems.
We will briefly touch upon the problem of loading of vehicle components in the
context of fatigue in Chap. 15.
All research and development on vehicles as well as on track reveals that rail-born
high-speed traffic has become a driving force for new developments in Europe.
1.4 Railway Industry in Europe
Rail vehicle dynamics at the beginning of the third millennium not only looked back
at 150 years of history and 50 years of government-backed research and development.
It had also to be acknowledged that the rail vehicle industry has changed dramatically
in the last 25 years. In 1997, the German Institute for Economic Research published
a research report [31] that is illuminating in many aspects of the development of
German rail vehicle construction since about 1980. By advancing this compilation
another five years, a dramatic development can be observed. It has to be included
that the Canadian rail technology company Bombardier begins by taking over DWA
(Deutsch Waggonbau, German rail car construction) and Talbot in Aachen, followed
by the takeover of Adtranz. In 1980, there were almost 50 companies in the business
of building rail vehicles in central Europe. Today, there remain only three big sup-
pliers (Siemens Transportation Systems, Bombardier Transportation, Alstom) plus a
relatively small number of smaller suppliers including CAF, Talgo, and Stadler. The
remaining companies have all been consolidated. Figure 1.5 shows this integration
process for the period between (about) 1980 and 1996. It remains to be seen whether
this integration will continue and how many rail vehicle manufacturers will remain
in Central Europe.
In recent years, the Asian manufacturers of railway vehicles became stronger and
also began to enter the European market more systematically. In 2015, the acquisition
by Hitachi of AnsaldoBreda and Finmeccanica’s 40 % stake in Ansaldo STS was
announced. In China, in 2015, CRRC Corp. Ltd. was officially launched as the
world’s largest railway manufacturer with the completion of the merger of rolling
stock manufacturers China CNR Corp and CSR Corp. CNR and CSR were already
the largest manufacturers of rail vehicles before the merger.
It seems that manufacturers of freight wagon bogies (such as ELH in Germany
but also in Italy) remain independent.
12 German Research Foundation.

12 1 Introduction
Konzentrationsprozess der Systemhäuser in der Schienenfahrzeugindustrie
1980 1982 1984 1986 1988 1990 1991 1992 1993 1994 1995
WSSB, D
KM, D Integra CH
DUEWAG, D SGP, A
Siemens, D
Siemens VT
Häni, CH SAEL, D JV / KVB
Friedmann, A Krupp,D
AEG, D MAN / MBB, D LEV, D Von Roll, CH

Westinghouse, USA AEG
Kiepe, D
CCC Group, E ABB
Kalmar Verstad, S Hänglund, S Comeng, AUS Daimler- Benz
Sura Traction, S Ascan Scania, DK Strömberg, SF Sorefame, P Transportation
ASEA, S BREL, GB ML Engineering, GB
BBC, CH ABB
Energiebau ELTA, P
Br. Wheelset, GB Dresden, D
Garrett, USA Waggon Union, D
Scandia Randers, DK Thyssen-Henschel, D
TCO, F L.M.E., S
Strommens Verkstaed, N
Franco
Belge, F
Chantiers de GEC, GB
l’Atlantique, F ACEC,B MTM, E
ATENSIA, E
Brissoneau CIMT, F SEL, D MEINFESA, E LHB, D
et Lotz, F GEC
Alsthom, F Carel Fouche` De Dietrich, F Alsthom
Creussot Loire, F Metcam, GB
Jeumont Schneider, F
Quellen: Siemens; Fortschreibung: DIW.
Fig. 1.5 Integration process for Central European rail vehicle manufacturers exemplified for the
years 1980–1996 according to [31, 32]. The period after 1996 is described in the text
1.5 Overview of This Book
1.5.1 Classification
The subsequent Chaps. 2–15 can be classified into six groups:
• The Chaps. 2 and 3 deal with modeling, beginning with modeling of the vehicle,
the track, and excitation and then the modeling of wheel/rail contact.
• In Chaps. 4–6, forced vertical vibrations are examined. There, the frequency
domain for deterministic to random vibration is the focus.
• Chapter 7 builds on this, dealing with investigations of ride comfort.
• Chapters 8–13 investigate free lateral vibrations. One could still talk about fre-
quency domain investigations if a broadening definition is used. The practical use
of investigations like these are investigations of stability of rail vehicles. Chapter
13 is concerned with nonlinear stability observations.
1.5 Overview of This Book 13
• Chapter 14 deals with curving behavior investigations, including wear processes

in curves.
• The concluding chapter, Chap. 15, deals with stress calculations, which in the
opinion of the authors will increase in significance in future.
1.5.2 Vertical and Lateral Vibrations
The unusual division into forced vertical and free lateral vibrations is due to practical
problems, in one case comfort, in the other stability. While the stability behavior of
the vehicle is governed almost exclusively by lateral dynamics, comfort (and loading)
are governed by vertical as well as lateral dynamics. The essential parts of a comfort
calculation, however, can be identified by looking only at the vertical behavior.
1.5.3 Curving Behavior
The authors consider the field of curving behavior to be tricky. The classic quasistatic
curving theory of Hermann Heumann13 [34] is–as the title reveals–a static theory that
does not fit very well in a textbook on rail vehicle dynamics. That is also true for
more recent quasistatic curving behavior investigations conducted by large software
systems [35]. Dynamic curving investigations demand either elaborate time-domain
calculations that have to be computer-based or complicated stochastic considerations
[36]. On the other hand, it cannot be denied that demands for high running stability
and good curving behavior are competing desiderata. In order to understand the
different demands that result from these contradictions, Chap. 14, which deals with
curving behavior, was added to this book.
1.5.4 Frequency- and Time-Domain Calculations
Although the ever increasing performance of computers is leading to the implemen-

tation of time-domain solutions, this book will focus mainly on frequency-domain
solutions.
13 Hermann Heumann was born in Neubauhof, Germany, in 1878. In 1905, he went to the Preussische
Staatsbahn. In 1910, he finished his dissertation in Danzig, and in about 1920, he became professor at
TU Aachen. Heumann’s scientific work was mainly dedicated to quasistatic curving investigations
[33, 34]. By a number of simplifications that in principle are valid in tight curves, he succeeded in
developing a calculation methodology that was used by the industry during the second half of the
twentieth century. It became obsolete only when computer simulation became standard. Heumann
died in 1967, in Grafrath, Germany.
14 1 Introduction
• They are more suitable for principal investigations into the vehicle/track system.
• Furthermore, the frequency-domain solutions are better suited for understanding
the basics of the theories behind the commercial multibody simulations of today.
• It is possible in most cases to transfer the approach and findings from frequency-
domain calculations to time-domain calculations.
• Finally, an introduction to time-domain solutions would require an introduction
to the corresponding part of numerical mathematics, which would go beyond the
scope of this book.
References
1. B. Bergander, Private communication. Guest lecture TU (Springer, Berlin, 1993)

2. CEN. Railway applications: testing for the acceptance of running characteristics of railway
vehicles. EN 14363, CEN, Brussels, June 2005
3. J.A. Elkins, A. Carter, Testing and analysis techniques for safety assessment of rail vehicles.
Veh. Syst. Dyn. 22, 185–208 (1993)
4. N. N. Testing and approval of railway vehicles from the point of view of their dynamic behavior
– Safety – Track fatigue – Ride quality. UIC Code 518, 4th edition, UIC,from January 1997,
UIC, September 2009
5. K. Popp, W. Schiehlen (eds.), System Dynamics and Long-Term Behavior of Railway Vehicles,
Track and Subgrade, vol. 6, Lecture Notes in Applied Mechanics (Springer, Berlin, 2002)
6. C.F. Dendy Marshall, A History of British Railways down to the Year 1830 (Oxford University
Press, Oxford, 1938), pp. 147–148
7. A.H. Wickens, The dynamics of railway vehicles - from Stephenson to Carter. Proc. Inst. Mech.
Eng. Part F 212, 209–217 (1998)
8. F.J. Redtenbacher, Die Gesetze des Locomotiv-Baues (The laws of design of locomotives)
(Bassermann, Mannheim, 1855)
9. J. Klingel, Über den Lauf von Eisenbahnwagen auf gerader Bahn (On the running of rail-
way vehicles on straight track). Organ für die Fortschritte des Eisenbahnwesens, Neue Folge
20:113–123, Tafel XXI (1883)
10. C. Boedecker, Die Wirkungen zwischen Rad und Schiene und ihre Einflüsse auf den Lauf und
den Bewegungswiderstand der Fahrzeuge in den Eisenbahnzügen (The effects between wheel
and rail and their influences on the running behavior and the resistance of vehicles in railway
trains). Hahn’sche Buchhandlung, Hannover (1887)
11. F.W. Carter, On the action of a locomotive driving wheel. Proc. R. Soc. Lond. A. 112, 151–157
(1926)
12. F.W. Carter, On the stability of running of locomotives. Proc. R. Soc. Lond. A 121, 585–610
(1928)
13. F.W. Carter, The electric locomotive. Proc. Inst. Civil Eng. 201, 221–252 (1916). Discussion
pages 253–289
14. H. Fromm. Berechnung des Schlupfes beim Rollen deformierbarer Scheiben (Analysis of
creepage during rolling of deformable disks). Z. Angew. Math. Mech. (also PhD thesis TH
Berlin, 1926), 7:27–58, 1927
15. Y. Rocard, Dynamique générale des vibrations. 2. éd., revue et augmentée (Masson, Paris,
1949)
16. Y. Rocard, L’instabilité en méchanique. Automobiles, avions, ponts suspendus. Masson, Paris,
English Translation by Crosby (Lockwood, London, 1957). 1957
17. Y. Rocard, General Dynamics of Vibrations (Crosby, London, 1960)
18. M. Julien, Y. Rocard, La stabilité de route des locomotives, deuxième partie (Hermann & Cie,
Paris, 1935)
References 15
19. Y. Rocard, La Stabilité de Route des Locomotives, Première Partie, avec une note de M. R.
Lévy (Hermann & Cie, Paris, 1935). Teil II siehe [18]
20. K. Knothe, F. Böhm, History of stability of railway and road vehicles. Veh. Syst. Dyn. 31(5–6),
283–323 (1999)
21. T. Matsudaira, Shimmy of axles with pair of wheels (in Japanese). J. Railw. Eng. Res. pp.
16–26 (1952)
22. C. Billy, 331 km, h en, J‘y étais. La View du Rail 1785, 49–51 (1955)
23. T. Matsudaira, Paper awarded prize in the competition sponsored by Office of Research and
Experiment (ORE) of the International Unions of Railways (UIC). ORE–Report RP2/SVA-C9,
UIC, Utrecht (1960)
24. T. Matsudaira, Hunting problem of high speed railway vehicles with special reference to bogie
design for the new Tokaido line. Proc. Inst. Mech. Eng. 180, 58–66 (1965)
25. A.H. Wickens, The dynamics of railway vehicles on straight track: fundamental consideration
of lateral stability. Proc. I. Mech. Eng. 180(3F), 1–16 (1965)
26. A.H. Wickens, The dynamics stability of a simplified four-wheeled vehicle having profiled
wheels. Int. J. Solids Struct. 1, 385–406 (1965)
27. CTh Müller, Das Schlingerproblem in der Sicht von Vergangenheit und Gegenwart (The hunting
problem in the past and present). Glas. Ann. ZEV 93(11), 329–336 (1969)
29. DEC Hrsg, Statusseminar III. Spurgeführter Fernverkehr. Rad/Schiene-Technik. Berichte. Bad
Kissingen, März 1976, Frankfurt/Main, Deutsche Eisenbahn Consulting (1976)
30. W. Michels, Theoretische Untersuchungen der Laufstabilität des Radsatzes (Theoretical inves-
tigations of the running stability of a wheelset.. In Statusseminar III. Spurgeführter Fernverkehr.
Rad/Schiene-Technik. Berichte. (In: Status seminar III. Track guided long-distance traffic.
Wheel/rail-technique. Reports) Bad Kissingen, März 1976, pp. 13.1–13.31, Frankfurt/Main,
Deutsche Eisenbahn Consulting (1976)
31. R. Hopf, H. Kufeld, H. Link, J. P. Weiß, H. Wessels, Lage und Perspektiven der deutschen
Schienenfahrzeugindustrie (Situation and perspectives of the German railway vehicle industry).
Berlin (1997)
32. R. Hopf, H. Kufeld, H. Link, J.P. Weiß, H. Wessels, Der deutsche Schienenfahrzeugbau im
Zeichen der Globalisierung der Märkte (The German construction of railway vehicles under
the influence of globalization of markets). DIW-Wochenbericht 63(9), 149–156 (1996)
33. H. Heumann, Grundzüge des Bogenlaufs von Eisenbahnfahrzeugen (Principles of curving of
railway vehicles). Die Lokomotive 39, (1942)
34. H. Heumann, Grundzüge der Führung der Schienenfahrzeuge. Elektrische Bahnen, 49–
52:Sonderdruck von Arbeiten aus den Jahren 1950 – 1953 (Principles of guidance of railway
vehicles. Reprint of publications from 1950 – 1953 and 1954) (1954)
35. C. Bußmann, Quasistatische Bogenlauftheorie und ihre Verifizierung durch Versuche mit dem
ICE (Quasistatic curving theory and its verification by experiments with the ICE). Fortschrit-
tberichte VDI, Reihe 12, Nr. 338 (also dissertation TU Berlin). VDI-Verlag, Düsseldorf (1997)
36. J. Zhang, Dynamisches Bogenlaufverhalten mit stochastischen Gleislagefehlern - Modell-
und Verfahrensentwicklung unter Verwendung der Methode der statistischen Linearisierung
(Dynamic curving behavior with stochastic track irregularities – Development of models and
processes based on the method of stochastic linearization). VDI Fortschritt–Berichte (also PhD
thesis at TU Berlin), Reihe 12, Nr. 304. VDI–Verlag, Düsseldorf (1997)
Chapter 2
Modeling of Vehicle, Track, and Excitation
2.1 Prior Considerations and Coordinate Systems
In order to be able to examine the dynamics of a vehicle, the vehicle must first be
transformed into a mechanical model. For comfort calculations and stability investi-
gations (sinusoidal motion), it is usually enough to create models that can reproduce
the eigenfrequency of the system in the low-frequency range accurately up to about
25 Hz. In those cases, it is enough to model wheel axles and bogies as rigid bodies
that are connected to the carbody and each other by springs and dampers. The car-
body is usually modeled with six rigid-body degrees of freedom and the first elastic
eigenmodes. One often tries to use the symmetries of the vehicle during these cal-
culations, although railcars and locomotives rarely are built to be truly symmetric.
Ignoring small imperfections in the symmetry gives the advantage that the vertical
model, that is, the vertical and longitudinal movement, and the lateral model can
each be examined separately.
Software systems such as A DAMS R AIL, G ENSYS, M EDYNA, S IMPACK, VAM -
PIRE , and VOCO not only facilitate the work of the engineer to formulate and solve the
equations of motions, but also provide postprocessing tools to calculate the measures
needed to judge the amount of wheel/rail forces and ride comfort. These software
systems all compute three-dimensional coupled vertical and lateral models. In this
monograph, which is meant as an introduction, we will (almost) strictly distinguish
between vertical and lateral dynamic phenomena.
Investigations of fatigue strength, the calculation of the forces between wheel and
rail in the medium-frequency (40–400 Hz) range, the examination of vehicles running
on corrugated rails (500–1500 Hz), and acoustic investigations require models that
have the range of several thousand hertz.
Figure 2.1 shows a vehicle consisting of
• one carbody,
• two bogies, and
• four wheel axles,

DOI 10.1007/978-3-319-45376-7_2
18 2 Modeling of Vehicle, Track and Excitation
Fig. 2.1 Coordinate systems; nomenclature of motions
which we consider rigid bodies for simplicity. We differentiate between a rail-bound

(fixed) (x,y,z)-coordinate system and an (x , y , z )-coordinate system with velocity
v0 that moves with the vehicle. In this reference coordinate system, displacements
and torsion are measured as well. These are considered to be parasitic movements:
uz vertical motion,
uy lateral motion,
ux longitudinal motion,
ϕz yawing,
ϕy pitching,
ϕx rolling (often in combination with u y lateral motion).
Rolling in combination with lateral motion is called sway or body rolling.
2.2 Vehicle Modeling
2.2.1 Bogie Frame Design, Primary Suspension,

and Bogie Guidance
Figure 2.2 shows one type of Minden–Deutz bogie, which the the Deutsche Bahn has
used in many of its advanced designs. The horizontal bogie linkage to the wheelsets is
realized through maintenance-free elastic leaf spring guidance. The primary vertical
suspension consists of coil springs. There are bogies in which rubber blocks are used
as suspension in all three dimensions. In the figure of the cross section of the bolster
(Fig. 2.3), the bolster beam as well as the central carbody support in a pivot bearing
are shown. The lateral auxiliary or slide supports can be designed in a way that does
not make them load-bearing but only roll or sway preventing in function.
They can also be load-bearing, in which case they are used as friction dampers.
The bolster beam is suspended by cylindrical coil springs against the hangers. The
2.2 Vehicle Modeling 19
lateral bump stop longitudinal bump stop
bolster beam bolster beam
vertical bolster beam

- vertical longitudinal
primary spring pendulum secondary wheelset guidance
suspension
Fig. 2.2 Passenger coach bogie with two wheelsets, type Minden–Deutz with disk brake (MD 36),
from [1]
dampers installed with an inclination result in effective damping in the vertical as

well as in the horizontal plane transverse to the moving direction. In order to prevent
the bolster beam from swinging too far, there are elastic buffers installed in the
longitudinal and transverse directions. The clearance of the bolster in the longitudinal
direction is usually 5 mm, and the transverse clearance can be up to 70 mm.
curve dependent lateral play

carbody ~ 30 mm
lateral bolster play ~
centre-bowl
side bearer
progressive spring
bogie frame
bump stop for bolster pendulum
pendulum with links
lateral bolster guidance
bolster trough
secondary spring
Fig. 2.3 Details of bolster pendulum arrangement of MD-bogie, from [2]
Fig. 2.4 Characteristics of 40

lateral secondary spring
20
force (kN)
-20
-40
30 0 30
lateral displacement (mm)
For the guidance of the bolster beam in the hangers, so-called transverse control
arms—sometimes also nonwearing blade springs—are used, which do not decrease
the freedom of movement of the bolster beam. Due to the limiting buffer and therefore
the additional progressive springs used, the characteristic of the lateral secondary
suspension becomes highly nonlinear (Fig. 2.4).
In bogies of Minden–Deutz type, there is in most cases a friction damper installed
between the carbody and the bolster. On straight track, the friction damper does
Fig. 2.5 Passenger car bogie SGP 400 for second-generation ICE trains. Four brake disks and eddy
current brake
not usually break out. This occurs first on entering a curve. In case of sinusoidal
movements strong enough to cause the friction damper to break out on straight
track, it will act in a damping manner. The friction damper will have a stabilizing
influence on the instability caused by hunting. For the second generation of the ICE,
Siemens Verkehrstechnik GmbH developed the high-performance bogie SGP 400;
cf. Fig. 2.5. The SGP 400 bogie does not contain any friction elements, removing
one strong nonlinearity. Naturally, there are a number of other connecting elements
(cf. Fig. 2.4) that contain nonlinearities. These, however, are unproblematic, since
they can be linearized for small displacements.
The most commonly used bogie type in freight traffic is the Y25 bogie, which is
shown in Fig. 2.6. The Y25 bogie contains a large number of friction elements and
behaves therefore highly nonlinearly. A linear treatment is therefore not justifiable.
suspended side-bearers centre-bowl

frictional yaw damping after + 1mm longitudinal play
roll limitation after 12mm vertical play
(spring)
brake linkage
centre-bowl Lenoir - link
block brake block brake

load dependent
primary suspension
load dependent primary damping (two stages)
vertical-lateral friction plates
vertical-longit. friction plates
after10mm lateral displacement
Fig. 2.6 Y25 freight wagon bogie with two wheelsets [1]
An overview of other bogies that is still worthwhile is given by a book from the
former GDR [3].
2.2.2 Mechanical Model of the Vehicle. Connecting Elements
The basic model of a passenger coach consists of carbody, two bogie frames, and
four rigid wheelsets. Between carbody and bogie frame, the secondary suspension is
installed, and between the bogie frame and the wheelsets is the primary suspension.
Many design variants of bogie vehicles are traceable to the basic model shown in
Fig. 2.7. Figure 2.8 shows typical suspension elements that are available for modeling
in multibody simulation tools such as S IMPACK , M EDYNA , and G ENSYS.
For high-speed vehicles with four axles, however, there are some additional elastic
eigenmodes to consider in order to describe the comfort level correctly up to 25 Hz;
cf. Fig. 2.9.
2.2.3 Elastic Carbodies
As already mentioned, it is in general not sufficient for carbodies to consider only

the six degrees of freedom of the rigid body.
That is especially true for lightweight carbody designs. Finite-element models
of this kind of carbody are relatively easy to realize with today’s technology. The
inclusion of an entire FE model in a model for studying vehicle dynamics is hardy
possible due to the enormous demands on calculation time and memory access.
According to ISO [4], however, the frequency range up to 80 Hz should be included.
There are two possibilities:
z z
x y
Fig. 2.7 Simple mechanical model of four-axle vehicle. Six carbody degrees of freedom, six degrees
of freedom each for the bogie frames (The origin of the carbody and bogie coordinate systems are
at the height of the top of the rails.)
description graphical symbol
traction link (elastically linked)
coil spring (coupled via link)
torsional bar; torsional spring
damper (viscous)
friction damper (Coulomb - element)
torsional damper
bump stop (elastic)
bump stop (viscous)
bump stop (with Coulomb friction)
spring / damper - in series
complete coupling element,

realized as:
coil spring (flexicoil - spring) body i
rubber chevron
silent block
leaf spring
elastic rod (with tensile-, bending-, shear- and

torsional elasticity)
body j
air spring
Fig. 2.8 Simple coupling elements for multibody algorithms (top) and complex element (bottom)
1. The carbody can be approximated in a model as elastic beam with bending and
shear resistance as well as torsional stiffness, as done by [5]. The generalized
masses and stiffnesses [6] have to be chosen in such a way that the lower vertical
and lateral bending eigenfrequencies as well as the torsional eigenfrequency of
Fig. 2.9 Eignenmodes of a lightweight carbody for ICE (courtesy of DUEWAG, Krefeld
Uerdingen). Bending coupled with cross-section deformation at 10 Hz (top), dominant cross section
deformation at 13 Hz (bottom)
the free beam are in accordance with values that can be obtained from a standing
sinusoidal test during which the bogie frame is softly supported.
2. In dealing with a lightweight carbody, this is not a viable approach, since the lower
eigenmodes no longer consist of pure bending eigenmodes but already contain
heavy local deformations, as can be seen in Fig. 2.9. In these cases, the eigenvalues
and eigenmodes must be calculated from an FE model. For dynamic calculations,
only the lowest eigenvalues and eigenmodes are considered (modal reduction [6]).
In practice, this is a bit more complex, since the carbody is connected to the bogie
through the secondary suspension, and the forces that are introduced through the
secondary suspension into the carbody influence the vibrations. Details about this
can be found, e.g., in [7]. In large software systems this is already included.
2.3 Modeling of Track and Excitation
2.3.1 Track Modeling
In vehicle dynamics investigations, most of the time the track is regarded as rigid
and fixed. If it is desired to include in the simulations the forces between wheel and
rail up to frequencies of 200 Hz, that is, forces due to, for example, wheel flats, then
simple track models exist. In Fig. 2.10, such a model is inserted under the wheelset.
z z
x y
v0
d3 x d3
d3 c1 c1
c1 c1 2
d4 d4 2 2
c2 c2 c2 d4 c2
2 2 2
Fig. 2.10 Track modeling for frequencies up to 100 Hz

2.3 Modeling of Track and Excitation 27
The basic elements that are inserted in the vertical direction consist of a spring, and
in parallel, a damper and a Voigt–Kelvin element (parallel connection of spring and
damper).
In order to capture frequencies up to 300 Hz accurately, several of the basic ele-
ments have to be connected in consecutive order [8]. The problem in this context
is to determine the parameters, which is possible only with track dynamic methods
[8–10]. In Table 2.1, the parameters for the vertical model for three different types
of track from the DB-network ((1) track with B70 sleepers in an overhauled line,
(2) newly built track with B70 sleepers, and (3) test track on the southern bypass of
Stendal with B75 sleepers) are compiled. Negative values for the stiffnesses are no
problem, since all models have only eigenvalues with negative real parts. In principle,
it is possible to formulate analogous models for the lateral and sway motions.
Table 2.1 Vertical rail model consisting of three, respectively four, basic elements arranged serially.
If the basic models are to be placed beneath every wheel contact point, the values have to be divided
in half. Figure 2.10 shows only one of these basic models
Overhauled line medium New line hard subsoil Southern bypass Stendal
subsoil hard subsoil
Rail UIC 60 Sleeper pad Rail UIC 60 Sleeper pad Rail UIC 60 Sleeper pad
Zw700 Sleeper B70 Zw700 Sleeper B70 Zwp 104 Sleeper B75
Distance between Distance between Distance between
sleepers 0.6 m Ballast sleepers 0.6 m Ballast sleepers 0.63 m Ballast
thickness 0.3 m Subsoil thickness 0.3 m Subsoil thickness 0.4 m Subsoil
cs = 150 ms cs = 300 ms cs = 300 ms
c0 +0.9968E+08 +0.2434E+09 +0.1888E+09
i =0 d0 +0.4046E+06 +0.3824E+06 +0.2359E+06
c1 +0.4487E+09 +0.6395E+09 +0.1756E+09
c2 −0.4314E+07 −0.5351E+07 −0.1697E+08
i =1 d3 +0.1962E+05 −0.2655E+05 +0.4667E+05
d4 −0.2000E+05 +0.2594E+05 −0.4680E+05
c1 −0.1165E+11 −0.8016E+10 +0.5296E+09
c2 +0.2705E+09 +0.5398E+10 −0.1032E+09
i =2 d3 +0.8651E+06 +0.9640E+07 +0.3032E+06
d4 −0.8387E+06 −0.2222E+07 −0.2738E+06
c1 +0.2400E+10 +0.5959E+09
c2 −0.9275E+09 −0.1864E+07
i =3 d3 +0.1897E+07 +0.3582E+05
d4 −0.1309E+07 −0.3643E+05
2.3.2 Modeling of the Excitation
The excitation of rail vehicles is mainly the result of track irregularities and out-of-
round wheels, which act as excitation in the contact of wheel and rail. Furthermore,
there are excitations from out-of-balance wheels or from aerodynamic forces. These
excitations can be divided into
• periodical excitations,
• general, deterministic excitations, and
• stochastic excitations.
The excitations resulting from out-of-round wheelsets are always periodic, since they
will be repeated after one revolution. The basic harmonic frequency is
v0
f1 = , (2.1)
2πr
which with a wheel circumference of 2πr ≈ 3 m and a velocity of v0 = 60 m/s still
falls within the frequency range that is relevant for the vertical dynamics. The higher
harmonic frequency, e.g., by polygonization, which occurs in the ICE, results in very
uncomfortable medium frequency noise and vibrations (“buzzing” at 100 Hz), which
can severely damage track and wheelset; see, e.g., [10–13].
Tracks are never installed perfectly; there are always track irregularities that will
become worse during service. Tracks can be described by four types of track irreg-
ularities (cf. Fig. 2.11). The track gauge fault is relevant for vehicle dynamics only
in nonlinear considerations. Therefore, we can concentrate on the first three irregu-
larities, which are all shown in Fig. 2.12.
Vertical errors in the position and level of the track can, like out-of-round wheel
defects, be periodic, for example if the rail is not continuously welded. The track
irregularity has to be determined as the lowering of the rail under static load, since that
is exactly the disturbance that the wheel will experience. At the rail joints, a kink or
even a jump will occur. Hanging sleepers also often occur periodically, which results
in near-periodic track irregularities. Also, production processes (e.g., production of
slab track) can yield periodic track irregularities.
Much more common are deterministic singular irregularities, which in most cases
result from irregularities in the track, such as passing over switches and crossings,
difference in stiffness when driving onto a bridge, passing over level crossings, or
several hanging sleepers, which act upon the wheelset as a height error even if no
height difference can be measured in the track (e.g., with laser measurement).
The most common, the random (stochastic) track irregularities, will be described
in Chap. 6 as spectral power densities (Fig. 2.13).
In connection with the track model introduced in Fig. 2.10, a modified excitation
model is usually introduced (Fig. 2.14). In doing so, the wheelset will not pass over
a rail with a profile irregularity; rather, a band of disturbances without mass, with
velocity v 0 , is pulled through between wheelset and rail. The carbody (for bogie
vehicles) or bogie frame (for a single bogie) is fixed in the rolling direction. The
Fig. 2.11 Track irregularities (as in Renger [14])
Fig. 2.12 Sketch of track irregularities

(a) z t (x)
x
~30 m
rail length
z t (x)
(b)
x
~ 2.40 m
z t (x)
(c)
500 m
Fig. 2.13 Schematic examples of a periodic track irregularity (rail joints), b deterministic single
irregularity (hanging sleepers), and c stochastic vertical alignment
S x
v0 0
x
Fig. 2.14 Model of a moving irregularity (For the sake of clarity, the origin of the track following
an (x , z )-coordinate system is situated in the bogie frame and not, as is usual, at the height of top
of the rail.)
results from models with moving wheelset as compared to moving profile irregularity
differ only when one is working with continuous track models for high vehicle speed
or high excitation frequencies [10].
For the excitation model of a moving profile irregularity, the (0; x, y, z)- coordi-
nate system moves with constant velocity v 0 against the rolling direction, while the
coordinate system (S, x , y , z ) is fixed. Special attention is required when accel-
eration comes into play (acceleration, braking, curving). In those cases, additional
forces of inertia have to be considered.
The excitation of a band of disturbance values is sufficient for most cases of
vehicle dynamics investigations. Only when one is simulating passing over bridges
does the model reach its limit [15, 16].
References
1. Deutsche Bahn AG. Merkbuch für Schienenfahrzeuge - Reisezugwagen - Band 3 (DS 939 03)
(German Railways AG. Wishbook for railway vehicles. - Coaches - Volume 3), no year
2. H. Bugarcic, Grundlagen spurgebundener Fahrzeuge I (Basics of railway vehicles running on
guideways, I), Lecture notes (Institute of Railway Vehicle Technology, TU Berlin, 1985)
3. W. Hanneforth and W. Fischer. Laufwerke (Bogies). transpress, Berlin, 1986
4. ISO. Mechanical vibration and shock - Evaluation of human exposure to whole body vibration
- Part 4: Guidelines for the evaluation of the effects of vibration and rotational motion on
passenger and crew comfort in fixed guideway transport systems. ISO 2631-4. International
Organization for Standardization, Genève, 2001
5. R. Gasch, W. Hauschild, W. Kik, K. Knothe, H. Steinborn, Stability and forced vibrations
of a 4-axled railway vehicle with elastic carbody, in The Dynamics of Vehicles on Roads
and on Tracks, Proceedings of the 5th IAVSD - 2nd IUTAM Symposium held at the Technical
University Vienna, Austria, September 1977, ed. by A. Slibar, H. Springer (Swets & Zeitlinger,
Amsterdam, 1977), pp. 464–480
6. R. Gasch K. Knothe. and R. Liebich Strukturdynamik, Diskrete Systeme und Kontinua (2.
neubearbeitete Auflage) (Structural Dynamics, Discrete Systems and Continua). Springer,
Berlin et al., 2012
7. St. Dietz. Vibration and Fatigue Analysis of Vehicle Systems Using Component Modes.
Fortschritt-Berichte VDI (also PhD thesis TU Berlin), Reihe 12, Nr. 401. VDI–Verlag, Düs-
seldorf, 1999
8. Y. Wu. Einfache Gleismodelle zur Simulation der mittel- und hochfrequenten
Fahrzeug/Fahrweg-Dynamik (Simple track models for the simulation of mean- and high-
frequency vehicle/track dynamics). In Fortschritt-Berichte VDI (also PhD thesis at TU Berlin),
Reihe 12, Nr. 325. VDI–Verlag, Düsseldorf, 1997
9. U. Gerstberger, K. Knothe, and Y. Wu. Combined modelling of discretely supported track
models and subgrade models - vertical and lateral dynamics, Volume 6 of Lecture Notes in
Applied Mechanics, pages 247–264. Springer, Berlin e.a, 2002
10. K. Knothe, Gleisdynamik (Track dynamics) (Ernst & Sohn, Berlin, 2001)
11. D.R. Ahlbeck, J.A. Hadden, Measurement and prediction of impact loads from worn railroad
wheel and rail surface profiles. ASME J. of Engng. for Industry 107, 197–205 (1985)
12. D. R. Ahlbeck. A study of dynamic load effects due to railroad wheel profile roughness. In
M. Apetaur, editor, The Dynamics of Vehicles on Roads and on Tracks. Proceedings of the 10th
IAVSD Symposium held in Prague, CSSR, August 24–28, 1987. Supplement to Vehicle System
Dynamics, vol. 17, pages 13–16, Lisse and Amsterdam, 1988. Swets & Zeitlinger
13. A. Böhmer, T. Klimpel und K. Knothe., Dynamik und Festigkeit von gummigefederten
Radreifen (Dynamics and strength of rubber suspended wheel rims). ZEV+DET Glasers.
Annalen 124(3), 223–230 (2000)
14. A. Renger. Laufdynamische Berechnung von Schienenfahrzeugen. Ein Konzept zum Einbau
des Verfahrens der statistischen Linearisierung in das MKS-Programmsystem MEDYNA (Run-
ning dynamics of railway vehicles – A concept for the integration of the process of statistic
linearization into the MBS program system MEDYNA). ILR-Mitt. 264, Institute for aeronautics
and astronautics, TU Berlin, 1991
15. L. Frýba, Response of a beam to a rolling mass in the presence of adhesion. Acta Technica
CSAV 19, 673–687 (1974)
16. L. Frýba, Dynamics of Railway Bridges (Thomas Telford, London, 1996)
Chapter 3
Modeling of Wheel/Rail Contact
All forces between wheel and rail (Fig. 3.1) act on a contact patch of a size of about
1.5 cm2 . The weight of the vehicle is translated through normal forces; the guidance
through large-radius curves is provided by tangential forces; and during acceleration
and braking, additional tangential forces in the circumferential direction of the wheel
arise.
Since the contact between wheel and rail can be replaced by a constraint, normal
contact mechanics is actually irrelevant for rail vehicle dynamics. In dealing with
loading, however, the normal contact problem is of considerable importance. In the
contact patch of heavy vehicles, the normal forces can exceed 100 kN. In those cases,
the normal stresses have to be known accurately. Furthermore, the solution of the
normal contact problem is a precondition for the calculation of the tangential forces
relevant for the running dynamics.
We will first deal with the profile geometry of wheel and rail, Sect. 3.1, followed
by the issues of kinematics in the contact of wheel and rail, Sect. 3.2. Section 3.3
covers the normal contact problem, and Sect. 3.4 the tangential contact problem. The
last two can generally be examined separately from each other.
In contrast to Chap. 2, regarding the modeling of vehicle and track, this chapter
will also present the equations that will be necessary later in formulating the equations
of motion.
3.1 Profile Geometry
The maximum loading that occurs in wheel and rail strongly depends on the profiles
of wheel and rail. But in addition, the lateral contact mechanics and with it the whole
rail vehicle dynamics depend on the matching of wheel and rail profiles. Today,
the profiles of wheels and rails in central Europe are to a great extent standardized
[1, 2]. If a track is newly built or rails replaced, the UIC 60 rail, which comes with a
DOI 10.1007/978-3-319-45376-7_3
34 3 Modeling of Wheel/Rail Contact
Fig. 3.1 Forces and

creepage in the wheel/rail
contact. Acting forces are the
normal force N, the
tangential forces Tξ and Tη ,
and the spin moment Mζ . v0
From the wheelset velocities
v ξ , v η , and the angular
velocities ωζ as well as from
the rail velocities
(v ξ , v η , ωζ ), the relative
velocities (creep velocities) T
v
are calculated T
v
M
N
N
M
T
T
v
v
standardized profile (Fig. 3.2), is in most cases the preferred rail profile. The running
surface consists of a curvature with a radius of 300 mm. At both sides, this curvature
is connected with curvatures of 80 mm radius. At the gauge corner, the curvature has
a radius of 13 mm. These curvatures fade gradually into straight sections.
This rail profile is not a purely worn profile, that is, a profile that would result
from constant traffic on a straight track. That has the disadvantage that even though
it is only a question of fractions of a millimeter, due to the jumps in the radii of
the curved profiles, nonelliptical contact patches occur. This, however, is a preview
of later sections see Fig. 3.12. The occurrence of jumps in the radius can result in
higher maximum loads and cause in consequence a more complicated treatment of
the contact mechanics.
Therefore, an effort was made to replace the conventional UIC 60 profile with a
profile without jumps in the curvature radii on the running surface. Moelle [3], for
example, introduced instead of the different circular radii in the running surface an
elliptical arc with a continuous transition from 300 mm to 80 mm. The profile DB
3.1 Profile Geometry 35
Fig. 3.2 Rail profile UIC 60 72

with measures
52,053
20,456
14,3
R 80
R
13
R 300
51
R3
1:2,
75
Fig. 3.3 Wheel profile running circle

S 1002 with measures
70
32,5
70°
10
+y A1
13
R
R
,5
20
20
+z ,5
R1
2
R
R1
60E2 [4] developed by DB AG has a continuous transition all the way to the gauge
corner.
Also, the rail cant and the wheel profile influence the running dynamics. In the
network of DB AG, the rail cant is 1:40, while in the network of SNCF, it is 1:20.
Because of this, there are different wheel profiles in service, conical ones for the
SNCF but wear profiles for the DB AG. For wear profiles, the intervals between
reprofiling (assuming that the same type of rail steel is used) are longer, which leads
to lower maintenance costs. The most commonly used wheel profile in Europe is
the S 1002 (see Fig. 3.3), which is a wear profile for which the running surface can
be described by higher-order polynomial terms or pointwise [5, 6]. When it is used
together with the DB 60E2, there will be no contact jumps in the running surface.
Figure 3.4 presents a good overview of the development of worn profiles over time.
It is a presentation of the measured wear profiles [7] of a US freight wagon with a
total weight of 60 t. The progressive wear with increasing service time is clearly
visible. In order to conduct running dynamic investigations like these, it is necessary
to measure and analyze wheel profiles. That is also necessary if the effect of wear
profiles in curves (Fig. 3.5) on the curving behavior and loading is to be determined.
new profile
worn profile nr.1
worn profile nr. 2
worn profile nr. 3
Fig. 3.4 Wear profiles of a US freight wagon with a weight of 70 tons [7]
Fig. 3.5 Comparison of a new profile

worn profile measured on the worn profile
outer rail in a curve (Radius 0
= 519 m, rail cant 1:40,
v max = 100 km/h; freight
-10
Höhe [mm]
traffic and some passenger

traffic) with a new UIC 60
profile [8] -20
-30
-40
-10 0 20 40 60 80
[mm]
In measuring wear profiles at different points in a curve, it can be observed that the
profiles are congruent. The profile is determined by the service conditions (including
the design of the most common bogies) and by the curve radius. These parameters
do not change if the radius of the curve is constant.
3.2 Contact Kinematics Between Wheel and Rail
In the kinematics of the wheel rail contact, there is first the question of which contact
points will develop between wheel and rail if the wheelset and the rails are regarded as
rigid bodies and no deformations are allowed in the contact area and how the contact
points change when relative displacements between wheelset and rails occur. If it
3.2 Contact Kinematics Between Wheel and Rail 37
is required that there be continuous contact between wheel and rail, there are two
kinematic constraints. If the rail is regarded as fixed, then four of the wheelset’s
six degrees of freedom will be independent variables, and two will be dependent
variables.
It is not necessary to treat the kinematic relations as an example of a rigid wheelset
and a rigid rail. Instead, it is possible to assume an elastic wheelset axle allowing both
wheels to displace independently from each other. Instead of the kinematic relations
of a rigid wheelset and a rigid rail, the kinematic relations for a wheel/rail connect-
ing element become necessary. The linearized kinematic relations for a wheel/rail
connection element like this can be found, e.g., in [9, 10]. This possibility is imple-
mented, e.g., in the software systems M EDYNA, A DAMS R AIL, and S IMPACK.
Initially, the case of conical or circular wheel and rail profiles is treated, Sect. 3.2.1.
In this case, analytical solutions are possible. Subsequently, kinematic relations for
general profiles are considered in Sect. 3.2.2 that can be described only numerically.
General profiles in which small displacements from the center positions lead to a
jump of the contact point or to a fast change of the profile radii are not treatable
linearly. Instead, Sect. 3.2.3 shows how the relations can be linearized. These, in
turn, allow a transition to equivalent circular profiles, which is further treated in
Sect. 3.2.4. Finally, the kinematic relations are widened to cover cases in which rail
irregularities are present (Sect. 3.2.5). Furthermore, creep equations are determined
(Sect. 3.2.6).
3.2.1 Contact Kinematics with Conical and Circular Profiles
The kinematic relations were developed by, among others, Wickens [11, 12] and Joly
[13] in order to perform running dynamic investigations. A good overview is given
by the report of an ORE committee [14]. Also, Knothe [15], Matsui [16], and Mauer
[9] deal with kinematics; the most thorough work is probably the last one mentioned.
For reasons of simplicity, it is assumed that both wheel profiles as well as both
rail profiles are symmetric with respect to the center of the wheelset and track,
respectively, and can be described by circular or conical profiles. Combinations of
circular profiles will be treated in detail; conical profiles constitute a special case.
The nomenclature for undisplaced and displaced positions for circular profiles is
given in Fig. 3.6. The nomenclature is explained in the following:
e0 half the distance between the circle planes containing the contact points with
the wheelset in central position on the track,
r0 rolling circle radius at central position,
δ0 inclination of the tangent in the wheel contact patch at central position,
RW radius of curvature for the wheel profile,
RR radius of curvature of the rail profile.
e0
z
= rotation about
vertical axis
left wheel
track axis
x y, uy
y= rotation about
wheelset axis
r0
RR
0 left rail
R
W
zR z
e0 yR
x
e L
S rL
uz
S0
y
L
uy
Fig. 3.6 Wheelset in centered and displaced position; terminology
The wheelset is heading toward the observer. The designation “left rail” (Index L) is
from the point of view of an observer traveling with the vehicle. For the calculation of
running dynamics, which will follow later, the following parameters are of interest:
• the radii of the right and left rolling circles, that is, rR and rL , for lateral displace-
ment of the wheelset,
• the tangential inclination of the right (left) wheel contact patch (tan δR and tan δL ),
and
• elevation of the center of gravity uz (uy , ϕz ) as a function of the lateral displacement
uy and the yawing angle ϕz .
Running dynamics investigations will show that the difference in rolling radius r =
rL −rR has a significant influence on the stability of the wheelset, while the difference
zR
~uy
z0
eL
eR
rL r0 0
r0 yR
rR
y0
L
uy
e0 e0
Fig. 3.7 Wheelset with double coned profiles
in angle of contact or the elevation of the center of gravity are measures for how the
mass is pushing the wheelset back to the central position while displaced.
For conical profiles, it can be determined without complicated calculations that
the contact angle and the difference of rolling radii changes proportionally to the
tangent of the cone’s angle (Fig. 3.7). The difference in rolling radii and the difference
in contact angle result with good approximation in the relations
rL r0 + tan δ0 uy (conical profile) , (3.1a)

tan δ0
δL δ0 + uy (conical profile) . (3.1b)
e0
For circular profiles, the sought values for rL and δL are calculated through a Taylor
expansion of the nonlinear equation up to linear terms:

∂rL
rL (uy , ϕz ) r0 + uy ,
∂uy 0

∂δL
δL (uy , ϕz ) δ0 + uy .
∂u y 0
Terms containing linear ϕz do not exist, since the (x, z)-plane is also the plane of
symmetry.
Still of interest is how the roll angle ϕx and the vertical displacement uz change
when the wheelset is displaced in the lateral direction (uy ) or rotated around the
vertical axis (ϕz ):
u y max
left wheel
right wheel H
rL
rR
left
track centre line

rail
MR
MW
right
rail
Fig. 3.8 Wheelset with circular profiles in displaced position. (Mechanism with four links, accord-
ing to an idea of Lutz Mauer)

∂ϕx
ϕx (uy , ϕz ) uy ,
∂uy 0

1 ∂ 2 uz 2 1 ∂ 2 uz 2
uz (uy , ϕz ) u − ϕ .
2 ∂uy2 0 y 2 ∂ϕ2z 0 z
The fact that the roll angle has only one linear term and that the elevation of the
center of gravity has only two quadratic terms depends on the (x, z)-plane being the
plane of symmetry. The calculation of these relations is a rather cumbersome task.
If circular profiles are used for wheel and rail and if the displacement is limited to
the (y, z)-plane, then it is surprisingly still possible to state the nonlinear relations
accurately. The whole state of displacement can—keeping the constraints—be visu-
alized through a four-bar mechanism (Fig. 3.8). The center points MW of the wheel
profiles (which are fixed permanently to the wheelset) and the center points of the
rails MR , which are permanently fixed to the rail profile, are for this purpose linked
with a bar. In Fig. 3.8, the position of the wheelset during a lateral displacement for
such a combination of profiles is shown. The track (both rails and sleeper) has up to
now been regarded as rigid. The model is also valid if the track is displaceable but
still rigid. That is the case, e.g., for a rolling test bench or if track irregularities are
specified (see Sect. 3.2.5).
In the following, the nonlinear kinematic relations between uy , uz , and ϕx for a
rigid undisplaceable track are given [9, 15]:
2 2 gW
2
sin2 ϕx
uy + hW sin ϕx = RW − RR
gW
2
− 2gW gR cos ϕx + gR 2
−gW
2
sin2 ϕx , (3.2)
2 2 (gW cos2 ϕx − gR )2
uz + hW cos ϕx + hR = RW − RR
gW
2
− 2gW gR cos ϕx + gR 2
−(gW cos ϕx − gR )2 , (3.3)
with the abbreviations1
gW = e0 + RW sin δ0 ,
hW = e0 + RW cos δ0 ,
gR = e0 + RR sin δ0 ,
hR = e0 + RR cos δ0 .
Through expansion in series, the linearized relations for the sought geometric values
and the angle ϕx can be obtained,
rL(R) r0 ± λ uy (3.4a)
1
tan δL(R) tan δ0 ± ε uy , (3.4b)
e0
1
ϕx σ uy . (3.4c)
e0
The coefficients of linearity λ, ε, and σ in this context are still only abbreviations.
How these are obtained still has to be shown. For uz , the first terms that are not zero
are quadratic terms:
1 1
uz ζuy2 − χ ϕ2z , (3.5)
2 2
where, assuming the nonlinear relations for a limitation of the displacement in the
(y, z)-plane (Fig. 3.8), only the quadratic term in uy can be determined. A linearized
relation is obtained for the differential
duz ζuy duy − χ ϕz dϕz . (3.6)
1 The rail profile is convex: the rails’ radius of curvature is mathematically positive, since the center
of curvature is located on the inner normal. The wheel profile is concave: the radius of curvature
is negative. The value RR in Fig. 3.6 is the radius of curvature of the rail profile (with accurate
algebraic sign), while RW is the absolute value of the radius of curvature of the wheel.
The designation of the linear coefficients follows in general the definitions of the
ORE committee C 116 [14, 17]:
rL − rR
λ conicity (coefficients of the
2uy
differences of rolling radii),
e0 (tan δL − tan δR )
ε normalised coefficient of the difference of the
2uy
angle of contact,
ϕx e0
σ normalised coefficient of the angle ϕx
uy
(running parameter),
∂uz /∂uy
ζ coefficient of the gravitational stiffness of the
uy
lateral displacement,
∂uz /∂ϕz
χ coefficient of the gravitational stiffness of the
ϕz
yawing angle.
The complete linear coefficients for large contact angles δ0 are given below:

RW sin δ0 e0 + RR sin δ0
λ= , (3.7a)
RW − RR e0 cos δ0 − r0 sin δ0
e0 (e0 + RW sin δ0 )
ε= , (3.7b)
(RW − RR ) (e0 cos δ0 − r0 sin δ0 )
e0 sin δ0
σ= , (3.7c)
e0 cos δ0 − r0 sin δ0
sin δ0
ζ=
e0 cos δ0 − r0 sin δ0

e0 + RW sin δ0 e0 + RR sin δ0
+ 2 , (3.7d)
RW − RR cos δ0 e0 cos δ0 − r0 sin δ0
2
e0 − r0 tan δ0
χ = tan δ0 e0 − r0 tan δ0 + . (3.7e)
RW − RR
If we are dealing with only small angles of contact δ0 , it is possible to put sin δ0 = δ0
and cos δ0 = 1 and to disregard δ0 with respect to 1. In that case, the result for small
angles of contact δ0 will be
for conical profiles for circular profiles
λ δ0 , (3.8a) λ RW /(RW − RR ) , (3.9a)

ε δ0 , (3.8b) ε e0 /(RW − RR ) , (3.9b)
σ δ0 , (3.8c) σ δ0 , (3.9c)
ζ 2δ0 /e0 , (3.8d) ζ 1/(RW − RR ) , (3.9d)
χ δ0 e0 ; (3.8e) χ δ0 e0 + e0 /(RW − RR ) . (3.9e)
Equation (3.5) and the variables ζ and χ require an explanation. For a lateral
displacement, the center of gravity will be elevated, while for a yaw movement, the
center of gravity will be lowered. In order to achieve (or prevent) this, forces are
required that are proportional to the respective displacements. This is the concept of
“gravitational stiffness.” Both effects can be clearly demonstrated with the double
cone model.
3.2.2 Contact Kinematics for Arbitrary Profiles
The profiles of wheel and rail do not usually have circular shapes, as can be seen
in Figs. 3.2 and 3.4. The question now is whether it is possible to linearize arbitrary
profiles like these. In order to determine this, we consider the difference in rolling
Fig. 3.9 Rolling radius r = r L - r R [mm]

difference r as a function
of lateral wheelset 8
displacement uy (profile
combination ORE 6
S 1002/UIC 60 and ORE
S 1002/DB 60E2) for rigid 4
wheel and rail profile.
Source DB AG, FTZ 2
0
u y [mm]
-2
UIC60, 1:40
-4
60E2impr, 1:40
-6
-8 -6 -4 -2 0 2 4 6 8
flange root contact

flange contact tread contact
wheel
lateral displacement
-8 -7 -6 -5 -4 - 3 -2 -1 0 1 2 3 4 5 6 7 8 u y [mm]
rail
Fig. 3.10 Wheel and rail contact point as a function of lateral wheelset displacement for rigid
profiles (ORE S1002 on UIC60, rail cant 1:40, track gauge 1435 mm
radius 2r = rL − rR , depending on the lateral displacement uy , which is shown in

Fig. 3.9 for the combination of a new UIC 60 rail profile with a rail cant of 1:40 and
an S 1002 wheel profile. It is clearly visible that there are several jumps in the rolling
radius difference depending on the lateral displacement. A linearization in terms of
a Taylor expansion is possible for a lateral displacement up to about 0.25 mm. In
the comparison, the difference in rolling radius in using DB 60E2 is also registered.
It can be noted that there are no jumps up to 6 mm. The cause for the jumps in the
difference in rolling radius for the UIC 60 profile is to be found in the jumps of
the contact points; see Fig. 3.10. In this figure, the positions of the contact points of
the left2 wheel and rail profile in relation to the lateral displacement are delineated
through the bulky beams. The amplitude of the lateral displacement is indicated in the
middle. One can observe that the contact point of wheel and rail profile are moving
in opposing directions during a displacement of the wheelset. Figure 3.10 feigns that
there is an area of the running surface between wheel and rail in which contact never
occurs. This illusion depends on the fact that the geometry software RSGEO was
used, in which rigid profiles were used. If elasticity of wheel and rail is taken into
account by the software, then contact patches will develop, which also use areas for
load transfer where no contact of rigid bodies is indicated. Normally, no jumps of
the contact points occur in that case; rather, a smooth transition of the contact points
over the wheel and rail profiles occurs.
2 In
this case, it is the left wheel from the point of view of the observer traveling with the wheelset.
The wheelset is moving out of the figure.
The first jump for rigid profiles occurs already at uy = 0.25 mm. After this jump,
the differences in rolling radius and contact angle proceed on an almost straight
line, which does not pass through the origin, up to a lateral displacement of 4.8 mm
(Fig. 3.10). With increasing uy , the contact point first jumps to the flange root of
the wheel, and with more than 6 mm lateral displacement, to the wheel flange. If the
displacement increases even further, the wheel will start flange climbing. In this case,
more or less ±6 mm lateral displacement constitutes the track clearance.
3.2.3 On the Determination of Equivalent Contact

Parameters by Means of Quasilinearization
As long as there is no flange contact, a pronounced sinusoidal motion can be observed

for the ORE S 1002/UIC 60 profile combination, which suggests quasilinear behav-
ior. In these cases, it is desired to obtain an equivalent conicity λe with which the
approximate relation between difference in rolling radius r and the amplitude of
lateral displacement uy0 can be determined:
r(uy ) = rL (uy ) − rR (uy ) 2λe uy . (3.10)
We will once again limit ourselves to the case that the profile pair is symmetric with
respect to the midplane of wheelset and track when determining the equivalent conic-
ity λe . The equivalent conicity will then be
determined by the requirement that during
a sinusoidal motion uy (t) = uy0 sin ωt , the linearized difference in rolling radius
integrated over one period deviates as little as possible from the quadratic mean,
2π
[r(uy0 sin τ ) − 2λe uy0 sin τ ]2 dτ = Min . (3.11)
0
Through differentiation with respect to λe , the result will be

π
1
λe = r(uy0 sin τ ) sin τ dτ . (3.12)
πuy0
0
This approach to convert a nonlinear relation for r(uy ) into a linear relation is
referred to as quasilinearization, in the special case of a harmonic motion also referred
to as harmonic linearization [18–20]. If arbitrary track irregularities are taken into
account, a stochastic motion is present. Those cases are then referred to as statistical
linearization [21, 22].
For the equivalent conicity λe in relation to an assumed amplitude of lateral
displacement uy0 for profiles that are symmetric with respect to the wheelset and
track center plane respectively, it is true that
π
1
λe = [rL (uy0 sin τ ) − rR (uy0 sin τ )] sin τ dτ . (3.13)
πuy0
0
In an analogous way, an equivalent coefficient εe of the difference in contact angle

can be formulated:
π
e0
εe = [tan δL (uy0 sin τ ) − tan δR (uy0 sin τ )] sin τ dτ . (3.14)
πuy0
0
Finally, an equivalent rolling angle coefficient σe (equivalent running parameter) can

be determined,
π
2e0
σe = ϕx (uy0 sin τ ) sin τ dτ . (3.15)
πuy0
0
Figure 3.11 shows the equivalent conicity λe for the profile combination UIC 60
and S 1002. At amplitudes below ûy0 = 0.25 mm, the equivalent conicity has an
extremely low value. It then increases to almost 0.8, and for amplitudes of about 3 mm,
λe = 0.2. At DB AG, 3 mm is considered a realistic value for common disturbance
movement. For even higher amplitudes, the equivalent conicity decreases further,
until at 4.8 mm and at slightly above 6 mm (jump of the contact point to the flange)
it increases sharply. For the profile DB 60E2, the equivalent conicity also starts out
Fig. 3.11 Equivalent e

conicity λe as function of
0.8
amplitude uy0 of the UIC60, 1:40
0.7
wheelset for the combination 60E2impr, 1:40
of a rigid S 1002 wheel
0.6
profile and a rigid UIC 60
rail profile
0.5
0.4
0.3
0.2
0.1
0
0 1 2 3 4 5 6 7
wheelset amplitude u y 0 [mm]
low and then increases with the amplitude of the lateral displacement uy0 until just
above 6 mm, where it reaches a value of about 0.15.
The result is therefore that even profiles that are not represented by a circular arc
can be treated within a linearized calculation. The cost, however, is a dependency of
the equivalent coefficient of contact on the amplitude uy0 .
Still unsatisfying is the fact that the equivalent conicity jumps to high values for
small amplitudes of lateral displacement for the profile combination UIC 60/S 1002.
This can be adjusted only if—as already hinted above—an approximation of the
elastic behavior of wheel and rail is taken into account, which has the consequence
that also the nonelliptic contact patch is taken into account. Since the calculation of
nonelliptic contact patches is rather cumbersome, see Sect. 3.3.4, Kik and Piotrowski
[23], among others, have developed an approximate solution that determines the
nonelliptic contact patch from a penetration plane. The center of gravity of the plane
fills the position of the contact point; the differences in rolling radii therefore change
continuously.
3.2.4 Conversion into Equivalent Circular Profiles
The calculation of contact mechanics, especially the determination of the creep force-
creep relation, is particularly easy for circular profiles. A suggestion of how arbitrary
profiles are converted to equivalent circular profiles has been presented by Mauer
[9]. With the assumption that e0 and r0 are constant, a circular profile combination
that results in the same values for the parameters λ, σ, and ε as the initial profile is
determined. By resolving the relation for σ after δ0 = δe , the equivalent angle of
contact is determined,
σe
tan δe = , (3.16)
1 + σe r0 /e0
which in turn allows for the determination of sin δe and cos δe . By formulating the
expression
e0 λe
= RW,e sin δe , (3.17)
εe − σ e
it is possible to calculate the equivalent curvature radius of the wheel profile
λe e0
RW,e = . (3.18)
(εe − σe ) sin δe
Finally,
e0 (λe − σe )
= RR,e sin δe (3.19)
εe
results in the equivalent curvature radius of the rail,
e0 (λe − σe ) 1
RR,e = . (3.20)
εe sin δe
This illustrates a procedure how the input of amplitude of lateral displacement uy0 for
arbitrary profiles can be used not only for determination of equivalent linearization
parameters but also for geometric parameters (tan δe , RW,e , RR,e ) and therefore an
equivalent combination of circular profiles.
Even for the case that no symmetric profile pairs are present (e.g., wheelsets in
curves), it should be possible to project arbitrary profile pairs on circular profiles at
which the curvature radii for left and right will differ.
3.2.5 Linearized Contact Kinematics with Track

Irregularities
Figure 2.11 shows four types of track irregularities: lateral alignment yt , vertical
alignment zt , cross level error ϕxt , and gauge error yt . Twisting of both rails relative
to the sleepers or the track is taken into account only as an effect on the rail gauge,
since track irregularities over time become inelastic deformations of the track bed and
relative twisting of both rails is comparatively small [21, 24]. In order to investigate
linear forced vibrations caused by track irregularities like these, one has to broaden the
kinematic relations for this case first. That is not complicated for profile combinations
that are symmetric with respect to the midplane of the track and excite only vertical
vibrations. The vertical displacement must be equal to the height error. If longitudinal
and lateral alignment errors occur, the kinematics become more complicated.
In the following consideration, we will ignore the gauge error also as a source of
excitation for forced vibrations. As a symmetric parameter of excitation, the gauge
error causes forced vertical vibrations only in case of nonlinear considerations.
However, no lateral vibrations are caused. This does not imply that the gauge error
lacks significance. Small variations in the gauge can cause distinctive changes of
the contact points and in consequence changes of the contact point parameters. As a
parameter therefore, the gauge change has to be considered. Numerical investigations
of the influence of the gauge can be found, e.g., in Nefzger’s paper [5].
The remaining excitation parameters are therefore
yt lateral alignment,
zt vertical alignment, and
ϕxt cross level error.
We still regard the track as fixed, now, however, with a track irregularity. The wheel
displacements uyR and ϕzR are regarded as independent values of displacement.
Sought are the relations with which the dependent values of displacement uzR and
rotation ϕxR can be determined. We limit ourselves to the circular profile combination.
If ϕxR is assumed to be small, which always is the case, then no further calculation
is necessary.
For the determination of the difference of rolling radius and other contact para-
meters, it is irrelevant whether the wheelset is displaced relative to the rail in the
positive direction by uy or there is a negative lateral alignment yt present with the
same absolute value. It depends on the relative displacement uy − yt . The cross level
error ϕxt can be included in this consideration as well. A small cross level error ϕxt
of the track has the consequence that at the elevation of the center of gravity of the
wheelset, not yt is active, but rather yt − r0 ϕxt . The active lateral displacement is
consequently
uy,rel = uy − (yt − r0 ϕxt ) .
After this initial consideration, it is possible to state the linearized kinematic rela-
tions under consideration of track irregularities, namely for rolling radii and contact
angles,

rL(R) = r0 ± λe uy − yt + r0 ϕxt , (3.21a)
1
tan δL(R) = tan δ0 ± ε uy − yt + r0 ϕxt , (3.21b)
e0
as well as the roll angle and the elevation of the center of gravity,
1
ϕx = ϕxt + σ uy − yt + r0 ϕxt , (3.22a)
e0
1 1 2 1
uz = zt − r0 (ϕxt )2 + ζ uy − yxt + r0 ϕxt − χ ϕ2z . (3.22b)
2 2 2
Comment
Looking at Eq. (3.22b), it is initially surprising that the cross level error ϕxt appears
twice on the right-hand side, indeed once with the initial factor r0 /2 and once in
parentheses with the initial factor ζ/2. The term with the factor r0 /2 is formed by
observing the contact constraints without relative lateral displacement (namely uy =
yt − r0 ϕxt ); the wheelset turns only with the rail. The term in parentheses describes
the exact effect of a relative lateral displacement. The terms ϕx in Eq. (3.22a) can be
interpreted analogously.
3.2.6 Creepage Calculation
Part of the kinematic relation is also the equations necessary to determine the creep-
age, that is, the normalized relative velocities of wheel and rail at the contact point.
A detailed derivation of these equations can be found, e.g., in [9, 19, 24, 25].
The relations are given here for the sake of completeness. Only linearity assump-
tions are introduced. In connection with the formulation of the equations of motion
for lateral dynamics of a single wheelset (Chap. 9), the simplified form of the creep
equations will be shown.
We consider a wheelset with a velocity V = v 0 + v and an angular velocity
= 0 + . The wheelset is displaced by uy and rotated by ϕz around the vertical
axis. A superimposed velocity u̇x in the longitudinal direction and an angular velocity
ϕ̇y will be considered. The effects of track irregularities and a possible longitudinal
displacement of the contact point will not be considered.
Our goal is the formulation of the linearized relations. If quadratic terms appear in
the variables uy and ϕz , they can be disregarded as nonlinear terms. For this reason,
influences like uz (uy , ϕz ) do not have to be considered.
First, one obtains the relative velocities at the left contact point relative to a contact
point fixed coordinate system (ξ, η, ζ). The origin OK of the coordinate system is
found in the center of the contact ellipse (see also Appendix):

v ξL = (v 0 + u̇x ) − 0 + ϕ̇y rL − ϕ̇z eL , (3.23a)
v ηL = u̇y cos δL + ϕ̇x (eL sin δL + rL cos δL )
(v 0 + u̇x ) + (0 + ϕ̇y ) rL
− ϕz cos δL , (3.23b)
2
ωζL = − 0 + ϕ̇y sin δL + ϕ̇z cos δL , (3.23c)
or, after introduction of the relations rL , δL and ϕx of the Eqs. (3.4a)–(3.4c), and
disregarding terms that contain uy , ϕz in quadratic form or as a product, one gets

v ξL = (v 0 + u̇x ) − 0 r0 + λe uy − ϕ̇y r0 − ϕ̇z e0 , (3.24a)
σ
v ηL = u̇y cos δ0 + u̇y (e0 sin δ0 + r0 cos δ0 )
e0
v 0 + 0 r0
− ϕz cos δ0 , (3.24b)

2
ε
ωζL = −0 sin δ0 + uy − ϕ̇y sin δ0 + ϕ̇z cos δ0 . (3.24c)
e0
In order to determine the creepage, a reference velocity is required. As in Kalker

[26], the mean velocity
v 0 + 0 r0
vm =
2
is used. Simultaneously, it is assumed that the angle δ0 is very small and can be
disregarded compared to 1. The resulting creepages are
0 λe 1
νξL = νξ0 − uy + u̇x − r0 ϕ̇y − e0 ϕ̇z , (3.25a)
vm vm

1 σ r0
νηL = −ϕz + 1+ u̇y , (3.25b)
vm e0
0 ε δ0 1
νζL = −νζ0 − uy − ϕ̇ + ϕ̇z , (3.25c)
e0 v m vm y vm
in which
v 0 − 0 r0
νξ0 = ,
vm
0 δ0
νζ0 = .
vm
3.3 Normal Contact Mechanics
3.3.1 Overview of the Calculation of Contact Point Stresses
As long as one is interested only in the position of the contact point and the relative
velocities at the contact point, it is usually sufficient to consider wheel and rail to be
rigid nondeformable bodies.
Under the influence of the normal forces, a contact patch is formed around the
contact point. The determination of the contact patch is a part of the normal contact
problem. As soon as relative velocities are present, tangential forces also begin to
appear. The determination of these forces is a part of the tangential contact problem.
In many cases, it is possible to treat the normal contact problem (determination of the
contact patch and the normal forces arising there) separate from the tangential contact
problem. The solution of the normal contact was already performed by Hertz (1882)
[27, 28]. The derivation of the solution of the normal contact problem is described
in a number of textbooks (e.g., [29–31]). We will limit ourselves to a discussion of
the assumptions, the statement of the solution, and a qualitative evaluation of the
solution.
3.3.2 Assumptions of the Normal Contact Problem
If the normal contact problem of wheel and rail is treated in the framework of Hertz
theory or with the program C ONTACT by Kalker [32], the following assumptions
have to be made:
1. Kinematic linearity: Linear kinematic relations apply.

2. Linearity of material, elasticity: The material is linear–elastic.
3. Half-space assumption: The two bodies in contact (wheel and rail) can be
regarded as half-spaces.
4. Assumption of smoothness: The surfaces are completely smooth.
5. Assumption of separation: Structural and contact mechanics can be treated
separately from each other.
Specifically for Hertz theory, the following holds:
6. Hertz assumption: The surfaces of both bodies can be described as planes

of second degree.
For the material of wheel and rail, the following additional assumptions apply:
7. Material equality: Wheel and rail material are identical.

8. Assumption of homogeneity and
9. assumption of isotropy: the material is homogeneous and isotropic.
The running speed must fulfill the following assumption:
10. Running velocity: The running speed is small compared to the smallest
velocity of wave propagation in the material.
For some of the assumptions, it is immediately plausible that they are fulfilled
with good approximation for the wheel/rail contact. For others, fundamental inves-
tigations are necessary in order to determine how much the reality deviates from the
assumptions and what consequences these deviations have. That will not be done
here.
For two bullet points, however, the consequences of these assumptions will be
stated explicitly:
3.3 Normal Contact Mechanics 53
• Due to assumption 10, forces of inertia for moved loads can be neglected; see also
[33]. The normal contact problem can always be formulated statically.
• Due to assumption 7 together with assumptions 3 and 5, the normal contact problem
can be treated separately from the tangential contact problem.
3.3.3 Nonelliptic Contact Patches
Since the curvature for the rail UIC 60 profile changes abruptly in the rolling contact
area, the Hertz assumption for this profile is violated. This results in nonelliptic con-
tact patches with nonellipsoidal stress distributions. To approximate the nonelliptic
contact patch with an elliptical contact can result in errors if one is interested in
the contact stress maxima independent of the approximation method. In that case, a
process must be used that allows for the nonelliptic contact patch to be calculated.
Le The [34] was probably the first to calculate nonelliptic contact patches like these
for the wheel/rail contact. A simple approximate solution for the contact patch shape
is given by Kik and Piotrowski in [23]. A more exact solution for general half-space
contact problems can be determined with the software C ONTACT [32, 35, 36].
Figure 3.12 shows the profile combinations for S 1002 and UIC 60 and compares
the nonelliptic contact patches and elliptical contacts that result from the radii in
the contact point. The nonelliptical contact patches were determined by means of
special boundary element procedures. The calculation of the contact ellipses was
done according to Sect. 3.3.6.
3.3.4 Hertzian Treatment of the Normal Contact Problem
Introductory Remark
All considerations concerning the basics of the normal contact problem where moved
to the appendix (Sect. 16.3) to simplify the main presentation. In the following,
the focus is to present the necessary relations with which basic stress calculations
can be performed, when required, for rail vehicle dynamics. We limit ourselves to
the treatment of contact problems for elliptical contact patches. In addition to the
assumptions already made, we also assume that the main curvature planes for both
bodies coincide with the planes fixed through the contact coordinate system.3 In
illustrating this, we follow Johnson [29] to a large extent.
The Task
With the assumptions above, under the influence of a normal force N, a contact
ellipse between both bodies is formed, with the contact radii a and b and an approach
3 That is not necessarily the case for forward displacements of the contact point in tight curves.
Fig. 3.12 Nonelliptical x s [ mm ]

contact patches on the right u y = - 2mm
rail as a function of the
lateral displacement for a u y = -2 mm
UIC 60 and S 1002
wheel/rail profile
combination (according to
Le The [34])
u y = -1 mm
ys , u y = 0 mm
20 10 -10 -20
u y = 1 mm
u y = 2 mm
δ between both bodies. In the contact patch, a stress distribution with a half-ellipsoidal
shape with the maximal value of p0 is formed.
For given surfaces of both bodies, that is, given curvature radii (R1ξ , R2ξ , R1η ,
R2η ), as well as given material properties, the following has to be calculated for a
known normal force N:
1. the contact radii a and b,
2. the elastic deformation δ, and
3. the maximum surface pressure p0 .
Designation of Curvature Radii
The designations for the longitudinal and lateral radii of the two bodies in contact
can be seen in Fig. 3.13. For a concave surface (center of curvature outside the body),
a negative radius of curvature has to be employed. That is usually the case for wheels
at the running surface area and in the flange root. New rail profiles, however, are all
convex, and the radii of curvature are therefore positive.
Equivalent Radii of Curvature
Both surfaces are combined to an equivalent surface through the addition of
z-coordinates that add the curvatures (Fig. 3.14). This corresponds to the formation
of equivalent radii of curvature:
M 1
wheel
body 1
1
R 2=
R
1
1
1
2 2
rail 2
body 2
2
R
1
R
M 2
R 2 b
a
M 1
Fig. 3.13 Curvatures in the wheel/rail contact. (In case of contact, the two coordinate systems
(ξ1 , η1 , ζ1 ) and (ξ2 , η2 , ζ2 ) coincide.) The radii are named according to Hertz, i.e., a > b
f1 R 1
f 1+ f 2
R*
d d
f2
R 2
Fig. 3.14 Description of the surfaces of two bodies in the vicinity of the contact with second-order
parabolas (f1 , f2 ) (left) and replacement with the case of a parabolic stamp on a plane (right). Shown
are the cuts with the (η, ζ)-plane. Here d is the approach of the two bodies
1 1 1
∗ = +
Rξ R1ξ R2ξ
1 1 1
∗
= + . (3.26)
Rη R1η R2η
Equivalent Mean Radius of Curvature

∗
The geometrical mean Rm can be formed from these two parameters,

∗
Rm = Rξ∗ Rη∗ , (3.27)
which can be interpreted as the radius of a sphere. We call it the equivalent mean
radius of curvature.
Equivalent Modulus of Elasticity
If both bodies are made of different materials, it is common to combine the modulus
from the individual bodies into one equivalent modulus of elasticity E ∗ . For bodies
of different materials, one has

−1
1 − ν12 1 − ν22
E∗ = + . (3.28)
E1 E2
For the case of identical materials that was assumed earlier, one has that (E1 = E2 =
E and ν1 = ν2 = ν), which leads to
E G
E∗ = = . (3.29)
2(1 − ν 2 ) 1−ν
3.3.5 Spherical and Point Contact
We will first deal with the contact of a sphere (radius R) with a plane, since the
calculation for this case is particularly easy and the same relations can be used for
ellipsoidal contact with good approximation. The result is a contact patch with the
shape of a circle and contact radius c:

3 3 1 − ν
3 1 2
c= NR =
3
NR (3.30)
4 E∗ 2 E
The structure of this equation results also from the examination of dimensions. Only
the initial factors require an accurate calculation.
The elastic deformation δ results in
c2
δ= . (3.31)
2R
Sought is finally the maximum compressive stress
3
p0 = N. (3.32)
2πc2
3.3.6 Ellipsoidal Contact
For the contact of two ellipsoids with different equivalent radii of curvature Rξ∗ and
Rη∗ but the same main plane of curvature, an elliptical contact patch with radii a and
b will form. For the calculation of a and b, we adopt the approach of Johnson [29];
see Fig. 3.15.
First, a mean contact radius c is calculated from the equivalent radius of curvature
∗
Rm , √
c = ab . (3.33)
For this mean contact radius c, the following relation is obtained:

3 1 3 (1 − ν 2 ) ∗
c= NR∗ F1 (e) = NRm F1 (e) .
3 3
(3.34)
4 E∗ m 2 E
That is, as expected, essentially the same relation as for the spherical contact, see
Eq. (3.30), except from the factor F1 (e), which is a combination of elliptic integrals
and is dependent on the eccentricity e of the ellipse. In Fig. 3.15, the solution is
presented in a way that F1 depends on the proportion g of the relative radii of
curvature. For Johnson [29] as for Hertz, this relation is formed in a way that ensures
that it is always greater than or equal to 1:
R
g = = max(Rξ /Rη , Rη /Rξ ) . (3.35)
R
For the case of spherical contact, g, and therefore also F1 , becomes equal to 1. Even
for R /R = 100, F1 is still larger than
√ 0.8. √
In Fig. 3.15, the relation of (b/a) R /R as a function of R√ /R is shown, with
which the semiaxes of the ellipse can be determined from c = ab:

a = c a/b and b = c b/a.
In order to determine the elastic deformation δ, that is, the approach of the two
bodies, an additional value F2 is needed:

2
∗
3 NRm 1 c2 F2
δ= ∗ F2 = .
3
(3.36)
4 E∗ Rm ∗
Rm F12
The maximum compressive stress is

1.0
0.9 2.0
F1
1
2
1 -3
b R 2
F
1 F2
=
a R
3
b R 2
F
0.8 1.8
a R
F1 F3
0.7 1.6
-1
R 6
F2 R
0.6 1.4
0.5 1.2
1
1.0 3.0 10 30
1
R 2
R
Fig. 3.15 Functions for the elliptic integrals for the normal contact calculation according to Johnson
(Fig. 4.4 in [29])

3 6 NE ∗ 2 1
p0 = N= 3
. (3.37)
2πc2 π 3 Rm
∗ 2
F12
3.3.7 Contact of Rollers, Line Contact
Contact of rollers is a limiting case that will be treated separately. The point load N
is substituted by the line load NL (unit N/m). The contact band has a width of 2a.
The mean radius of curvature is still denoted by

−1
1 1 1
= + .
R∗ R1 R2
As before, an equivalent modulus of elasticity

−1
1 1 − ν12 1 − ν22
= +
E∗ E1 E2
Fig. 3.16 Analogous model N1

of a linear contact spring for N
the normal contact problem
and derivation of the Hertz N0
spring stiffness from the
slope of the tangent
is introduced. As before, for the same materials, one has
∗ E G
E = = .
2(1 − ν 2 ) 1−ν
For half the width of the contact band, a is calculated to be

4 1 8 (1 − ν 2 )
a = NL R∗ = NL R∗ , (3.38)
π E∗ π E
and for the maximum compressive stress p0 ,

2 NL E
p0 = NL = . (3.39)
πa 2(1 − ν 2 ) π R∗
During the derivation of the half-space theory, it becomes clear that it is impossible to
determine the elastic deformation δ in a similarly simple manner as for the ellipsoidal
contact (point contact).
3.3.8 Linearized Replacement Model
For system dynamic investigations in the field of vehicle and track interaction, it is in
most cases sufficient (and for investigations in the frequency domain it is necessary)
to replace the nonlinear relation between approach δ and normal force by a linear
spring. The stiffness of the spring is derived by linearizing a static reference normal
force N0 (Fig. 3.16).
For the three-dimensional contact, the resulting replacement stiffness is

dN 3 N0
ch = = . (3.40)
dδ δ0 2 δ0
3.4 Tangential Contact Mechanics
3.4.1 Introduction to the Tangential Contact Problem
Historical Remark
The task of tangential contact mechanics for rail vehicle dynamics consists in deter-
mining the tangential forces in the contact patch, the so-called creep forces, in relation
to the relative velocities at the contact point, the so-called creep velocities. In the
first correct lateral dynamic calculation in 1916 [37], Carter used a linear relation
between the creep forces and the creepage:
creep force = creep coefficient × creepage.
Surprisingly, there exists a work from the nineteenth century that deals with tan-
gential rolling contact mechanics. It was written by Reynolds [38]. Reynolds noted
during trials with rubber rollers that the resulting contact patch consisted of static fric-
tion zones and creep zones. Carter’s investigations of tangential contact mechanics
between 1916 and 1926 were probably influenced by the work of Reynolds [39].
The solution to the tangential contact problem with creepage during rolling of a
wheel on a rail for the simple case of a roller on a plane was given first by Carter [40]
in 1926. At the same time, in Berlin, Fromm treated the same problem, where he
considered not only the case of contact between two elastic half- spaces but also the
contact between two cylinders and the contact of a cylinder inside a hollow cylinder
(conformal contact, roller bearing). The dissertation dated 1926 was published in
ZAMM in 1927 [41]. Much later, in 1950, Poritsky [42] once again considered the
tangential contact problem for the rolling of two rollers of the same material. An
extension for rollers of different materials was published by Bufler (1959) [43]; the
solution for rollers with lateral creepage was given by Heinrich and Desoyer (1967)
[44]. A good overview by Johnson is given in his book [29]. In his dissertation in
1967, Kalker investigated the problem of rolling contact for elliptical contact patches
with longitudinal and lateral creep [45]. Since that time, the most general form of the
solution is usually called Kalker’s theory. An extensive presentation can be found
in Kalker’s book [26]. For calculations, Kalker’s software C ONTACT [35] is usually
the software of choice these days.
Definition of the Sign
Without derivation, we aim to explain the creep force creep relation using the example
of a braked wheel. A braked wheel is considered because the creepage and the creep
force become positive with our sign convention.
In Fig. 3.17a, an unloaded wheel is rolling without relative velocity between wheel
and support,
v rel ≡ v 0 − 0 r0 = 0 (pure rolling) .
3.4 Tangential Contact Mechanics 61
(a) (b) (c)

0 MB MB
Q B
r0 v0 B v0
Q Q
T
v0 = 0 r 0 v0 - r0 > 0 N
( pure rolling ) ( braking ) N
T
Fig. 3.17 Cylinder on plane. Pure rolling (a), rolling with braking (b), and free body diagram of
braked cylinder (c)
Next to it, a braked wheel is shown. Initially, a wheel is shown that is separated
from the bogie frame and the disk brake (Fig. 3.17b). The active vertical force Q
(which also includes the weight of the wheel) can be recognized as well as the force
B that results from the braking process and the braking torque MB . On the right-hand
side, a free body diagram of the wheel also reveals the normal force N and the creep
force T (Fig. 3.17c).
With the help of this example, the definition of the sign can be explained as well.
Creep forces are positive if they act on the rail in the positive coordinate direction
at a contact point fixed coordinate system. For the creep force T in Fig. 3.17, that is
the case. In order to define the algebraic sign of the creep, first the velocities of the
contact point on the wheel and the rail side are needed.4 On the wheel and rail side,
the contact point velocities are positive in the direction of the contact point fixed
coordinate system. The differences between the wheel and rail velocities constitute
the creep velocities. Divided by a reference velocity, this results in the creepages.
Phenomena that Occur with Friction
During the transfer of tangential forces in the contact patch, as already noted by
Reynolds, elastic deformations of both bodies occur, and at least in parts of the
contact patch, sliding occurs. These sliding occurrences are phenomenologically
complicated, and there is so far no complete mathematical–mechanical treatment for
this problem available. In the following, the most important, in some cases competing,
influence factors in the boundary layers of the friction partners are listed; see also
[46, 47].
• Through mechanical impact such as wear, plastic deformation, and hardening as
well as chemical activation of the running surfaces, the material properties change.
• By physical absorption of foreign particles, e.g., oxygen accumulation, a covering
layer can build up on the running surfaces.
• In these situations, chemical reaction layers form (fretting corrosion).
4 Velocities on the rail occur, for instance, on a roller rig but also during investigation in the medium-
and high-frequency ranges when the rail is no longer regarded as rigid and fixed.
• The physical properties of these reaction layers can be further altered due to
mechanical influence or the influx of even more foreign particles (absorption of
water).
• Finally, the resulting covering layer can initiate a chemical reaction with the base
material (e.g., electrochemical corrosion).
For all these phenomena, there are no generally accepted physical theories.
A mathematical mechanical description is possible for dry friction of two solid
bodies without third body parts or covering layers. The above-mentioned factors are
not taken into account, however. Most cases are limited to elastic deformation, and
thus no plasticizing processes are included. The usual argument for this is that after
running in of the surface, the plastic deformation of the wheel-rail surfaces is to a
large extent finished. Through the formation of residual stresses, work hardening,
and plastic redistribution, the profiles have adjusted to each other in a way that the
stresses during subsequent overrollings are absorbed elastically (Melans shakedown
theorem [48]).
Additional Assumptions
Let us again recall that it was assumed for the normal contact problem that both
bodies were made from the same material (Assumption 7) and that the running
velocity is small enough (Assumption 10). Because of these assumptions, the normal
and tangential contact problems can be treated separately. For the tangential contact
problem, it is not necessary to consider inertial forces (see also [33]).
Nevertheless, a number of additional assumptions are needed, which are presented
in the following:
11. Dry friction: It is assumed as explained above that friction occurs under
dry conditions.
12. Coulomb assumption: It is assumed that the globally valid Coulomb law
is transferable to local processes in the contact patch. The coefficient of
adhesion and coefficient of sliding friction are equal. The coefficient of
sliding friction is not dependent on state variables or on the sliding direction.
13. Steady-state assumption: The diameter of the contact in the rolling direc-
tion is very much smaller than the characteristic wavelengths of relevance
for the movement process. Therefore, the rolling contact can be treated as
a steady-state process.
For assumptions 12 and 13, further explanations are needed: the transferability
of Coulomb’s law to local processes in the contact patch is, as long as nothing
more is known of these processes, a plausible solution. It is possible, however, that
the coefficient of sliding friction depends on other state variables (normal stress,
temperature) and the direction of sliding (longitudinal or lateral) in the contact patch.
The assumption of steady state is valid for low-frequency rail vehicle dynamics
(typical case: sinusoidal motion). It is not valid for rolling over corrugation, since
the contact diameter (about 1 cm) has the same order of magnitude as the wavelength
(2–8 cm) [10, 49, 50].
In using the software C ONTACT developed by Kalker [35], a number of these
assumptions can be ignored. The computing time increases considerably, however.
3.4.2 Analytical Solution for Rolling Contact (Line Contact)
The steady-state tangential contact problem for rolling of a roller on a plane with
longitudinal creepage (elliptically distributed surface pressure) was solved by Carter
[40] and Fromm [41] in 1926. In the following, we will refer to Carter, since the
solution by Fromm, who introduced fewer preconditions, is more complicated. The
roller is a model of a wheel, and the plane is a model of a rail. Carter’s result was that
the tangential stress distribution can be represented as the difference of two circles
(or ellipses); see Fig. 3.18. Details of Carter’s solution can be found in the appendix
(Sect. 16.3.3).
In order to simplify the presentation, an additional coordinate system is introduced:
ξ ∗ = ξ + a0 − a0∗ . (3.41)
The center of one of the circles is O, the center of the other one is O∗ ; see Fig. 3.18.
We consider the case of braking. The distribution of the tangential stress in the
contact patch is illustrated in Fig. 3.18. The direction of travel of the wheel is from
right to left, and q1 (ξ) is the maximum possible tangential stress, that is, the normal
stress p(ξ) multiplied by the friction coefficient μ. For the illustration, q1 (ξ) was
normalized in a way that it results in a semicircle. A particle enters the contact patch
at the border and sticks initially if the creepage is not too large. In the adhesion area,
the tangential stress increases continuously, since a constant rigid body creep νξ is
present. When the maximum possible tangential stress is reached, the particle starts to
slide and remains in this condition in the sliding area until it leaves the contact patch.
This phenomenon was already known to Reynolds qualitatively. Carter showed (see
appendix, Sect. 16.3.3) that the tangential stress distribution q(ξ) is obtained from
subtracting a semicircle q2 (ξ) from q1 (ξ) in the adhesion area. The tangential stress
q(ξ) can then be formulated as
q(ξ) = q1 (ξ) + q2 (ξ ∗ ) , (3.42a)

μp0
a2 − ξ 2 for − a < ξ < a ,
q1 (ξ) = a (3.42b)
0 in all other cases ;
μp0
− a a∗2 − ξ ∗2 for − a∗ < ξ ∗ < a∗ ,
q2 (ξ ∗ ) = (3.42c)
0 in all other cases .
Fig. 3.18 Tangential stress v

distribution for the
two-dimensional contact q1( )
problem according to Carter
q( )
leading trailing
edge edge
a0 0
* a*0 0* - a*0 -a0
q2( *)
adhesion sliding
area area
Fig. 3.19 Local creepage v

sξ (x) in the Carter solution
s( )
leading- trailing
edge edge
a 0 , a*0 - a*0 -a0
, *
adhesion sliding
area area
The relation for the so-called local creep can also be taken from the appen-
dix (Sect. 16.3.3, Eq. (16.18)). Since we consider steady-state processes, the term
∂uξ (ξ, t)/∂t can be set to zero. The local creep in steady state remains
duξ (ξ)
sξ (ξ) = νξ + . (3.43)
dξ
The local creepage is a combination of the rigid body creepage νξ and surface par-
allel strain duξ (ξ)/dξ in the rolling direction, where uξ is the difference between the
displacements of wheel and rail respectively. The local creepage sξ (ξ) is shown in
Fig. 3.19. In the adhesion area, sξ (ξ) turns to zero. In the sliding zone, the local creep-
age increases in value until it reaches its maximum at the border. Outside the contact
patch, the local creepage approaches the rigid body creepage νξ asymptotically. The
following relation results for the rigid body creepage:

μp0 a∗
νξ = 4(1 − ν 2 ) 1− . (3.44)
E a
If the rigid body creepage νξ is known, the size of the adhesion area (2a∗ ) in relation
to the total contact patch (2a) can be calculated with Eq. (3.44). The quantity a∗
cannot be larger than a. The maximum creepage where an increase of the tangential
force can be noted or at which (as it is also called) saturation occurs, that is, total
sliding, assumes the following value:
μp0 μa
νξ,sat = 4(1 − ν 2 ) = . (3.45)
E r0
Since for the central position of the wheelset in the rail, a is approximately 5 mm
and r0 500 mm, it follows that νξ,sat μ/100. The saturation creep is therefore
usually substantially lower than 1 %. As can be seen from Eq. (3.45), it depends on
the friction coefficient μ, the normal force NL relating to the length, and the modulus
of elasticity E.
The tangential stresses can be integrated to a tangential line load TLξ (longitudinal
creep force with unit N/m):

TLξ (t) = qξ (ξ, t) dx . (3.46)

After some transformation, this results in

2
TLξ νξ
=μ 1− 1− ∀ 0 ≤ νξ ≤ νξ,sat ,
NL νξ , sat (3.47)
TLξ
=μ ∀ νξ ≥ νξ,sat .
NL
The initial slope of Carter’s curve provides the so-called creep coefficient:

∂TLξ 2 2NR πEa πa G
= μ NL = = = . (3.48)
∂νξ νξ =0 νξ,sat a 4 (1 − ν 2 ) 2 1−ν
This initial slope does not depend on the coefficient of friction but solely on the
loading and the elasticity properties of the materials.
In Fig. 3.20 the graph for the creepage–creep force relation for a selected number
of creepages is shown as well as the tangential stresses in the contact patch. With
increasing rigid body creep νξ , the sliding area increases too.
TL / N L linear
theory
s
1
a s
a s
1 / max
normalized creep
Fig. 3.20 Creepage–creep force relation in the Carter solution
3.4.3 Kalker’s Theory of Rolling Contact for Ellipsoidal

Contact
A closed analytical solution is not possible for a roller/plane contact even if in addition
to the longitudinal creepage also νξ , lateral creepage νη is included [44]. For rail
vehicle dynamics, the spin creepage νζ must be considered as well. Furthermore, the
contact problem is always three-dimensional, with elliptical (or even nonelliptical)
contact patches.
For the general case, the nonlinear creepage–creep force relations must be deter-
mined:
Tξ = Tξ (νξ , νη , νζ ) , (3.49a)
Tη = Tη (νξ , νη , νζ ) , (3.49b)
Mζ = Mζ (νξ , νη , νζ ). (3.49c)
For parts of the problem, approximate solutions were derived by Johnson in 1958
[51, 52]. The first complete numerical solution was published by Kalker in his disser-
tation [45]. More recent suggestions for solutions are discussed in detail in Kalker’s
monograph [26]. Regarding the formulation of linearized differential equations of
motion, the first interesting problem is that of linearized relations with respect to the
creepage-free reference state (νξ , νη , νζ ) = (0, 0, 0),
⎧ ⎫ ⎡ ⎤ ⎧ ⎫
⎨ Tξ ⎬ Tξ,ξ Tξ,η Tξ,ζ ⎨ νξ ⎬
Tη = ⎣ Tη,ξ Tη,η Tη,ζ ⎦ νη , (3.50)
⎩ ⎭ ⎩ ⎭
Mζ lin Tζ,ξ Tζ,η Tζ,ζ ν=0 νζ
where Ti,j = ∂Ti /∂νj .

These relatively complicated calculations were also first conducted by Kalker

[45]. The result can be presented as follows:
⎧ ⎫ ⎡ ⎤⎧ ⎫
⎨ Tξ ⎬ C11 0 √ 0 ⎨ νξ ⎬
Tη = Gab ⎣ 0 C
√ 22 ab C23 ⎦ νη . (3.51)
⎩ ⎭ ⎩ ⎭
Mζ lin 0 − ab C23 ab C33 νζ
The result is an extension of Eq. (3.48) for rolling contact. The coefficients Cik
are called creep coefficients or Kalker coefficients. They are specified in Table 3.1
in dependence on the semiaxis relation g and Poisson ratio ν. We would like to
stress explicitly that according to Kalker’s theory, the ξ-axis always points in rolling
direction, which means that the semiaxis a as well is in the rolling direction, even
though it is the minor semiaxis of the ellipse.
The linearized creep-free state can in some cases lead to significant errors, e.g.,
when one is investigating running dynamics that are superimposed on traction and
braking or steady-state curving. Even the fact that due to the contact angle, a steady-
state spin is present requires a linearization with respect to this spin. Usually this is
not done, however.
For Eq. (3.51), it is at first surprising that a spin generates a lateral creep force
and a lateral creep a spin moment. This will be discussed in connection with the
simplified theory of rolling contact, Sect. 3.4.5.
3.4.4 Approximate Solutions According

to Vermeulen–Johnson and Shen–Hedrick–Elkins
The Approximate Solutions by Johnson [51] and by Vermeulen

and Johnson [53]
The basic idea of Carter’s solution is that the initially unknown distribution of tan-
gential stress in the contact patch can be illustrated as the difference between two
semicircles (more precisely: two semiellipses). By doing this exactly, analytical solu-
tions for all interesting state variables can be obtained. Johnson [51] asked himself
whether it is possible to derive analytical solutions also for three-dimensional cases.
He considers a sphere that is rolling on a plane with longitudinal (or lateral) creep.
The contact patch for a sphere on a plane is a circle. Johnson assumes that the
adhesion area also is a circle that can be found at the entering boundary and that the
distribution of tangential stresses is analogous to Carter’s illustration of the difference
between two hemispheres; see Fig. 3.21. The advantage is that the integral can be
solved analytically. The drawback is that the solution is no longer exact, but only
approximate. By this means, an approximate solution is obtained for pure longitudinal
and lateral creep. The approximate solution for the pure longitudinal solution is, e.g.,
16
Table 3.1 Kalker coefficients according to [26] (with = ln and g = min (a/b, b/a)). Note: radius a in the rolling direction
68
g2
ν C11 C22 C23 C33

0.00 0.25 0.50 0.00 0.25 0.50 0.00 0.25 0.50 0.00 0.25 0.50
4π 2 π2 √ 4π 2
π g
g = 0.0 3(1−ν) 1 + ν 21 + ln 4 − 5
1−ν 4 16 (1 − ν) g
g = a/b 0.1 2.51 3.31 4.85 2.51 2.52 2.53 0.33 0.47 0.73 6.42 8.28 11.70
0.2 2.59 3.37 4.81 2.59 2.63 2.66 0.48 0.60 0.81 3.46 4.27 5.66
0.3 2.68 3.44 4.80 2.68 2.75 2.81 0.61 0.72 0.89 2.49 2.96 3.72
0.4 2.78 3.53 4.82 2.78 2.88 2.98 0.72 0.82 0.98 2.02 2.32 2.77
0.5 2.88 3.62 4.83 2.88 3.01 3.14 0.83 0.93 1.07 1.74 1.93 2.22
0.6 2.98 3.72 4.91 2.98 3.14 3.31 0.93 1.03 1.18 1.56 1.68 1.86
0.7 3.09 3.81 4.97 3.09 3.28 3.48 1.03 1.14 1.29 1.43 1.50 1.60
0.8 3.19 3.91 5.05 3.19 3.41 3.65 1.13 1.25 1.40 1.34 1.37 1.42
0.9 3.29 4.01 5.12 3.29 3.54 3.82 1.23 1.36 1.51 1.27 1.27 1.27
1.0 3.40 4.12 5.20 3.40 3.67 3.98 1.33 1.47 1.63 1.21 1.19 1.16
g = b/a 0.9 3.61 4.22 5.30 3.51 3.81 4.16 1.44 1.59 1.77 1.16 1.11 1.06
0.8 3.65 4.36 5.42 3.65 3.99 4.39 1.58 1.75 1.94 1.10 1.04 0.95
0.7 3.82 4.54 5.58 3.82 4.21 4.67 1.76 1.95 2.18 1.05 0.97 0.85
0.6 4.06 4.78 5.80 4.06 4.50 5.04 2.01 2.23 2.50 1.01 0.89 0.75
0.5 4.37 5.10 6.11 4.37 4.90 5.56 2.35 2.62 2.96 0.96 0.82 0.65
0.4 4.84 5.57 6.57 4.84 5.48 6.31 2.88 3.24 3.70 0.91 0.75 0.55
0.3 5.57 6.34 7.34 5.57 6.40 7.51 3.79 4.32 5.01 0.87 0.67 0.45
0.2 6.96 7.78 8.82 6.96 8.14 9.79 5.72 6.63 7.89 0.83 0.60 0.34
0.1 10.70 11.70 12.90 10.70 12.80 16.00 12.20 14.60 18.00 0.80 0.53 0.23

2π (1−ν)(3−ln 4)

2π 3 − ln 4 g 1+ (1−ν)−2ν 2π 1 π (ν − 2)
g = 0.0 1+ √ 1−
( − 2ν) g − 2ν (1 − ν) + 2ν 2g g (1 − ν) − 2 + 4ν 4 (1 − ν) − 2 + 4ν
3 Modeling of Wheel/Rail Contact
Fig. 3.21 Adhesion and slip q ( , = 0)

zone for rolling of a sphere
on a plane according to
approximation by p( )
Johnson [51]
q ( )
a a
a* a*
a
sliding area
adhesion area

3μ N (4 − 3ν) 1 − 34 ν a
νξ,sat = = μ , (3.52)
16 G a2 (1 − ν) R

νξ 3
Tξ = μ N 1− 1− , (3.53)
νξ,sat

∂Tξ 16 G a2
= . (3.54)
∂νξ νξ =0 4 − 3ν
Vermeulen and Johnson [53] generalized this solution to the case of an elliptical
contact patch. The adhesion area is found at the leading edge of the contact patch.
The only difference between this and the solution by Johnson is that the equations
for the saturation creepages νξ,sat and νη,sat contain complete elliptic integrals that
depend on the semiaxis relation of the ellipse. The result for the saturation creepage
νξ,sat is, for instance,
3μN
νξ,sat = . (3.55)
G abπ
The function , which is a combination of complete elliptic integrals, can be found in

the appendix, Sect. 16.4. While Carter’s solution results in a quadratic parabola, the
Vermeulen–Johnson solution results in a cubic parabola. Regarding further details,
the reader is referred to the appendix (Sect. 16.4).
The approximate solution by Vermeulen and Johnson has the indisputable advan-
tage that it is an analytical solution. It is a disadvantage, however, that already at the
initial slope, quantitative deviations from the more exact numerical solution arise
and that the spin creep is not considered.
Approximate Solution by Shen–Hedrick–Elkins
Already in 1967, Hobbs had suggested [54] using the Cik -values from Kalker’s theory
[45] in Vermeulen and Johnson’s approximation for the initial slope of the creepage–
creep force relation. Shen, Hedrick, and Elkins picked up this suggestion [55] and
derived the relations for combined longitudinal and lateral creep under consideration
of spin creep.
1. Step: First, the linearized tangential forces are calculated with Kalker’s creep
coefficients: ⎧ ⎫
lin ⎨ νξ ⎬
Tξ C11 0 √ 0
= Gab ν . (3.56)
Tηlin 0 C22 ab C23 ⎩ η ⎭
νζ
2. Step: For these linearized tangential forces, a resulting linearized tangential force
is determined:
2 2
T =
lin
Tξlin + Tηlin . (3.57)
3. Step: The resulting linearized tangential force is too high. It has to be reduced.
The reduction factor α is chosen in a way that the resulting tangential force behaves
as described by Johnson respectively Vermeulen–Johnson:
⎧
lin
2
⎪
⎪ 1 T lin
⎨1 − 1 T + for T lin ≤ 3μN ,
α = 3 μN 27 μN (3.58)
⎪
⎪ μN
⎩ for T lin ≥ 3μN .
T lin
4. Step: By means of the reduction factor, the following is calculated:
"
Tξ Tξlin
= α . (3.59)
Tη Tηlin
• The direction of the tangential forces results from the relation by Shen, Hedrick,
and Elkins according to the linear theory of Kalker.
• The reduction factor α ensures that the resulting creep force behaves according to
the theory of Johnson and Vermeulen and never exceeds μN.
• The influence of the spin creepage on the lateral creep force is included by means
of the linear theory. The influence of the spin torque is not taken into account.
Comparing the solutions by Johnson with the solution by Shen, Hedrick, and
Elkins for the case of pure spherical contact, e.g., regarding the saturation creep,
yields
3μ N 1
νξ,sat
SHE
= 2
· , (3.60)
G a C11
3μ N 4 − 3ν
νξ,sat
Johnson
= · . (3.61)
G a2 16
In consequence, the C11 -value of Johnson’s solution can be read:
16
Johnson
C11 = . (3.62)
4 − 3ν
The solution by Shen, Hedrick, and Elkins still has two more drawbacks:
1. In case of fully developed sliding and large longitudinal and lateral creepages,
the resulting creep force should in case of isotropic friction conditions have the
direction of the resulting creep velocity. That is not the case.
2. For very large spin creep alone, the spin moment will provide saturation. This
effect is not accounted for.
3.4.5 Simplified Theory of Rolling Contact [54]
Initial Remark
Kalker’s complete theory of rolling contact, which was presented briefly in Sect. 3.4.3,
is computationally costly. It is therefore understandable that a simplified but fast the-
ory of rolling contact was looked for. The initial ideas at least regarding the wheel/rail
contact once again stem from Kalker [56]. A similar theory is used for the rolling
contact in road vehicle dynamics. Wheel and ground consist in this simplified the-
ory of rigid base bodies with bristles or knobs continuously applied (model of road
vehicle dynamics) or an elastic skin cover (Kalker’s idea of a model). Both models
are mathematically equal.
Simplified Theory for Pure Longitudinal Creep
For the case of pure longitudinal creep, such a model is shown in Fig. 3.22. For
simplicity, bristles attached to the wheel (0’–10’) are regarded as rigid, while the
bristles attached to the ground (0–10) can be elastic. With such a model, all the
essential effects can be covered. From a mechanics point of view, this model is a
Winkler foundation. The constitutive equations of this Winkler foundation replace the
constitutive equation of the elastic half-space. For this simplified model, a different
v0
v0 - 0
r
(a)
r
leading edge
trailing edge
10 9
8 7 6 5 4 3 2 1 0
1 10 9 8 7 6 5 4 3 2 1 0
,u
u( )
(b)
u = (v0 - r)
vm0
1
,u vm = (v0 + 0
r)
2
u( )
(c)
=-a =a
sliding area adhesion area

maximal possible
(d) q ( ) tangential stress p
tangential stress x
Fig. 3.22 Tangential stresses in the simplified theory for pure longitudinal creepage (braking). Bris-
tle model (a), corresponding discrete displacements of the rail bristles (b), continuous displacement
of an elastic bed (c), tangential stress (d)
distribution of normal stresses is obtained from that of the half-space model. The
distribution of normal stresses is parabolic, which can be formulated as

ξ2 2N
p(ξ) = p0 1 − 2 with p0 = . (3.63)
a abπ
The value of the elliptical semiaxis a is adopted from the half-space theory, which is
synonymous with determining the vertical flexibility of the knobs in a suitable way.
The tangential contact problem in the simplified theory can be described by the
equations given below. As for Carter’s theory, the following abbreviations are used:
v1 − v2
u(ξ) = uξ1 (ξ) − uξ2 (ξ) , νξ = .
vm
adhesion area sliding area

|qξ | ≤ μ p (3.64a) |qξ | = μ p (3.64b)
kinematic relations
sξ = νξ + u (ξ) < 0 ∀ νξ < 0
νξ + u (ξ) = 0 (3.65a) (3.65b)
sξ = νξ + u (ξ) > 0 ∀ νξ > 0
constitutive equations
u(ξ)
qξ (ξ) = (3.66a)
Lξ
The tangential flexibility of the bristles Lξ is unknown in this case. Through integra-
tion of the kinematic relations and inserting the result into the constitutive equation,
the distribution of the tangential stresses in the adhesion area is obtained:
ξ
a(1 − a )νξ
qξ (ξ) = (3.67a)
Lξ
and in the sliding area

ξ 2 νξ
qξ (ξ) = p0 1 − 2 . (3.67b)
a |νξ |
This is explained further with the help of Fig. 3.22. We assume that the wheel is being
braked and the velocity v 0 is larger than the product r0 0 .
A bristle entering the contact patch of the rail is unloaded while entering and
therefore not displaced. If the wheel is moved forward by the distance between two
bristles (ξ), every bristle is subjected to a small relative displacement, which can
be calculated from the kinematic relation
u (v 1 − v 2 )
=− .
ξ vm
In our special case,
1
v 1 = v 0 − 0 r, v 2 = 0, vm = (v 0 + 0 r) ,
2
which implies that the relative displacement becomes
v 0 − 0 r
u = −2ξ .
v 0 + 0 r
This difference in displacement has to be summed during each step. Since we have
assumed that only the bristles on the rail side are deformable, all rail bristles in our
example will be displaced by the value of u in the positive direction of ξ during
a forward movement of the wheel. In a steady-state process, for small values of ξ,
bristle 2 is displaced by u relative to bristle 1, and bristle 3 is displaced by 2u
relative to bristle 1. The state of displacement that occurs to bristles 1–10 in this
manner is shown in Fig. 3.22b.
Infinitesimally dense bristles or an equivalent thin elastic layer on the rigid rail
body can now be considered. It does not matter whether only one or both contacting
bodies are covered with such a thin elastic skin. In Fig. 3.22c, it is shown how these
infinitesimal densely applied bristles are displaced.
The tangential stress in the bristles is proportional to the relative displacement
uξ (ξ). For rigid wheel bristles, therefore,
uξ2 (ξ)
qξ (ξ) = . (3.68)
Lξ
In the entry boundary of the adhesion zone, linearly increasing tangential stresses
arise that act in the positive ξ-direction on the rail and in the negative ξ-direction on
the wheel. The linear increase can continue only until the tangential stress reaches
the product of the friction coefficient and the normal force μp(ξ). After this, the
tangential stress coincides with the μp(ξ)-curve and is therefore equal to zero at the
trailing edge. There is always a small band at the trailing edge where sliding occurs.
In the framework of the linear theory, it is assumed that creepages remain small
enough that this sliding area can be neglected.
It can be recognized clearly that the relative velocities between the rigid base body
of the wheel and the rigid base body of the ground can be realized in two ways. One
way of realization is that two particles from wheel and ground respectively actually
slide on one another; in the case of adhesion, the relative velocities are absorbed by
elastic deformations. This particular form of creep is in the older literature called
quasi creepage or pseudo creepage.
Pure Spin Creepage in the Framework of the Simplified Theory
In the general case, besides longitudinal creepage there is also lateral creepage and
spin creepage. Lateral creepage can, for instance, occur during the contact of two
rollers with skew axis. Spin creepage occurs if the contact patches of both rollers
have an inclination angle of δ0 .
In the framework of the simplified theory, both cases are relatively easy to explain,
especially if adhesion is assumed in the whole contact patch. All equations necessary
for the treatment of the problem are compiled in the appendix (Sect. 16.5). In the case
of adhesion, the following two relations are obtained:
(a) (b) (c)

A D
v0 - r0 0
E
B
y
0
F
x x
C
Fig. 3.23 Relative speeds (a), elastic deformations (b), and stresses (c) in the contact patch for
pure spin
∂uξ
(νξ − νζ η) + =0, (3.69a)
∂ξ
∂uη
(νη + νζ ξ) + =0, (3.69b)
∂ξ
from which uξ and uη can be determined through integration. The integration con-
stants are defined from the consideration that there is no tangential stress at the
leading edge ξ = ξr , that is, qξ (ξ = ξr ) = 0 and qη (ξ = ξr ) = 0. With the constitu-
tive equation from the simplified theory, one has for the displacement that
uξ (ξ = ξr ) = 0 , (3.70a)
uη (ξ = ξr ) = 0 . (3.70b)
This leads to the relative displacements in the contact patch
uξ = −(νξ − νζ η)(ξ − ξr ) , (3.71a)

1
uη = −νη (ξ − ξr ) + νζ (ξ 2 − ξr2 ) . (3.71b)
2
This rather formal presentation will now be explained for the case of pure spin
creepage with the help of Fig. 3.23. For the sake of clarity, once again we assume
that only the rail is covered with infinitely densely arranged bristles. Onto this cover
of bristles, a rectangular grid is applied, whose deformation is to be examined.
Figure 3.23b shows displacements of the rigid wheel in different positions of the
contact patch relative to the ground for pure spin creepage. A line that enters the
contact patch in position A will be displaced in the η-direction due to the relative
velocity prevailing there. Position B, which entered the contact patch earlier, has
already assumed its maximum lateral displacement. For all points between B and C,
the relative velocity is negative, so that the line B-C is bent in the negative η-direction
before leaving the contact patch. A similar displacement in the η-direction happens
to the line D-E-F. For the displacements in the ξ-direction, analogous considerations
can be made. All points that are situated on the line A-B-C are displaced in the
negative ξ-direction due to the negative relative velocity that acts in the ξ-direction.
Points on the line D-E-F, in contrast, are displaced in the positive ξ-direction.
The whole state of displacement is shown in Fig. 3.23b. It can be recognized
that discontinuities arise at the trailing edge of the contact patch. The bristle layer
can handle these discontinuities. However, we have to be clear of the fact that the
discontinuities are a consequence of the (unrealistic) assumption that adhesion occurs
throughout the whole contact patch. Since creepage always occurs at the trailing edge
due to the normal pressure decreasing to zero, the tangential displacements of the
bristle cover also become zero as soon as such creep processes are allowed.
Figure 3.23b clearly shows the disadvantage of the bristle model. Independent of
whether sliding in the contact patch is allowed, all bristles outside the contact patch
remain undeformed. For the half-space model, this is different: material entering the
contact patch is already distorted and also leaves the contact patch distorted. For
non-steady-state processes, this can lead to considerable errors.
Figure 3.23c shows the tangential stresses for two lines η = constant, and for one
line ξ = constant. In the same way they would act on the rail or roller 2. Since a creep
force is considered positive if it acts on the rail (roller 2) in the positive direction
of a coordinate axis or is rotating around a positive coordinate axis, a resulting
positive spin moment Mζ and a resulting positive tangential force Tη belong to the
tangential stresses of Fig. 3.23c. A resulting tangential force Tξ does not occur, due
to symmetry. We have therefore proven the population of the third column of the
matrix of Eq. (3.51) at least concerning the signs.
3.4.6 Adaptation of the Theory to Reality
For a long time, it was doubted whether Kalker’s theory and the approximate solu-
tions that follow describe the creepage–creep force relation and the creep processes
correctly. Especially in England, extensive testing was conducted in order to assess
Kalker’s theory. The results are compiled in [54]. The tests by C. Th. Müller from
Minden can be found there as well. The result is that under ideal conditions (rolling
surfaces uncontaminated and as smooth as possible), the rolling contact processes
can be described qualitatively and quantitatively by Kalker’s theory.
Nevertheless, in reality there are some deviations, since ideal conditions never
can be achieved on an actual track. In recent times, it has been shown [57, 58] that
surfaces with micro roughness also theoretically can lead to a decrease of the initial
slope (and in consequence to a decrease of the creep coefficient). This justifies the
approach by British Rail. At BR, the full Cik -coefficients were never used; instead,
only a fraction of them (in most cases 60 %) was used.
The following procedure is suggested:
• The creepage–creep force relation of Kalker (or a simplified derivation of it) is

adopted qualitatively.
• The friction coefficient μ has to be known.
• An initial factor that can be interpreted as the shear modulus G can be obtained
by tests. In case of doubt, the modulus of creep has to be determined from tests as
well.
References
1. DIN. Oberbau: Schienen – Symmetrische Breitfußschienen ab 46 kg/m (Superstructure: Rails

- Symmetric bottom rails from 46 kg/m.). DIN EN 13647-1, German Institute for Standard-
ization, Standards Committee Iron and Steel, Standards Committee Railway Vehicle, Berlin
(1999)
2. N.N. Radprofile, Breite 135 und 140 mm (Wheel profiles, width 135 and 140 mm). DIN 5573,
Deutsches Institut für Normung e.V., Normenausschuss Schienenfahrzeuge, Berlin (1995)
3. R. Gasch, D. Moelle, K. Knothe, The effects of non-linearities on the limit-cycles of railway
vehicles, in The Dynamics of Vehicles on Roads and on Tracks, Proceedings of the 8th IAVSD-
Symposium held at Cambridge, Mass./USA, August 1983, ed. by J.K. Hedrick (Swets and
Zeitlinger, Lisse/Amsterdam, 1984), pp. 207–224
4. D. Nicklisch, European Patents. EP 1 091 044 A1, European Patent office, Paris (2001)
5. A. Nefzger, Geometrie der Berührung zwischen Radsatz und Gleis (Geometry of contact
between wheelset and track). ETR 23, 113–122 (1974)
6. ORE, A. Nefzger, Entwicklung des Radprofils S 1002 auf dem Streckennetz der DB (Devel-
opment of the wheel profile 1002 on the network of DB). C 116/DT81, ORE, Utrecht (1979)
7. N.K. Cooperrider, E.H. Law, R. Hull, P.S. Kadala, J.M. Tuten, Analytical and Experimen-
tal Determination of Nonlinear Wheel/Rail Geometric Constraints. Interim Report No. FRA-
OR&D, US Department of Transportation (1975)
8. A. Theiler, Abschlussbericht der Arbeitsgruppe am ILR der TU Berlin zum Projekt Optikon
(Final report of the working group at ILR of TU Berlin in the project Optikon). Technical
report, Institute for aeronautics and astronautics, TU Berlin (2002)
9. L. Mauer, Die modulare Beschreibung des Rad/Schiene-Kontaktes im linearen Mehrkörperfor-
malismus (Modular description of wheel/rail contact for a linear multibody formalism). Ph.D.
thesis, Technical University Berlin (1988)
10. S. Müller, Linearized wheel-rail dynamics – stability and corrugation, in Fortschritt-Berichte
VDI (also Ph.D. thesis TU Berlin), Reihe 12, Nr. 368. (VDI–Verlag, Düsseldorf, 1998)
11. A.H. Wickens, The dynamic stability of simplified four-wheeled vehicles having conical
13. R. Joly, Untersuchungen der Querstabiliät eines Eisenbahnfahrzeuges bei höheren
Geschwindigkeiten (Investigation of lateral stability of a railway vehicle at higher speeds).
Rail International - Schienen der Welt 3, 168–204 (1972)
14. ORE, Equations of motion of railway vehicles. ORE-report to question C 116/RP4, UIC, ORE,
Utrecht (1974)
15. K. Knothe, Die geometrisch nichtlinearen Beziehungen für einen starren Radsatz, der auf einer
starren Schiene quer verschoben wird (The geometric nonlinear relations for a rigid wheelset
that is laterally displaced on a rigid track). ILR–Mitt. 17, Institut für Luft- und Raumfahrt, TU
Berlin (1975)
16. N. Matsui, A re-examination of the wheel/rail contact geometry and its application to the
hunting analysis of railway bogie vehicles having profiled wheels, in Proceedings 7th IAVSD
Symposium on Dynamics of Vehicles on Roads and on Tracks, Cambridge, UK, September,
1981, ed. by A. Wickens (Swets and Zeitlinger, Amsterdam, 1982), pp. 468–480
17. ORE, Geometri of contact between wheelset and track. Part 1. Measurement methods and
evaluation. ORE-report to question C 116/RP3, UIC, ORE, Utrecht (1973)
18. N.K. Cooperrider, J.K. Hedrick, E.H. Law, C.W. Malstrom, The application of quasilineariza-
tion techniques to the prediction of nonlinear railway vehicle response, in The Dynamics of
Vehicles on Roads and on Tracks. Proceedings of the IUTAM Symposium held at Delft, The
Netherlands, August 1975, ed. by H.B. Pacejka (Swets and Zeitlinger, Amsterdam, 1976), pp.
314–325
19. W. Hauschild, Grenzzykelberechnung am nichtlinearen Rad–Schiene–System mit Hilfe der
Quasilinearisierung (Limit cycle analysis for nonlinear wheel-rail-systems based on quasilin-
earization.). Ph.D. thesis, Technical University Berlin (1981)
20. J.K. Hedrick, N.K. Cooperrider, E.H. Law, The Application of Quasi-linearization Techniques
to Rail Vehicle Dynamics Analysis. Final Report No. DOT-TSC-902, US Department of Trans-
portation (1978)
21. A. Renger, Laufdynamik von Schienenfahrzeugen – Beitrag zur Querdynamik von vierach-
sigen Schienentriebfahrzeugen (Running dynamics of railway vehicles – Contribution to the
lateral dynamics of four-axled railway vehicles). Report R–Mech–02/85, Akademie der Wis-
senschaften der DDR, Institut für Mechanik, Berlin (1985)
22. J. Zhang, Dynamisches Bogenlaufverhalten mit stochastischen Gleislagefehlern - Modell-
und Verfahrensentwicklung unter Verwendung der Methode der statistischen Linearisierung
(Dynamic curving behavior with stochastic track irregularities – Development of models and
processes based on the method of stochastic linearization). VDI Fortschritt–Berichte (also
Ph.D. thesis at TU Berlin), Reihe 12, Nr. 304. VDI–Verlag, Düsseldorf (1997)
23. W. Kik, J. Piotrowski, A fast, approximate method to calculate normal load at contact between
wheel and rail and creep forces during rolling, in Proceedings of the 2nd Mini Conference on
Contact Mechanics and Wear of Rail/Wheel Systems held at Budapest, July 29–31, 1996, ed.
by I. Zobory (Technical University of Budapest, 1996), pp. 52–61
24. A. Renger, Theorie der Bewegung eines Radsatzes auf geradem Gleis mit stochastischen Gleis-
lagefehlern (Theory of motion of a wheelset on straight track with stochastic track irregulari-
ties). Z. Angew. Math. Mech. 62, 141–169 (1982)
25. A.D. de Pater, The exact theory of the motion of a single wheelset moving on a purely straight
track. Report No. 648, Delft University of Technology, Delft (1979)
26. J.J. Kalker, Three Dimensional Elastic Bodies in Rolling Contact (Kluwer Academic Publish-
ers, Dordrecht, 1990)
27. H. Hertz, Über die Berührung fester, elastischer Körper. (On the contact of solid elastic bodies).
Journal für die reine und angewandte Mathematik 92, 156–171 (1882)
28. H. Hertz, Über die Berührung fester elastischer Körper (On the contact of solid elastic bodies),
in Heinrich Hertz, Collected Works. Band 1, Leipzig (1895), pp. 155–173
29. K.L. Johnson, Contact Mechanics (Cambridge University Press, Cambridge, 1985)
30. A.E.H. Love, A Treatise on the Mathematical Theory of Elasticity, 4th edn. (Dover, New York,
1926)
31. A.I. Lurje, Räumliche Probleme der Elastizitätstheorie (Three-Dimensional Problems of Elas-
ticity) (Akademie-Verlag, Berlin, 1963)
32. J.J. Kalker, User’s Manual of CONTACT, Version CONPS93. Technical report, TH Delft, NL
(1993)
33. G. Wang, K. Knothe, The influence of inertia forces on steady-state rolling contact between
two elastic cylinders. Acta Mech. 79, 221–232 (1989)
34. H. Le The, Normal- und Tangentialspannungsberechnung beim rollenden Kontakt für Rota-
tionskörper mit nichtelliptischen Kontaktflächen. (Analysis of normal and tangential stresses
in rolling contact for rotational bodies with nonelliptical contact patches), in VDI Fortschritt–
Berichte (also Ph.D. thesis TU Berlin), Reihe 12, Nr. 87. (VDI–Verlag, Düsseldorf, 1987)
35. E. Vollebregt, User Guide for CONTACT. Technical report, VORtech BV (2012)
36. E. Vollebregt, Numerical modelling of measured railway creep versus creep-force curves with
CONTACT. Wear 314, 87–95 (2014)
References 79
37. F.W. Carter, The electric locomotive. Proc. Inst. Civil Eng. 201, 221–252 (1916) (Discussion
pp. 253–289)
38. O. Reynolds, On rolling friction. Philos. Trans. R. Soc. Lond. 166(I), 155–174 (1876)
39. A.O. Gilchrist, The long road to solution of the railway hunting and curving problem. Proc.
Inst. Mech. Eng. 212, 219–226 (1998)
40. F.W. Carter, On the action of a locomotive driving wheel. Proc. R. Soc. Lond. A 112, 151–157
(1926)
41. H. Fromm, Berechnung des Schlupfes beim Rollen deformierbarer Scheiben (Analysis of
creepage during rolling of deformable discs). Z. Angew. Math. Mech. (also Ph.D. thesis TH
Berlin, 1926) 7, 27–58 (1927)
42. H. Poritsky, Stresses and deflections of cylindrical bodies in contact with application to contact
of gears and locomotive wheels. J. Appl. Mech. 17, 191–201 (1950) (Diskussion hierzu in [20])
43. H. Bufler, Zur Theorie der rollenden Reibung (On the theory of rolling friction). Ing. Arch.
XXVII, 137–152 (1959)
44. G. Heinrich, K. Desoyer, Rollreibung mit axialem Schub (Rolling friction with axial shear
force). Ing.-Arch. 36, 48–72 (1967)
45. J.J. Kalker, On the rolling contact of two elastic bodies in the presence of dry friction. Disser-
tation, TH Delft, Delft (1967)
46. M. Fink, Physikalisch-chemische Vorgänge zwischen Rad und Schiene (Physical-chemical
processes between rail and wheel). Glasers Annalen 75, 207–210 (1951)
47. B.N.J. Persson, Sliding Friction - Physical Principles and Applications (Springer, Berlin, 1998)
48. E. Melan, Der Spannungszustand eines “Mises–Henckyschen” Kontinuums bei veränderlicher
Belastung (Stress state of a Mises–Hencky continuum for variable load). Sitzber. Akad. Wiss.
Wien, Abt. IIa 147, 73–87 (1938)
49. A. Groß-Thebing, Lineare Modellierung des instationären Rollkontaktes von Rad und Schiene
(Linear modeling of non-steady rolling contact of rail and wheel). VDI Fortschritt–Berichte
(also Ph.D. thesis TU Berlin), Reihe 12, Nr. 199. VDI–Verlag, Düsseldorf (1993)
50. K. Hempelmann, Short Pitch Corrugation on Railway Rails – A Linear Model for Prediction,
in VDI Fortschritt–Berichte (also Ph.D. thesis TU Berlin), Reihe 12, Nr. 231. VDI–Verlag,
Düsseldorf (1994)
51. K.L. Johnson, The effect of a tangential contact force upon the rolling motion of an elastic
sphere on a plane. J. Appl. Mech. 25, 339–346 (1958)
52. K.L. Johnson, The effect of spin upon the rolling motion of an elastic sphere on a plane. J.
Appl. Mech. 25, 332–338 (1958)
53. J. Vermeulen, K.L. Johnson, Contact of nonspherical elastic bodies transmitting tangential
forces. J. Appl. Mech. 31, 338–340 (1964)
54. A.E.W. Hobbs, A Survey on Creep. Technical Note TN DYN 52, British Rail Research, Derby
(1967) (reprinted 1976)
55. Z.Y. Shen, J.K. Hedrick, J.A. Elkins, A comparison of alternative creep-force models for rail
vehicle dynamic analysis, in The Dynamics of Vehicles on Roads and Tracks. Proceedings
of the 8th IAVSD Symposium held at MIT Cambridge/MA, August 15–19, 1983, ed. by J.K.
Hedrick (Swets and Zeitlinger, Lisse, 1984), pp. 591–605
56. J.J. Kalker, Simplified theory of rolling contact, in Delft Progress Report, Series C, Mechanical
and Aeronautical Engineering and Shipbuilding (Delft University Press, Delft, 1973), pp. 1–10
57. A. Lünenschloß, F. Bucher, K. Knothe, Normalkontakt zweier Körper mit rauen Oberflächen
(Normal contact of two bodies with rough surfaces). Fortschritt–Berichte VDI, Reihe 2, Nr.
596, Düsseldorf, 2002. VDI–Verlag (2002)
58. A. Lünenschloß, F. Bucher, K. Knothe, Numerische Behandlung des quaistatischen Tangen-
tialkontaktproblems zweier Körper mit rauen Oberflächen (Investigation of the quasistatic tan-
gential contact problems of two bodies with rough surfaces). Fortschritt–Berichte VDI, Reihe
2, Nr. 616, Düsseldorf, 2002. VDI–Verlag (2002)
Chapter 4
Vertical Dynamics, Equations of Motion,
and Free Vibrations
In the present chapter, the differential equations for the motion of a rail vehicle with
two axles will be examined in detail, first for longitudinal and lateral motion with
the principle of linear and angular momentum. The reader who is familiar with these
concepts can skip Sect. 4.2. Then the possibility to formulate the equations of motion
by means of virtual displacements is discussed (Sects. 4.3 and 4.4). In this manner,
it is also possible to derive the equations of motion for a vehicle with elastic carbody
(Sect. 4.5). Finally, in Sect. 4.6, the solution of a two-axle vehicle (free vibration
case) is discussed.
4.1 Notation and Assumptions
The mechanical model of a rail vehicle as shown in Fig. 4.1 is used as a basis. In this
model, the following notation is used:
m CB carbody mass;
CB moment of inertia of the carbody for rotations around the y-axis;
mW wheelset mass;
W moment of inertia of the wheelset for rotations around the y-axis;
cx , cz spring stiffness between carbody and wheelset in longitudinal and vertical
directions;
dx , dz damping constants between carbody and wheelset;
ex , ez coordinates for the connection of the springs;
ep coordinate for the point of load application.
The following assumptions are made:
1. The vehicle moves with a constant velocity v 0 on a straight track. With the equa-
tions of motion, the deviations from this uniform movement (disturbances) are
captured.

DOI 10.1007/978-3-319-45376-7_4
82 4 Vertical Dynamics, Equations of Motion …
ep
ex ex
u zC
S u xC x
P (t) v = const
rear (index t) front (index l) 0
mC g C
cz dz dz cz ez
dx z dx
cx m Wg m Wg cx ideal
S
(non-disturbed)
zt 0 track x
zl
real disturbed track
Fig. 4.1 Mechanical model of a two-axle vehicle with springs and dampers as coupling elements
2. All kinematic relations and all spring and damper values can be linearized with
respect to a reference state.
3. Car body and wheelsets are rigid bodies. The track is rigid and fixed.
4. In the vertical direction, there is always contact between wheel and rail. This is a
geometric constraint.
5. For undisturbed as well as for movements with superimposed disturbances, pure
rolling of the wheels is taking place. No relative velocities (creep) occur between
wheel and rail.
4.2 Equations of Motion with the Principle of Linear

and Angular Momentum
The formulation of the equations of motion is performed by means of the principle

of linear and angular momentum. The following steps are needed.
Step 1: It has to be determined what displacements and rotations can occur and
which of these variables are the unknown degrees of freedom, i.e., how
many equations of motion have to be formulated (Sects. 4.2.1 and 4.2.2).
Step 2: The changes in length in the springs and the changes of velocity in the
dampers for the displacements from step 1 have to be determined. Due to the
assumption of linearity (2), the displacements can be applied individually and
be superimposed. As a result of these superimpositions, the resulting spring
forces are dependent on the given displacements and rotations; Sect. 4.2.3.
The derivation of the damping forces is analogous.
4.2 Equations of Motion with the Principle of Linear and Angular Momentum 83
Step 3: Derivation of free-body diagrams of the masses (Sect. 4.2.4). The spring
and damper forces determined in step 2 are applied as reaction forces on
the free masses. At the wheel contact point, additional constraint forces are
introduced.
Step 4: For the free oscillating masses, the principle of linear and angular momentum
can be formulated; Sect. 4.2.5.
Step 5: Under consideration of the geometric constraints, the constraint forces can
be eliminated; Sect. 4.2.6. The result is the equations of motion.
4.2.1 Displacement Degrees of Freedom for a Two-Axle

Vehicle
The movements of a rigid body in a plane can be described by two displacements and
a rotation around the center of gravity. The system here consists of three rigid bodies:
the carbody and two wheelsets; i.e., nine displacement variables (displacements or
rotations) are needed to describe the system.
The displacements of the carbody are not described in a fixed coordinate system
(O; x, y, z) but rather in a coordinate system (S; x , y , z ) that is moving with
velocity v 0 (see Fig. 4.1):
u xCB longitudinal displacement of the carbody center of gravity (x -direction);
u zCB vertical displacement of the carbody center of gravity (z -direction);
ϕCB rotation of the carbody center of gravity around the y -axis.
The displacements of the two wheelsets are not described in a fixed coordinate
system either, but in a coordinate system with a constant velocity v 0 that is moving
with the center of gravity of the wheelset and rotating with a constant angular velocity
ϕ0 = v 0 /r0 . Longitudinal displacements and rotation of the wheelset are therefore
displacements relative to the uniform rolling movement. For the leading wheelset
(index l), the following notation is used
u xl longitudinal displacement of the center of gravity of the leading wheelset;
u zl vertical displacement of the center of gravity of the leading wheelset;
u l relative rotation of the leading wheelset around its own axle.
Three displacements are defined for the trailing wheelset analogously (index t).
4.2.2 Constraints
Not all of these displacements are independent (free) displacements. Due to assump-
tion 4, there are vertical displacements of the wheelset from the vertical rail position.
The vertical displacements are equal to the height errors. The given track irregulari-
ties are marked by an overbar; hence z l and z t . It is therefore valid that
u zl = z̄ l , (4.1)
u zt = z̄ t . (4.2)
The given track irregularities (disturbances) z̄ l and z̄ t are initially dependent on the
position. They can be formulated as time-dependent, since x = v t. Moreover, z̄ l and
z̄ t are not independent from each other but only phase-shifted. Assume that at point
t = 0, axle z of the fixed and axle z of the vehicle following coordinate system
coincide. Then we can write

ex
u zl = z̄ t + , (4.3a)
v0

ex
u zt = z̄ t − . (4.3b)
v0
For undisturbed rolling movements and a superimposed disturbance, the rotation of

the wheelset around its axle and the displacement of the wheelset in the longitudinal
direction (x-direction) are connected by the requirement that the wheelset be rolling
on the rail without slip (assumption 5). Therefore, we can write
u xl
ϕl = , (4.4)
r0
u xt
ϕt = . (4.5)
r0
The longitudinal displacements of the wheelsets are introduced as independent dis-

placements with corresponding equations of motion. Due to Eqs. (4.4) and (4.5), the
wheelset rotation can then be expressed by wheelset longitudinal displacements. Of
the nine displacements, therefore, only five independent displacement remain, i.e.,
we have a system with five mechanical degrees of freedom.
In Fig. 4.2, the undisplaced state of the carbody is illustrated by a solid line; the
displaced state, separately for each respective displacement, is illustrated by dashed
lines. In an analogous way, the displacements can be illustrated for the wheelset
(Fig. 4.3).
z z z
u zC x
x
u xC x
C
Fig. 4.2 Displacement states of the carbody

zl zl zl
u zl
xl xl
xl
uxl l
Fig. 4.3 Displacement states of the leading wheelset
ex ex
C =1
x v = const
rear (index t) front (index l) P(t) 0
ez
Fig. 4.4 Unit displacement state ϕCB = 1
4.2.3 Forces in the Spring and Damper Elements
In the following, the forces in the spring elements will be determined. The determina-
tion of the forces in the damper elements will not cause any additional complications,
since one spring is always in parallel with one damper. The spring forces are denoted
by Fc . The index c marks the spring force; additional indices mark the position of
the springs (cf. Fig. 4.4). A spring force is positive if there is tension in the spring.
The change in spring length is denoted by v. With the spring stiffness c, we get
Fc = c v . (4.6)
For the damper forces, denoted by Fd , the following relation for the force in a damper
applies:
dl
Fd = d = d v̇ . (4.7)
dt
Determination of the Spring Elongation

First it is important to determine the spring elongation v in the four springs (v xl , v zl ,
v xt , v zt ) as a function of the nine degrees of freedom u xCB , u zCB , ϕCB ; u xl , u zl , ϕl ; u xt ,
u zt , ϕt . Due to assumption 2, the principle of superposition can be used. The nine
variables are treated separately in determining the spring elongation. Subsequently,
all parts from all variables are superposed.
In Fig. 4.4, a rotation ϕCB around the center of gravity of the carbody is applied. All
other displacements are set to zero. The nondisplaced state (solid) and the displaced
state (dashed) are illustrated. The springs are inserted in the displaced state, which
is between the undisplaced center of gravity of the wheelset and the undisplaced
connecting point of the carbody. By applying the notation from Fig. 4.4, the spring
elongations are obtained:
v xl = − ez ϕCB , (4.8a)
v zl = − ex ϕCB , (4.8b)
v xt = + ez ϕCB , (4.8c)
v zt = + ex ϕCB . (4.8d)
Similar considerations can be made for all nine displacements. The spring elongations
in dependence on these nine displacements are compiled in Table 4.1.
Determination of Spring and Damper Forces
For the spring stiffness, one has (Fig. 4.1)
cxl = cxt = cx , (4.9a)

czl = czt = cz . (4.9b)
The resulting spring forces due to the rotation ϕCB are
Fcxl = −cx ez ϕCB , (4.10a)

Fczl = −cz ex ϕCB , (4.10b)
Fcxt = +cx ez ϕCB , (4.10c)
Fczt = +cz ex ϕCB . (4.10d)
Table 4.1 Elongations in the four springs as a result of unit displacements

u xCB u zCB ϕCB u xl u zl ϕl u xt u zt ϕt
v xl 1 0 −ez −1 0 0 0 0 0
v zl 0 1 −ex 0 −1 0 0 0 0
v xt −1 0 +ez 0 0 0 1 0 0
v zt 0 1 +ex 0 0 0 0 −1 0
Table 4.2 Forces in the four springs as a result of unit displacements (index c: spring force, indices
x, z: spring direction, index l: leading, index t: trailing)
u xCB u zCB ϕCB u xl u zl ϕl u xt u zt ϕt
Fcxl cx 0 −cx ez −cx 0 0 0 0 0
Fczl 0 cz −cz ex 0 −cz 0 0 0 0
Fcxt −cx 0 +cx ez 0 0 0 cx 0 0
Fczt 0 cz +cz ex 0 0 0 0 −cz 0
Table 4.2 compiles the four spring forces in dependence on the nine possible dis-
placements.
The force Fd in a damper is proportional to the velocity v̇; the factor of pro-
portionality is the damping constant d. The calculation of the damping forces is in
principle the same as for the spring forces except that instead of the displacements
u xCB , u zCB , ϕCB , etc., now the corresponding velocities u̇ xCB , u̇ zCB , ϕ̇CB are used.
Since the example uses parallel springs and dampers, it is also possible to write
F = Fc + Fd = c v + d v̇ . (4.11)
4.2.4 Free-Body Diagrams of the Masses
In Sect. 4.2.5, the principle of linear and angular momentum for the three masses
will be formulated. The forces acting on the free masses are:
• the spring forces,
• unknown forces between wheel and rail, and
• the gravitational forces at the centers of gravity and the excitation force P(t).
Difficulties for the beginner usually arise regarding the signs (that is the direction)
of the forces. The following hold:
1. For the forces in springs and dampers, there is already a definition of the signs.
The reaction force that has to be applied to a positive spring or damper force is
then also positive.
2. Positive forces between wheel and rail in Fig. 3.1 are assigned with respect to the
forces with which they act on the rail. The corresponding reaction forces are then
also positive.
Figure 4.5 shows the free-body diagram with the forces that act on the bodies. The
rail and the forces that act on it are also shown. The normal forces are positive as
compressive forces; the tangential forces Tξ act on the rail in the positive ξ-direction.
In addition to the four forces from springs and dampers and the forces acting between
wheel and rail, there are also external forces acting on the center of gravity of
the carbody as well as the centers of gravity of the wheelsets. Finally, there is a
ep
ex ex
S
P (t)
x
rear (index t) mC g front (index l)
ez
Fzt Fz l
Fx t Fz l Fx l
Fzt
mWg mW g r0
Fx t Fx l
T t T l
Nt Nl
Nt Nl
T t T l
Fig. 4.5 Forces acting on the bodies
time-dependent excitation force P(t), which can come from an excitation of imbal-
ance from, e.g., a monitoring car.
The forces between wheelset and rail, Nxl , Nxt , Tξl , and Tξt , are unknown. That is
no problem, since besides the equations of motion (two principles of linear momen-
tum and one of angular momentum for each of the three masses), there are also four
geometric constraints at hand.
4.2.5 Principle of Linear and Angular Momentum

for the Formulation of the Equation System
For each of the free masses, three scalar equations can be formulated by means of
the principle of linear and angular momentum:
The changes of linear momentum (m ü x , m ü z ) are equal to the sum of all forces
acting in the respective direction; changes in angular momentum (ϕ̈) equal
the sum of all moments around the y-axis that pass through the center of gravity
S.
Using the forces and dimensions given in Fig. 4.5, this results in
m CB ü xCB = −Fxl + Fxt , (4.12)

m CB ü zCB = −Fzl − Fzt − P(t) − mg , (4.13)
CB ϕ̈CB = Fxl ez − Fxt ez + Fzl ex − Fzt ex + P(t)ep (4.14)
for the carbody and

m W ü xl = Fxl − Tξl , (4.15)
m W ü zl = Fzl − m W g + Nl , (4.16)
W ϕ̈l = Tξl r0 , (4.17)
for the leading wheelset. Corresponding equations for the trailing wheelset can be
derived.
4.2.6 Elimination of the Constrained Forces. Final System

of Equations
Equations (4.15)–(4.17) will be converted through the introduction of the geometric

constraints. First, it has to be considered that the displacements u zl and ϕl are not
independent displacements, due to Eqs. (4.4) and (4.1).
With Eq. (4.16), the unknown vertical force (normal force) between wheel and
rail can be determined,
Nl = m W z̄¨ l + m W g − Fzl , (4.18)
and for the trailing wheelset,
Nt = m W z̄¨ t + m W g − Fzt . (4.19)
With Eq. (4.17), Tξl can be determined:

W
Tξl = − ü xl . (4.20)
r02
Equation (4.20) can be introduced into Eq. (4.15). Rearranged, this results in
W
(m W + )ü xl = Fxl . (4.21)
r02
A corresponding equation is valid for the trailing wheelset:
W
(m W + )ü xt = −Fxt . (4.22)
r02
The Eqs. (4.12)–(4.14) and (4.21)–(4.22) yield the equations of motion. It has to be
kept in mind that for the leading and trailing wheelsets, u zl = z̄ l and u zt = z̄ t . The
remaining forces F have to be replaced by displacements of the center of gravity by

means of Table 4.2.
As equations of motion for the carbody we can therefore write
m CB ü xCB = −2cx u xCB + 2cx ez ϕCB + cx u xl + cx u xt , (4.23)

m CB ü zCB = −2cz u zCB + cz z̄ l + cz z̄ t − P(t) − mg , (4.24)
CB ϕ̈zCB = 2cx ez u xCB − 2cx ez ϕCB − 2cz ex ϕCB − cx ez u xl − cx ez u xt
2 2
−cz ex z̄ l + cz ex z̄ t + P(t)ep , (4.25)
and as equations of motion for the two wheelsets,

CB
m W + 2 ü xl = cx (u xCB − u xl ) − cx ez ϕCB , (4.26)
r0

CB
m W + 2 ü xt = cx (u xCB − u xt ) − cx ez ϕCB . (4.27)
r0
Equations (4.23)–(4.27) are not very well arranged. Therefore, it is usual to represent
the equations as matrices. With the displacement vector uT = {u xCB , u zCB , ϕCB , u xl ,
u xt }, this results in
⎡ ⎤⎧ ⎫
m CB 0 0 0 0 ⎪
⎢ 0 ⎥⎪⎪ü xCB ⎪
⎪
⎪
⎢ m CB 0 0 0 ⎥⎪⎨ü zCB ⎪
⎬
⎢ 0 0 CB 0 0 ⎥
⎢ ⎥ ϕ̈CB +
⎢ 0 0 mW + W ⎥⎪ ⎪
⎣ 0 r02
0 ⎦⎪⎪
⎪ ü xl ⎪
⎪
⎪
W ⎩ ⎭
0 0 0 0 mW + 2
r0
ü xt

mass forces
⎡ ⎤⎧ ⎫
2cx 0 −2cx ez −cx −cx ⎪ ⎪ u xCB ⎪
⎪
⎢ 0 ⎪ ⎪
⎢ 2cz 0 0 0 ⎥ ⎪
⎥ ⎨u zCB ⎬
⎪
+⎢
⎢−2cx ez 0 2cx ez2 + 2cz ex2 cx ez ⎥
cx ez ⎥ ϕCB
⎪ ⎪
⎣ −cx 0 cx ez cx 0 ⎦⎪ ⎪ u xl ⎪
⎪ ⎪
⎪
⎩ ⎭
−cx 0 cx ez 0 cx u xt

spring forces
⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎪
⎪ 0 ⎪ ⎪ ⎪
⎪ 0 ⎪ ⎪ ⎪
⎪ 0 ⎪
⎪
⎪
⎪ ⎪ ⎪ ⎪ ⎪ ⎪
⎨ −m CB g ⎪
⎬ ⎪⎨ −P(t) ⎪ ⎨ cz (z l + z t ) ⎪
⎬ ⎪ ⎬
= 0 + P(t)ep + −cz (z l − z t )ex . (4.28)
⎪
⎪ ⎪ ⎪ ⎪ ⎪ ⎪
⎪
⎪ 0 ⎪ ⎪
⎪ ⎪ 0 ⎪
⎪ ⎪ ⎪
⎪ 0 ⎪
⎪
⎩ ⎭ ⎪⎩ ⎪
⎭ ⎪
⎩ ⎪
⎭
0 0 0

gravity forces force excitation base excitation
The terms with unknown accelerations and displacements are found to the left of
the equal sign. The mass matrix multiplied by the acceleration vector ü is denoted by
M; the stiffness matrix multiplied by the displacement vector u is denoted by S. To
the right of the equal sign are the known forces, divided into gravitational forces, force
excitation, and base excitation. The weight contribution does not change with time;
the corresponding vector is denoted by p0 . The force excitation and base excitation
are time-dependent. The corresponding vectors are marked with the indices I and II.
Equation (4.28) can then be abbreviated to
M ü(t) + Su(t) = p̄0 + p̄I (t) + p̄II (t) . (4.29)
If damping forces are considered, one term is added:
M ü(t) + D u̇(t) + S u(t) = p0 + p̄I (t) + p̄II (t) . (4.30)
The matrix D is the damping matrix of the system. The vector p̄II now also contains
damping parts. The matrix D as well as the parts of damping of p̄II are as follows:
⎡ ⎤ ⎧ ⎫
2dx 0 −2dx ez −dx −dx ⎪
⎪ 0 ⎪
⎪
⎢ 0 ⎪ ⎪
⎢ 2dz 0 0 0 ⎥ ⎥
⎪
⎨ dz (ż l + ż t ) ⎪
⎬
D=⎢
⎢ −2dx ez 0 2dx ez2 + 2dz ex2 dx ez ⎥
dx ez ⎥ ; p̃d = −dz (ż l − ż t )ex .
⎣ −dx ⎪
⎪ ⎪
⎪
0 dx ez dx 0 ⎦ ⎪
⎪ 0 ⎪
⎪
⎩ ⎭
−dx 0 dx ez 0 dx 0
In Eq. (4.30), it is clearly marked which variables are time-dependent. For vectors,
this will be indicated by the symbol˜thus ũ = u(t). Equation (4.30) then reads
M ũ¨ + D ũ˙ + S ũ = p0 + ˜p̄I + ˜p̄II . (4.31a)
4.3 Principle of Virtual Displacements

for Rigid-Body Systems
4.3.1 Initial Remark
There are different possibilities for deriving the equations of motion: for each body,
the principle of linear and angular momentum can be formulated or the equations of
motion can be formulated by means of the principle of virtual displacements (or, as
it is also called, the principle of d’Alembert in the version of Lagrange).
4.3.2 Formulation of the Principle of Virtual Displacements
The principle of virtual displacements replaces the condition of balance (condition

of force balance and condition of moment balance). The principle of virtual dis-
placements for the rigid-body system in Fig. 4.6 can be formulated in the following
way:
For a virtual, i.e., for a geometrically possible, displacement state, the virtual
deformation energy is equal to the virtual work of the external forces (mass
forces, loads, nonconservative reaction forces, i.e., damping forces and creep
forces) and mass forces:
δVint = δWext + δWm . (4.32)
The virtual strain energy δVint is defined as the product of real spring forces Fc j
and virtual spring elongation δv j , summed over all springs j. The virtual work of
the external forces δWext is defined as the product of external forces and virtual dis-
placements in the direction of the forces, summed over all forces. For the present
case that is, besides the gravitational forces, the excitation force P(t). The virtual
work of the mass inertia forces finally is defined as the mass inertial forces (respec-
tively the inertia moments) multiplied by the corresponding virtual displacements
(respectively angular displacements), again summed over all bodies.
body 1 ep
ex ex
u zC
u xC x v = const
rear (index t) front (index l) P (t) 0
mC g C
Di 4 Di 2 ez
cz dz
dx
m Wg
mWg u l
Di 3 cx Di 1
T T l u l
t
Nt Nl
Fig. 4.6 Two-axle vehicle with springs and dampers in parallel (Di = coupling element)
4.3 Principle of Virtual Displacements for Rigid-Body Systems 93
The definition of the “virtual state of displacement” includes that the state has to
be possible from a kinematic point of view. It is a state of displacement, superimposed
on the real displacements, that has to fulfill the geometric constraints. If the vehicle
in Fig. 4.6 is put on a track, then geometric constraints in the normal direction as well
as rolling constraints arise.
The principle of virtual displacement holds an advantage in formulating the equi-
librium conditions or the fundamental dynamic equations for a system in which
the geometric constraints appear in a complicated form. Since the state of virtual
displacements satisfies the geometric constraints, forces in the direction of those
geometric constraints (normal forces at the point of contact and slip forces for the
wheelset) do not contribute to the virtual displacements. We do not have to eliminate
the constraints, which can be tedious. Another advantage is that the rigid carbody
can easily be replaced by an elastic one.
The principle of virtual displacement is now stated formally for a multibody
system with I bodies, J springs, and L forces (weight, outer loads):

J
L

δv j F j = δu xl Pxl + δu zl Pzl
j=1 l=1

I

− δu xi m i ü xi + δu zi m i ü zi + δϕi i ϕ̈i . (4.33)
i=1
If dampers are included in parallel to the springs, then
F j = Fc j + Fd j . (4.34)
Example: Vehicle with Two Wheelsets

As an example, Fig. 4.6 shows once again the vehicle detached from the rail including
all forces acting on the vehicle and between rail and wheelsets. These are as follows:
• the normal forces at the leading and trailing wheelsets, Nl and Nt , combined for
the right and left wheels;
• the longitudinal creep forces (2Tξl and 2Tξt ), also combined for both wheels;
• the weight m CB g of the carbody as well as the weights of both wheelsets (m W g);
• the time-dependent force P(t);
• the d’Alembertian inertial forces and the inertia moments for the carbody (m ü xCB ,
m ü zCB , ϕ̈yCB ) and both wheelsets.
A remark on the direction of the forces:
• The normal forces are positive as compressive forces on the bodies.
• For the creep forces, it has already been stated that they are positive if they act on
the rail in the direction of the ξ-coordinate axis.
• The d’Alembertian forces of inertia are positive in the negative direction of dis-
placement.
Now we will consider the vehicle in Fig. 4.6. First we look at the free-body diagram
without geometric constraints. Then the normal forces and the wheel vertical forces
Tξ will be considered, and one obtains the principle of virtual displacement

4
δv j F j = +δu zC B (−m CB g) + δu zl (−m W g) + δu zt (−m W g)
j=1
+ δu zP (−P(t))
+ δu ζl Nl + δu ζt Nt − δu ξl Tξl − δu ξl Tξl

− δu xC B m CB ü xC B + δu zC B m CB ü zC B + δϕCB CB ϕ̈CB

− δu xl m W ü xl + δu zl m W ü zl + δϕl W ϕ̈l

− δu xt m W ü xt + δu zt m W ü zt + δϕt W ϕ̈t . (4.35)
4.3.3 Inclusion of Geometric Constraints in the Principle

Of Virtual Displacement Exemplified on a Two-Axle
Vehicle
If the vehicle is put on the track with the requirement that the contact and rolling
constraints be met, then we can write for the real displacements
u ζl = u zl = z l ,
u ζt = u zt = z t ; (4.36)
u ξl = u xl − r ϕl = 0,
u ξt = u xt − r ϕt = 0, (4.37)
and for the virtual displacements,
δu ζl = δu zl = 0,
δu ζt = δu zt = 0 ; (4.38)
δu ξl = δu xl − r δϕl = 0,
δu ξt = δu xt − r δϕt = 0. (4.39)
The six terms become zero. Both normal forces, the forces Tξ , and the wheelset mass
m W are no longer part of the principle of virtual displacement. The following version
of the principle remains:
4.3 Principle of Virtual Displacements for Rigid-Body Systems 95

4
δv j F j = +δu zCB (−m CB g) +
j=1
+ δu zP (−P(t))

− δu xCB m CB ü xCB + δu zCB m CB ü z1 + δϕCB CB ϕ̈CB

− δu xl m W ü xl + δu zl m W ü zl + δϕl W ϕ̈l

− δu xt m W ü xt + δu zt m W ü zt + δϕt W ϕ̈t . (4.40)
4.4 Formalized Derivation of the Equations of Motion

and the Principle of Virtual Displacements
Technically, with Eq. (4.40), all is done. We would like to present one more algorithm,
for which we use matrix syntax. The development of such an algorithm of multibody
systems based on the principle of virtual displacement, Eq. (4.35) or (4.40), is a
standard procedure, which is explained in detail in, e.g., [1] pp. 259ff. For clarity,
we would like to take a more particular path but one that is also simpler. As for the
case of formulation by means of the principle of impulse and momentum, we will
proceed systematically step by step:
• Specification of the displacement vector with the degrees of freedom of the cut
free system;
• Specification of the relation between spring elongation and displacement;
• Specification of the spring laws and formulation of the strain energy;
• Specification of the mass matrix and the virtual work of the inertial forces;
• Formulation of the equation of motion in vector–matrix syntax for the free system;
• Specification of the constraints and formulation of the equations of motion of the
suspended system.
4.4.1 Displacement Vector with the Degrees of Freedom

of the Free System
The displacement vector for the free system can be written as

uTf = u xCB , u zCB , ϕCB , u xl , u zl , ϕl , u xt , u zt , ϕt .
Consequently, the free system has nine degrees of freedom.

4.4.2 Relation Between Spring Elongation and System

Displacement
The elongations for the four springs can also be summarized in a vector,

v T = v xl , v zl , v xt , v zt .
The relationship between spring elongation and displacement is given in the form of
v = T v uf , (4.41)
where the matrix T v does not have to be derived, since it is already given in Table 4.1:
⎡ ⎤
1 0 −ez −1 0 0 0 0 0
⎢ 0 1 −ex 0 −1 0 0 0 0⎥
Tv = ⎢
⎣ −1
⎥. (4.42)
0 +ez 0 0 0 1 0 0⎦
0 1 +ex 0 0 0 0 −1 0
For the relative velocities in the dampers parallel to the springs, one has
v̇ = T v u̇f . (4.43)
4.4.3 Spring Forces and the Formulation of the Virtual

Strain Energy
There is a linear relationship between the spring forces and spring elongations,
⎧ ⎫ ⎡ ⎤⎧ ⎫
⎪
⎪ Fcxl ⎪
⎪ cx 0 0 0 ⎪ ⎪ v xl ⎪
⎨ ⎬ ⎢0 ⎨ ⎪ ⎬
Fczl cz 0 0⎥ ⎥ v zl
= ⎢
⎣0
⎪ ⎪
⎪ Fcxt ⎪ 0 cx 0⎦⎪ ⎪ v xt ⎪
⎩ ⎭ ⎩ ⎪ ⎭
Fczt 0 0 0 cz v zt
or shortened in vector–matrix syntax,
F c = Cv. (4.44)
In the same way, the forces in the dampers parallel to the springs can be summarized
as
F d = Dv̇. (4.45)
The result for spring–damper connection elements is therefore

4.4 Formalized Derivation of the Equations of Motion and the Principle … 97
F = F c + F d = Cv + Dv̇. (4.46)
Still required are the virtual spring elongations. Analogous to Eq. (4.41), one has
δv = T v δuf . (4.47)
Now the virtual strain energy of the free system can be expressed by the system
displacement:
δVint,f = δuTf T Tv C T v uf . (4.48)

stiffness matrix Sf
of the free system
The compilation of the dampers is done correspondingly.
4.4.4 Mass Matrix and Formulation of the Virtual

Work of the Inertia Forces
The virtual work of the inertia forces can be written in a very compact way as well
if the masses and moments of inertia are summarized in a matrix:
⎡ ⎤
m CB 0 0 0 0 0 0 0 0
⎢ 0 m CB 0 0 0 0 0 0 0 ⎥
⎢ ⎥
⎢ 0 0 CB 0 0 0 0 0 0 ⎥
⎢ ⎥
⎢ 0 0 0 m Wl 0 0 0 0 0 ⎥
⎢ ⎥
Mf = ⎢ ⎢ 0 0 0 0 m Wl 0 0 0 0 ⎥
⎥. (4.49)
⎢ 0 0 0 0 0 Wl 0 0 0 ⎥
⎢ ⎥
⎢ 0 0 0 0 0 0 m Wt 0 0 ⎥
⎢ ⎥
⎣ 0 0 0 0 0 0 0 m Wt 0 ⎦
0 0 0 0 0 0 0 0 Wt
The virtual work of the inertial forces of the free system can then be written as
δWm,f = − δuTf M f üf . (4.50)
4.4.5 External and Constraint Forces
The virtual work of the external forces can also be written in vector format. If a load
vector of the free system,

p̄T0 f (t) + p̄TIf (t) = 0, −m CB g − P(t), ep P(t); 0, −m W g, 0; 0, −m W g, 0 ,
(4.51)
and a vector c(t) of the constraint forces of the free system,

cT (t) = 0, 0, 0; −Tξl , Nl , 0; −Tξl , Nl , 0 , (4.52)
are introduced, the result is

δWe,f = δuTf p̄0f (t) + p̄If (t) + c(t) . (4.53)
4.4.6 Equations of Motion for the Free System. Introduction

of Constraints
The principle of virtual displacement for the free system using vector–matrix syntax is

δuTf Sf uf (t) = δuTf p̄0f (t) + p̄If (t) + c(t) − δuTf M f üf (t). (4.54)
The displacement of the free system can be expressed by means of the constraints
through the remaining displacements. Summarized in matrix syntax, the constraints
read
⎧ ⎫ ⎡ ⎤
⎪ u xCB ⎪ 1 0 0 0 0 ⎡ ⎤
⎪
⎪ ⎪
⎪ ⎢ 0 1 0 0 0 ⎥ 0 0
⎪
⎪ u ⎪
⎪
⎪
⎪
zCB
⎪ ⎢
⎪ ⎥⎧ ⎫ ⎢ ⎥
⎪
⎪ ϕCB , ⎪
⎪ ⎢ 0 0 1 0 0 ⎥ ⎪ u xCB ⎪ ⎢ 0 0 ⎥
⎪
⎪ ⎪ ⎢
⎪ ⎢ 0 0 0 1 0 ⎥⎪ ⎥ ⎪
⎪ ⎪ ⎢ 0 0 ⎥
⎪ ⎢
⎨ u xl ⎪
⎪ ⎬ ⎢ ⎪ ⎥
⎥ ⎨ u zCB ⎬ ⎢ 0 0 ⎥ z̄ l
u zl = ⎢ ⎥
0 0 0 0 0 ⎥ ϕCB , + ⎢ ⎢ ⎥ , (4.55)
⎪
⎪ ⎪ ⎢
⎪ ⎢ 1 ⎥ ⎪
⎪ ⎪
⎪ ⎢ 1 0 ⎥ ⎥ z̄ t
⎪
⎪ ϕ ⎪
⎪ 0 0 0 0 ⎪ u ⎪
⎪ ⎢ ⎥⎪ ⎪ ⎢ ⎥
l xl
⎪
⎪ ⎢ 0 0 0 0 1 ⎥ ⎩ u xt ⎭ ⎢ 0 0 ⎥
r0
⎪
⎪ u xt ⎪
⎪
⎪ ⎢ ⎥ ⎣
⎪
⎪ ⎪ ⎣ 0 0 ⎦
⎪
⎪ u zt ⎪
⎪
⎪ 0 0 0 0 0 ⎦
⎩ ⎭ 0 1
ϕt 0 0 0 0 r10
which can be shortened to
uf (t) = T kin u(t) + T z z̄(t)

δuf (t) = δT kin u(t). (4.56)
If these relations are included in Eq. (4.54), the constraint forces will disappear (to
be checked!). Initially, this results, in matrix syntax, in the principle of virtual dis-
placements for the vehicle rolling on the rail,
δuT T Tkin Sf T kin u(t) + δuT T Tkin Sf T z z̄(t)

= δuT T Tkin p̄0f (t) + p̄If (t) + c(t) , − δuT T Tkin M f T kin ü . (4.57)
4.4 Formalized Derivation of the Equations of Motion and the Principle … 99
Using further abbreviations, we get

δuT S u(t) − δuT p̄II = δuT p̄0 − p̄I (t) − δuT M ü(t). (4.58)
Since the virtual displacements are arbitrary, it follows that the equations of motion
for vehicle rolling on the rail can be written as
S ũ + M ũ¨ = ˜p̄0 + ˜p̄I + ˜p̄II , (4.59)
where once again, the symbol ˜ is used to mark the time-dependency. If the matrix
operations are performed, the result is of course the system of differential equations
(4.28).
4.5 Equations of Motion for Elastic Car Bodies
The equations of motion for vehicles with elastic carbodies are formulated with the
assumption that in addition to the degrees of freedom for the rigid bodies, degrees of
freedom for the elastic carbody have to be considered in the form of amplitudes of
eigenmodes of the carbody. We will restrict ourselves to pure vertical behavior and
vertical eigenmodes of the free carbody. The displacement vector for the free system
then reads

ũTf = u xCB , u zCB , ϕCB , q1 , q2 , q3 , u xl , u zl , ϕl , u xt , u zt , ϕt .
Instead of nine degrees of freedom, there now are twelve. It would now be neces-
sary to formulate the principle of virtual displacement for a system with rigid bodies
and elastic bodies. The calculation of the elastic bodies is done by means of finite
element software. We will not consider that here but refer to the relevant literature
[1, 2]. Such a calculation can, for instance, be performed for a completely free
structure.
Due to this extension, there are several changes:
1. The mass matrix M CB,f of the free carbody now contains additional terms, called
generalized masses. Also, the stiffness matrix SCB,f is occupied on the diagonal
at places where a multiplication by q1 , q2 , and q3 or the corresponding virtual
displacements occurs. These variables are called generalized stiffnesses.
2. For spring elongations, influences of the elastic degrees of freedom q1 , q2 , and
q3 have to be taken into account. As a consequence, further terms appear in the
stiffness matrix, which will be multiplied by q1 , q2 , and q3 and the corresponding
virtual displacements.
The mass matrix for a free carbody becomes

⎡ ⎤
m CB 0 0 0 0 0
⎢ 0 m CB 0 0 0 0 ⎥
⎢ ⎥
⎢ 0 0 CB 0 0 0 ⎥
M CB,f = ⎢
⎢ 0
⎥. (4.60)
⎢ 0 0 m gen,1 0 0 ⎥ ⎥
⎣ 0 0 0 0 m gen,2 0 ⎦
0 0 0 0 0 m gen,3
The result for the stiffness matrix SW,f of the free carbody can be written as
⎡ ⎤
0 0 0 0 0 0
⎢ 0 0 0 0 0 0 ⎥
⎢ ⎥
⎢ 0 0 0 0 0 0 ⎥
SCB,f = ⎢
⎢
⎥. (4.61)
⎢ 0 0 0 sgen,1 0 0 ⎥ ⎥
⎣ 0 0 0 0 sgen,2 0 ⎦
0 0 0 0 0 sgen,3
The first three diagonal positions are zero, since vertical motion, longitudinal move-
ment, and pitching are not suspended for the free carbody. Generalized masses and
stiffnesses have to be calculated, for instance with finite element software.
The only things missing now are the influences of the elastic degrees of freedom
on spring elongation and damper relative velocities. In order to obtain these, all that
needs to be done is to rewrite the matrix T v of Eq. (4.42):
⎡ ⎤
1 0 −ez ϕxl,1 ϕxl,2 ϕxl,3 −1 0 0 0 0 0
⎢ 0 1 −ex ϕzl,1 ϕzl,2 ϕzl,3 0 −1 0 0 0 0⎥
Tv = ⎢
⎣ −1
⎥. (4.62)
0 +ez ϕxt,1 ϕxt,2 ϕxt,3 0 0 0 1 0 0⎦
0 1 +ex ϕzt,1 ϕzt,2 ϕzt,3 0 0 0 0 −1 0
The additionally used symbols ϕxl,1 to ϕzt,3 capture the spring elongation due to
elastic eigenmodes. For example, ϕxl,1 is the elongation of the spring between the
leading wheelset and the vehicle in the x-direction due to the first elastic eigenmode
(q1 ). The rest is performed as for the rigid carbody; see Sect. 4.4.3, with the slight
exception that the transformation matrix Tv from Eq. (4.62) has to be used. Then the
stiffness matrix of the springs is obtained, which has to be superimposed with the
stiffness matrix for the free carbody, Eq. (4.61).
In implementing this concept in a software system, it becomes clear that a large
number of eigenmodes is necessary in order to reach convergence. The reason for this
can easily be illustrated. For simplicity, the carbody is approximated as a beam. From
the vertical springs, discrete forces act on this beam. The bending moment then has
a kink at the spring connection point. This discontinuity has to be approximated by
the elastic eigenmodes, which have a continuous course, which never fully succeeds.
In order to obtain an acceptable approximation, a large number of eigenmodes is
necessary. Workarounds for this dilemma can be found in [1]. For rail vehicles, a
very successful concept was realized by Dietz [3].
4.6 Solution for Free Vibrations 101
4.6 Solution for Free Vibrations
The solution of the homogeneous system of equations (4.30),
M ũ¨ + D ũ˙ + S ũ = 0, (4.63)
describes the free vibrations with the initial conditions
ũ(t = 0) = u0 ˙ = 0) = u̇0 .
ũ(t (4.64)
Usually, an approach of the form

ũ = u eλt (4.65)
is introduced that transforms the system of differential equations into an (algebraic)

eigenvalue problem
[ λ2 M + λ D + S ] u = 0. (4.66)
The eigenvalues can be obtained with software such as MATLAB. Corresponding

to the number j of degrees of freedom, this results, disregarding abnormal cases, in
exactly j pairs of eigenvalues
λ j = −δ j iω j , (4.67)
which in most cases are complex conjugates.1 The associated pairs of eigenvectors
are then also complex conjugates.
The complete solution is a combination of the conjugate complex parts. In this
superposition, free coefficients appear. These have to be complex conjugates ( Ai and
Āi ) in order for the complete solution to be real. One obtains, therefore,
N

ũh = A j u j eλ j t + Ā j ū j eλ̄ j t . (4.68)
j=1
With the eigenvalues from Eq. (4.67), it is possible to write

N

ũh = e−δ j t A j u j eiω j t + Ā j ū j e−iω j t , (4.69)
j=1
with δ j as decay factor ,

ω j as angular eigenfrequency ,
u j as eigenvector ,
ū j as complex conjugate eigenvector to u j .
1 In special cases, real eigenvalues appear.

For the undamped system the eigenvalues are imaginary,
λ j = iω j ,
and the eigenvectors u j = ū j are real. The coefficients, however, are still complex
conjugates. The complete homogeneous solution for the undamped system is

N

ũh = u j A j eiω j t + Ā j e−iω j t , (4.70)
j=1
since ū j = u j . The real solution writes

N
ũh = u j (E j cos ω j t + F j sin ω j t) , (4.71)
j=1
where E j = 2 Re(A j ) and F j = 2 Im(A j ). The unknown coefficients E j and F j

are calculated from the initial condition (4.64) for t = 0.
For the undamped system of five degrees of freedom, there are only imaginary
eigenvalues (angular eigenfrequency). In Table 4.3, the corresponding eigenvectors
ui are summarized, illustrated, and interpreted:
The shape of the first natural vibration rolling can be determined without cal-
culation. Also, for the second natural vibration, vertical motion, the determination
of the analytical solution without numerical calculation is possible, since the pure
vertical vibration of the carbody is decoupled from all other degrees of freedom. This
is due to the fact that the model is completely symmetric with regard to the z -axis.
An analytical solution can also be found for the inversely phased vibration of both
wheelsets (fifth natural vibration), once again due to symmetry.
The third natural vibration is a form for which the pitching of the carbody dom-
inates coupled with small longitudinal movements. For the fourth natural vibra-
tion there is an in-phase longitudinal vibration of both wheelsets in the positive
x -direction. The carbody moves in the negative x -direction, that is, out of phase
compared to the wheelsets. In these cases, the eigenfrequencies are approximate,
based on the assumption that the wheelset masses are considerably smaller than the
carbody mass. A complete decoupling of the pitching movement from the longitudi-
nal movement would occur if the centers of gravity of the wheelsets and the carbody
were at the same height.
Vertical and pitching vibrations are relevant for comfort calculations. Rolling
motion and longitudinal vibration are relevant if the vehicle hits an obstacle (buffer
stop) or if the behavior of several coupled vehicles is to be investigated during heavy
braking. The longitudinal vibration has only limited importance during acceleration.
For this, however, a system of equations under consideration of the degrees of freedom
of the engine must be derived.
Table 4.3 Eigenvalues and eigenmodes (to be seen as examples)
Eigen-value λi Eigen vector ui Graphical representation Interpretation
⎧ ⎫ ⎧ ⎫ z Rolling i.e. “oscillating” with frequency ω1 = 0
λ1 = 0 ± i0 ⎪ u xC B ⎪ ⎪ 2.5 ⎪
⎪
⎪ ⎪
⎪ ⎪ ⎪ ⎪
⎪
⎪
⎪ ⎪ ⎪ ⎪ x
⎨ u zC B ⎪
⎬ ⎪ ⎨ 0 ⎪ ⎬
ϕC B = 0 = uC B
4.6 Solution for Free Vibrations
⎪
⎪ ⎪ ⎪ ⎪
⎪ u xl ⎪
⎪ ⎪ ⎪
⎪ ⎪ 2.5 ⎪
⎪ ⎪
⎪
⎪
⎩ ⎪
⎭ ⎪ ⎩ ⎪
⎭
u xt 2.5
z
⎧ ⎫ ⎧ ⎫ 2cz
λ2 = 0 ± iω2 ⎪ u xC B ⎪ ⎪ 0 ⎪ Bouncing ω2 =
⎪
⎪ ⎪ ⎪
⎪ ⎪
⎪ x mC B
⎪
⎪ ⎪
⎪ ⎪ ⎪
⎨ u zC B ⎪
⎬ ⎪ ⎨ 2.7 ⎪⎬
ϕC B = 0 = u2
⎪
⎪ ⎪ ⎪ ⎪
⎪ u xl ⎪
⎪ ⎪ ⎪
⎪ ⎪ 0 ⎪
⎪ ⎪
⎪
⎪
⎩ ⎪
⎭ ⎪ ⎩ ⎪
⎭
u xt 0
z
⎧ ⎫ ⎧ ⎫ 2cx ez2 + 2cz e2x
λ3 = 0 ± iω3 ⎪ Pitch with some longitudinal motion ω3
⎪ u xC B ⎪
⎪ ⎪ ⎪
⎪ ⎪ −0.02 ⎪
⎪ ⎪
⎪ x C B
⎪
⎪ ⎪ ⎪ ⎪
⎨ u zC B ⎪
⎬ ⎪ ⎨ 0 ⎪ ⎬
ϕC B = 5.7 = u3
⎪
⎪ ⎪
⎪ ⎪ ⎪ ⎪
⎪
⎪
⎪ u xl ⎪⎪ ⎪
⎪ 0.15 ⎪
⎪
⎪
⎩ ⎪
⎭ ⎩ ⎪ ⎪
⎭
u xt 0.15
(continued)
103
Table 4.3 (continued)
104
Eigen-value λi Eigen vector ui Graphical representation Interpretation

z
⎧ ⎫ ⎧ ⎫ cx
λ4 = 0 ± iω4 ⎪ u xC B ⎪ ⎪ −0.25 ⎪ longitudinal eigenmode with some pitch ω4
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ x m W + W /r02
⎪u
⎪ ⎪ ⎪ ⎪ ⎪
⎨ zC B ⎪ ⎬ ⎨ 0 ⎪ ⎬
ϕC B = ∼0 = u4
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ u xl ⎪⎪ ⎪
⎪ 2.3 ⎪ ⎪
⎪
⎩ ⎪
⎭ ⎪ ⎩ ⎪
⎭
u xt 2.3
z
λ5 = ⎧ ⎫ ⎧ ⎫ cx
⎪
⎪ u xC B ⎪
⎪ ⎪
⎪ 0 ⎪⎪ Oscillation of wheelsets ω5 =
0 ± iω5 ⎪
⎪u ⎪
⎪ ⎪ ⎪ ⎪
⎪ x m W + W /r02
⎪
⎨ zC B ⎪ ⎬ ⎪ ⎨ 0 ⎪ ⎬
ϕC B = 0 = u5
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ u xl ⎪⎪ ⎪ ⎪
⎪ −1.3 ⎪
⎪
⎩ ⎪
⎭ ⎪ ⎩ ⎪
⎭
u xt 1.3
4 Vertical Dynamics, Equations of Motion …
4.7 Exercises for This Chapter 105
4.7 Exercises for This Chapter
4.7.1 Constraint Forces when Constraints are Met
Show that the constraint forces from the equations of motion (4.54) are equal to zero
if the constraints are met.
4.7.2 Validity of the Rolling Condition
The rolling condition is valid without limitation only for pure rolling motion (first
eigenform from Table 4.3). In the third, fourth, and fifth eigenforms, there will always
be forces arising, among them creep forces Tξl and Tξt . Since creep forces (see
Sect. 3.4) always are connected to creepage, the rolling condition is violated. For a
free wheelset, the system of differential equations is to be formulated and solved.
The rolling condition is now no longer valid; u x and ϕy are independent of each
other.
The creepage–creep force relation is formulated as follows:
Tξ = f ξ νξ .
• What is the relation for the determination of the longitudinal creep νξ ?

• Formulate the system of equations of motion.
• Which eigenvalues will be obtained for the limiting case v 0 → 0?
• Interpret the result.
References
1. R. Gasch K. Knothe, R. Liebich, Strukturdynamik, Diskrete Systeme und Kontinua (2. neubear-
beitete Auflage) (Structural Dynamics, Discrete Systems and Continua) (Springer, Berlin 2012)
2. K. Knothe, H. Wessels, Finite Elemente–Eine Einführung für Ingenieure, 3., überarb. und erw.
Aufl. (Finite elements–An introduction for engineers. 3rd ed.) Springer, Berlin, Heidelberg, New
York, Barcelona, Hong Kong, London, Mailand, Paris, Singapur, Tokio, 1999
3. St. Dietz, Vibration and Fatigue Analysis of Vehicle Systems Using Component Modes.
Fortschritt-Berichte VDI (also PhD thesis TU Berlin), Reihe 12, Nr. 401. VDI–Verlag, Düs-
seldorf (1999)
Chapter 5
Forced Vertical Vibrations for Excitation
with Harmonic and Periodic Track
Irregularities (Frequency Domain Solution)
On the basis of the equations of motion for the two-axle vehicle with its five degrees
of freedom, Eq. (4.28), it can be shown that for the carbody, the degree of freedom of
the vertical motion is not coupled with the other four degrees of freedom (pitching
and longitudinal vibration of the carbody and wheelsets). The main features for the
calculation of forced vibration can therefore be illustrated on a system of one degree
of freedom that, excited by track irregularities, performs vertical motions; Fig. 5.1.
The equations of motion for the vertical motion of a vehicle with springs and
dampers can be written as
m ũ¨ z + d ũ˙ z + cũ z = cz̃ + d z̃˙ . (5.1)
Even if the track excitation z̃ is introduced as pure cosine excitation, Fig. 5.1, the
solution ũ z from Eq. (5.1) contains cosine as well as sine terms,
ũ z = u zc cos t + u zs sin t . (5.2)
Therefore, two coupled equations for the determination of the two unknown ampli-
tudes u zc and u zs are obtained. The determination of the solution is, as for all systems
with several degrees of freedom, much easier in a complex formulation.
First the basic rules for the complex calculation will be introduced (Sect. 5.1). Then
the solutions for vertical vibrations of a rail with a cosine-shaped track irregularity
will be determined and interpreted (Sect. 5.2). Finally, we will consider the general
periodic solution (Sect. 5.3).

DOI 10.1007/978-3-319-45376-7_5
108 5 Forced Vertical Vibrations for Excitation …
~
uz
m
z(x)
x=vt 2
z(x) = z c cos x
L
x
L
Fig. 5.1 Notation for a one-degree-of-freedom system
5.1 Complex Syntax
In order to apply the solution from Eq. (5.1) in complex form, some calculation rules
are needed that can be found in many mathematics textbooks (e.g., [1]):
Rule 1
Every complex number can be written either as
z = x + iy , (5.3)
or in polar form with the absolute value |z| = r and a rotation angle ϕ:
z = r (cos ϕ + i sin ϕ) ; (5.4)
see Fig. 5.2.

For the transformation from r and ϕ to x and y, one has
x = r cos ϕ, y = r sin ϕ. (5.5a,b)
In reverse order,

r= x 2 + y2 , (5.6)
Fig. 5.2 Polar form of a

imaginary part
complex number
z = x + iy or z = re i
r
y = r sin
x = r cos real part

5.1 Complex Syntax 109
is obtained for the absolute value. The angle ϕ is also called the argument of z,
ϕ = arg z , (5.7)
in which ϕ is uniquely determined only for −π < ϕ ≤ π. This is called the principal
value of the argument. It is
⎧
⎨ arccos √ x
if y ≥ 0
x 2 +y 2
ϕ= (5.8)
⎩ − arccos √ x2 2 if y < 0 .
x +y
Instead, it can be written

x
ϕ = arccos · sgn y ,
x2 + y2
where sgn y is the so-called signum function. In principle, the calculation is possible
by means of arctan; the distinction of the cases, however, becomes more complicated.
Rule 2
Using Euler’s formula we can write this in shortened form as
eiϕ = cos ϕ + i sin ϕ , (5.9)
where always
|eiϕ | = 1 . (5.10)
Rule 3
Complex numbers in polar form are multiplied by multiplying the absolute values
and adding the arguments. For
z = |z|eiϕ , (5.11)
w = |w|e , iψ
(5.12)
one has
z · w = |z||w|ei(ϕ+ψ) (5.13)
and
z |z| i(ϕ−ψ)
= e . (5.14)
w |w|
(a) (b) Im
Im
i t*
e
û z
1
t*
t*- t*
1 Re
Re e
i t* u^ z Re
1 (t)=cos
f (t) t f (t)
u^ z
cos t*
t* u^ z cos ( t*- )
1 (
+ )
2
1 (
+ )
1 (3
+ )
2
t t
Fig. 5.3 Cosine vibration without a and with b phase angle as the real part of a vector in the
complex plane
Rule 4
The function cos t can now be interpreted as the real part of a vector eit rotating
in the plane of complex numbers; see Fig. 5.3a. We can write
cos t = Re {eit } . (5.15)
One oscillation ũ z = û z cos (t − γ) with û z ∈ R can accordingly be considered to

be the real part of a complex rotating pointer û z ei(t−γ) ; see Fig. 5.3b:
ũ z = û z Re {ei(t−γ) } . (5.16)
5.1 Complex Syntax 111
Rule 5
A calculation in complex form becomes possible by adding an imaginary oscil-
lation
ũ z,im = i û z sin (t − γ) ,
to a real oscillation
ũ z,re = û z cos (t − γ) ,
to obtain a complex oscillation
ũ z = û z ei(t−γ) .
After finishing the complex calculation, one can return to the real numbers by taking
the real part.
5.2 Vertical Vibrations on a Track with Cosine-Shaped

Track Irregularity (Cosine Track)
5.2.1 Track Irregularities and Base-Point Excitations
The track irregularity z(x) is initially dependent on the longitudinal track coordi-
nate x. For a cosine track, we can write
2π
z(x) = z c cos ( )x (5.17)
L
for the track irregularity.
The base-point excitation z̃ in Eq. (5.1) varies with time. The base-point excitation,
which varies with time, can be obtained from the track irregularity by replacing the
spatial coordinate x with time t, that is, x = vt:
z̃ = z c cos t , (5.18)
where the angular excitation frequency and the excitation frequency f can be
written as
2πv v
= , f = . (5.19)
L L
For the velocity and acceleration, which are needed as well, we get
z̃˙ = −z c sin t , (5.20)

z̃¨ = −z c 2 cos t . (5.21)
frequency [Hz] wavelength [m]

of excitation
m
0.1
L= sleeper
2 distance
10 m 1
0.5
L=
1m
L= wheel
circumference
10 10
track irregularity
4-8 Hz
high m
sensitivity 10
L=
bounce
eigenfrequency 1 100
0m
10
L=
500
10 20 40 60 80 100 vehicle speed [m/s]
Fig. 5.4 Relationship between frequency, wavelength, and speed
It is meaningful to obtain an impression of the frequencies and wavelengths that

are important in rail vehicle dynamics. Human beings are particularly sensitive to
frequencies between 4 and 8 Hz, since the eigenfrequency of the stomach is in this
interval; cf. Chap. 7. The equilibrium organs of human beings are also sensitive to
frequencies below 0.5 Hz. The lower limit should therefore be a frequency of 0.2 Hz.
In rail vehicle dynamics, frequencies above 25 Hz are seldom considered. The interval
of interest therefore is from 0.2 to 25 Hz.
In the diagram in Fig. 5.4, frequencies are seen on the vertical axis and vehicle
speed on the horizontal axis. Constant wavelengths can be seen in the diagram as
inclined lines, and to the right of the diagram is indicated which frequencies are
excited by which track irregularity at different speeds v. If the focus of interest is the
frequency interval from 0.2 to 25 Hz, then velocities between 10 and 100 m/s yield
excitations in the wavelength range from 40 cm (approximate sleeper distance) to
500 m. One can imagine that measuring track irregularities with wavelengths in such
a large interval is not a trivial task.
5.2.2 Solution for the Vertical Motion
The vertical motion of a two-axle vehicle was described by the differential equation
Eq. (5.1). Below, the equation is given again:
m ũ¨ z + d ũ˙ z + cũ z = cz̃ + d z̃˙ . (5.22)

5.2 Vertical Vibrations on a Track with Cosine-Shaped … 113
We are interested only in steady-state solutions. The homogenous solution and

initial conditions are therefore unnecessary.
Step 1.
The base-point excitation z̃ itself can consist of a cosine and a sine part,
z̃ = z c cos t + z s sin t . (5.23)
The base-point excitation is now formulated as a phase-shifted cosine oscillation:
z̃ = ẑ cos (t − β) . (5.24)
Further, ẑ is the absolute value (amplitude) and β the phase angle of the base-point
excitation z̃: zc
ẑ = z c2 + z s2 , β = arccos · sgnz s . (5.25)
z c + z s2
2
Step 2.
The base-point excitation z̃ is regarded as the real part of a complex number
according to rule 4:
z̃ = Re{ẑei(t−β) } . (5.26)
And according to rule 5, it is transformed to a complex oscillation, which is put on

the right-hand side of Eq. (5.1) respectively (5.22):
z̃ = ẑei(t−β) . (5.27)
Step 3.
In order to solve Eq. (5.22), a common mode approach of the form ũ z = û z eit
is introduced, resulting in û z turning out to be complex. We therefore no longer use
the symbolˆ. This yields the relation
(−m2 + id + c)u z eit = (id + c)ẑei(t−β) . (5.28)
Step 4.
The expressions in parentheses can now be expressed in terms of absolute value
and phase. The result for the left-hand side can be found from the vector diagram
(Fig. 5.5):
Fig. 5.5 Vector diagram for Im 2

the forces acting on mass m m
as result of a unit
d) 2
displacement
+(
2
2
)
d
-m
(c
c- m 2 Re
c

(c − m2 )2 + (d)2 eiγ u z eit = c2 + (d)2 e−iα ẑei(t−β) , (5.29)
where the phase angles γ and α can be determined according to rule 1. For γ, one
has
c − m2
γ = arccos · sgnd . (5.30)
(c − m2 )2 + (d)2
The signs of the phase angles are chosen in such a way that they all are negative in
Eq. (5.31).
The expression sgn(d) can be omitted if it is assumed that d is always positive.
The vector diagram can be interpreted as an equilibrium condition among all forces
acting on the mass m under the condition of a harmonically varying displacement z̃
of amplitude 1. A similar transformation is performed on the right-hand side. The
result is an algebraic equation to determine the complex amplitude u z .
Step 5.
The solution for the complex amplitude u z can be stated by means of rule 3,

1 + (2Dη)2
ũ z = ẑei(t−α−β−γ) , (5.31)
[1 − η 2 ]2 + (2Dη)2
where the following abbreviations are introduced:

c
ω=
m
is the eigenfrequency of the undamped system,

η=
ω
is the dimensionless excitation frequency, and
d
D= √
2 cm
is the degree of damping. All that remains is to transform the complex solution back
to a real solution. In order to do that, the real part of the complex solution must
be taken. The real solution can then be expressed as a phase-shifted cosine-shaped
oscillation:

1 + (2Dη)2
ũ z = ẑ cos (t − α − β − γ) , (5.32)
[1 − η 2 ]2 + (2Dη)2

phase
displacement amplitude (absolute value)
where a phase angle β occurs for the base-point excitation,
z̃ = ẑ cos (t − β), (5.33)
and because of the conversion of the base-point excitation into a force, the phase
angle α is introduced:
1
α = − arccos · sgn2Dη . (5.34)
1 + (2Dη)2
The phase shift caused by the system finally is captured with the phase angle γ:
1 − η2
γ = arccos · sgn2Dη . (5.35)
(1 − η 2 )2 + (2Dη)2
From the solution for the displacement result, Eq. (5.32),

1 + (2Dη)2
ũ z = ẑ cos (t − α − β − γ) ,
[1 − η 2 ]2 + (2Dη)2
the acceleration can be determined:

ü̃ z = −2 1 + (2Dη)2
ẑ cos (t − α − β − γ) . (5.36)
[1 − η 2 ]2 + (2Dη)2

phase
acceleration amplitude (absolute value)
5.2.3 Interpretation of the Solution for Different Suspensions

Between Wheelset and Carbody
Solutions for three suspension types are compared with regard to their equations in
Fig. 5.6 and graphically in Fig. 5.7.
These are as follows:
• an unsprung vehicle where the wheel bearing and the carbody mass are rigidly
connected,
• a vehicle with spring and damper between wheel and carbody (passive system),
as well as
• a vehicle that in addition to spring and damper, also has an active control for the
oscillation behavior.
The actively controlled vehicle is explained further in the following: The absolute
acceleration ũ¨ z in the carbody is measured through a sensor (accelerometer). The
signal is on the one hand directly evaluated, and on the other hand integrated in order
to obtain the oscillation velocity ũ˙ z . Both signals are multiplied by suitable factors
a and b and superimposed in order to steer an actuator. By means of the actuator, a
force
F = −a ũ¨ z − bũ˙ z (5.37)
is applied between carbody and wheel.

All three vehicles run on a track with a cosine-shaped track irregularity. Displace-
ments and accelerations in the carbody are compared. Here ũ z is the vertical degree
of freedom for the carbody, and z̃ is the base-point excitation. The solution for the
vehicle with additional control can be stated analogously to the solution of the vehicle
with spring and damper.
In Fig. 5.7, the transfer functions are shown in the second row, where the dis-
placement amplitude û z of the vehicle in relation to the excitation amplitude ẑ can
be seen. Illustrating the acceleration result üˆ z , there are several possibilities. If the
acceleration amplitude ẑ2 is chosen as a reference value (last row), then the same
curves are obtained as for the displacement transfer function. More obvious is the
reference to a constant acceleration, where for the rigid vehicle, gravity is chosen,
and for the other two cases, ẑω 2 (second row).
Some conclusions:
• In order to avoid the extreme acceleration amplitudes that can occur for the
L g
unsprung vehicle (for v> 2π z
, the vehicle lifts), springs and dampers have
to be used.
• It becomes apparent that—compared to an unsprung vehicle—the passive √ √ sus-
pension is of benefit only for higher velocities, that
√ is, when 2πv/L≥ 2 c/m.
Therefore, the eigenfrequency of a vehicle ω = c/m should be designed to be
as low as possible (goal: f < 1 Hz).
vehicle types
rigid vehicle suspended vehicle (passive) suspended vehicle (active)

~
u z
~ dt
u z
~ ~ m a b
u z v uz v
m m d c
z(x) +
~
u
z(x) z(x) d c z z
z z x
x x controlled actuator with force F
L L ( see legend )
accelerometer ( piezo-quartz )
provides absolute acceleration
excitation
2πx 2πvt 2πv

z(x) = ẑ cos L
; z̃ = ẑ cos L
= ẑ cos Ωt Ω = L
equations of motion
ũz = z̃ ; ũ¨z = z̃¨ mũ¨z + dũ˙ z + cũz = cz̄˜ + dz̃˙ (m + a)ũ¨z + (d + b)ũ˙ z + cũz
= cz̃ + dz̃˙
solutions for displacements and accelerations
special case d=0

ũz = ũz =
√
ẑ 1+(2Dη)2
ũz = ẑ cos Ωt √ cos(Ωt−α−γ) √ ẑ
cos (Ωt−γ)
(1−η 2 )2 +(2Dη)2 (1−η 2 )2 +(2Dη)2
mit η = Ωω η = Ωω
c
ω = mc ω = m+a
D= √d
2 cm
D= √ b
2 c(m+a)
1−η 2 1−η 2
γ = arccos √ γ = arccos √
(1−η 2 )2 +(2Dη)2 (1−η 2 )2 +(2Dη)2
α = − arccos √ 1
1+(2Dη)2
ũ¨z = ũ¨z =
ũ¨z = −Ω2 ẑ cos Ωt = −Ω2 ûz cos(Ωt − γ − α) = = −Ω2 ûz cos(Ωt − γ) =
2
√ 2
√−η 21+(2Dη)
2 2
ẑω 2 √ −η 2
ẑω 2
(1−η ) +(2Dη) (1−η 2 )2 +(2Dη)2
cos (Ωt−α−γ) cos (Ωt−γ)

= üz cos (Ωt + π − γ − α)
ˆ = üz cos (Ωt + π − γ)
ˆ
Fig. 5.6 Vehicle responses for displacements and accelerations for a rigid vehicle, a passively
suspended vehicle, and an actively suspended vehicle
dt
^
u v
z rigid passively actively a b
vehicle c d suspended suspended c
vehicle vehicle
transfer function (displ.)
^
u ^
u
z z
u^ z z^ D= 0 z^
z^ D= 0.2 D= 0
u^ z / z^
D= 0.2
D= 0.5 D= 0.5
1 1 1
= 2 2 3 1 2 3
^ ^
at high speed uz uz
a c c e le r a t io n a m p lit u d e
^ contact is lost
uz z^ ^z
z /g
g D= 0
in v e h ic le u
^
D= 0.5
1
D= 0.2 D= 0.2
D= 0
1 1
D= 0.5
g 2
= = 1 2 2 3 1 2 3
z^
lim
L
normalized acceleration
^ ^
uz uz
^ D= 0
uz
amplitude in vehicle
z^
2
z^
2
^z 2 D= 0.2 D= 0
D= 0.2
D= 0.5 D= 0.5
1 1 1
1 2 2 3 1 2 3
-
2
z^
z^
2
- normalized acceleration amplitude like displacement amplitude!
Fig. 5.7 Graphical representation of transfer function for a rigid, passively respectively actively
suspended vehicle
• A lower limit for ω results from the fact that a static deflection due to the vehicle
weight
mg
u stat =
c
cannot become too large. For a linear spring, it can be estimated as

1 g
f min . (5.38)
2π u stat,max
In order to achieve f min = 1 Hz, it must therefore be the case that u stat,max = 0.25 m.
• The lower ω is, the more likely it is that the carbody amplitudes drop below
| uzz | = 1, and in consequence, the carbody accelerations become smaller.
5.2 Vertical Vibrations on a Track with Cosine-Shaped Track Irregularity (Cosine Track) 119
√
• The damping of a vehicle with passive damping
√ is useful below η = 2, especially
in the resonance (η = 1); above η = 2, passive damping is harmful. As a
compromise,
0.1 ≤ D ≤ 0.3 ,
is often chosen.
• Active damping is useful in the whole velocity range v > 0. The higher the better.
The parameter a acts as a mass. With a, it is possible to influence the resonance
position. The parameter b acts as a damping against a track following a fixed
point (sky hook). With its help, the level of comfort can be increased. It should be
mentioned, however, that the system considered here is a simplified (ideal) active
one. In reality, e.g., the time delay between measured acceleration and acting force
reduces the possibility of improvement for the active system.
5.3 Vehicle on a General Periodic Track
A general periodic track irregularity as shown in Fig. 5.8 can be described by a Fourier
series. A first kind of Fourier series presentation of z(x) can be written as
∞
z0 c 2π 2π
z(x) = + [z n cos n( )x + z ns sin n( )x] (5.39)
2 n=1
L L
with z 0 = 0 and
L L
2 2
z nc = ¯ 1 x d x z ns =
z(x) cos n ¯ 1 x d x, n = 1, 2, . . . ,
z(x) sin n
L 0 L 0
(5.40)
and
2π
( ¯
)= as base angular displacement frequency.
L
For the track irregularity shown in Fig. 5.8 in the range −L/2 < x < L/2,
~
u
v
z
h
x
Fig. 5.8 Vehicle on general period track irregularity

rail joint
Fig. 5.9 Track irregularities due to track joints
2hx
z(x) = , (5.41)
L
which leads to the following Fourier series:
2h 1 1
z(x) = ¯ 1 x − sin 2
[sin 1 ¯ 1 x + sin 3
¯ 1 x − . . .] . (5.42)
π 2 3
Typical periodic track irregularities are obtained, for example, for unwelded rails
at the rail joints. When the wheelset is passing, the rail is displaced more in the area
of the joint. This can be realized approximately with the track irregularity shown in
Fig. 5.9.
For comfort calculations, a different Fourier series, in which all Fourier terms are
given as phase-shifted cosine functions, is more convenient. Generally we can write
∞
z0
z(x) = + ¯ 1 x − βn ),
ẑ n cos (n (5.43)
2 n=1
with
z nc
ẑ n = (z nc )2 + (z ns )2 , βn = arccos · sgn z ns . (5.44)
(z nc )2 + (z ns )2
For the example from Fig. 5.8, we get
∞
2h 2π π
z(x) = cos n( )x − βn , βn = (−1)n−1 . (5.45)
n=1
πn L 2
This different type of presentation also provides an amplitude spectrum and a phase
spectrum of the track, which is shown in Fig. 5.10 for the example from Fig. 5.8.
Setting x = vt results in the time-dependent spectrum and the motions of the
base point
5.3 Vehicle on a General Periodic Track 121
Fig. 5.10 Amplitude and amplitude spectrum of track

^
phase spectrum of track Zn
2h
n 2
1= n
2 3 4 5 6 L
1 1 1 1 1 1
phase spectrum
n
90°
0 n 1
- 90°
∞

2π
z̃ = ẑ n cos n( )vt − βn . (5.46)
n=1
L
The ẑ n represent the discrete spectrum of the amplitudes. The relation between angu-
lar frequency and the spatial angular frequency ¯ is
¯ 1 v = 1 .
(5.47)
For harmonic excitation by the cosine track according to Eq. (5.32), we get

1 + (2Dη)2
u z (t) = ũ z = ẑ cos (t − α − β − γ) . (5.48)
(1 − η 2 )2 + (2Dη)2
Through generalization, the vehicle response is readily obtained if several fre-

quencies of excitation n1 (n = 1, 2, 3, . . .) act on the vehicle at the same time.
With η1 = 1 /ω, we obtain
∞
1 + (2Dnη1 )2
ũ z = ẑ n cos (n1 t − βn − αn − γn ) , (5.49)
n=0 [1 − (nη1 )2 ]2 + (2Dnη1 )2
1 − (nη1 )2
γn = arccos · sgn2Dnη1 , (5.50)
(1 − (nη1 )2 )2 + (2Dnη1 )2
1
αn = − arccos · sgn2Dnη1 . (5.51)
1 + (2Dnη1 )2
Compactly written, we have
∞

ũ z = V ( = n1 )ẑ n cos (n1 t − βn − αn − γn ) , (5.52)
n=0
or
∞

ũ z (t) = û zn cos (n1 t − βn − αn − γn ). (5.53)
n=0
The spectrum of initial amplitudes ẑ n is multiplied by the transfer function

V ( = n1 ) and results in the spectrum û zn of the vehicle amplitudes; see
Fig. 5.11.
The phase spectrum of the result originates from the superposition of phase angles
of the base-point excitation (βn ), the load conversion (αn ), and the system (γn ). All
Fig. 5.11 Principal ^

Z n
procedure to derive the
response spectrum
2h
n 1
V( )
D ~ 0.2
^
u zn
2h
n 1
5.3 Vehicle on a General Periodic Track 123
phase angles are thereby defined with negative sign in relation to a cosine-shaped
oscillation. Regarding further details, in particular the derivation of the equations,
we refer to [2].
The numerical cost of calculating the vehicle response by means of Eq. (5.52) can
be quite considerable. If we assume a disturbance variable that repeats after 200 m
and disturbances up to a frequency that corresponds to the sleeper distance, then 333
Fourier terms are needed. The alternative is to perform time step integration. As long
as we deal with linear equations of motion in the form of
M ũ¨ + D ũ˙ + S ũ = p̃, (5.54)
there is a very elegant integration method that is stable for every case. With the
help of modal analysis, the system is transferred from N degrees of freedom into
N oscillators with one degree of freedom. These are then integrated with a method
shown in [2] (pp. 175ff and 207ff). For nonlinear cases, usually classical time step
integration methods are applied.
5.4 Solution for a Vehicle with Elastic Carbody Modes
Without going into detail, we would like to study the example of the vertical transfer
function for the vehicle. A four-axle vehicle is examined. In Fig. 5.12 (for the sake
of clarity, only two axles are shown), the carbody is approximated with a beam
and modeled with the first and second bending eigenmodes [3]. The derivation of
the equations of motion is described in Sect. 4.5. The vehicle is excited by vertical
disturbances with a wavelength of 3.63 m. The vehicle runs at different speeds that
excite different frequencies; see Fig. 5.13. The calculation was performed first for
a vehicle modeled with rigid bodies and then with the two bending eigenmodes
included.
For the carbody with additional elastic degrees of freedom (Fig. 5.13), maxima in
vertical acceleration occur in the center of the carbody (dots) as well as above the
z2 z1
v0
speed
u zc
u xc
w
z l , z t : excitation
z 1 , z 2 : measurement
z t (t) z l (t)
Fig. 5.12 Sketch of four-axle vehicle negotiating vertical track irregularities

a z [ m s -2 ]
1000
500
0
0 4 8 12 16 frequency 20 [ Hz ]
1. vertical bending mode
vertical displacement of bogie frame
a z [ m s -2 ]
1000
500
0
0 4 8 12 16 frequency 20 [ Hz ]
0 30 60 speed [ m s -1 ]
Fig. 5.13 Vertical transfer functions for the vehicle in Fig. 5.12 for excitation with vertical track
irregularities with wavelength 3.63 m. Measurements of vertical accelerations in the center of the
vehicle (•) and above the bogies (×). Top figure with elastic carbody modes; bottom figure carbody
modeled as a rigid body
bogies (crosses). The reason for this is that the eigenfrequencies of the carbody for
the first vertical bending eigenmode are located here. For a rigid-body model, these
distinct maxima disappear. There are, however, some weaker maxima for a slightly
lower frequency. These are due to the bouncing eigenfrequency of the bogie, which
is between 10 and 11 Hz.
5.5.1 Two-Axle Vehicle on Cosine Track
Examine the resonance behavior of a two-axle vehicle, the superposition of vertical

and pitch motion, as well as the influence of wavelength and running speed.
For the calculation, the two-axle vehicle in Fig. 4.4 will be the basis. The carbody
is assumed to be rigid. In order to facilitate hand calculation, it is assumed that
the center of gravity for the carbody in Fig. 4.4 is at the same level as the axles
(low loader), that is, ez = 0. Due to this exception, vertical and pitch motion are
decoupled.
Determine the steady-state oscillation resulting from the vertical motion. For
which excitation wavelength is the excitation maximal? For which is it minimal?
Choose typical parameters for a two-axle freight wagon: m = 13000 kg, ex =
4 m, cz = 3 · 106 N/m, dz = 4 · 104 Nm/s, v = 100 km/h. Sketch the transfer function
for the vertical motion as a function of running speed for L = 8 m, respectively as
function of wavelength for v = 100 km/h. Interpret both diagrams.
5.5.2 Two-Axle Vehicle on Generally Periodic Track
The two-axle vehicle from the exercise of Sect. 5.5.1 is now traveling on a track
with rail joints. Maximum track deflection occurs when a wheel is exactly above
the joint, and minimal deflection when a wheel is between two joints. The deflec-
tion acts as a disturbance to the vehicle and can be approximated roughly with a
parabolic excitation

1
with an amplitude of ±0.5 mm and a wavelength L = 50 m,
2
2x
z(x) = z max 1 − . Choose the same parameters as above and determine
l
the amplitude and phase spectrum for the vertical vehicle response.
References
1. K. Meyberg, P. Vachenhauer, Höhere Mathematik, Band 1: Differential- und Integralrechnung

Vektor- und Matrizenrechnung, Band 2: Differentialgleichungen, Funktionentheorie, Fourier-
Analysis, Variationsrechnung (Higher mathematics, Vol. 1: Differential- and integral analysis,
vector and matrix analysis. Vol. 2: Differential equations, complex analysis, Fourier analysis,
calculus of variations). Springer Verlag, Berlin, 1990 (Bd.1), 1991 (Bd.2)
2. R. Gasch, K. Knothe, R. Liebich, Strukturdynamik, Diskrete Systeme und Kontinua (2. neubear-
beitete Auflage) (Structural Dynamics, Discrete Systems and Continua). Springer, Berlin (2012)
3. R. Gasch, W. Hauschild, W. Kik, K. Knothe, H. Steinborn, Stability and forced vibrations of
a 4-axled railway vehicle with elastic carbody, in The Dynamics of Vehicles on Roads and on
Tracks, Proceedings of the 5th IAVSD - 2nd IUTAM Symposium held at the Technical University
Vienna, Austria, September 1977, ed. by A. Slibar, H. Springer (Swets & Zeitlinger, Amsterdam,
1977), pp. 464–480
1 For more accurate studies, track dynamics have also to be taken into account.
Chapter 6
Random Vibrations due to Stochastic Track
Irregularities
To characterize stochastic track irregularities and the resulting random vibrations,

different methods have to be used from those used for harmonic or general periodic
vibrations.
6.1 Characterization of an Irregular Track with Help

of Root Mean Square Value and Power Spectrum
value z¯q (or the effective

The arithmetic (linear) mean value z̄ and the quadratic mean
value derived from the quadratic mean value, z eff = z¯q , which also is called the
root mean square (rms) value) characterize the track irregularities in a rough manner:
L
1
arithmetic (linear) mean value z̄ = z(x) d x , (6.1)
L 0
L
1
quadratic mean value z¯q = (z(x) − z̄)2 d x , (6.2)
L 0

effective value z eff = z¯q . (6.3)
For periodic processes, it is enough to integrate z(x) over one period.

For harmonic track excitation (Fig. 6.1, left) the quadratic mean value z¯q and
the square root, i.e., the effective value, are direct measures for the vibration
amplitude ẑ 1 :

1 L 2 1
z¯q = ẑ 1 cos ¯1 x − β d x = ẑ 12
L 0 2

z eff = z¯q = 0.707 ẑ 1 . (6.4)

DOI 10.1007/978-3-319-45376-7_6
128 6 Random Vibrations due to Stochastic …
z (x) z (x) z (x) L

L
x x x
L L

z(x) = ẑ1 cos (Ω̄1 x − β1 ) z(x) = n ẑn cos (nΩ̄1 x − βn ) random vibration
Fig. 6.1 Harmonic, periodic, and stochastic track excitation; “example functions” of length L
For general periodic excitation (Fig. 6.1, middle), only the squares of the different
amplitudes influence the quadratic mean value. The resulting effective value is a
rough measure of the vibration level:
∞
1 2 1 2
z¯q = [ẑ 1 + ẑ 22 + ẑ 32 + · · · ] = ẑ ,
2 2 n=1 n

∞

z eff = z¯q = 0.707 ẑ 12 + ẑ 22 + ẑ 32 + . . . = 0.707
ẑ n2 . (6.5)
n=1
Since the phase is lost in the calculation of the quadratic mean value, totally differ-
ent vibration processes z(x) lead to the same quadratic mean value as long as the
amplitudes of the harmonic contributions ẑ k are the same.
Uniform random processes (Fig. 6.1, right) are uniform regarding the amplitudes but
random regarding the phase shifts. For the evaluation, an extremely long period is
chosen, L → ∞. If different samples of length L from a measurement result in the
same amplitudes ẑ k even though the phases are different, then a process is uniformly
random and statistically stationary. In that case, the spectra of the vehicle response
can be calculated from track excitation spectra.
The longer the period L is chosen, the smaller the basic frequency ¯ 1 = 2π
L
becomes. The amplitude spectrum becomes more and more “dense,” while the
amplitudes get smaller and smaller. It seems natural to distribute the squares of the
amplitudes ẑ n2 /2 continuously on the angular frequency (with respect to the length
coordinate) ¯=
¯ 1 (Fig. 6.2).
We define
1 2
lim ¯ = S().
ẑ (n) ¯ (6.6)
¯
→0,n ¯
→ ¯ ¯
2
The numerical evaluation is usually performed in the discrete form, however. One
must choose L large enough that ¯ becomes sufficiently small.
6.1 Characterization of an Irregular Track with Help of Root … 129
1 ^2 1 ^2 1 ^* 2 1 ^* 2 S( (
z ( or zn ) z ( or zn )
2 n 2 2 n 2
period L period L * = 2L period L
S( (
S( (
1 2 1 3 1
*
1
*2 *
3
*
4
*
5
*
6
=2 * = 2 / L*
L
spectrum of squares of amplitude power spectrum
Fig. 6.2 Amplitude spectrum for varying period (left, middle) and transformation into continuous
power spectrum (right)
z (s) z (s) z(s) L

L
s s s
L
L
displacement z (s) z (s) = z^ 1 cos ( 1 s- 1) z (s) = zn cos ( n 1 s- n)

n-1
with 1 = 2 /L
z^ n z^ n
z^ 1
1
phase shift for each
1 1 2 13 1 4 1 1 position of L
n n
differs
1= 0
1
1
arithm. mean value z z=0 z=0 z=0

1 ^2
spectrum of squares of 1 2
z^ n z
2 n S( ) 1 ^2 *
2
2z
amplitudes = ^ ( 1)
power spectral density
= PSD
1
2 *
1 13 1 4 1 1 1
1 2
S ( *1) = 2 z^ ( 1=
*1 )
1 2 1 ^2
quadratic mean value z q = 2 z^ 1 zq = 2 zn
zq = S( )d
n
=0
= S (n )
n
2
effective value z eff z eff = 0.707 z^ 1 z eff = 0.707 z^ n z eff = S( ) d
n
(root mean square) =0
Fig. 6.3 Determination of effective value for harmonic, general periodic, and stochastic (“random”)
track irregularities
The power spectrum (= spectrum of squares of amplitudes) provides informa-

tion about the contribution of each frequency to the quadratic mean value.
In a similar way as for harmonic and periodic vibrations, the quadratic mean value
and standard deviation can be determined from the power spectrum (cf. Fig. 6.3).
Examples of power spectra of track irregularities are shown in Figs. 6.8 and 6.10.
6.2 Determination of Vehicle Response for Stochastic Track

Excitation
We recapitulate the determination of vehicle response and the spectrum of squares

of the amplitudes (power spectrum) for general periodic excitation (Fig. 5.8).
The vehicle response was found to be (cf. Sect. 5.3)
∞

u z (t) ≡ ũ z = û zn cos (n1 t − βn − αn − γn ), (6.7)
n=0
with
û zn = V (n)ẑ n . (6.8)
The quadratic mean value can be written as
∞ 2
1 T
1 T
ū q = ũ 2z dt = û zn cos (n1 t − βn − αn − γn ) dt, (6.9)
T 0 T 0 n=0
which results after integration in
∞
1 2 1 2
ū q = [û z1 + û 2z2 + û 2z3 + . . .] = V ( = n1 )ẑ n2 . (6.10)
2 2 n=0
This can be expressed as follows:
The power spectrum of the vehicle response (which builds up the quadratic mean
value with the contributions from each frequency) is derived by multiplication
of the discrete power spectrum of the excitation 21 ẑ n2 by the square of the transfer
function V 2 (1 ).
6.2 Determination of Vehicle Response for Stochastic Track Excitation 131
1 ^2
2 zn power spectrum
of excitation
n 1
1
2
V ( )
square of
transfer function
1
1u
^2
2 zn 1u
^ 2 = V 2 ( ) 1 z^ 2
2 zn 2 n
power spectrum of
vehicle response
n 1
Fig. 6.4 Process of determination of power spectrum of vehicle response for general periodic
excitation
This is illustrated in Fig. 6.4. The effective value can be written as

∞
1
u z,eff = ū q =
V 2 ( = n1 )ẑ n2 . (6.11)
2 n=0
For random statistically stationary excitation, the excitation spectrum is continuous.

Formally, this means that the terms with sums can be transformed into integrals:
∞
=∞
ū q = lim V (n )S(n ) =
2
V 2 ()S() d . (6.12)
→0 0
n=0
The excitation spectrum S() and the spectrum of the vehicle response Su () are
again connected via the square of the transfer function V ():
Su () = V 2 ()S() . (6.13)
A graphical interpretation is given in Fig. 6.5.

The continuous response spectrum again gives the contributions of the different
frequencies to the quadratic mean value u z,eff :

∞

u z,eff =
V 2 ()S() d . (6.14)
=0
S( ) S( ) ^ 1 z^ 2
= 2
power spectrum
of excitation
2
V ( )
square of
transfer function
Su
2
Su ( ) = V ( ) S ( )
power spectrum of
vehicle response
Fig. 6.5 Process of determination of the spectrum of the vehicle response for stochastic excitation
~
~ u z
u z
m
t -
z(x) p ( u z)
= u z q = u z,eff
x
Fig. 6.6 Single-degree-of-freedom system on stochastic excitation
For a Gaussian distribution of track irregularities, also the vehicle response is

Gaussian distributed. Since a Gaussian distribution is completely defined by the
linear mean value (zero in this case) and the quadratic mean value, the probability
density function of the instantaneous position of the system
is also given
(Fig. 6.6).
If the mean value is zero, the oscillation is between + ū q and − ū q for 63.3 %
of the time. In the standard UIC 518 [1] or the more recent European standard [2], the
following procedure is prescribed: To start with, the track to be evaluated is divided
into a number of sections. For each section, the mean value is calculated. For each
section, then the so-called 99.85 % value is calculated. That means that 99.85 % of
all values are lower. A statistical maximum is determined in the following way: from
the 99.85 % values of all sections, the mean value is calculated; for safety-relevant
quantities (e.g., safety against derailment), three times the standard deviation 3 · σ
is added to that mean value; for other quantities, only 2.2 · σ is added to the mean
value.
6.3 Power Spectra of Track Irregularities 133
6.3 Power Spectra of Track Irregularities
6.3.1 How to Derive Power Spectra of Track Irregularities
Discrete power spectra ( 21 ẑ k2 -lines) and continuous spectra S() can be derived by
narrowband filter technology from the irregularity as a function of the length coor-
dinate z(x) or of time z(t).
Derivation of Power Spectrum from Correlation Function
Before describing the derivation of the power spectral density in more detail, some
basic comments shall be made. The power spectrum is defined as the Fourier trans-
form of the correlation function (cf. Khintchine [4]). This means that calculation of
the correlation function is a method to determine the power spectrum. To do this,
the input signal z(t) needs to be stored, displaced with a time instant t ∗ , multiplied
by itself, and finally integrated. Additionally, t ∗ needs to be varied continuously.
This effort is, even with today’s computer capacity, in most cases not meaning-
ful. Therefore, the procedure will probably not be much used. The power spectrum
is instead determined directly. Here older analog and newer digital methods exist.
We will discuss only some basic issues. For more detailed descriptions, we refer
to [5, 6].
Analog Determination of Power Spectrum
Figure 6.7 illustrates how to process an input signal to determine the power spectrum.
It is assumed that the input signal z (t) already is low-pass filtered. By low-pass
filtering, aliasing effects will be avoided, which can cause low-frequency parts of the
signal to be influenced by high-frequency content.
1. The signal is stored. It is processed in parallel or sequentially by a number of
selective filters.
tape recorder band pass filter multiplier (square) integrator

2
Z y )t)
z t 2
Z (f m ) =
f T
z (t ) A z (t, f m ) [ z (t, fm) ] 2 =
1 [ z (t, f ) ] 2 dt
m
T
0
T
0 t fm f t 0 t
2
Z (f m )
power spectral density S (f m ) =
f
T
root mean square value

2 1 2
S (f ) df
= z (t ) dt =
T
0 0
Fig. 6.7 Derivation of power spectrum from a time series (according to Buxbaum [3])
2. The selective filters have bandpass characteristics. The excitation signal can pass
only in the working area of the filter with mean frequency f m and bandwidth f .
The quality of the filter is determined by the order of its transfer function, where
high quality means high selectivity. Here is the biggest disadvantage of analog
measuring technology. High-quality filters are very expensive.
3. The filtered signal is sent to the two inputs of a multiplier, where the square of
the input signal is calculated.
4. The new signal z 2 (t, f m ) is sent to an integrator, where the mean value is calcu-
lated. Regarding measuring technology, the integrator is realized by a first-order
filter that can be described by the following differential equation:
dy
τ + y = z 2 (t, f m ) . (6.15)
dt
Here z 2 is the input signal and y the output signal of the filter. The parameter τ is
a time constant, which should be chosen depending on the mean frequency f m of
the selective filter. Before this is discussed in more detail, another thought will be
explained. The solution of the differential equation (6.15) results in the following
integral expression, which is equal to one point of the spectrum:

1 T
1 T
τ
y dt = z 2 dt + [y(0) − y(T )] . (6.16)
T 0 T 0 T
If the time T is chosen infinitely large, then the calculation of the mean value of
the output signal y and the input signal z 2 results in the same value. But T has to
be limited to get results in a reasonable time. Therefore, for the time constant τ ,
initially low start values will be chosen to minimize the influence of the expression
in brackets. Thus the choice of τ also depends on the mean frequency f m and the
bandwidth f of the preceding bandpass filter. Details can be found in [5, 6].
Digital Determination of Power Spectrum
In recent decades, analog signal processing methods have lost more and more of
their importance as digital methods have become standard. The difficulties associ-
ated with digital signal processing, however, remain. The commonly used method
in today’s measuring equipment to determine power spectra is based on the fast
Fourier transform (FFT). Fourier transforms quantize and digitize the measuring
values and store them in so-called transient memories. The transforms then calculate
the Fourier transform of the input signal. By multiplying the Fourier transform by its
complex conjugate, the power spectral density is determined. This procedure seems
simple. One must not forget, however, that this power spectral density is just an
estimate. In contrast to the analog methods, these methods do not a priori suppress
the aliasing effect. Today, this is more and more done with digital filters as well. For
example, the aliasing effect can be reduced if the signals are discretized with a much
higher sampling rate than demanded from Shannon’s theorem.
6.3.2 Power Spectra for the DB Network
Power Spectra for the Design of ICE 1

For the network of the former Deutsche Bundesbahn, in the report of the working
group Rheine–Freren “Definitionsphase Rad-Schiene-Demonstrationsfahrzeug R/S-
VD” [7], analytic expressions for power spectral densities for the most important
track irregularities can be found (cf. Fig. 2.11):
• Longitudinal level (vertical track irregularity)
AV 2c
¯ =
Sz () (6.17)
¯2
( ¯ 2 + 2c )
+ 2r )(
• Alignment (lateral track irregularity)
AA 2c
¯ =
Sy () (6.18)
¯2
( ¯ 2 + 2c )
+ 2r )(
• Cant irregularity in radians
AC /a 2 · 2c ¯2
¯ =
Sϕ () (6.19)
¯ 2 + 2r )(
( ¯ 2 + 2c )(
¯ 2 + 2s )
These analytic expressions can also be found in [8] and are today used in countries
other than Germany as well. In the USA, very similar expressions are used. They can
be found in [9]. Also in China, power spectra are used to describe track irregularities;
see, for example, [10]. More about analytic descriptions of power spectra and track
irregularities in general can be found in [11]. The unit of spectral density is always the
square of the quantity under consideration divided by the frequency. The longitudinal
level and the alignment are given in meters. Therefore, the unit of the corresponding
power spectral density is m 2 /(rad/m). The cant irregularity is given as an angle. The
unit of the spectral density is 1/(rad/m).
For a conventional track (no high-speed line) with relatively good maintenance
status (good track), the following coefficients will be used:
s = 0.4380 rad/m ,
c = 0.8246 rad/m ,
r = 0.0206 rad/m ,
AV = AA = AC = 5.9233E − 7 m · rad ,
a = 0.75 m .
−2
Longitudinal level −2
Alignment
10 10
−3 −3
10 10
−4 −4
10 10
S(Ω̄) [m3 /rad]
S(Ω̄) [m3 /rad]

−5 −5
10 10
−6 −6
10 10
−7 −7
10 10
−8 −8
10 10
−9 −9
10 10
−10 −10
10 10
−2 −1 0 1 −2 −1 0 1
10 10 10 10 10 10 10 10
Frequency [rad/m] Frequency [rad/m]
−2
Cant irregularity
10
−3
10
−4
10
S(Ω̄) [m3 /rad]
−5
10
−6
10
−7
10
−8
10
−9
10
−10
10
−2 −1 0 1
10 10 10 10
Frequency [rad/m]
Fig. 6.8 Spectral densities of longitudinal level, alignment and cant irregularity as a function of
¯
spatial frequency
At the beginning, the ICE trains were dimensioned with the spectra that are shown
in Fig. 6.8. We want to add, however, that already in 1971, an ORE committee dealt
with power spectral densities of track irregularities [12].
Design Calculations for ICE 2.2
For the design calculation of the ICE 2.2, times series of measured track irregularities
were used instead of power spectral densities [13] (cf. vertical track irregularity on
the Göttingen–Hannover high-speed line; Fig. 6.9). These track irregularities were
used as input to time step integrations. Stichel [14] calculated power spectra of this
time series over a length of 1 km (Fig. 6.10) and performed a polynomial fit with a
fractional rational function. The result was more or less independent of the kilometer
of the line from which the power spectrum was calculated. Only between 4.5 and
5.5 km did the results differ. This can been seen already in the time series. At 5 km,
single defects dominate.
Since other sections also result in similar power spectra, it is likely that the power
spectrum in Fig. 6.10 is representative for track irregularities on a relatively well
maintained German high-speed track.
10
5
[ mm]
0
-5
-10
1000 2000 3000 4000 5000

longitudinal coordinate [m]
Fig. 6.9 Vertical level irregularity on Göttingen–Hannover high-speed line according to [13]
Fig. 6.10 Two-sided power

spectral density for section PSD (km 2.0 - 3.0)
2–3 km of vertical level Polynomial
irregularity from Fig. 6.9
with polynomial fit with a
fractional rational function
¯ ) [m3 /rad]
S(Ω
Frequency [rad/m]
If the power spectra from Fig. 6.8 are compared with those of high-speed tracks
(Fig. 6.10), it can be seen that the decay rate of the curve in the frequency range that
is covered in measurements, i.e., between roughly 0.1 rad/m (62.8 m wavelength)
and 1 rad/m (6.28 m wavelength), is lower in the power spectra of the high-speed
tracks. For higher frequencies, the decay rate, however, is higher. Since the very
low frequency areas are not covered with measured values, this conclusion has to be
handled with care. The maximum value of the power spectrum in Fig. 6.8 is signifi-
cantly higher than the maximum values of the power spectrum for the high-speed line
(Fig. 6.10). There are several reasons for this. On the one hand, in Fig. 6.8 one-sided
power spectra are given. This means that only values for positive frequencies exist.
In Fig. 6.10, two-sided power spectra are shown. To recalculate one-sided power
spectra into two-sided power spectra, all values have to be divided by a factor of 2.
Furthermore, track irregularities usually are higher for conventional tracks than on
high-speed lines. Thirdly, as indicated above, for wavelengths above approximately
70 m, the results are not reliable.
power spectrum
displacement

∞
u(t) = ûn cos (nΩ1 t − βn )
1 ^2 n=0
u
2 n
quadratic mean value :

1 2
∞
1 T 2
ūq = u (t) = û
n 1 T 0 2 n=0 n
1 2 1 3 1 4 1
power spectrum
velocity

∞
u̇(t) = −nΩ1 ûn sin (nΩ1 t − βn )
1 2 ^2
2 n un n=0

∞
1
u̇¯q = (nΩ1 )2 û2n
2
n=0
Square of vibration
n 1 velocity amplitude.
1 2 1 3 1 4 1
power spectrum
acceleration

∞
1 4 ^2 ü(t) = − (nΩ1 )2 ûn cos (nΩ1 t − βn )
2 n un
n=0

∞
1
ü¯q = (nΩ1 )4 û2n
2
n=0
Square of vibration
n 1 acceleration amplitude.
1 2 1 3 1 4 1
Fig. 6.11 Amplitude spectrum for distance, velocity and acceleration

6.4 Supplementary Comments Regarding the Relationship … 139
6.4 Supplementary Comments Regarding the Relationship

Between Measured Spectra of Spatial Angular
Frequencies and Spectra of Angular Frequencies
as Function of Time
¯ have to be trans-
For calculation, the spectra found from measurement results S()
formed into spectra as functions of angular frequencies S(). With = 2π ¯
v = v
L
¯
and thus d = vd , we can write
1 1
¯
S() ¯ = S(
¯ = ) , (6.20)
v v
S()
i.e.,
1
¯
S() = S() . (6.21)
v
For the comfort calculation, the squares of amplitude spectra (= power spectrum)
of vibration velocity and acceleration are also needed. For periodic excitation (dis-
crete spectra), the expressions for the power spectra of vibration, vibration velocity,
and vibration acceleration shown in Fig. 6.11 apply.
For the continuous spectrum of stationary random vibrations with 12 û 2n =S()
ˆ ,
we can write
vibration distance power spectrum Su = Su () , (6.22)

vibration velocity power spectrum Sv = 2 Su () , (6.23)
vibration acceleration power spectrum Sa = 4 Su () . (6.24)
6.5 Interpretation of Response Spectrum
The power spectrum of the vehicle response Su () is found from the power spectrum
of the track excitation S() with the square of the vehicle transfer function:
Su () = V 2 ()S(). (6.25)
• It contains the contributions of the different frequencies to the quadratic mean

value ∞
ū q = Su () d. (6.26)
0
• The power spectrum of the vehicle response Su () is the basis for the ride comfort
evaluation.
• Based on the power spectra of the vehicle response or the corresponding quadratic
mean values, procedures to describe the load history and with this the fatigue life
calculation can also be derived [14]; cf. Chap. 15.
References
1. N. N. Testing and approval of railway vehicles from the point of view of their dynamic behaviour
– Safety – Track fatigue – Ride quality. UIC Code 518, 4th edition, UIC,from January 1997,
UIC, September 2009
2. CEN. Railway applications - Testing for the acceptance of running characteristics of railway
vehicles. EN 14363, CEN, Brussels (2005)
3. O. Buxbaum, Auflage (Variable amplitude fatigue analysis. Safe and economic design of devices
with risk of fatigue failure), 2nd edn., Betriebsfestigkeit. Sichere und wirtschaftliche Bemessung
schwingbruchgefährdeter Bauteile (Stahleisen mbH, Düsseldorf, 1992)
4. A. Khintchine, Korrelationstheorie der stationären stochastischen Prozesse (Correlation theory
of steady state stochastic processes. Math. Ann. 109, 604–615 (1934)
5. S.H. Crandall, W.D. Mark, Random Vibration in Mechanical Systems (Academic Press, New
York, 1963)
6. W. Giloi, Simulation und Analyse stochastischer Vorgänge (R. Oldenbourg, München-Wien,
1967)
7. R/S-VD. Arbeitsgemeinschaft Rheine-Freren, Rad/Schiene-Versuchs- und Demonstra-
tionsfahrzeug; Definitionsphase R/S-VD; Ergebnisbericht der Arbeitsgruppe Lauftechnik
(Wheel/Rail-Experimental and Demonstration-Vehicle. Definition-Phase R/S-VD. Result
report of working group running technique) (1980)
8. ERRI. B176, Bogies with steered or steering wheelsets. Rp1 Volume 1: Preliminary studies
and specifications. Volume 2: Specifications for a bogie with improved curving characteristics.
Volume 3. Specifications for a bogie with improved curving characteristics designed for carbody
tilt (1989)
9. J. Corbin, Statistical representation of track geometry. Tech. report, ENSCO 1112 under con-
tract DOT-TSC-1211. Final report May 1976 - April 1978 (1978)
10. X. Liu, H. Li, W. Wang, Analysis on the track irregularity power spectral density of the Beijing–
Shanghai high speed railway, in Proceedings of the Second International Conference on Rail-
way Technology. Research, Development and Maintenance, April 8-11. (Ajaccio, France 2014)
11. A. Haigermoser, B. Luber, J. Rauh, G. Gräfe, Road and track irregularities: measurement,
assessment and simulation. Veh. Syst. Dyn. 53(7), 878–957 (2015)
12. ORE. Track irregularities (spectral density). ORE-report to question C 116 (Interaction between
vehicle and track. Report Nr.1), UIC, ORE, Utrecht (1971)
13. N.N. Spezifikation für den Hochgeschwindigkeits-Triebzug ICE 2.2 (Specification for the high
speed trainset ICE 2.2) (1994)
14. S. Stichel, Betriebsfestigkeitsberechnung bei Schienenfahrzeugen anhand von Simulation-
srechnungen (Variable amplitude fatigue analysis of railway vehicles based on simulations).
VDI Fortschritt–Berichte (also PhD thesis TU Berlin), Reihe 12, Nr. 288. VDI, Düsseldorf
(1996)
Chapter 7
Human Perception of Vibrations - Ride
Comfort
To assess the ride comfort of a road or rail vehicle, first the vibration response has to
be measured or calculated. The assumption is that the comfort felt is determined by
acceleration. It is not only the acceleration amplitudes that are important, however,
but also the frequencies, since a human being can be regarded as a vibration system
with resonances at certain frequencies; cf. Fig. 7.1.
The calculated (“objective”) acceleration therefore has to be multiplied by a
frequency-dependent weighting function in the frequency domain. The result is
(“subjective”) accelerations relevant for the comfort assessment. This is equiva-
lent to filtering in the time domain. The weighting function then is transformed into
an equation of motion. The input (“right-hand side”) to the equation of motion is the
objective acceleration. The result is the subjective acceleration. For periodic objec-
tive accelerations the same subjective accelerations should arise as in the case of
multiplication in the frequency domain.
The weighting functions for the human perception of vibration are difficult to
determine, since they depend on a number of unquantifiable factors (e.g., from the
environment or the current state of well-being). Therefore, there exists a variety of
weighting functions, some used more frequently than others. Furthermore, there are
different opinions on whether the mean level of the vibration or occasional peak
values should determine the level of ride comfort. Newer ride comfort criteria tend
to give more attention to peak values than do older ones.
Some of the criteria we regard as important will be described below. The principal
approach tends to be similar in all cases. The original German criterion “Wertungsz-
iffer” is described in more detail. For further studies of the topic we recommend
Griffin’s textbook [2].
The problem of measuring vibrations in the vicinity of railway tracks and the
evaluation of such measurements is treated in a German DIN standard [3, 4].

DOI 10.1007/978-3-319-45376-7_7
142 7 Human Perception of Vibrations - Ride Comfort
(ca 25 Hz )
( 30 - 80 Hz)
(ca 60 Hz )
(16 - 30 Hz)
(10 - 12 Hz)
( 4 - 8 Hz )
(50 - 200 Hz)
( 2 Hz - 20 Hz )
Fig. 7.1 A human being as a system able to vibrate [1]. The frequencies shown give the impression
of resonance frequencies of different body parts. For example, it can be concluded that the frequency
of the stomach is between 4 and 8 Hz
7.1 Wertungsziffer According to Sperling
In a journal paper from 1941 [5], Sperling describes experiments carried out in the lab-
oratories of the German Railways (Reichsbahn Versuchsamt) in Berlin-Grunewald.
The main goal was to develop a consistent procedure to assess the ride comfort of
passenger and freight vehicles. Since a system of comfort values (in German, Wer-
tungsziffern) already existed, the new one should be in line with that system. The
result of the experiments was a generalized system of comfort numbers (abbreviated
Wz) that is still in use today to assess ride comfort.
In the experiments, the test subjects had to sit on a bench solidly connected to a
shaking table. The table was excited in either the vertical or lateral direction. The
excitations where nearly sinusoidal vibrations with frequencies between 1 and 12 Hz
and oscillation widths between 0.0001 and 0.025 m. To characterize the perception,
a scale of comfort values (in German, Empfindungswerte) E according to Table 7.1
was used.
7.1 Wertungsziffer According to Sperling 143
Table 7.1 Transformation of E Description in words

verbally described perception
into a scale of comfort values 1.0 Just noticeable
2.0 Clearly noticeable
3.0 Pronounced but not unpleasant, tolerable
3.25 Highly noticeable, still tolerable
3.5 Very strong and unpleasant, not tolerable after a
longer period
4.0 Extremely strong and unpleasant, harmful after a
longer period
20
E= E
E = 3,25 = 3, E =
E= 3 5 4
10 2,5
frequency f [Hz]
8 E
=2
6
4
1
0,01 0,1 1 10
vibration amplitude x^ [cm]
Fig. 7.2 Dependence of comfort values E on the vibration amplitude x̂ and the frequency f for
horizontal vibrations according to [5]
The assessment had to be given after an excitation period of 2–10 min. The result-
ing comfort values for horizontal vibrations are given in Fig. 7.2. In the chosen double
logarithmic scale, the points of equal perception lie almost on a straight line.
After some calculation, one can get as equation for the comfort values E(x̂, f ):

E = 3.1 10 x̂ 3 f 5 . (7.1)
Since the results for the vertical and lateral directions where very similar, this rela-
tionship was used for vibrations in both directions. To adapt Eq. (7.1) to the existing
system of comfort values Wz, the constant was slightly changed:

Wz = 2.7 10 x̂ 3 f 5 . (7.2)
If the acceleration amplitude is introduced instead of the vibration amplitude,
â = (2π f )2 x̂ ,
we get the expression

â 3
Wz = 0.896 10 (7.3)
f
for Wz as a function of the acceleration.

In a journal paper from 1956 [6], Sperling introduced a frequency weighting
function F( f ) to improve the reproduction of human perception:

â 3
Wz = 0.896 10 F( f ) . (7.4)
f
Usually, the frequency weighting function F( f ) is replaced by a weighting function

B( f ) that can be applied directly to acceleration amplitudes. We can write

Wz = 10
B( f )3 â 3 . (7.5)
For vertical vibrations, the expression

1.911 f 2 + (0.25 f 2 )2
Bvert ( f ) = 58.8 (7.6)
(1 − 0.277 f 2 )2 + (1.563 f − 0.0368 f 3 )2
is generally used, and

Blat ( f ) = 1.25 · Bvert ( f ) (7.7)
is used for lateral vibrations.

The comfort values Wz and E are very similar. cf. Eqs. (7.1) and (7.2). The verbal
meaning of the comfort values is given in Table 7.1. A comfort value Wz between 3
and 3.5 is regarded as just tolerable. In Fig. 7.3, lines of equal perception resulting
from Eq. (7.5) are given, expressed by Wz, for vibrations in the vertical and lateral
directions. It can be seen that vibrations with frequencies between 5 and 6 Hz are
regarded as most disturbing. Further, we can make the following simple calculation:
at 1 Hz, an increase of a vertical acceleration amplitude of 0.2 m/s2 by a factor of 4
to 0.8 m/s2 changes the comfort value from Wz = 2 to Wz = 3.
With a somewhat different weighting function, the comfort value is also used to
assess the general running behavior of rail vehicles, mostly freight vehicles. Vertical
and lateral vibrations here are multiplied by the same weighting function:

[(1 − 0.056 f 2 )2 + (0.645 f )2 ](3.55 f 2 )
BGv,l ( f ) = 114 .
[(1 − 0.252 f 2 )2 + (1.547 f − 0.0044 f 3 )2 ](1 + 3.55 f 2 )
(7.8)
Also, the verbal description is somewhat different, as can be seen in Table 7.2. As
limit value for the operation of freight wagons, WzG = 4.25 is usually given.
W z lateral W z vertical
5 5
4 4
3 3
acceleration amplitude [m/s 2 ]

acceleration amplitude [m/s 2 ]
2 Wz
Wz 2 =4
=4 Wz
Wz = 3,5
1 = 3,5 1
Wz
Wz =3
=3
0,5 0,5 Wz
0,4 Wz 0,4 = 2,
= 2, 5
0,3 5 0,3
0,2 0,2
Wz
Wz =2
=2
0,1 0,1
0,05 0,05
0.5 1 2 3 4 5 10 20 0.5 1 2 3 4 5 10 20
frequency [Hz] frequency [Hz]
Fig. 7.3 Curves of equal perception (equal Wz-value) depending on frequency and vibration ampli-
tude in the lateral direction (left) and the vertical direction (right)
Table 7.2 Verbal description W zG Description of running behaviour in words

of W zG-values for the
general assessment of the 1.0 Very good
running behavior of freight 2.0 Good
wagons 3.0 Tolerable
3.5 Just tolerable
4.0 Operable
4.5 Not operable
5.0 Dangerous
7.1.1 Periodic Vibrations
Equation (7.5) is of course valid only on a track with a cosine excitation. Is it possible
to extend it to general periodic vibrations? To do that, we rewrite Eq. (7.5),
Wz = [(B( f )â)2 ]0.15 , (7.9)
so that the effective value of the acceleration

1 2
aeff = â
2
can be introduced. Now we can write
Wz i = [2(B( f i )aieff )2 ]0.15 . (7.10)
According to Chap. 6, the effective value of the acceleration for general periodic
vibrations can be calculated as

1 2
aeff = (â + â22 + . . .)
2 1
or
aeff = (a1,eff
2
+ a2,eff
2
+ . . .) .
To assess ride comfort for a general periodic vibration, the effective values of the
accelerations are first weighted:
2
B( f 1 )a1,eff + ... .
Then, as in Eq. (7.10), the Wz-value is calculated:

Wz = [2 (B( f 1 )a1,eff )2 + (B( f 2 )a2,eff )2 + . . . ]0.15 . (7.11)
In Fig. 7.4, this is shown graphically.
Fig. 7.4 Weighting of 2 1 2

a i,eff = 2 a i B 2 ( f 1)
squares of accelerations
(power spectrum) by B( f )
(squared) to calculate the
Wz-value
f
f1 f2 f3 f4
2
a i,eff B 2 ( f i )
f
f1 f2 f3 f4
Fig. 7.5 Power spectrum of

Sa ( ) Sa ( i)
accelerations
7.1.2 Random Vibrations
For statistically stationary excitation, the power spectrum of the accelerations

becomes continuous, since it contains all frequencies; cf. Fig. 7.5. The acceleration
spectrum of the vehicle response Sa i for one frequency i and one frequency band
d can be recalculated to a line in the discrete spectrum of squares of accelerations
2
ai,eff = Sa (i )d , (7.12)
where Sa is the acceleration spectrum of the vehicle response. Since the weighting
function is given as B( f ), the power spectrum has to be transformed. Because of
= 2π f,
we can write
d = 2πd f,
and then
2
ai,eff = Sa (2π f i )2π d f i = Sa ( f i )d f i . (7.13)

Sa ( f i )
The effective value of the acceleration is the integral over the acceleration spectrum
of the vehicle response (cf. Chap. 6):
∞
2
aeff = Sa d f .
0
The frequency-weighted effective value can be calculated as

∞
B( f )2 Sa d f .
0
discrete excitation continuous excitation
1 ^2 1 ^2 ^
z S( ) z = S( )
2 n 2
track power spectrum
of displacements
n 1
1
2 2
V ( ) V ( )
vehicle square of
transfer function
-
^2
1u Su
vehicle response 2 zn ^ 2 = V 2 ( ) 1 z^ 2
1u 2
2 zn 2 n Su( ) = V ( ) S ( )
power spectrum of
displacements
n 1
power spectrum of 1 4 u^ 2 Sa 4
2 zn Sa ( ) = Su ( )
accelerations ^2
a
(basis for comfort calculation) zn
n 1
comfort calculation B (f) B (f)

weighting curves
=2 f
=2 f
single value ( Wz i ) Wz n 0,15
2
2 2 0,15 Wz (f) = B (f) Sa(f)
Wz n = 2 (B (fn ) a zn f(
=2 f
n 1= n2 f
20 3 20 3
global Wz-value Wz = Wz n 3 20 Wz = Wz (f) 3 df 20
n
0
Fig. 7.6 Procedure for calculation of W z-value for stochastic and statistically stationary excitation
The comfort value is finally written as

∞ 0.15
Wz = B( f )2 Sa d f . (7.14)
0
The procedure to determine the Wz-value for periodic and random vibrations
is illustrated in Fig. 7.6. For the transition from the calculation of the acceleration
response to the ride comfort calculation, we have to use = 2π f .
7.2 ISO 2631
ISO 2631 [7] is an international standard providing methods to assess vibrations with
respect to comfort, health, and motion sickness. The standard was not developed
exclusively for rail vehicles, but is a general standard for vehicles and machines.
7.2 ISO 2631 149
The standard does not provide any limit values. Instead, an evaluation method is
provided that can be a basis for the definition of limit values. For rail vehicles, for
example, accelerations evaluated according to ISO 2631 are used in UIC 513 [8] and
the European standard EN 12299 [9] (cf. Sect. 7.3).
For the development of the weighting curves, test persons sit or lie on shaking
tables that are excited with different amplitudes and frequencies. The frequency-
weighted effective value of the accelerations is, according to ISO 2631, calculated
to be

1 T 2
aeff,ISO = a (t) dt ; (7.15)
T 0 ISO
aISO is the ISO weighted instantaneous acceleration. Three basic weighting filters
exist:
wk : vertical between 0.5 and 80 Hz;

wd : lateral and longitudinal between 0.5 and 80 Hz;
wf : vertical for motion sickness between 0.1 and 0.5 Hz.
The weighting curves are shown in Fig. 7.7. The accelerations have to be deter-
mined for the assessment of ride comfort for sitting passengers. For vertical accel-
erations, in a supplement to ISO 2631 from 2001, the weighting curve wb is given.
This curve is said to be more representative for rail vehicles than wk and is usually
chosen today. In Fig. 7.8, the two weighting curves for vertical ride comfort are com-
pared. Comparing these two weighting curves with the curves of equal perception
10
0
-10
wf wk
-20
weighting curve
wd
-30
-40
-50
-60
-70
-80
-90
0,016 0,0315 0,063 0,125 0,25 0,5 1 2 4 8 16 31,5 63 125 250
frequency f [Hz]
Fig. 7.7 Weighting curves according to ISO 2631 [10]

0
wk
weighting curve
-5
-10
wba
-15
-20
-25
-30
0,1 1 10 100
frequency f [Hz]
Fig. 7.8 Comparison of two weighting curves according to ISO 2631. Today, for rail vehicles,
usually wb is used
(Wz-values), Fig. 7.3, one can conclude that in Fig. 7.7 the highest weight is between
4 and 8 Hz. This is equivalent to the minima in Fig. 7.3.
The equations for vertical and horizontal comfort filters are given below; the
equations for the filters wk and wf can be found in the standard.
The vertical transfer function Hb (i) can be calculated as
Hb (i) = Ha (i) · Hb0 (i) ,
with
(i)2 4π 2 1002
Ha (i) = ,
[(i)2 + 2π0.4
0.71
i + 4π 2 0, 42 ][(i)2 + 2π100
0.71
i + 4π 2 1002 ]
and
(i + 2π16)((i)2 + 2π2.5

0,8
· i + 4π 2 2.52 )
Hb0 (i) = ×
[(i)2 + 2π16
0.63
i + 4π 2 162 ][(i)2 + 2π4
0.8
i + 4π 2 42 ]
2π · 0.4 · 16 · 42
2
× .
16 · 2.52
The horizontal transfer function Hd (i) can be written as
Hd (i) = Ha (i) · Hd0 (i) ,

7.2 ISO 2631 151
with Ha (i) as above and
i + 2π2 2π. 22
Hd0 (i) = · .
(i)2 + 0.63
2π2
i + 4π 2 22 2
7.3 CEN Standard EN 12299
An assessment as described in the sections above involving simply calculating the

mean value over a longer piece of track was regarded as insufficient in the long run.
Not only the mean vibration levels are important for the perceived ride comfort.
Rare but significant single disturbances can influence the perception of the vibration
level significantly. Therefore, the ERRI committee B153 [11] was given the task of
developing more appropriate comfort criteria. Their work forms the basis for the
limit values given in the European standard EN 12299 [9]. EN 12299, in turn, is
a successor of UIC 513 [8]. Some of the comfort criteria given in EN 12299 are
explained below.
7.3.1 Simplified Criterion for Mean Comfort: NMV
The only mandatory comfort criterion in the CEN standard is NMV . Moreover, this
criterion is aimed at the assessment of mean comfort, even though, as can be seen
below, singular disturbances are judged more harshly than in older comfort criteria.
For the use of this evaluation method, the following assumptions are made:
• The uncompensated track plane acceleration in a curve does not exceed 1 m/s2 .
• There are no stronger accelerations in the longitudinal direction.
A consequence of these assumptions is that NMV cannot be used for trains with tilt
mechanisms or for commuter trains with frequent acceleration and deceleration.
For commuter trains, two other optional criteria are defined: NVA for seated pas-
sengers and NVD for standing passengers.
The basis for NMV is the assessment of the accelerations with the ISO-filters wb
for vertical and wd for longitudinal and lateral vibrations introduced in Sect. 7.2.
The accelerations are measured on the carbody floor, both above the bogies and
at the center of the vehicle. As can be seen in Eq. (7.16), effective values in all
three coordinate directions of at least 60 consecutive five-second intervals should be
calculated:
1 5(i) 2
aeff,ISO,i = a (t) dt i = 1, 60 . (7.16)
5 5(i−1) ISO,i
Fig. 7.9 Example for the number of sections: 60

determination of the 12 minimum value: 0.005 m/s
2
2
95 %-value of the effective maximum value: 0.089 m/s
frequency of events [ % [
values of the accelerations 10
distribution function:
50%: 0.029 m/s 2
8 95%: 0.064 m/s 2
standard deviation: 0.016 m/s2
6 width of class: 0.002 m/s 2
0
0.00 0.02 0.04 0.06 0.08 0.10
acceleration [m/s 2 [
Table 7.3 Verbal description NMV Description in words

of NMV -values
NMV < 1 Very comfortable
1 ≤ NMV < 2 Comfortable
2 ≤ NMV < 4 Medium
4 ≤ NMV < 5 Uncomfortable
NMV ≥ 5 Very uncomfortable
From these 60 effective values, the 95 %-value1 should be determined (cf. the example
in Fig. 7.9). These 95 %-values are used to determine the ride comfort value

NMV = 6 · (ax95,wd )2 + (ay95,wd )2 + (az95,wb )2 . (7.17)
A new feature of this criterion is also that vibrations in all coordinate directions are
used for a common comfort criterion. Also, for NMV , there exists a verbal description
of the comfort values; cf. Table 7.3.
A typical limit value in customer specifications for high-speed traffic is a 95 %-
value of 0.2 m/s2 for lateral and vertical accelerations. Accelerations in the longitu-
dinal direction are often neglected. Transformed into NMV , this means that

NMV = 6 · (0.2)2 + (0.2)2 ≈ 1.7 .
1 The 95 %-value means that 95 % of the obtained values are below this value.
7.3 CEN Standard EN 12299 153
7.3.2 Comfort on Curve Transitions: PCT
Curve transitions can make a significant contribution to the discomfort felt by pas-
sengers in a rail vehicle. Therefore, in the appendix of EN 12299, another possible
comfort criterion is given, called PCT (“Percentage Disturbed from Curve Transi-
tions”). It is applicable for vehicles with and without carbody tilt. It is to be evaluated
on the carbody floor in the center of the carbody and above the leading bogie. The
criterion is written as
...
PCT = (A ü y,max + B u y,max − C ) + D ϕ̇x,max
E
[%] , (7.18)
where:
ü y,max is the maximum value of the lateral acceleration on the carbody floor in the
time interval from the beginning of the transition curve until 1.6 s after exiting
the transition curve,
...
u y,max is the maximum jerk in the time interval for 1 s before entering the transition
curve until the end of the transition curve,
ϕ̇x,max is the maximum value of the absolute roll velocity in the time interval from the
beginning to the end of the transition curve measured in degrees per second.
The term in parentheses is taken into account only when it is positive.

The measures to be evaluated are first low-pass filtered with a cutoff frequency
of 2 Hz:
i + 2π · 2 2π · 22
H(i) = · .
(i)2 + 2π·2
0.63
i + 4π 2 · 22 2
The determination of the necessary quantities is illustrated in Figs. 7.10 and 7.11.
The constants to be applied are given in Table 7.4.
7.3.3 Comfort on Discrete Events: PDE
Another ride comfort criterion suggested in the CEN standard is called PDE (“Per-
centage disturbed from discrete events”). It can be used if it is expected that discrete
events such as passing a switch have a significant influence on the perceived ride
comfort. The NMV criterion described above does not account for this type of distur-
bance in a satisfying way because of the calculation of mean values. The calculation
of PDE is as follows:
PDE = a ü y,p + b ü y,mean − c [%] . (7.19)

uy
1s
lateral acceleration
u y,max
u y,max
u y1
u y1
time t
1s transition curve 1,6 s
calculation of u y,max
calculation of u y,max
...
Fig. 7.10 Determination of the quantities ü y,max and u y,max for the calculation of PCT according
to EN 12299 [9]
time t
transition curve
calculation of x
start
end
Fig. 7.11 Determination of ϕ̇x,max for the calculation of PCT according to EN 12299 [9]
The quantities used are as follows:
ü y,p the difference between the maximum value ÿmax and the minimum value ÿmin
during a sliding time interval of two seconds, low-pass filtered with the ISO-
filter wd given in Sect. 7.2.
ü y,m also low-pass-filtered mean value in the same time interval.
7.3 CEN Standard EN 12299 155
Table 7.4 Constants A to E for the calculation of PCT

A B C D E
Standing 28.54 20.69 11.1 0.185 2.283
Seating 8.97 9.68 5.9 0.120 1.626
uy
lateral acceleration in carbody
u y,p
u y,max
u y,m
u y, min
time t
2s
Fig. 7.12 Determination of necessary quantities for the calculation of PDE , according to EN
12299 [9]
Table 7.5 Constants a to c for the calculation of PDE

a b c
Standing 16.62 27.01 37.0
Sitting 8.46 13.05 21.7
The determination of ü y,p and ü y,m is shown in Fig. 7.12. The constants to be used
in Eq. (7.19) are given in Table 7.5.
7.4 Final Remarks
7.4.1 Measure or Simulate
The standards explicitly or implicitly assume that the accelerations are measured.
During the design process of a vehicle, ride comfort values are simulated. To get
realistic values, the track layout and the track irregularities have to be known. Alter-
natively, the simulations can be performed on an agreed standard track. Also, some of
the flexible eigenmodes of the carbody should be included in the simulation model.
For the final approval of a vehicle today, however, customers demand acceleration
measurements.
7.4.2 Ride Comfort as System Property
A ride comfort value, no matter whether it is measured or simulated, is always

a system property. The value depends on the vehicle properties and on the track
properties, track irregularities, and operational conditions. The specification of the
running behavior of a vehicle is meaningful only together with the definition of
vehicle speed and track quality, especially the amount of track irregularities. This
can cause discussions in a contract between the operator and the manufacturer of a
vehicle.
7.4.3 Duration of Exposure of Vibration
It is obvious that the duration of exposure influences the perceived passenger comfort.
In spite of this, it is not mentioned in this chapter.
In one section of ISO 2631, the duration of exposure is addressed. The standard
mentions exposure time. In practice, however, the issue of exposure time is treated
in a way that for long-distance trains, where passengers spend more time, tougher
ride comfort criteria are set than for regional or commuter trains, where passengers
spend less time.
7.5.1 Calculation of Wz-Values According to Sperling
Look at the 1-DOF system in Fig. 5.1 and determine the transfer function for carbody
bounce with the input values m = 40.000 kg, c = 2 · 106 N/m, and d = 1 · 105 Ns/m.
Calculate the Wz-value for the vertical oscillations of the vehicle with the spectral
density given in Eq. (6.17) as excitation.
Extend the model with a bogie frame with mass m = 5000 kg and primary sus-
pension cp = 1 · 107 N/m and dp = 5 · 104 Ns/m. It is difficult to derive an analytical
solution now. Recalculate the W z-value and compare with the solution above. Try
to improve the ride comfort by “playing” with the vehicle parameters.
References 157
References
1. H. Bodén, U. Carlsson, R. Glav, H.P. Wallin, M. Åbom, Ljud- och vibrationslära. Lecture Notes
(in Swedish). The figure is taken from information materials of the company Brüel & Kjaer
GmbH), Markus Wallenberg Laboratory for Noise and Vibration Research (1999)
2. M.J. Griffin, Handbook of Human Vibration (Academic Press, London, 1990)
3. DIN. Schwingungsmessungen in der Umgebung von Schienenverkehrswegen – Meßverfahren
(Vibration measurement near railways. Measurement methods). DIN 45672 - Teil 1, German
Institute for Standardization, Beuth Verlag, Berlin (1991)
4. DIN. Schwingungsmessungen in der Umgebung von Schienenverkehrswegen – Auswertev-
erfahren (Vibration measurement near railways. Evaluation methods). DIN 45672 - Teil 2
(Entwurf), German Institute for Standardization, Beuth Verlag, Berlin (1993)
5. W. Helbig, E. Sperling, Verfahren zur Beurteilung der Laufeigenschaften von Eisenbahnwa-
gen (Process for the evaluation of the running behavior of railway vehicles). Organ für die
Fortschritte des Eisenbahnwesens 96(12), 177–187 (1941)
6. E. Sperling, Beitrag zur Beurteilung des Fahrkomforts in Schienenfahrzeugen (Contribution
to the evaluation of ride comfort in rail vehicles). Glasers Annalen 80, 314–320 (1956)
- Part 1: General requirements. ISO 2631-1, International Organization for Standardization,
Genève (1997)
8. N.N. Guidelines for evaluating passenger comfort in relation to vibration in railway vehicles.
UIC code 513, UIC, Paris (1994)
9. CEN. Railway Applications - Ride comfort for passengers - Measurement and evaluation, EN
12299 (prepared by TC256 WG7) (2009)
- Part 4: Guidelines for the evaluation of the effects of vibration and rotational motion on
passenger and crew comfort in fixed guideway transport systems. ISO 2631-4. International
Organization for Standardization, Genève (2001)
11. ERRI. B153/RP 10, Vibratory comfort: Drawing up weighting curves (1994)
Chapter 8
Introduction to Lateral Dynamics of Railway
Vehicles
8.1 Preliminary Remark
The vibration behavior of rail vehicles in the lateral direction is to a great extent
determined by the behavior of the wheelset and thus by the processes in the contact
between wheel and rail. The motions of a free wheelset, i.e., a wheelset that is not
connected to a bogie frame or carbody by springs and dampers, are shown in Fig. 8.1.
If a wheelset with conical profiles running with speed v 0 is put on the track with a
small angle around the vertical axis (ϕz = ϕz0 ), a sinusoidal motion begins. In
the starting position, both wheel radii are equal. When starting the rolling motion,
the wheelset moves slightly into the positive u y -direction. The left rolling radius
increases, whereas the right one decreases. The left wheel is now moving somewhat
faster than the right one, and “catches up.” This continues until the wheelset is
perpendicular to the track (ϕz = 0). The radius of the right wheel now has reached
its maximum value, whereas the radius of the left wheel has obtained its minimum
value. The lateral displacement of the center of gravity of the wheelset to the left has
reached its maximum value u y0 . When continuing the rolling motion, now the left
wheel passes the right one, until again the central position of the center of gravity
of the wheelset is reached (u y = 0). The angle around the vertical axis again has
reached a maximum value, now, however, in the negative direction (ϕz = −ϕz0 ). For
small lateral displacements, the motion of the center of gravity can be approximated
with a sine function. Therefore, we speak of sinusoidal motion (or hunting motion)
of the wheelset.
This motion is a kinematic process. The wavelength L of the sinusoidal motion
(cf. Fig. 8.1) can be determined by the geometric properties of the wheelset, namely
• the rolling radius in the centered position r0 ;
• the lateral distance of the wheel contact points 2e0 ;
• the cone angle δ0 (the conicity).
The equation will be derived in Sect. 8.2. The relationship between speed v 0 , wave-
length L, of the sinusoidal motion, and the period of the oscillation is

DOI 10.1007/978-3-319-45376-7_8
160 8 Introduction to Lateral Dynamics of Railway Vehicles
(a) right z left

wheel wheel
e0 0
r0 y,uy
(b)
z0
du y
dx 0
u y0
L /2
v0
x
Fig. 8.1 Sinusoidal motion of a wheelset
L
v0 = , (8.1)
T
which results in the equation for the frequency f or the angular frequency ω of the
sinusoidal motion
8.1 Preliminary Remark 161
v0 2πv 0
f = oder ω= . (8.2)
L L
This sinusoidal (or hunting) motion was investigated in 1883 by Johannes Klingel, of
Karlsruhe [1]. The motivation for these investigations was hunting motions of railway
vehicles that significantly influenced the vibration comfort. Until the beginning of the
1970s, in Germany and other countries it was assumed that wheelset or bogie hunting
acts as a base-point excitation of the carbody with frequency ω. Since the wavelength
L is constant for a conical profile on circular profiles for the rails (cf. Fig. 8.3) or
for a knife-edge bearing (Fig. 8.4), according to Eq. (8.2), the frequency depends
on the running speed. The task of the designer is then to choose the primary and
secondary suspension in a suitable way. The eigenfrequencies of the carbody modes
should either be sufficiently far away from the hunting frequency in the typical speed
range of the vehicle or adequately damped. Practical suggestions on how this can be
achieved were made by Klingel [1].
With the above interpretation, not all observed phenomena can be explained. For
example, it cannot be explained that the amplitude of the lateral displacement of
the wheelset depends on the speed: at low speeds, the lateral motion is small and
caused by track irregularities. Even after large disturbances (e.g., curve entrance), the
oscillation amplitude decreases again. Above a critical speed, however, the oscillation
increases to relatively large amplitudes. The wheel flange hits the rail head.
In Fig. 8.2, the lateral motions of the wheelset center of gravity due to an initial
disturbance are shown. The track is assumed to be straight and without any irregular-
ities. Linear motions are postulated. At low speeds, the oscillation of the wheelset is
damped (Fig. 8.2a). At higher speeds, the lateral motions due to an initial disturbance
can increase (Fig. 8.2c). The lateral movements are limited only by the wheel flanges
hitting the rail heads. Between these two states, there is a critical speed at which the
wheelset performs an undamped periodic oscillation (Fig. 8.2b). Below this speed,
the oscillations are stable, and above the critical speed, they are unstable.
In 1887, Boedecker claimed [2] that the oscillation of a wheelset can become
unstable, and he tried to analyze this phenomenon (Fig. 1.3). He succeeded only
partly, because he could not describe the contact mechanics accurately.1 In 1916,
Carter presented investigations on the stability of locomotives for the Institution of
Civil Engineers in London that remain correct with today’s knowledge [3].2 Carter
used a linear relationship between creepage and creep force. It was in the 1950s, with
increasing running speeds, that the necessity of stability investigations was realized,
first in Japan [4], somewhat later in France and England [5], and also in Germany [6].
1 He described the unstable motion correctly qualitatively. Today, we know that it is mainly contact
mechanics that are responsible for the existence of instability.
2 During the discussion, Carter was mocked: “The mathematical demonstration which the author
had given might be perfectly correct, but not many years ago, before the introduction of the motor-
bicycle, it was demonstrated mathematically by an eminent French mathematician that a motor-
bicycle would be an absurdity; and there were many other instances in engineering history to show
that mathematics might occasionally go astray, and that practice was worth a great deal in matters
of that kind.”
(a)
Damped oscillations
below critical speed
(b)
Perodic, undamped oscillations
at the limit for critical speed
(c)
Increasing oscillation amplitudes
above critical speed. The amplitudes
are limited by the wheel flanges
hitting the rail heads.
Fig. 8.2 Lateral movement of a wheelset on straight track for different speeds
An investigation of this type of eigenbehavior, especially the detection of the stability

limit, will be done in Chap. 10.
8.2 Hunting and Klingel Equation
Sinusoidal or hunting motions as shown in Fig. 8.1 were investigated by Klingel [1];
cf. Fig. 1.2. In the following, we will not show the derivation that Klingel gave, but
a different one that has the advantage of being applicable to arbitrary profiles. We
are looking for the equation of motion of the trajectory of the kinematic motion
in Fig. 8.1. We assume pure rolling, i.e., that no sliding in the wheel–rail contact
occurs. Further inertial and gravitational forces are neglected, and the wheelset is
unsuspended. This means that no spring or damper forces are present. Figure 8.3
illustrates the rolling motion. In Fig. 8.3a, the wheelset is in a centered position
with the following notation: r0 is the rolling radius, 2e0 is the lateral distance of
contact points, and δ0 is the contact angle. In Fig. 8.3b, the wheelset is shown at
its maximum lateral displacement. The two rolling radii rL and rR are respectively
smaller and larger than r0 , while the center of gravity of the wheelset travels at speed
v 0 = 0 r0 . (8.3)
8.2 Hunting and Klingel Equation 163
(a) z
0
r0 r0 y,uy
0 0
2e 0
(b)
0
rR rL
R L
uy 0
(c) eR eL
rR r0 rL
vR v0 vL y,u y
M
Fig. 8.3 Geometric relations for the derivation of the Klingel equation
For pure rolling, the velocities of the two points marked on the wheelset axle
(Fig. 8.3c) are
v L = 0 rL , (8.4a)
v R = 0 rR . (8.4b)
In an infinitesimal neighborhood, the wheelset center of gravity describes a circle

of radius ρ. According to the theorem of intersecting lines, ρ can be determined from
Fig. 8.3c:
r0 rL (u y ) − rR (u y )
= . (8.5)
ρ(u y ) eL + eR
The same relationship holds for an arbitrary position of the wheelset, also in the case
of a small angle of attack of the wheelset. For a central position rL = rR = r0 , the
angle of attack is at its maximum, and the radius of curvature approaches infinity,
ρ → ∞. Equation (8.5) is already the equation of motion of the trajectory. This
becomes clearer when the radius of curvature is replaced with
1 d2 u y
− , (8.6)
ρ(u y ) dx 2
which is possible for small lateral displacements u y .

The equation of the kinematic motion for general profiles is now known:
d2 u y r (u y )
+ = 0. (8.7)
dx 2 r0 (eL + eR )
What is missing is a relationship for the difference in rolling radius r = rL − rR

and for eL +eR . Instead of using the expressions from Sect. 3.2.1, we shall derive them
from the very simple example of a double cone on a knife-edge bearing (Fig. 8.4).
From Fig. 8.4, we derive
rL − r0 +u y tan δ0 , (8.8a)
rR − r0 −u y tan δ0 , (8.8b)
which means for the rolling radius difference rL − rR = r ,
zR
z0 ~uy
rL r0 0
r0 yR
rR
y0
uy
e0 e0
Fig. 8.4 Kinematic relationships for rL , rR for a knife-edge bearing

8.2 Hunting and Klingel Equation 165
rL − rR 2u y tan δ0 2u y δ0 . (8.9)
From Fig. 8.4, we further conclude that eL + eR remains constant,
eL + eR 2e0 . (8.10)
As the linearized equation for the kinematic wheelset motion, it can be written as
d2 u y δ0 u y
2
+ = 0. (8.11)
dx r0 e0
For the initial conditions in Fig. 8.1, we get for the solution
2πx
u y = u y0 sin , (8.12)
L
with the wavelength L of the hunting motion

e0 r0
L = 2π . (8.13)
δ0
Equation (8.12) justifies our speaking of sinusoidal motion. With Eq. (8.2), we can
write
v0 δ0 δ0
f = and ω = v 0 (8.14)
2π e0 r0 e0 r0
for the frequency and angular frequency of the sinusoidal motion. Expression (8.14)
is known as the Klingel equation. The Klingel frequency is proportional to the vehicle
speed. The square of the Klingel frequency is proportional to the contact angle δ0
and inversely proportional to the rolling radius r0 and to half the lateral distance of
the contact points e0 .
Equation (8.13) indicates how the wavelength and the frequency of the hunting
motion can be influenced. The lateral distance between the wheelsets 2e0 is fixed. The
wavelength can be increased, or the frequency decreased, by increasing the rolling
radius r0 or decreasing the contact angle δ0 . For cylindrical profiles, the wavelength
becomes infinite; there is no sinusoidal motion. For real profiles like the S1002
wheel profile, δ0 has to be replaced by the equivalent conicity λe ; the wavelength
can decrease significantly in this case.
8.3 Assumptions for Derivation of the Klingel Equation
For the derivation of the Klingel equation, it is assumed that the wheelset hunting can
be described as kinematic motion in which both wheels perform pure rolling. This
means that only normal forces and no tangential forces are transmitted between wheel
u y (t)
x,v
L/2
Fig. 8.5 Kinematic hunting
and rail. Further, we assumed that inertial and gravitational forces are neglected and
that the wheelset is unsuspended, i.e., we do not have spring or damper forces.
Additional assumptions are listed below:
1. The lateral displacements u y are so small that the equations can be linearized.
2. The rotation about the vertical axis ϕz remains so small that the problem can be
treated in the (y, z)-plane.
3. The angle δ0 has to be small, too.
4. Finally, conical profiles on a knife-edge bearing were assumed. Both assumptions
do not result in a loss of generality, as will be shown below.
Summary: The kinematic hunting motion (sinusoidal motion) of a wheelset

(Fig. 8.5) is governed by two equations for the lateral displacement u y (t) and
the rotation about the vertical axis ϕz (t):
u y = u y0 sin ωt
ω
ϕz = u y0 cos ωt .
v0
The wavelength and the angular frequency of the hunting motion (Klingel fre-
quency) are determined by

e0 r0 λe
L = 2π , (8.13) and ω = v0 . (8.14)
λe e0 r0
The quantity λe is the equivalent conicity. For conical profiles, we have
λe = δ0 (Fig. 8.6, left), while for circular profiles, an analytic equation can be
given for λe [7] (Fig. 8.6, right).
8.3 Assumptions for Derivation of the Klingel Equation 167
e0 e0
r0
r0
RW
e~ 0 RR e~ 0
R W- R R
RW
Fig. 8.6 Conicity and equivalent conicity for circular profiles
In Sect. 3.2.4, it was shown how arbitrary profiles can be transformed into equiv-
alent circular profiles. The relationship between wavelength and equivalent conicity
is valid also for this case.
8.4 Determination of Equivalent Conicity with Eq. (8.13)
Equation (8.13) can be used in a different way if we assume that the wavelength of the
hunting motion is known from measurements or as a result of a time step integration
with realistic profiles. In this case, we can resolve Eq. (8.13) for λe :
2
2π
λe = e0 r0 . (8.15)
L
The wavelength L of a more or less sinusoidal motion can be determined, for example,
by integrating the nonlinear equation of motion (8.7). With eL + eR = 2e0 , the
equation can be written
d2 u y r (u y )
2
+ = 0. (8.16)
dx 2 r0 e0
The rolling radius difference depends nonlinearly on u y (Fig. 3.9). The method
described is implemented today in UIC code 519 [9] to estimate the equivalent
conicity for an arbitrary wheel–rail profile combination.3
3 It
is not known who was the first to give the equation of motion (8.16) for kinematic hunting of
arbitrary profiles. It can be found, for example, in [9].
8.4.1 Final Remark to This Chapter
With the Klingel equation, the wavelength and the frequency of the hunting motion
can be determined. Deciding whether the hunting is stable (decreasing amplitudes)
or unstable (increasing amplitudes) can be done only by deriving and solving the
complete equations of motion including inertial forces, spring and damper forces,
and creep forces. This is done in Chaps. 9 and 10 for a single wheelset, in Chap. 11 for
a bogie, and in Chap. 12 for an entire vehicle. Chapter 13 gives a short introduction
to nonlinear phenomena.
References
1. J. Klingel, Über den Lauf von Eisenbahnwagen auf gerader Bahn (On the running of railway
vehicles on straight track). Organ für die Fortschritte des Eisenbahnwesens, Neue Folge 20:113–
123, Tafel XXI (1883)
2. Chr. Boedecker. Die Wirkungen zwischen Rad und Schiene und ihre Einflüsse auf den Lauf und
and rail and their influences on the running behaviour and the resistance of vehicles in railway
trains). Hahn’sche Buchhandlung, Hannover (1887)
3. F.W. Carter, The electric locomotive. Proc. Inst. Civ. Eng. 201, 221–252 (1916). Discussion pp.
253–289
4. T. Matsudaira. Shimmy of axles with pair of wheels (in Japanese). J. Railw. Eng. Res. pp. 16–26
(1952)
6. CTh Müller, Das Schlingerproblem in der Sicht von Vergangenheit und Gegenwart (The hunting
problem in the past and presence). Glas. Ann. - ZEV 93(11), 329–336 (1969)
7. K. Knothe, Die geometrisch nichtlinearen Beziehungen für einen starren Radsatz, der auf einer
starren Schiene quer verschoben wird (The geometric nonlinear relations for a rigid wheelset
that is laterally displaced on a rigid track). ILR–Mitt. 17, Institut für Luft- und Raumfahrt, TU
Berlin (1975)
8. N.N. Guidelines for evaluating passenger comfort in relation to vibration in railway vehicles.
UIC code 513, UIC, Paris (1994)
9. CTh Müller, Radreifenverschleiß und Fahrzeuglauf (Wear of wheel rims and vehicle running).
Österr. Ing.-Z. 109, 215–224 (1964)
Chapter 9
Derivation of Equations of Motion
for Lateral Dynamics
9.1 Principle of Virtual Displacements for a Suspended

Wheelset with All Degrees of Freedom
In deriving the equations of motion for vertical dynamics, we have seen that either
the principle of linear and angular momentum (Sect. 4.2) or the principle of vir-
tual displacements, or in other words, the principle of d’Alembert in the version of
Lagrange (Sect. 4.2) can be used. For general problems with constraints and pre-
loads,1 it is often more convenient to use the principle of virtual displacements, even
though the choice is also a matter of personal preference.
9.1.1 Evaluated System and Acting Forces
In Fig. 9.1, a view from behind (Fig. 9.1a, b) and a top view of the wheelset (Fig. 9.1c)
are shown. All forces acting on the wheelset are indicated.
These are as follows:
• suspension forces (spring and damper forces), i.e., on the left side FxL , FyL , and
FzL and the same on the right side (index R);
• the creep forces TξL , TξR , TL , and TR as well as the spin moments MζL and MζR ;
• the normal forces NR and NL at the wheel contact point;
• the vertical load 2Q acting at the center of gravity of the wheelset; the weight of
the wheelset and the weight resting on the wheelset are added;
• d’Alembert’s inertial forces m ü x , m ü y , and m ü z and the moments of inertia2 z ϕ̈x ,
y ϕ̈y , and z ϕ̈z .
1 Preloads are loads that exist at the beginning of the oscillation process, typically the weight.
2 The moments of inertia
z and x are equal by symmetry. Therefore, x = z is always assumed.

DOI 10.1007/978-3-319-45376-7_9
170 9 Derivation of Equations of Motion for Lateral Dynamics
2e z dz cz
(a) 2 2
2Q dy
2
cy
2
2e0
z
(b)
FzR uz FzL
Qy
.. z
FyL
..
y FyR mu y
R uy L
y uy y
T R
.. L
mu z
NR NL
R T L
Qz z
M R M L
(c) Qz x
..
T mux T
R L
Qy
..
y FyL
..
y FyR mu y
y uy y
ux
FxR FxL
2e x
R L
x
..
x
Fig. 9.1 System of wheelset with notation (a) as well as view from behind (b) and top view (c)
(Springs in the x-direction are in the direction of traveling of the wheelset)
Regarding sign and direction of forces, the following apply:

• Spring forces are positive if they act as tension forces in the springs.
• For the creep forces, it was already decided that they are positive when they act in
the direction of the ξ- and η-axes on the rail.
• The normal forces at the wheel contact point are positive as compressive forces.
• D’Alembert’s inertial forces are positive in the direction of negative displacements.
9.1 Principle of Virtual Displacements for a Suspended … 171
9.1.2 Formulation of Principle of Virtual Displacements
The principle of virtual displacements is the following:
For a virtual, i.e., for a geometrically possible, displacement state, the virtual
deformation energy is equal to the virtual work of the external forces (mass
forces, loads, nonconservative reaction forces, i.e., damping forces and creep
forces), and mass forces:
δVint = δWext + δWm . (9.1)
It is convenient to use the principle of virtual displacements if the system contains

geometric constraints. In the case of the wheelset, the two contact conditions are such
constraints. It is demanded from the virtual displacement that it be geometrically
possible, i.e.,that the geometric constraints not be violated. As a result, the forces
in the direction of the geometric constraints (in the case of the wheelset, these are
the normal forces at the wheel–rail contact points) no longer exist in the principle
of virtual displacements, since there are no displacements in the direction of the
geometric constraints. In this way, the elaborate elimination of the constraint forces
is unnecessary. Still, the kinematic relations have to be formulated. This is necessary
for all methods in any case.
The principle of virtual displacements for the wheelset in Fig. 9.1 can now be
given:
+ FcxL δv xL + FcyL δv yL + FczL δv zL

+ FcxR δv xR + FcyR δv yR + FczR δv zR
= + (−m ü x )δu x + (−m ü y )δu y + (−m ü z )δu z
+ (−z ϕ̈x )δϕx + (−y ϕ̈y )δϕy + (−z ϕ̈z )δϕz
+ (−TξL )δu ξL + (−TηL )δu ηL + (−MζL )δϕζL + NL δu ζL
+ (−TξR )δu ξR + (−TηR )δu ηR + (−MζR )δϕζR + NR δu ζR
+ (−2Q)δu z , (9.2)
where firstly, the normal forces multiplied by a virtual displacement δu ζ are terms
in the equation. To the left of the equal sign we find the virtual deformation energy,
i.e., the product of spring forces and virtual deformations of the spring,3 while on
the right-hand side are all other forces acting on the wheelset that are carrying out
virtual work on the respective virtual displacements. Since the virtual displacement
state has to satisfy the geometric constraints,
3 For example, δv is the virtual spring deformation of the spring in the x-direction on the left-hand
xL
side.
δu ζL = 0 and (9.3a)
δu ζR = 0, (9.3b)
two terms disappear, and the unknown normal forces NL and NR are no longer part
of the principle of virtual displacements. For the creep forces, however, the normal
forces are still needed.
In addition, we assume that the wheelset has a constant forward speed v 0 and
a constant angular speed 0 . Small additional vibrations u x and ϕy that might be
superposed to these motions are neglected. With this simplifying assumption, we get
ü x = 0 and (9.4a)
ϕ̈y = 0 , (9.4b)
i.e., two inertia terms in Eq. (9.2) disappear.
9.1.3 Determination of the Virtual Displacements
Even though the real displacements u x and ϕy are set to zero, the virtual displacements
δu x and δϕy will be taken into account. The result is an equilibrium condition in the
x-direction and a moment equation about the y-axis.
The remaining task is to express all virtual displacements from Eq. (9.2) with the
independent virtual displacements δu x , δu y , δϕy , and δϕz , whereby the geometric
constraints (9.3a) and (9.3b) have to be satisfied. The resulting expressions can be
found in Table 9.1. For derivation of these expressions, the following three assump-
tions were made:
• The angle δ0 is small so that the approximations sin δ0 δ0 and cos δ0 1 are
valid; δ0 can be neglected in relation to 1.
• In the following, we are interested only in the linear equations of motion. If the
forces are already proportional to a displacement, no terms have to be taken into
account in the expressions for the corresponding virtual displacements that are
proportional to a real displacement. The only force that is constant is the weight 2Q.
In the expression for the corresponding virtual displacement δu z , therefore, terms
have also to be taken into account that are proportional to the real displacements
u y and ϕz . Therefore, the expression for u z has to be given up to quadratic terms
in u y and ϕz .
• It is possible that the creep forces also include a constant term, for example for a
driven wheelset. In this case, a linearization around the traction creep force might
be performed. Also, the spring forces might include constant terms, i.e., the load
of bogie and vehicle on the vertical primary spring. The calculation of virtual
displacements is rather complicated in these cases [1].
Table 9.1 Compilation of the virtual displacements and rotations as well as the forces that perform
work at these
Type Force Equations for the related virtual displacements or
rotations
Creep −TξL(R) δu ξL(R) = δu x ± e0 δϕz − r0 δϕy
forces −TηL(R) δu ηL(R) = δu y + rL(R) δϕx ∓ eL(R) δL(R) δϕx
δu y (1 + r0 σ)
−MζL(R) δϕζL(R) = δϕz ± δ0 δϕy
Weight −2Q δu z = ζu y δu y − ϕz δϕz
Spring FcxL(R) = + 21 cx v xL(R) δv xL(R) = −δu x ∓ ex δϕz
forces FcyL(R) = + 21 cy v yL(R) δv yL(R) = ±δu y
FczL(R) = + 21 cz v zL(R) δv zL(R) = −δu z ± ex δϕx −δu z ± ex σδu y
To derive the real spring elongations only the symbols
δ need to be omitted
Damping FdxL(R) = + 21 dx v̇ xL(R) Analog to the virtual spring displacements
forces
Mass −m ü x = 0
forces −m ü y δu y
−m ü z δu z = ζ u y δu y − χ ϕz δϕz
−x ϕ̈x δϕx = σδu y
−y ϕ̈y = 0
−x ϕ̈z δϕz
For the determination of a virtual displacement δu i , one can start with the equation
for the real displacement u i , as long as that equation is explicitly known. The virtual
displacement then can be written as
∂u i ∂u i ∂u i ∂u i
δu i = δu x + δu y + δϕy + δϕz . (9.5)
∂u x ∂u y ∂ϕy ∂ϕz
The second possibility is to prescribe small virtual displacements δu x , δu y , etc.,

that are projected in the direction of the virtual displacement δu i being sought. The
first alternative was used to determine δu z , the second one to find δu ηL . All virtual
displacements and the forces acting on them are given in Table 9.1.
Table 9.1 does not include the linear expressions for the creep forces and the
creepages. Both are copied from Sect. 3.4.3. We make two additional assumptions:
1. For the calculation of creep forces, a reference state without creep forces is
assumed, i.e., we assume pure rolling (v 0 = 0 r0 ). This means that we neglect
the small amount of spin creepage existing because of the inclined wheel tread,
since it is possible in this case to use the linearized Kalker equation.
2. Also, for the calculation of creepages we assume pure rolling. Additionally, we

neglect u̇ x and ϕ̇ y . The velocity with which all creep velocities are normalized
can then be written as v m = v0 +r20 0 = v 0 .
The creep forces linearized for a reference state without creepages can then be
written as ⎧ ⎫ ⎡ ⎤⎧ ⎫
⎨ Tξ ⎬ C11 0 √ 0 ⎨ νξ ⎬
Tη = Gab ⎣ 0 C
√ 22 ab C 23 ⎦ νη (9.6)
⎩ ⎭ ⎩ ⎭
Mζ lin 0 − ab C23 ab C33 νζ
With the additional assumptions above, the creepages can be calculated from the
displacements and velocities:
⎡ ⎤
⎧ ⎫ ⎧ ⎫ λ
⎨ νξ ⎬ ⎪
⎨
0 ⎪
⎬ ⎢ r0± 0 ⎥
0 ⎢ ⎥ uy
νη = + ⎢ 0 −1 ⎥
⎩ ⎭ ⎪ δ ⎪ ⎣ γ ⎦ ϕz
νζ L(R) ⎩ ± ⎭
0
r0 − 0
r0
⎡ ⎤
0 ±e0
1 ⎣ u̇ y
+ 1 + σr0 0 ⎦ . (9.7)
v0 ϕ̇z
0 1
Equation (9.7) is valid for v 0 = 0 r0 , i.e., there are no traction or braking forces.
The remaining work involves inserting and sorting operations. The result is an
expression of the form
δu x [· · · ] + δu z [· · · ] + δϕy [· · · ] + δϕz [· · · ] = 0 . (9.8)
Since the virtual displacements are arbitrary, the square brackets must equal zero,
which results in the equations of motion.
9.1.4 Equilibrium Conditions in the x-Direction and Around

the y-Axis
The square brackets belonging to δu x and δϕy result in the force equilibrium in the
x-direction and the moment equilibrium around the y-axis:
− Tξ L − Tξ R + FcxL + FcxR = 0 , (9.9)
and
(Tξ L + Tξ R )r0 + (Mζ L − Mζ R )δ0 = 0 . (9.10)
If each expression is replaced in the other, the sum of the longitudinal creep forces
is eliminated, (Tξ L + Tξ R ). The new equation can be written
δ0
FcxL + FcxR = −(Mζ L − Mζ R ) ,
r0
or if Eqs. (9.6) and (9.7) are used for the spin moments,

δ0 2
FcxL + FcxR = 2G ab C33 . (9.11)
r0
This equation can be interpreted mechanically: the sum of the two forces FxL + FxR
is the force exerted from the bogie frame on the wheelset in the x-direction. It is
necessary to keep a constant speed v 0 . The force is necessary to overcome the very
small part of the rolling resistance that arises from the spin creepage, which in turn
is a result of the nonzero contact angle.
9.1.5 Equilibrium Conditions in the y-Direction and Around

the z-Axis
The remaining square brackets from Eq. (9.8) belong to δu y and δϕz . They result
is the system of equations of motion for the suspended wheelset we are looking for,
Eq. (9.12). For the sake of clarity, different physical effects are separated in the
damping and stiffness matrix:

m + z σ 2 0 ü y 2Qζ 0 uy
+ +
0 z ϕ̈z 0 −2Qχ ϕz

dy + dx ex2 σ 2 0 u̇ y cy + cx ex2 σ 2 0 uy
+ +
0 ex2 dx ϕ̇z 0 ex2 cx ϕz
⎡ ⎤
2 (1 + r0 σ)2 2 (1 + r0 σ)
⎢ 3 ⎥
⎢ GabC22 G (ab) 2 C23 ⎥
⎢ ⎥
⎢ ⎥ u̇ y
1/v 0 ⎢ ⎥ +
⎢ 2e 2 Gab ⎥ ϕ̇z
⎢ −2 (1 + r0 σ) 0 ⎥
⎣ 3 ab
⎦
G (ab) 2 C23 C11 + C33 e2
0
⎡ ⎤ .
−2 (1 + r0 σ) rε −2 (1 + r0 σ)
0
⎢ 3
GabC22 ⎥
⎢ G (ab) 2 C23 ⎥
⎢ ⎥
⎢ ⎥ uy 0
⎢ ⎥ =
⎢ e λ
+2 r0 Gab ⎥ ϕz 0
⎢ ⎥
⎣ 0
2G (ab) 2 C23 ⎦
3
C11 + C33 2ab
e0
(9.12)
In the first row we find the mass terms and the weight proportional terms of the
stiffness matrix; in the second row are the terms from the primary suspension (springs
and dampers); in the third row are the creep force terms of the damping matrix, and
in the fourth row, the creep force terms of the stiffness matrix.
The creep force terms in Eq. (9.12) are quite confusing. Two secondary effects
can be neglected, though:
• The C33 -proportional terms resulting from the spin moment can be neglected, since
ab/e02 1.
• Since the contact angles are small on the running tread (δ0 1), because of
σ = δ0 , the σ-proportional terms can also be neglected compared to 1.
With these assumptions, Eq. (9.12) can be simplified. As simplified equations of

motion of a suspended wheelset, we get Eq. (9.13). If also the C33 -terms are neglected,
we get the equations of motion in the formulation by Carter [2] and later Wickens
[3–5]. Both authors, however, use simplified expressions for the creep force terms:

m 0 ü y 2Qζ 0 uy
+ +

dy 0 u̇ y cy 0 uy
+ +
0 ex2 dx ϕ̇z 0 ex2 cx ϕz
3

1 2GabC22 2G (ab) 2 C23 u̇ y
v 0 −2G (ab) 23 C 2GabC e2 +
23 11 0 ϕ̇ z
3

−2G (ab) 2 C23 −2GabC22 uy 0
=
2 er00λ GabC11 2G (ab) 2 C23 ϕz
3
0
(9.13)
9.2.1 Interpretation of Creep Force terms in Eq. (9.13)
While the terms from mass, weight, springs, and dampers in Eq. (9.13) are obvious,
the creep force terms are not. Interpret the eight terms with the prefactor Gab in
Eq. (9.13).
9.2.2 Rolling Resistance Due to Spin Creepage
√ the rolling resistance from spin creepage for a contact angle δ0 = 1/25 and
Calculate
c = ab = 5 mm. Try to find information about the total resistance of a rail vehicle
from the literature, e.g., [6, 7]. What is the percentage of the rolling resistance from
spin creepage?
9.2.3 Equations of Motions for Forced Lateral Vibrations
With Eq. (9.13), the equations of motion for free vibrations in the lateral direction are
given. What equations are obtained for forced vibrations due to track irregularities
(lateral alignment, cant)? What expressions do we get for cylindrical wheels and only
lateral alignment as excitation?
9.2.4 Rolling Resistance in the Simplified Theory
Try to solve the following contradiction without calculation: In the linear theory from
Kalker, adhesion in the whole contact patch is assumed. Further, all necessary dis-
placements are achieved by elastic deformations. Even though linear elastic material
behavior is assumed, energy dissipation occurs, as can be seen from Eq. (9.11). How
can this be explained?
9.2.5 Equations of Motion of a Suspended Wheelset

with Numbers
Derive the equations of motion (9.13) for the wheelset with the data given in Table 9.2
in matrix form. The different contributions should be separated as in Eq. (9.13). Add
the units.
9.2.6 Creep Forces Under the Assumption of a Reference

State with Creep Forces
In Eq. (9.6), the creep forces are given with the assumption of a reference state
without creepages. What would change if it were assumed that at least the spin
moment basically always is different from zero because of the contact angle δ0 ?
Table 9.2 Input data for a suspended wheelset (Characteristic values that should be used only for
studies of the principles)
Notation Description Dimension Value Comment
m Mass kg 1200
x = z Inertia moments kg m2 450
dy /2 Damping of primary suspension (lateral) N s/m 3 · 102
dx /2 Damping of primary suspension N s/m 3 · 102
(longitudinal)
cy /2 Stiffness of primary suspension (lateral) N/m 5 · 106
cx /2 Stiffness of primary suspension N/m 5 · 107 Minden-Deutz
(longitudinal)
P Axle load N 1 · 105
e0 Half lateral distance between wheel contact m 0.75
points
r0 Rolling radius m 0.46
ex Half lateral distance between springs m 0.95
λ Equivalent conicity – 0.20 Wear profile
γ Coefficient of contact angle difference 1/m 15.0 R 0.36 m
ζ 1. Coefficient of gravitational stiffness 1/m 15.0 R 0.29 m
χ 2. Coefficient of gravitational stiffness m 0.03 δ0 = 0.04
2GabC22 Contact stiffness (lateral) N 2.43 · 107 E, ν as above
2GabC11 Contact stiffness (longitudinal) N 2.00 · 107
3
2G(ab) 2 C23 Nm 0.38 · 105
References
1. L. Mauer, Die modulare Beschreibung des Rad/Schiene-Kontaktes im linearen Mehrkörperfor-

malismus (Modular description of wheel/rail contact for a linear multibody formalism). Ph.D.
thesis, Technical University Berlin (1988)
2. F.W. Carter, The electric locomotive. Proc. Inst. Civ. Eng. 201, 221–252 (1916). Discussion pp.
253–289
3. A.H. Wickens, The dynamic stability of simplified four-wheeled vehicles having conical wheels.
Int. J. Solids Struct. 1, 319–341 (1965)
6. G. Sauvage, J.P. Fortin, Resistance to forward movement of railway vehicles. Fr. Railw. Rev. 1,
161–168 (1983)
7. J. Elkins, R.A. Allen, N.G. Wilson, Effect of wheel/rail contact on train rolling resistance,
in Contact Mechanics and Wear of Rail/Wheel Systems. Proceedings of the 2nd International
Conference held at the University of Rhode Island, Kingston, R.I., ed. by Gladwell G.M.L, H.
Ghonem, J. Kalousek (1986), pp. 213–228
Chapter 10
Lateral Eigenbehavior and Stability
of a Wheelset on Straight Track
With Eqs. (9.12) and (9.13), we have a general system of equations of motion of the
form
M ü + D u̇ + Su = 0. (10.1)
The damping matrix D and stiffness matrix S can be separated into different parts.
The indices used here are as follows:
F = contributions from primary springs;
D = contributions from viscous dampers;
C = contributions from creepages;
P = contributions from masses.
The equations of motion can then be reformulated as

1
M ü + DD + DC u̇ + [ SP + SF + SC ] u = 0. (10.2)
v0
It is important to notice that there is a prefactor (1/v0 ) in front of the matrix DC .

This means that the amount of damping that this matrix contributes decreases with
increased vehicle speed.
10.1 Determination of Eigenvalues and Eigenvectors
A free wheelset with the equations of motion (10.2) exhibits natural oscillations after
an initial disturbance. Since we have linear homogeneous differential equations, the
solution is described as

yy x1 λt
= e or u = x eλt . (10.3)
ϕz x2

DOI 10.1007/978-3-319-45376-7_10
180 10 Lateral Eigenbehavior and Stability of a Wheelset on Straight Track
Substituting Eq. (10.3) into Eq. (10.1), we obtain the homogeneous, algebraic equa-
tion
[ λ2 M + λ D + S ] x = 0, (10.4)
which has a nontrivial, nonzero solution x only if
det [ λ2 M + λ D + S ] = 0. (10.5)
The matrices of Eq. (10.4) have the dimension 2 × 2, since we have a system of two
degrees of freedom. The determinant, Eq. (10.5), results in a characteristic polyno-
mial of order 4 with respect to λ:
a0 + a1 λ + a2 λ2 + a3 λ3 + a4 λ4 = 0. (10.6)
Generally, it is impossible to determine the roots of the characteristic equation (10.6),

the so-called eigenvalues of the equation of motion, analytically. A numerical solution
is inevitable. To determine the eigenvalues, in many cases the characteristic polyno-
mial (10.6) is not even used, but the system of equations of motion is transformed
into a first-order system with twice the number of differential equations:

M 0 d u̇ D S u̇ 0
+ = (10.7)
0 I dt u −I 0 u 0
or
I 0 d u̇ −M −1 D −M −1 S u̇ 0
− = . (10.8)
0 I dt u I 0 u 0
The new vector of unknowns v T = {u̇T , uT } is also called a state vector. The first row
in Eq. (10.7) is equivalent to Eq. (10.1); the second row is the identity du/dt = u̇.
Obtaining the inverse of the matrix M, which is necessary for the transition from
(10.7) to (10.8), is usually possible and rather simple, since the matrix M is a diagonal
matrix. Abbreviated, Eq. (10.8) can be written as
I v̇ − A v = 0. (10.9)
Again introducing
v = y eλt ,
we obtain an algebraic system of equations
[λ I − A] y = 0. (10.10)
By demanding that the determinant be equal to zero, we again obtain the characteristic
equation (10.6). The transformation is meaningful because standard algorithms exist
10.1 Determination of Eigenvalues and Eigenvectors 181
for the eigenvalue problem (10.10). Especially, the HQR algorithm is recommended,
which can be found in program libraries and in the literature [1].
Regarding the eigenvalues, the following can be concluded:
1. Altogether, there are four eigenvalues: λ1 to λ4 .
2. The eigenvalues are either complex conjugates or real-valued.
The ansatz from Eq. (10.3) was therefore too short. If we assume that all eigenvalues
are complex conjugates, we have to write

2

u= qk x k e(αk +iωk )t + q̄k x̄ k e(αk −iωk )t , (10.11)
k=1
where x k is a complex eigenvector and x̄ k the corresponding complex conjugate

eigenvector
x k = x Re,k + ix Im,k x̄ k = x Re,k − ix Im,k .
The two complex conjugate prefactors q k and q̄ k in Eq. (10.11) guarantee, as should
be the case, that the final displacement vector u is real-valued.
From Eq. (10.11) follows the solution in the time domain e(αk +iωk )t = eαk t eiωk t .
The solution consists of an oscillating term (according to Euler, eiωk t = cos ωk t +
i sin ωk t) and an exponentially decreasing term eαk t . The exponential term makes the
oscillation increasing or decreasing. For positive values of αk , the exponential term
is not decreasing but increasing. For stable running behavior of the wheelset, the real
part αk of all eigenvalues must be negative.
10.2 Root Loci Curves
The inverse of the vehicle speed 1/v0 is a prefactor of the damping matrix containing
terms that relate to creep effects in the wheel–rail contact; cf. Eq. (10.2). This means
that the coefficients ai of the characteristic polynomial (10.6) also depend on v0 , and
in turn, that the eigenvalues λi must also be a function of vehicle speed.
Often, the eigenvalues λk = αk ± iωk are shown as dots in the complex (α, ω)-
plane. The curve connecting the dots for the vehicle speed v0 as parameter is called
a root loci curve.
A single wheelset has four eigenvalues, usually two complex conjugate pairs. The
root loci curves are shown in Fig. 10.1. Since we are interested in only the principal
character of the the root loci curve for the moment, no values for masses, stiffnesses,
etc., are given.
Only the root loci curve resulting in weakly damped eigenmodes is critical. The
motions of the wheelset belonging to this eigenmode are almost sinusoidal. For
v0 = 0, the root loci curve begins at the origin of the (α, ω)-coordinate system. For
increasing speed, it first moves into the half-plane with negative real part (damped
v
crit
v = v crit
slightly damped
heavily damped
root loci curve
root loci curve v
v=0
wheelset hunting at
v = v crit
L/2
v0
x
Fig. 10.1 Root loci curves for a single wheelset and eigenmode for the weakly damped root loci
curve
oscillation). For higher speeds, it changes direction and moves into the half-plane with
positive real part. The vehicle speed where the real part is zero is called the critical
speed vcrit or linear critical speed. The corresponding frequency is denoted by ωcrit . An
investigation of the motion of the wheelset for different speeds on the root loci curve
gives that all movements are almost sinusoidal. At the point on the imaginary axis
(v0 = vcrit ), the eigenmode is a harmonic sinusoidal motion. For speeds v0 < vcrit ,
the sinusoidal motion is damped, while for v0 > vcrit , the amplitude of the sinusoidal
motion is increasing.
10.2 Root Loci Curves 183
The eigenvalues belonging to the higher damped eigenmode of Fig. 10.1 are
always negative. No critical speed exists for this eigenmode. The reason is that
the eigenmode involves very high creepages in the wheel–rail contact.
10.3 Approximate Solution for Low Speeds
It is advisable to look at the coefficients of the differential equation before starting

with the approximation for low speeds. This cannot be done by writing the equation
numerically, since the displacement vector u contains displacements and rotations
and since the mass matrix is multiplied by accelerations, the damping matrix by
velocities, and the stiffness matrix by displacements.
A solution for this dilemma is to use a dimensionless formulation. Therefore, a
dimensionless time τ is introduced by dividing the time t by the vibration period T0
of the sinusoidal motion:
t ∂ ∂ 1 ∂
= τ and there by = = ω0 , (10.12)
T0 ∂t ∂τ T0 ∂τ
with
λe
ω0 = v . (10.13)
e0 r0
Accordingly, dimensionless amplitudes are introduced by relating u y and ϕz to the

amplitude of the sinusoidal motion of the free wheelset. This results in multiplying
the matrices by a transformation matrix from the right. A similar multiplication from
the left is performed. The transformation itself will be carried out in Sect. 10.7.2.
The introduction of the dimensionless time τ results in a prefactor v 2 ahead of the
mass matrix and a factor v in front of the damping matrix from the viscous dampers.
The prefactor v1 in front of the damping matrix from the creepages disappears, how-
ever. By choosing, for example, a vehicle speed of 100 km/h (27 m/s), the following
can be concluded:
1. The forces from the primary springs are dominating, especially for a rotation with
respect to the vertical axis. This is true for the Minden-Deutz bogie, on which the
calculation was based and assuming an undisplaceable bogie frame. For a softer
wheelset guidance, the longitudinal primary stiffness cx can be roughly equal to
the primary lateral stiffness cy .
2. For low speeds, the inertial forces can be neglected. The amount of the inertial
force for a displacement in the lateral direction at a speed of 27 m/s is only about
10 % of the force from the lateral primary springs. For higher speeds, it is no
longer impossible to neglect the inertial forces.
3. Also, the damping forces from the primary suspension are negligible.
4. The C23 -terms of the stiffness and damping matrix with the creepage terms are
small and are partly compensated by the gravitational stiffness.
In summary, at low speeds, the influence of the mass matrix as well as the matrix
with the forces from viscous dampers D D can be neglected.
The original equation for the eigenvalues can be written as

m 11 0 d11D + v10 d11C 1
d
v0 12C
det λ2 +λ
0 m 22 + v10 d21C d22D + v10 d22C

s11PF + s11C s12C
+ = 0. (10.14)
s21C s22PF + s22C
With the simplifications described above, for low speeds we get the following equa-
tion instead:

λ d11C d12C s + s11C s12C
det + 11PF = 0. (10.15)
v0 d21C d22C s21C s22PF + s22C
The corresponding characteristic equation can be given without difficulty:

λ 2
det DC
v

λ
+ d11C (s22PF + s22C ) + d22C (s11PF + s11C )
v
− d21C s12C − d12C s21C + det S = 0, (10.16)
which is equivalent to the characteristic equation of a single mass oscillator with

mass m, damping constant d, and spring stiffness s,
mλ2 + dλ + s = 0. (10.17)
The solution of Eq. (10.17) is

d s d2
λ1,2 =− ±i − . (10.18)
2m m 4m 2
As long as the damping of the single-mass system is not too high, the result is two
complex conjugate eigenvalues
λ1,2 = −δ ± iω, (10.19)
with
10.3 Approximate Solution for Low Speeds 185
d
δ= and (10.20a)
2m

s d2
ω= − . (10.20b)
m 4m 2
Only for relatively large damping can two purely real valued eigenvalues be found.
This case will not be discussed here, however.
In order to be able to apply solutions (10.19) and (10.20a) as well as (10.20b) to our
characteristic equation (10.16), it has to be ensured that both “mass” (m = det DC )
and “stiffness” (s = det S) are positive. For m, this is always the case, and in our
example for s as well. The eigenvalues obtained for a damped sinusoidal eigenmode
of the wheelset is
[· · · ] iv
λ1,2 = − v ± 4 det DC det S − [· · · ]2 . (10.21)
2 det DC 2 det DC

damping δ frequency iω
If we let SPF = 0 in the equation for the frequency,

v 4 det DC det S − [· · · ]2
ω= , (10.22)
2 det DC
and neglect the C23 -proportional terms that contain spin creepage, the result is the
Klingel equation introduced above, Eq. (8.14). Equation (10.22) is thus a more gen-
eral form of the Klingel equation.
Some additional conclusions can be drawn from Eq. (10.21). Since damping d
always has to be positive, the term in square brackets, which is the term from
Eq. (10.16), has to remain positive:
d11C (s22PF + s22C ) + d22C (s11PF + s11C ) − d21C s12C − d12C s21C > 0. (10.23)
If Eq. (10.23) is satisfied, then the root loci curve moves into the half-plane with
negative real part. If the term in square brackets is negative, the root loci curve from
the origin moves directly into the half-plane with positive real part. In the first case,
the motion is stable for small speeds, while in the second case it is unstable.
The coefficients in Eq. (10.23) can be expressed by physical values (spring stiff-
ness, contact stiffness, etc.). If this conversion is made, the following relation is
obtained:

2Q C22 ex2 (ab)3/2 Rδ0
cy + + cx > 2G 1+ . (10.24)
R − R C11 e02 r0 (RW − R R ) e0
The more the term on the left-hand side of Eq. (10.24) dominates, the more damped
is the sinusoidal motion of the wheelset. This means that from this equation, we can
understand what quantities stabilize the wheelset at low speeds:
• The spring stiffnesses cx and cy are stabilizing.

• An increase in the axle load 2Q increases the term on the left-hand side. At the
same time, the term on the right-hand side is increased via the dimensions a and
b of the contact ellipse.
• Also, the difference RW − R R in the radii of curvature can be found on both sides
of Eq. (10.24). In addition, the mean radius of the ellipse c = (ab)1/2 is increased
by decreasing RW − R R . This indicates that damping of the sinusoidal motion
decreases the more worn the profiles become: RW − R R −→ 0.
• Theoretically, increasing the rolling radius r0 reduces the term on the right-hand
side. This does not, however, take the influence on a and b into account. Anyway,
there are other limitations on r0 . Therefore, in reality, this influence cannot be
used.
Conclusion: For low speeds, the amount of damping of the sinusoidal motion is
increased by increasing the primary stiffnesses. Worn, i.e., more conformal, profiles,
however, reduce damping.
To explain the phenomenon in Fig. 10.1, i.e., that a root loci curve can move from
the negative half-plane into the positive half-plane, the influence of the masses has
to be taken into account. This is done in Sect. 10.4.
10.4 Stability Investigation with Hurwitz Criterion
In many cases, one is not interested in the whole root loci curve but only in the
critical value vcrit (Fig. 10.1). To calculate vcrit , the characteristic polynomial (or the
eigenvalue problem) has to be solved for different v0 -values to find the speed at
which the real part αk of the eigenvalues is zero. For systems with only a few degrees
of freedom, an elaborate numerical zero determination is unnecessary. Very simple
conditions exist under which it is possible to determine whether a motion is stable.
As example, we look at the eigenmode of the wheelset where the mass terms
are neglected. The corresponding characteristic equation is (10.16), which is equiva-
lent to the characteristic equation of a spring–mass-damper–system, Eq. (10.17). For
Eq. (10.17), we can directly conclude that the natural vibration is stable only if the
damping coefficient d is positive. A negative damping coefficient would equate to an
active element that would exert a force on the mass in the direction of the velocity that
is used to determine the force. The stability condition for the system of Eq. (10.17)
is therefore
d > 0.
The correctness of this statement can also be understood from the eigenvalues of
Eq. (10.18): the damping coefficient is proportional to the real part δ.
This fact was used in the last chapter, though not for the critical speed, but “only”
to determine the initial inclination of the root loci curve. In the following, we will
investigate whether it is possible for characteristic equations of higher order also to
develop simple stability criteria.
10.4 Stability Investigation with Hurwitz Criterion 187
A closed expression for the stability criterion would make it very easy to discuss
the influence of system parameters. It can be shown that it is possible to formulate
analytic stability criteria for characteristic polynomials up to order 6.
For a mechanical system with N degrees of freedom and positive definite mass
matrix, the eigenvalues are the roots of a polynomial of the order 2N :
a0 + a1 λ + a2 λ2 + · · · a2N λ2N = 0. (10.25)
The basic idea of the closed expressions for the stability criteria is very simple: At
the stability limit, the system performs harmonic oscillations with a frequency that
could be called a critical frequency or stability frequency ωcrit . If now λ = iωcrit is
substituted into the characteristic equation (10.25), the characteristic equation can
be separated into one part for the imaginary parts and one equation for the real parts:
a1 − a3 ωcrit
2
+ a5 ωcrit
4
− · · · = 0, (10.26a)
a0 − a2 ωcrit
2
+ a4 ωcrit
4
− · · · = 0. (10.26b)
For the second-order characteristic equation from Eq. (10.26b), we obtain the critical
frequency
a0
ωcrit
2
= , (10.27)
a2
and from (10.26a), the stability limit condition is a1 = 0. This is, in fact, equivalent
to what is expected mechanically for the single-degree-of-freedom system:
s1 + λd1 + λ2 m 1 = 0.
The coefficient a1 is equal to the damping d1 . The system is stable as long as
a1 = d1 ≥ 0.
For the third-order characteristic equation, the critical frequency can be formulated
in two ways:
ωcrit
2
= a0 /a2 or ωcrit
2
= a1 /a3 . (10.28a, b)
If ωcrit
2
is eliminated, the stability limit condition can be written as
a1 a2 − a0 a3 = 0.
Whether the sign less than (≤) or the sign greater than (≥) means stable running can
be resolved by the following consideration: For the second-order system, an increase
in a1 increases stability. The stability condition has to be valid also when a3 is small
and the third-order system basically behaves like the second-order system regarding
the influence of a1 . Therefore, the stability condition can be written as
a1 a2 − a0 a3 ≥ 0. (10.29)
For higher-order characteristic equations, the easiest way is to refer to the stabil-
ity criteria given by Adolf Hurwitz,1 which are given in the appendix, Sect. 16.6.
The mathematically correct procedure of Hurwitz to define stability criteria has the
advantage that no criteria are forgotten. In Table 16.1 in the appendix (Sect. 16.6), the
stability criteria for characteristic equations up to order six are given. It is mentioned
from which equation the critical frequency is derived. It has to be pointed out that
for a polynomial of order N , all stability criteria for lower-order polynomials also
have to be met. Further, all coefficients of the characteristic polynomial have to be
positive and different from zero.
As an example, for the stability condition for the second-order characteristic
equation (10.29), it was assumed implicitly that the coefficients a0 and a2 are positive.
Regarding the coefficient a2 , this is obvious for an oscillator, since the mass is always
positive. For a0 , it is not obvious. For an upside down pendulum, the stiffness is
negative, and therefore it is monotonically unstable. The same behavior is shown by
an unsuspended wheelset with regard to the yaw motion.
10.5 Critical Speed of a Wheelset
The system of differential equations for a single wheelset, Eq. (10.1), results in a
fourth-order characteristic polynomial (10.6). This polynomial is now explicitly
given and studied with the stability criterion of Table 16.6. The same abbreviations as
in Eq. (10.14) are used. Viscous dampers are neglected, and the stiffness contributions
are combined:

2 m 11 0 λ d11C d12C s11 s12
det λ + + = 0. (10.30)
0 m 22 v d21C d22C s21 s22
As before, in the damping matrix so-called gyroscopic effects are disregarded. They
become significant only at speeds above 250 km/h.
The coefficients of the characteristic equation can be written as
a4 = m 11 m 22 , (10.31a)
1 1
a3 = [m 11 d22C + m 22 d11C ] = a31 , (10.31b)
v v
1
a2 = m 11 s22 + m 22 s11 + 2 [d11C d22C − d12C d21C ]
v
1
= a20 + 2 a22 , (10.31c)
v
1 Adolf Hurwitz (1859–1919) was inspired by Aurel B. Stodola (1859–1942) to develop criteria for
the stability of equations of motion. The Hurwitz criterion was published in 1895.
10.5 Critical Speed of a Wheelset 189
1 1
a1 = [d11C s22 + d22C s11 − d12C s21 − d21C s12 ] = a11 , (10.31d)
v v
a0 = s11 s22 − s21 s12 . (10.31e)
The terms are already divided into speed-dependent and speed-independent terms.
We substitute the coefficients into the stability criterion. Without proof, we assume
that in our case, the criterion
a1 a2 a3 − a0 a32 − a4 a12 ≥ 0 (10.32)
is decisive. If Eqs. (10.31a)–(10.31e) are substituted into this equation, we obtain

1 1 2 1 2 1
a11 a20 + a22 2 a31 2 − a0 a31 − a4 a11 ≥ 0. (10.33)
v v v2 v2
If in Eq. (10.33) the equal sign is used instead, the critical speed vcrit is found:
1
a4 a11
a31
+ a0 a31
a11
− a20
= . (10.34)
vcrit
2 a22
The frequency ωcrit at the stability limit is found from the relation
ωcrit
2
= a1 /a3 = a11 /a31 . (10.35)
The defined task is thus solved: Eq. (10.34) determines the critical speed, i.e., the
speed at the stability limit.
10.6 Interpretation of Stability Criteria of a Single

Wheelset
In this section, the stability limit condition of the single wheelset will be evaluated
and interpreted. In Eqs. (10.34) and (10.35), therefore, the expressions (10.31a) to
(10.31e) are introduced, and afterward, the abbreviations for the matrix coefficients
are replaced by the expressions from Eq. (10.30). These calculations are quite tedious.
For simplification, first the nondiagonal elements of the creep–damping matrix are
neglected, since they are small. The equation for ωcrit then can be written as
d11C s22 + d22C s11

ωcrit
2
= . (10.36)
d11C m 22 + d22C m 11
To rearrange Eq. (10.34) is more difficult. In the following, the process will be shown
in detail.
First, the expressions for a22 and a0 from Eqs. (10.31c) and (10.31e) are intro-
duced. The resulting equation is somewhat rearranged to more clearly illustrate the
main effects and the side effects:
d11C d22C
vcrit
2
= .
2
(−s12 s21 ) aa31
11
1+ 1
(−s12 s21 )
a a11
4 a31 + s s
11 22 − a a11
20 a31
In a next step, ahead of the curly braces, as abbreviation the frequency ωcrit from
Eq. (10.35) is introduced. The expression in square brackets can be significantly
simplified with a tedious calculation. We arrive at
d11C d22C 1
vcrit
2
= ωcrit
2
, (10.37)
(−s12 s21 ) 1 − k
where k is usually a small correction term:

2
d11C d22C m 22 s11 − m 11 s22
k= . (10.38)
(−s12 s21 ) m 11 d22C + m 22 d11C
Equations (10.36)–(10.38) finally are surprisingly “handy”. Once again, we make

the following simplifying assumptions:
• Primary damping,
• neglect of gyroscopic effects,
• neglect of nondiagonal terms d12C of the damping matrix.
In the appendix, it is shown that it is justified to neglect the nondiagonal terms
d12C of the damping matrix (Sect. 16.7).
Introduction of Geometric and Physical Measures
In Eqs. (10.36–10.38), geometric and physical measures will be introduced. Before
this, another simplification is made: the longitudinal stiffness cx is assumed to be
so high that it dominates the term s22 in the stiffness matrix; i.e., the effects from
creepage (C23 ) and gravity (−χ2Q) can be neglected.
First, the coefficients from Eq. (9.13) are inserted into Eq. (10.37):
e0 r0 1
vcrit
2
= ωcrit
2
. (10.39)
λ 1−k
This is on the one hand a surprising result, and on the other hand a nice one. At
the stability limit, the relationship between critical speed and frequency is almost
the same as in the Klingel equation. This equation just needs to be modified by
the correction factor k. Then the frequency at the stability limit ωcrit and thus the
critical speed vcrit can be determined. The expressions for d11C , s22 , d22C , s11 , m 22 ,
and m 11 from Eq. (9.13) are substituted into Eq. (10.36). Taking into account the
simplifications introduced above, after some rearrangement we obtain
10.6 Interpretation of Stability Criteria of a Single Wheelset 191
2
2 1+ e0 C11 2Qζ
+
cy
− 2εG(ab)3/2 C23
ex cx C22 ex C22 cx cx r0 cx
ωcrit
2
= C22
. (10.40)
e0 m C11 1+ me02 C11
What is missing is the expression for k from Eq. (10.38):

2 2
1 ex cx r0 ex
k=
λ 2GabC11 e0 e0
⎡ 2 ⎤2
cy 2εG(ab)3/2 C23
e0
1 + 2Qζ
− − 1
⎢ me02 ex cx cy r0 cy ⎥
⎣ C22 ⎦ . (10.41)
1 + me2 C11
0
The different terms in Eq. (10.41) are already arranged in a way that they are as near
as possible to 1. Such an estimate is impossible, however, with the terms
2
1 ex cx
λ 2GabC11
and
2Qζ 2εG(ab)3/2 C23
1+ − .
cy r0 cy
Now the values from Table 9.2 (Sect. 9.2.5) are replaced in the first term. This yields
2
1 ex cx cx 2
, (10.42)
λ 2GabC11 1.79.107
where the longitudinal primary stiffness still is arbitrary. The second term is rewritten
so that the contributions from contact stiffness, gravitational stiffness, and lateral
spring stiffness are directly visible:

1 ε 2G(ab)3/2 C23
cy +2Qζ 1 − . (10.43)
cy ζ 2Qr0

lateral spring stiffness gravitational stiffness contr. contact stiffness contr.
If also the expressions from Hertz’s theory are introduced, it becomes obvious that
the term in parentheses does not depend on the shear modulus G or on the axle
load 2Q. If the values for the set of parentheses are determined as a function of
the relationship between the semiaxes a/b (0.1 < a/b < 10), only minor changes
are detected. Because of the reducing influence of the C23 -term, the expression in
parentheses takes values between approximately 0.22 (for a/b = 10) and almost zero
(for a/b = 1/10).
Estimation of k and of the Critical Speed

Now we are going to look at three different cases for the estimation of k from
Eq. (10.41):
(a) The wheelset is almost not suspended, i.e., the spring stiffnesses are very small
(< 105 N/m). In this case, for worn profiles, the spring stiffnesses can be
neglected compared to the contact stiffness and the gravitational stiffness. The
value k is much smaller than 1 (k 1).
(b) One of the spring stiffnesses is chosen relatively high relative to the other one,
e.g., cx > 107 N/m (cy cx ). In that case, Eq. (10.41) can be simplified to
2 2
1 ex cx r0 ex 1
k 2 .
λ 2GabC11 e0 e0 C22
1+ me02 C11
We observe that k can be made as high as we want to by increasing cx . According

to Eq. (10.39), it is not, however, meaningful to exceed k = 1. For k = 1, the
critical speed vcrit is already infinite. As maximum value for the longitudinal
stiffness, we obtain

2GabC11 e0 e0 λ C22
cx,max = 1+ .
ex ex r0 me02 C11
A similar result is obtained if not cx but cy is increased continuously.

(c) If, however, both spring stiffnesses are increased at the same time in a way that
cy
= 1,
mex2 cx
then it is impossible to increase k to 1, and instead, k remains at the same value

as in case (a). The critical speed vcrit does not reach infinity.
Numerical Determination and Graphical Illustration of the Critical Speed
The results above from analytic calculations are obtained as well if the critical speed is
determined purely numerically. In Fig. 10.2, the critical speed is shown as a function
of the spring stiffnesses cx and cy . A logarithmic scale is chosen for all axes. The
three effects discussed above are clearly shown:
• For low stiffnesses, we recognize a “plateau”. The critical speed depends only on
the gravitational stiffness and the contact stiffness and remains quite low.
• If either the longitudinal stiffness cx or the lateral stiffness cy is increased, at certain
values the critical speed rises very fast.
• If both stiffnesses are increased simultaneously, however, a valley is observed.
The necessary high stiffnesses can be achieved without difficulty. Still, it is impossible
in reality to achieve infinite critical speed. In deriving the equations of motion, we
10.6 Interpretation of Stability Criteria of a Single Wheelset 193
plateau valley steep face

Vcrit
100 m /s
cx 10 m/s
cy
[ N/ m ] 10 5
10
6
4 106
[ N/ m ]
10 3
10 4 10 5
10 2 10 3
10 10 2
Fig. 10.2 Critical speed as a function of cx and cy
assumed that the bogie frame does just follow the track with speed v0 without any
of its own degrees of freedom. This is, of course, not true. In reality, the bogie frame
is displaced laterally and yaws about the vertical axis. Only if these motions are
blocked, e.g., by a friction yaw damper, is it possible to achieve high critical speeds
like those calculated for a stiffly suspended wheelset in this chapter. This is realized
in the Minden-Deutz (MD52) bogie.
The friction yaw damping is not a very elegant solution, especially if Coulomb
friction elements are used. On entering a curve, a bogie with friction yaw damping
would like to continue in a straight direction. The right position of the wheelsets in
the curve is achieved first by flange contact of one of the wheelsets. This will lead
to extensive wear. A vehicle optimized for curving will therefore probably not be
equipped with friction yaw damping. For the stability investigation in that case, it is
impossible to look at only one wheelset.
10.7.1 Characteristic Equation
Using the separation of matrices from Eqs. (10.2), (10.5) then can be written as

m 11 0 d11D + v10 d11C 1
d
v0 12C
det λ2 +λ
0 m 22 + v10 d21C d22D + v10 d22C

s11PF + s11C s12C
+ = 0.
s21C s22PF + s22C
For this formulation, give the coefficients a0 to a4 of the characteristic equation.

10.7.2 Transformation of Wheelset Equations of Motion
Carry out the transformation of the equations of motion indicated in Sect. 10.3,
especially in Eq. (10.12). The angular natural frequency of the sinusoidal motion,
∗
Eq. (10.13), should be introduced. For the natural vibration, the ansatz x = eλ τ
should be used. After this transformation, the inverse value of the speed v does
not appear as a prefactor in front of the matrix DC . Instead, a new prefactor, λ∗ v,
shows up in front of the damping matrix D D , and a prefactor λ∗2 v 2 in front of the
mass matrix M. Give the numerical values for the different matrix coefficients in an
equation similar to that in Sect. 9.2.5.
10.7.3 Graphical Illustration of the Root Loci Curves

of a Suspended Wheelset and Determination
of the Critical Speed
Starting with the equation of motion of the suspended wheelset in Sect. 9.2.5, inves-
tigate the stability of the wheelset for different parameters and illustrate it as root
loci curves. Sketch eigenmodes.
The input data for the wheelset can be taken from the table in Sect. 9.2.5. First,
perform the stability investigation for the given parameters. After that, show and
discuss the influence of different parameters on the result. A suggestion is to look at
the relationship between the primary stiffnesses cx and cy .
10.7.4 Independently Rotating Wheels
How do the equations and the result change under the assumption of independently
rotating wheels, i.e., both wheels can rotate freely on the wheelset axle?
Reference
1. W.H. Press, S.A. Teukolsky, W.T. Vetterling, B.P. Flannery, Numerical Recipes, 2nd edn. (Press
Syndicate of the University of Cambridge, Cambridge, 1992)
Chapter 11
Lateral Eigenbehaviour and Stability
of Bogies
Up to now, we have looked at the natural vibration behavior of one wheelset. In

this chapter, bogies will be investigated. It can be expected that it will be even more
difficult to find general conclusions in this case. In Sect. 11.1, we will perform a
numerical investigation and plot the eigenvalues derived as a function of vehicle
speed v in root loci curves in the complex plane.
These numerical calculations of eigenvalues and thus the calculation of the critical
speed are time-consuming. For the very first design stages, it would be helpful to have
some approximation formulas or at least some qualitative results. How to achieve such
approximations will be discussed in Sect. 11.2. Finally, in Chap. 12, the calculation of
the natural vibration behavior and the critical speed v crit of vehicles will be covered.
Stability investigations of bogies and four-axle vehicles with elastically supported
wheelsets have been performed by, among others, Matsudaira [1, 2], Wickens [3–5],
Joly [6], and Keizer [7, 8].
11.1 Numerical Calculation of Eigenvalues

and Critical Speed
We begin by looking at a “classical,” nonpowered bogie with a bogie frame and two
wheelsets (Fig. 11.1).
For the calculation, the following assumptions are made:
(1) As in the investigation of a single wheelset, the rails are regarded as rigid and
fixed.
(2) Both wheelsets and the bogie frame are regarded as rigid bodies.
(3) The carbody is an infinitely high mass that is following the track without its own
degrees of freedom.

DOI 10.1007/978-3-319-45376-7_11
196 11 Lateral Eigenbehaviour and Stability of Bogies
front view
1 1
2 d zF 2 c zF
1
2cz
1
2dz
1 1
2 dy 2 c y
1 1
2 c y 2 dy
1
2cx
1
2dx 1
1 2 c zF
1 2 c yF
2 dyF 1
1 2 d zF
2 d xF
1 1
v0 2 c xF 2 c xF
1
2 d xF
1
2 dy
1
2cy
top view side view
Fig. 11.1 Bogie, system, and notation (Bogie frame with two wheelsets)
(4) For the calculation of eigenvalues, only the degrees of freedom for the wheelsets
and the bogie frame need to be introduced. All other masses are added to these
masses. For a powered bogie, it is usually necessary to introduce the motors
individual bodies with their own degrees of freedom.
(5) The wheel–rail contact is described with linear relations, as in the investigations
of the single wheelset above.
(6) Each wheelset is connected to the bogie frame via a primary suspension consist-
ing of springs and dampers. Further, a spring and damper connection between
bogie frame and the guided carbody exists (secondary suspension).
(7) All suspension elements are linear. This excludes elements with play or yaw
friction damping between bogie frame and carbody. Such an element would
need to be described as a Coulomb friction element.
(8) Full symmetry with respect to the (x, z)-plane is postulated.
11.1 Numerical Calculation of Eigenvalues and Critical Speed 197
Based on the last assumption, the motions in the vertical and lateral directions can
be investigated separately. In this chapter, as in the case of the single wheelset, only
lateral dynamics are considered.
The equations of motion for the lateral running behavior of the bogie in Fig. 11.1
will be derived in an exercise. With the assumptions above, seven degrees of freedom
are needed to describe the system. Abbreviated, the equations can again be written
as
M ü + D u̇ + Su = 0 , (11.1)
where M, D, and S are 7 × 7 matrices. As for the single wheelset, we introduce for
the natural oscillations the ansatz
u = xeλt (11.2)
and get 2
λ M + λD + S x = 0 . (11.3)
By demanding
det λ2 M + λ D + S = 0 , (11.4)
the eigenvalues λk can be calculated. How many eigenvalues do we get for the system
of equations of motion in Eq. (11.1)?
The root loci curves for a bogie with two wheelsets are shown in Fig. 11.2. As long
as the eigenvalues are complex conjugates, seven root loci curves are determined
if we restrict ourselves to the positive imaginary parts. Since we are interested in
general conclusions regarding the root loci curves, no input values such as masses
and stiffness values are given. Therefore, the coordinate axes do not have units. The
root loci curves are given for speeds between 0 m/s (circles) und 50 m/s (filled dots).
The are numbered from 1 to 7.
The root loci curves can be characterized by the following three properties:
Speed-Dependence:
Because of the processes in the wheel–rail contact, all root loci curves are speed-
dependent. Some root loci curves are strongly speed-dependent, i.e., curves 1–3 and
5–7, while curve 4 is weakly speed-dependent. In this example, there are no root loci
curve that are independent of speed.
Origin of Root Loci Curves:
Root loci curve 7 begins at the origin of the coordinate system (δ = 0, ω = 0), while
curves 4–6 begin at points in the complex plane with an angular frequency ω and
a negative real part α = −δ.1 The root loci curves 1–3 begin with infinitely high
negative real parts (α = −δ = −∞, ω = 0).
1 Forthe real part, we use α or δ; see Figs. 11.2 and 11.3. Here δ is negative for positive real part α.
For negative real part, that means for a damped vibration, δ is positive.
rad
sec
6
1
2
4
3 7
-
rad
v = 0 [m/ sec ] , v = 50 [m/ sec] sec
Fig. 11.2 Root loci curves for a bogie (Bogie frame with two wheelsets)
Tendency to Instability:
For some root loci curves, eigenvalues with positive real part can occur within the
investigated speed area, i.e., eigenvalues with unstable eigenmodes. This applies only
to curve 7. For a further increase in speed, eigenvalues with positive real part might
arise, for example, also for curve 6.
For further interpretation of the root loci curves, the eigenmodes belonging to
the respective curves also have to be determined. Before we show the mode shapes,
however, two other thoughts are presented.
For the first reflection, the root loci curves and the corresponding mode shapes
for a single wheelset are shown again in Fig. 11.3.
Two of the speed-dependent eigenvalues of the rigid wheelset on rigid track have
positive imaginary parts, while two have negative imaginary parts. For one of the
curves with positive imaginary parts, for v = 0 the eigenvalues begin at (δ = 0,
ω = 0). The eigenvalues of the other curve begin at (α = −δ = −∞, ω = 0).
For the second root loci curve, all eigenvalues are strongly damped. For the first
curve, the eigenvalues are only weakly damped or undamped, and when they exceed
the critical speed v crit , they are even negatively damped. In Fig. 11.3, the eigenmode
for the first root loci curve at the stability limit is shown. The eigenmode is a pure
hunting motion. The yaw motion ϕz has a phase shift of 90◦ with respect to the lateral
displacement u y . For speeds below the critical speed, the eigenmode is damped. This
mode is sometimes called the kinematic mode.
In a second reflection, the mode shapes of the bogie frame are looked at. For this,
we block the two wheelsets and investigate the eigenvalues and eigenmodes that
arise. Since the only speed-dependent terms stem from the wheel–rail contact, the
eigenvalues now have to be independent of the vehicle speed. The bogie frame has
11.1 Numerical Calculation of Eigenvalues and Critical Speed 199
v
crit
v = v crit
heavily damped slightly damped

root loci curve root loci curve
v
v=0
wheelset hunting at
v = v crit
Fig. 11.3 Root loci curves and mode shapes for a single wheelset
uy
sway mode lateral mode yaw mode
Fig. 11.4 Displacements for mode shapes of bogie frame
three degrees of freedom, and therefore, we get three complex conjugate pairs of
eigenvalues. The results can be found in Fig. 11.2, namely the root loci curves 4, 5,
and 6. The corresponding mode shapes are shown in Fig. 11.4.
For eigenmodes 5, 6, and 7, the movement of bogie frame is a combination

of lateral displacement (u yF ) and yaw motion (ϕzF ). As long as the wheelsets are
blocked, the yaw eigenmode of the frame appears isolated (Fig. 11.4). The coupling
between lateral displacement and yaw is introduced by the wheelsets.
For root loci curve 4, the eigenmode is a sway motion (ϕxF ). Usually, the sway
motion (ϕxF ) is coupled with the lateral displacement. The motion is called roll if the
center of rotation is above the center of gravity (in this case, of the bogie frame) and
sway if the center of rotation is below the center of gravity of the body. Sometimes,
the terms lower sway and upper sway are used. In the example here, we cannot
see very much roll or sway. The reason is the strong roll dampers in the secondary
suspension. In eigenmodes 5, 6, and 7, which do not contain significant sway, the
system tries to avoid the damping influence of the roll dampers. Only root loci curve
4, which is strongly damped, however, contains significant sway.
After these initial thoughts, the eigenmodes belonging to the root loci curves in
Fig. 11.2 can be explained. For low speeds, the root loci curves 1, 2, and 3 begin on
the negative real axis. The corresponding modes therefore are initially very strongly
damped or even aperiodically damped. For curves 1, 2, and 3, the single wheelsets
therefore do not perform kinematic motions—with almost no creep—at low speeds,
but rather motions where large creepages occur in the wheel–rail contact. For typical
speeds, these modes are so heavily damped that they are not interesting for further
study. As a remark, it should be said, however, that this may change if the primary
springs (especially the longitudinal ones) are very soft and the damping low. In that
case, the wheelset modes can also become unstable at realistic speeds.
The eigenmode belonging to root loci curve 4 is a strongly damped sway motion
of the bogie frame even when the whole bogie is investigated. The wheelset motions
are of minor importance.
The eigenmode for root loci curves 5, 6, and 7 are shown in Fig. 11.5. A kinematic
motion of both wheelsets belongs to mode shape 7. The yaw motion ϕz0 and the lateral
displacement u y0 of the wheelsets have a phase shift of 90◦ . The bogie frame moves
almost in phase with the kinematic motion of the wheelsets. It is certainly possible
that there are small phase shifts between the lateral displacement of the wheelsets
and the bogie frame and between the yaw angles. Optically, the impression that
they move in phase dominates. The amplitudes of wheelsets and bogie frame can be
different as well.
A kinematic mode of the wheelsets belongs also to eigenmode 6. The wheelsets
move essentially in phase. The movements of the bogie frame are almost out of
phase. In our example for this mode shape, the lateral displacement of the bogie
frame dominates.
For eigenmode 5, both wheelsets perform a kinematic motion out of phase. The
bogie frame mainly performs a yaw motion with 90- and 270-degree phase shifts to
the wheelsets.
11.2 Analytic Approximations for Bogies 201
rlc 7 rlc 5 rlc 6
Fig. 11.5 Eigenmodes for root loci curves 7, 5, and 6 from Fig. 11.2
11.2 Analytic Approximations for Bogies
It is impossible here to describe all possibilities on how to derive approximate solu-

tions for the critical speed of bogie vehicles in detail. We refer to the relevant litera-
ture, for example to the studies by Rocard from the 1930s [9, 10] and to the work by
Wickens [4, 5, 11], Scheffel [12, 13], and Keizer [7]. The following text describes
only one of the many possibilities to simplify the investigation of bogies. It is oriented
at Keizer [7] and should motivate further reading.
Firstly, the equations of motion of a bogie (frame with three degrees of free-
dom, two wheelsets with two degrees of freedom each) are derived. The following
simplified assumptions are made:
(1) A bogie without carbody is considered. No secondary suspension (springs or
dampers) is present.
(2) Primary damping is neglected.
(3) The simplified wheelset matrices from Eq. (9.13) are taken as a basis. In addition,
the C23 -proportional terms and the gravitational stiffness are neglected.
For this bogie, the result is the system of Eq. (11.5). The notation used is again
summarized below:
cx Spring stiffness of one wheelset in the x-direction,

cy spring stiffness of one wheelset in the y-direction,
2e0 lateral distance between wheel contact points,
2ex lateral distance between attachment points of longitudinal springs,
r0 rolling radius,
2b wheelset distance,
λe equivalent conicity,
v vehicle speed,
m wheelset mass,
wheelset moment of inertia about the vertical axis,
mF bogie frame mass,
zF bogie frame moment of inertia about vertical axis,
xF bogie frame moment of inertia about the longitudinal axis.
Further, the two abbreviations
f 1 = 2GabC11 , f 2 = 2GabC22
are introduced.
⎡ ⎤⎧ ⎫
m 0 0 0 0 0 0 ⎪ ⎪ ü y1 ⎪
⎪
⎢0 ⎪
⎪ ϕ̈z1 ⎪
⎢ 0 0 0 0 0 ⎥ ⎥⎪⎪
⎪
⎪
⎪
⎢0 ⎪ ⎪
⎢ 0 m 0 0 0 0 ⎥ ⎥
⎪
⎨ ü ⎪
y2 ⎬
⎢0 0 ⎥ ϕ̈
⎢ 0 0 0 0 ⎥ ⎪ z2 ⎪ +
⎢0 0 ⎥ ⎪ ü yF ⎪
⎢ 0 0 0 mF 0 ⎥⎪⎪
⎪
⎪
⎪
⎪
⎣0 0 0 0 0 F0 ⎦⎪ ⎪
⎪ ϕ̈ ⎪
zF ⎪
⎪
⎩ ⎭
0 0 0 0 0 0F ϕ̈xF
⎡1 ⎤⎧ ⎫
f
v 2
0 0 0 000 ⎪ ⎪ u̇ y1 ⎪
⎪
⎢ 0 1 f 1 e2 0 ⎪ ⎪
⎢ v
0 0 0 0⎥ ⎥
⎪
⎪
⎪ ϕ̇z1 ⎪⎪
⎪
⎢ 0
0
⎪ ⎪
⎢ 0 v f2 0 0 0 0 ⎥
1
⎥
⎪
⎨ u̇ y2 ⎪⎬
⎢ 0 0 v f 1 e02 0 0 0 ⎥⎥ ⎪ ϕ̇z2 ⎪ +
1
⎢ 0
⎢ 0 0 0 0 0⎥ ⎪ u̇ yF ⎪
⎢ 0 0 ⎥⎪ ⎪
⎪
⎪
⎪
⎪
⎣ 0 0 0 0 0 0 0⎦⎪ ⎪
⎪ ϕ̇ ⎪
zF ⎪
⎪
⎩ ⎭
0 0 0 0 000 ϕ̇xF
⎡ ⎤⎧ ⎫
cy − f2 0 0 −cy −cy b × ⎪ u y1 ⎪
⎢ λ 0 f 1 e2 cx 0 ⎪ ⎪
e
⎢ r0 0 0 −ex2 cx 0 ⎥ ⎥⎪
⎪
⎪ ϕz1 ⎪
⎪
⎪
⎢ 0
x
⎥ ⎪
⎪ ⎪
⎢ 0 cy − f 2 −cy cy b × ⎥ ⎨ ⎪ u y2 ⎪
⎪
⎬
⎢ 0 0 λ f 1 ex2 cx 0 −ex2 cx 0 ⎥
⎢ ⎥ ϕz2
⎢ −c 0 −cy 0 2cy 0 0 ⎥ ⎪
⎪ u yF ⎪
⎪
⎢ ⎥⎪ ⎪
⎥⎪ ⎪
y
⎢ 2ex2 cx ⎪
⎪ ϕzF ⎪⎪
⎣ −cy b −ex cx cy b −ex cx 0 +2b2 cy 0 ⎦ ⎪
2 2
⎪ ⎪
⎪
⎩ ⎭
× 0 × 0 0 0 × ϕ xF
=0. (11.5)
In Eq. (11.5), the last row and column of the matrix are not completed, but the
elements different from zero are only marked with crosses (×). It is not necessary to
introduce these values, since the next assumption is that the roll motion of the bogie
frame (rotation about the longitudinal axis) can be neglected.
This assumption is justified by looking at the root loci curves in Fig. 11.2. Curve
4, belonging to the roll motion, is far away from the curves of the kinematic motions
5, 6, and 7.
11.2.1 Coordinate Transformation to Introduce Generalized

Degrees of Freedom
The components of our displacement vector u have been up to now the displacements
and rotations of the two wheelsets and the bogie frame. The displacement and rotation
of the bogie frame are kept as degrees of freedom for the moment, while for the
wheelsets, new degrees of freedom are introduced, and with them the equations of
motion are formulated. The displacement states for the new wheelset degrees of
freedom are shown in Fig. 11.6.
The unknown amplitudes (degrees of freedom) of these new displacement states of
the wheelsets that are given in Fig. 11.6 are denoted by qi . They can be characterized
as follows:
q1 = 1: The lateral movements of both wheelsets have the same amplitude but are
out of phase.
q2 = 1: The yaw motions of the wheelsets have the same amplitudes but are out of
phase.
q3 = 1: The lateral movements of the wheelsets have the same amplitude and are in
phase.
q1 q2 q3 q4
1 1
1
1 1
Fig. 11.6 New generalized displacement states of the wheelsets (see Eq. (11.6))
q4 = 1: The yaw motions of the wheelsets have the same amplitudes and are in
phase.
The new degrees of freedom qi are linear combinations of the original unknown
displacements and rotations of the wheelsets. The introduction of generalized coor-
dinates is justified by the fact that the motions of the wheelsets in the eigenmodes of
the bogie with two wheelsets in Fig. 11.5 are dominant in the combinations chosen
above. There is a linear relation between the new generalized displacements qi and
the original displacements and rotations of the wheelsets; see Fig. 11.6. If we also
include the degrees of freedom of the bogie in the transformation, we can write
⎧ ⎫ ⎡ ⎤
⎪
⎪ u y1 ⎪
⎪ 1 0 1 0 0 0 ⎧ ⎫
⎪
⎪ ⎪ ⎢ ⎪
⎪ q1 ⎪
⎪
⎪ ϕz1 ⎪⎪
⎪ ⎢ 0 1 0 1 0 0⎥⎥⎪⎪
⎪
⎪
⎪
⎪ ⎪
⎪ ⎢ ⎥ ⎪
⎪ q2 ⎪
⎪
⎪
⎨ y2 ⎬ ⎢
u −1 0 1 0 0 0 ⎥ ⎨ q3 ⎬
ϕz2 = ⎢ ⎢ 0 −1 0 1 0 0 ⎥
⎥ ⎪ q4 ⎪ . (11.6)
⎪
⎪ ⎪
⎪ ⎢ ⎥ ⎪ ⎪
⎪ yF ⎪
⎪ u ⎪ ⎢ 0 0 0 0 1 0 ⎥⎪⎪ ⎪ ⎪
⎪
⎪ ⎪ ⎣ ⎪ q5 ⎪
⎪ ϕzF ⎪
⎪ ⎪
⎪ 0 0 0 0 0 1⎦⎪ ⎩ ⎪ ⎭
⎩ ⎭ q6
ϕxF 0 0 0 0 0 0
The lateral displacement of the bogie frame and the rotation of the frame about
the vertical bogie axis are kept as generalized degrees of freedom q5 and q6 . The
last equation includes that the roll motion of the bogie is neglected. Abbreviated,
Eq. (11.6) can be written as
u = T q. (11.6a)
This relation is replaced into the equations of motion (11.1):
M ü + D u̇ + Su = 0,
with the result

M T q̈ + DT q̇ + ST q = 0 . (11.7)
This equation is multiplied by the transpose T T of the transformation matrix T of

Eq. (11.6a) from the left. This additional matrix multiplication is done in order not to
destroy any symmetric properties of the matrices. With the abbreviations introduced
below, we have
M∗ = T T M T , (11.8a)
∗
D = T DT ,
T
(11.8b)
S∗ = T T ST (11.8c)
we can finally write

M ∗ q̈ + D∗ q̇ + S∗ q = 0 . (11.9)
The matrices M ∗ , D∗ , and S∗ have the dimensions 6 × 6. The system of equations

that we obtain after these matrix multiplications is Eq. (11.10).
⎡ ⎤
⎧ ⎫
⎢ 2m F ⎪
⎪ q̈1 ⎪
⎢ 0 0 0 00 ⎥ ⎥⎪ ⎪
⎪
⎪
⎢ 0 2zF ⎪ q̈2 ⎪
⎪
⎢ 0 0 00 ⎥ ⎪
⎥ ⎨ q̈3 ⎬
⎪
⎢ 0 0 2m F 0 00 ⎥ +
⎢ ⎥ ⎪ q̈4 ⎪
⎢ 0 0 ⎥ ⎪ ⎪
⎢ 0 0 2zF 0 ⎥⎪ ⎪ ⎪
⎪ q̈5 ⎪
⎪
⎣ 0 0 0 0 0 ⎦⎪
mF ⎩ ⎪ ⎭
q̈6
0 0 0 0 zF
0
⎡ ⎤
⎧ ⎫
⎢ 2 f2 0 0 0 0 0 ⎥⎪ ⎪ q̇1 ⎪
⎪ ⎪ ⎪
⎢ ⎥⎪ ⎪ ⎪
⎢ 0 0⎥ ⎪ 2⎪
q̇ ⎪
⎥ ⎨ q̇3 ⎬
2
⎢ 2 f 1 e0 0 0 0
1 ⎢ 0 ⎥
⎢ 0 2 f 2 0 0 0 ⎥ ⎪ q̇4 ⎪ + (11.10)
v ⎢ 0 ⎥ ⎪ ⎪
⎢ 0 0 2 f e
1 0
2
0 0 ⎥⎪ ⎪ ⎪ ⎪
⎪ q̇5 ⎪
⎣ 0 0 0 0 0 0⎦⎪ ⎩ ⎪ ⎭
q̇6
0 0 0 0 0 0
⎡ ⎤
⎧ ⎫
⎢ 2cy −2 f 2 0 −2cy b ⎥ ⎪
⎪ q1 ⎪
⎢ 0 0 ⎥ ⎪ ⎪ ⎪
⎢ 2λe e0 f 1 2ex2 cx ⎥⎪ ⎪ q ⎪
2⎪
⎢ r0
0 0 0 0 ⎥⎪ ⎨ ⎪ ⎬
⎢ −2 f 2 −2cy 0 ⎥ q3
⎢ 0 0 2cy ⎥ =0.
⎢ 0 −2ex2 cx ⎥ ⎪ q4 ⎪
⎢ 0 0 2λe er00 f 1 2ex2 cx ⎥⎪ ⎪
⎪
⎪
⎪
⎢ 0 ⎥ ⎪ q5 ⎪
⎪
⎣ 0 0 −2cy 0 2cy
2
⎦⎪ ⎩ ⎪
q6
⎭
2ex cx
−2cy b 0 0 −2ex cx 0
2
+2b2 cy
At this stage, it is not easy to see why Eq. (11.10) is easier to handle than Eq. (11.5).
Another transformation is necessary, which usually is called static condensation.
The last two equations of (11.10) are
m F q̈5 − cy q3 + cy q5 = 0 , (11.11a)
zF q̈6 − cy bq1 − ex2 cx q4 + (ex2 cx + b cy )q6 = 0 .
2
(11.11b)
Equation (11.11a) is the principle of linear momentum for the bogie frame, and
Eq. (11.11b) is the principle of angular momentum. In these two equations, the iner-
tia terms are neglected to eliminate degrees of freedom (q̈5 = 0, q̈6 = 0). Two
equilibrium conditions, one for lateral displacement and one for yaw rotation, are
formulated without the influence of the inertia terms. Then from Eq. (11.11a), the
generalized displacement q5 can be expressed by q3 , and with the help of Eq. (11.11b),
the generalized displacement q6 can be expressed by q1 and q4 :
q5 = q3 , (11.12a)
bcy ex2 cx
q6 = q 1 + q4 . (11.12b)
ex2 cx + b2 cy ex2 cx + b2 cy
The generalized degrees of freedom q5 and q6 are now no longer independent; the
lateral displacement and the yaw rotation of the bogie frame are coupled to the
motions of the wheelsets. Equation (11.12a) can be interpreted in a way that for a
displacement state q3 for a lateral displacement of the wheelsets in phase, the bogie
frame follows in the same way. The lateral displacement of wheelsets and bogie
frame can be regarded as rigid-body motion. Accordingly Eq. (11.12b) describes the
rotation of the bogie frame about the vertical axis for lateral displacements of the
wheelset out of phase (q1 ) or yaw rotations in phase (q4 ). No displacement or rotation
of the bogie frame belongs to displacement state q2 .
After the static condensation, only the first four out of the six generalized dis-
placements q1 to q6 are independent; q5 and q6 are dependent. This can again be
described by a transformation:
⎧ ⎫ ⎡ ⎤
⎪ q1 ⎪
⎪ ⎪ 1 0 0 0 ⎧ ⎫
⎪
⎪ ⎪
⎪ ⎢0
⎪ q
⎪ ⎬ 2 ⎪
⎪ ⎢ 1 0 0⎥⎥⎪⎪ q1red ⎪
⎪
⎨ ⎢0 ⎨ ⎬
q3 0 1 0⎥⎥ q2red ,
=⎢
⎢ (11.13)
⎪ q4 ⎪
⎪ ⎪ ⎢0 0 0 1⎥⎥⎪⎪ q3red ⎪
⎪
⎪
⎪ ⎪
⎪ ⎣0 ⎩ ⎭
⎪ q5 ⎪
⎪ ⎪ 0 1 0 ⎦ q4red
⎩ ⎭
q6 α 0 0 β
with the abbreviations

bcy
α= ,
ex2 cx + b2 cy
ex2 cx
β= .
ex2 cx + b2 cy
In symbolic form, for Eq. (11.13) we can write
q = T red q red . (11.13a)
Since for the transformation Eq. (11.13), where the number of degrees of freedom
was reduced by two, only static relations were used, it is justified to call this reduction
process static condensation. The equation of motion for the generalized displace-
ments, Eq. (11.9), is now treated with the transformation (11.13a) analogously to the
transformations above. The result is
T
T red M ∗ T red q̈ red + T red
T
D∗ T red q̇ red + T red
T
S∗ T red q red = 0 . (11.14)
Again abbreviations for the matrix products can be introduced, so that the equation
can be written as
M ∗red q̈ red + D∗red q̇ red + S∗red q red = 0 . (11.14a)
The matrices of the system of Eq. (11.14) are of size 4 × 4.

The mass and moment of inertia of the bogie frame have not disappeared from
the equations. The motions of the bogie frame are now coupled to the motions of
the wheelsets. The mass and moment of inertia of the bogie frame are added to the
masses and the moments of inertia of the wheelsets. With the formal transformation
applied, we do not need to think of how mass and moment of inertia ZF of the bogie
frame have to be distributed on the wheelsets.
We will now take a closer look at the matrices of the condensed system of
Eq. (11.14a). For the stiffness matrix, we get
⎡ ⎤
ex2 cx cy ex2 cx cy b
⎢ e2 cx +b2 cy − f 2 0 − e2 cx +b2 cy ⎥
⎢ x x ⎥
⎢ ⎥
⎢ λ e0 f e2 c ⎥
⎢ e r0 1 x x 0 0 ⎥
S∗red =⎢
⎢
⎥ .
⎥ (11.15)
⎢ − f2 ⎥
⎢ 0 0 0 ⎥
⎢ ⎥
⎣ e2 cx cy b ex2 cx cy b2 ⎦
− 2 0 λe er00 f 1 2
x
ex cx +b cy
2 ex cx +b cy
2
From this matrix, two interesting effects can be observed:

• Firstly, the third diagonal element is equal to zero, since a rigid-body motion of
wheelsets and bogie frame belongs to q3red .
• Secondly, we can conclude that the two primary stiffnesses cx and cy always appear
in a certain combination, which we shall denote by cb and cs :
cb = ex2 cx , (11.16a)
e2 cx cy
cs = 2 x . (11.16b)
ex cx + b2 cy
The term cb is called the bending stiffness and cs is called the shear stiffness of the
two wheelsets in the bogie frame.
The terms bending stiffness and shear stiffness are explained in Fig. 11.7. In
Fig. 11.7a, the undeformed bogie is shown. First, the two wheelsets will be “bent”
against each other like the cross sections of a beam (Fig. 11.7b). This causes displace-
ments in only the longitudinal springs cx , not in the lateral springs cy . The lateral
springs are placed somewhat angularly, as can be seen in the figure. The length
remains constant for small displacements. Therefore, no spring forces in the lateral
direction arise. The resistance of the longitudinal springs to such “bending” of the
wheelsets against each other is called bending stiffness.
In the next step, the wheelsets are “ sheared against each other,” again as in the
case of a beam where the end cross sections are given a transverse displacement
by application of a shear force. The corresponding displacement state can be seen
(a) (b) (c)
Fig. 11.7 Explanation of the terms bending and shear stiffness
in Fig. 11.7c. For a shear deformation, the bogie frame is also placed slightly at an
angle. This causes deformations both in the longitudinal springs cx and in the lateral
springs cy . The resistance of the bogies to such a “shear” deformation is called shear
stiffness. With the corresponding abbreviations, the reduced stiffness matrix can be
rewritten as ⎡ ⎤
⎢ cs − f2 0 −cs b ⎥
⎢ ⎥
⎢ ⎥
⎢ λ e0 f 0 ⎥
⎢ e r0 1 cb 0 ⎥
S∗red =⎢
⎢
⎥ .
⎥ (11.15a)
⎢ 0 − f2 ⎥
⎢ 0 0 ⎥
⎢ ⎥
⎣ ⎦
−cs b 0 λe er00 f 1 cs b2
For the reduced damping matrix, which includes only terms with creepages, we get
⎡ ⎤
⎢ f2 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 f e2 0 0 ⎥
⎢ 1 0 ⎥
D∗red =⎢
⎢
⎥ .
⎥ (11.15b)
⎢0 0 f2 0 ⎥
⎢ ⎥
⎢ ⎥
⎣ ⎦
2
0 0 0 f 1 e0
Finally, the reduced mass matrix can be written

⎡ ⎤
⎢m + α zF αβzF
1 2 1
0 0 ⎥
⎢ 2 2
⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0 0 ⎥
M ∗red =⎢
⎢
⎥ .
⎥ (11.15c)
⎢ 0 m+ 1 ⎥
⎢ 0 m
2 F
0 ⎥
⎢ ⎥
⎣ ⎦
1
2
αβzF 0 0 + 21 β 2 zF
Even though a number of simplifying assumptions were made, we still have four
degrees of freedom in the equations of motion. The corresponding characteristic
equation is an eighth-degree polynomial. It is impossible to find closed forms for the
roots of this characteristic polynomial. The eigenvalues would need to be calculated
numerically. Even if we restrict ourselves to a stability investigation, it is cumbersome
to define stability criteria. In Table 16.6 in Chap. 16, therefore only polynomials up to
the degree six are given.
This means that Eq. (11.14a) has to be simplified further. Before we do so, another
transformation is applied. We already stated that the third diagonal element is equal
to zero. This suggests that we further combine displacements q1red and q4red in a
way that results in another rigid-body motion with a rigid-body rotation of the bogie
frame together with both wheelsets. Also, here the primary springs are not deformed;
another zero has to appear on the diagonal axis of the stiffness matrix.
This transformation will be performed before further simplifications are intro-
duced. The displacements that are introduced are shown in Fig. 11.8.
The transformation matrix with which the reduced displacement vector qred is
calculated from a new displacement vector q red has the following form:
q 1 red q 2 red q 3 red q 4 red
1 1
1 1
Fig. 11.8 New displacements

⎧ ⎫ ⎡ ⎤⎧ ⎫
⎪
⎪ q1red ⎪
⎪ 1 0 01 ⎪ ⎪ q 1red ⎪
⎪
⎨ ⎬ ⎢ ⎨ ⎬
q2red 0 1 0 0⎥ ⎥ q 2red .
=⎢
⎣0 (11.17)
⎪
⎪ q3red ⎪
⎪ 0 1 0⎦⎪ ⎪ q ⎪
⎩ ⎭ ⎩ 3red ⎪ ⎭
q4red 0 0 0 b1 q 4red
The transformation itself is routine. A new system of equations results:
M red q̈ red + Dred q̇ red + Sred q red = 0 . (11.18)
What do the matrices of this equation system look like? The stiffness matrix consists
of two parts. In one part, the properties of the primary suspension are described;
the other part contains creepage terms. As expected, we find another zero diagonal
element in the matrix of the spring stiffnesses. Further bending and shear stiffness are
nicely separated now and appear only in one diagonal element of the spring stiffness
matrix: ⎡ ⎤
cs 0 0 0
⎢ 0 cb 0 0 ⎥
SFred = ⎢⎣ 0 0 0 0⎦ .
⎥ (11.19a)
0 0 00
⎡ ⎤
⎢ 0 − f 2 λe er00 f1 ⎥ 0
⎢ ⎥
⎢ ⎥
⎢ λ e0 f λe r 0 f 1 ⎥
e0
⎢ e r0 1 0 0 ⎥
SCred =⎢
⎢
⎥.
⎥ (11.19b)
⎢ 0 − b f2 ⎥
1
⎢ 0 0 ⎥
⎢ ⎥
⎣ ⎦
0 0 λe r 0 b f 1 0
e0
The damping matrix has become more complicated, though. It is no longer a diagonal
matrix: ⎡ ⎤
⎢ f2 0 0 f2 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 f e2 ⎥
1 1⎢ 1 0 0 0 ⎥
Dred = ⎢ ⎥. (11.19c)
v v⎢
⎢0 0
⎥
⎥
⎢ f2 0 ⎥
⎢ ⎥
⎣ ⎦
e2
f2 0 0 f 2 + f 1 b02
The mass matrix can be written as

⎡ ⎤
⎢m + α zF m+ αzF
1 2 1
0 0 ⎥
⎢ 2 2b
⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0 0 ⎥
M red =⎢
⎢
⎥,
⎥ (11.19d)
⎢ 0 m+ 1 ⎥
⎢ 0 m
2 F
0 ⎥
⎢ ⎥
⎣ ⎦
m + 2b
1
αzF 0 0 1
b2
+ mb2 + 21 zF
with the abbreviation

bcy
α= .
ex2 cx + b2 cy
If then the parameter α is replaced by the relationship between bending stiffness and
shear stiffness,
bcs
α= ,
cb
the mass matrix can be expressed as

⎡ ⎤
2
b 2 cs
⎢m + m+
cs
2 c2 zF
0 0 2cb zF ⎥
⎢ b ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0 0 ⎥
M red =⎢
⎢
⎥.
⎥ (11.19e)
⎢ ⎥
⎢ 0 0 m+ 1
m
2 F
0 ⎥
⎢ ⎥
⎣ ⎦
m + 2ccsb zF 0 0 1
b2
+ mb2 + 21 zF
The third diagonal element of the mass matrix is half the total mass m B of the bogie
and the fourth diagonal element is, apart from the prefactor (1/b2 ), half the total
moment of inertia B of the bogie. In the following, the abbreviations
m B = 2m + m F , B = 2 + 2mb2 + zF
will be used for the sake of simplicity.
11.2.2 Bogie with Infinitely High Bending

and Shear Stiffness
First we look at the case in which wheelsets are rigidly connected to the bogie
frame. In our system of equations of motion, Eq. (11.18), with the matrices from
Eqs. (11.19a)–(11.19d) this special case is achieved by choosing infinite values for
the bending stiffness cb and the shear stiffness cs . The number of degrees of freedom
∗ ∗
is reduced, and only q3red and q4red remain. This can be easily understood by looking
∗ ∗
at Fig. 11.8. The displacements belonging to q1red and q2red are not present for infinite
bending and shear stiffness. The resulting equation for calculating the eigenvalues is
1
m
2 B
0 λ f2 0 0 − b1 f 2 q3
λ 2
+ e02 +
0 1
2b 2 B v 0 f 2 + 2 f 1 λ e
1 e0
b r0
f 1 0 q 4
b

0
= . (11.20)
0
Formally, this is the same task as for the single wheelset; cf. Eq. (9.13). All results
from the single wheelset therefore directly apply to the bogie with infinitely high
bending and shear stiffness.
First, we are going to determine the relationship corresponding to the Klingel
equation. Again we have to look at low speed, neglect the inertial forces, and use
λ = iω for the eigenvalue. Then the angular frequency is calculated to be
ω 2 λe
= , (11.21a)
v e0 r0 1 + f 2 b2
f 1 e02
or with 2
2π ω 2
= ,
L v
the wavelength of the wheelset hunting motion becomes

f 2 b2
2 e0 r 0
L = (2π)
2
1+ . (11.21b)
λe f 1 e02
Also, for the rigid bogie, for low speeds there is a linear relationship between fre-
quency and vehicle speed. The proportionality factor is similar to the one in the
Klingel equation for the single wheelset, Eq. (8.14).
The relationship for the wavelength is similar, too. In the equation, however, an
additional term appears:
f 2 b2
1+ ,
f 1 e02
which always is greater than 1. For the frequency, Eq. (11.21a), the same term is in the
denominator. Hence the wavelength of a rigid free rolling bogie is always larger than
that of a free rolling single wheelset at the same speed. The hunting frequency, on
the other hand, is smaller. The kinematic relations for the bogie were first described
by Heumann [14].
Fig. 11.9 Definition of

lateral and yaw stiffness
between bogie frame and
carbody
c^ c
The eigenbehavior of the bogie with infinitely high bending and shear stiffness
is now investigated by also taking the secondary suspension into account. We still
assume that the terms from the gravitational stiffness and the term proportional to
C23 can be neglected. The bogie frame is connected to a track-following carbody via
a secondary lateral spring stiffness c and a secondary yaw stiffness ĉ (cf. Fig. 11.9).
Both stiffnesses have to appear as additional diagonal terms in the stiffness matrix.
Starting from the principle of impulse and momentum for the whole bogie, it is
obvious that prefactors are necessary:
1
m
2 B
0 λ f2 0
λ2 + e2
0 1

2b2 B v 0 f 2 + b02 f 1

c
− b1 f 2 q3 0
+ 2 = . (11.22)
λe b1 er00 f 1 2bĉ 2 q4 0
The secondary lateral stiffness c and the secondary yaw stiffness ĉ will probably be
high enough to justify neglecting the effects from the gravitational stiffness and from
the term C23 of both wheelsets. Starting with Eq. (11.22), a stability investigation for
the bogie is carried out. First, the low-speed case (v 0) is again treated, where
the inertia terms can be disregarded. As equation for the eigenvalues, after some
rearrangement, we obtain
2
λ f 2 b2 λ ĉ f 2 b2 c
1+ + + 1+
v f 1 e02 v 2 f 1 e02 f 1 e02 2 f 2

c ĉ λe
+ + = 0 . (11.23)
2 f 2 2 f 1 e02 e0 r0
Similarly to the case of the single wheelset, we get
λ1,2 = αiω , (11.24)
where the expressions for the real part and the imaginary part of the eigenvalues are
⎡ ⎤
v⎣ c 1 ĉ ⎦
α = −δ = − + 2
(11.25a)
2 2 f2 1 + f2 b2 2 f 1 e0
2
f 1 e0
and
! # $2
! ĉ 2
− 1 + ff2 be2 2cf2
! λe
ω = v!
2 f 1 e02
− .
1 0
" f 2 b2
2 (11.25b)
e0 r0 1 + f 1 e02
f 2 b2
4 1 + f e2
1 0
From Eq. (11.25a), it can be seen that with the assumptions made here, the oscillations
of a bogie without secondary suspension (c = 0, ĉ = 0) are always undamped. If
the gravitational stiffness and the C23 -term of the wheelsets are taken into account,
the oscillations can even become unstable. The secondary suspension is usually stiff
enough to compensate for destabilizing effects from the wheelsets. If realistic values
for the secondary suspension are assumed, the motion is alway damped.
If one of the stiffnesses is large enough, the oscillation is aperiodically damped
at low speeds. In this case, the second term in the root expression dominates.
What is the resulting critical speed for the rigidly suspended bogie based on
Eq. (11.22)? Since the structure of the equation is exactly the same as the equation
of motion for the single wheelset, we can copy the solution from Eq. (10.39):

f 2 b2
e0 r0 1 + f 1 e02 1
v crit
2
= ωcrit
2
, (11.26)
λe 1−k
i.e., apart from a correction factor k, a very formal coincidence with the Heumann
equation for the single wheelset, Eq. (11.21a), is obtained. Again the frequency at
the stability limit can be calculated:
2

ĉ + cb2 1 + ff2 be2
ωcrit
2
= .
1 0
2
(11.27)
B + m B b 1 + ff2 be2
2
1 0
The expression for the correction factor k remains to be given. We get

⎡ ⎤2
2
e0 r0 f 2 b2 c 1 − c
ĉm B
k= 1+ ⎣ B
⎦ . (11.28)
λe f 1 e02 2 f2 2
1 + bmBB 1 + f 2 b2
f 1 e02
Interpretation of the equations gives the same results as for the suspended single
wheelset. If a guided carbody is assumed, the critical speed can become infinite by
choosing either the secondary lateral stiffness or the yaw stiffness high enough. The
concept of friction yaw damping is a step in this direction. Unfortunately, we do not
get rid of all difficulties: The next step would be to investigate the eigenbehavior of
a whole vehicle, whereby degrees of freedom are assigned also to the carbody.
Without doubt, a better solution would result if a stability increase were possible
by measures in the primary suspension. Is it possible to increase the critical speed
by an optimal choice of bending and shear stiffness? With regard to this question,
in the following section, two special cases will be investigated, i.e., infinitely high
bending stiffness and infinitely high shear stiffness.
11.2.3 Realization of Infinitely High Bending

Stiffness or Shear Stiffness
First, we will investigate whether it is possible to realize infinitely high bending

stiffness or shear stiffness at all. The first case, of infinitely high bending stiffness
cb , is achieved if the primary longitudinal stiffness cx is one or several orders of
magnitude greater than the primary lateral stiffness cy . This is the case if the wheelsets
in the longitudinal direction are coupled to the bogie frame with links (bogies of type
Minden-Deutz).
An infinitely high shear stiffness cs cannot be achieved by choosing one of the
primary stiffnesses to be infinitely high. For the bogie in Fig. 11.7a, it is impossible
to achieve an infinitely high shear stiffness together with a finite bending stiffness. To
achieve this combination, a different bogie design is necessary, i.e., a cross-coupled
bogie. In such a bogie, in addition to the primary suspension between the wheelsets
and the bogie, a coupling between the two wheelsets is introduced, as indicated in
Fig. 11.10.
In addition to the coupling in the lateral direction, a yaw coupling could also be
introduced. The calculation of the corresponding bending and shear stiffnesses will
be performed in an exercise.
As soon as the bending and shear stiffnesses of the cross-coupled bogie are known,
the eigenbehavior of the conventional bogie in Fig. 11.7a and that of the cross-coupled
bogie, Fig. 11.10, can be described by the same system of equations of motion with
the matrices (11.15).
Fig. 11.10 Schematic view

of a cross-coupled bogie
11.2.4 Bogies with Infinite Shear Stiffness
We begin with a bogie with infinite shear stiffness cs . The equations of motion in
matrix form are obtained by letting the value cs in the matrices of Eq. (11.19a) become
∗
infinitely large. The displacement q1red then vanishes, and the remaining equations
for the eigenvalues are
⎧ ⎡ ⎤ ⎡ ⎤
⎨ 0 0 f 1 e02 0 0
λ
λ2 ⎣ 0 21 m B 0 ⎦ + ⎣ 0 f2 0 ⎦+
⎩ v e02
0 0 1
2b 2 B 0 0 f 2 + b2 1
f
⎡ λe e0 ⎤⎫ ⎧ ⎫ ⎧ ⎫
cb 0 r0
f 1 ⎬ ⎨ q2 ⎬ ⎨ 0 ⎬
+⎣ 0 0 − f 2 /b ⎦ q3 = 0 . (11.29)
⎭⎩ ⎭ ⎩ ⎭
− f 2 λbe er00 f 1 0 q4 0
We shall now investigate the eigenvalues for high speeds (v → ∞). In this case, the v1 -
proportional matrix from Eq. (11.29) can be omitted. As equation for the eigenvalues,
one still has

m B B 4 m B B λe e0 m B λe e0
λ6 + λ cb + λ 2
f f
1 2 + f f
1 2
2 2b2 2 2b2 r0 b2 2 r0
λe e0
+ cb f 1 f 2 = 0 . (11.30)
r0 b2
Keizer [7, 8] dealt intensively with the question whether it is possible to achieve
infinite critical speed for a bogie. He begins with a dimensionless equation for the
eigenvalues:
m m e02

F F F
λ6K +1 + +1 +
2m m F e02 m F b2 2m me02 b2
cb m F m e02

F F
+ λ4K 2 +1 + + 1 +
αK 2m m F b2 2m me02 b2

mF e02 cb e02
+ λ2K + + 1 + = 0, (11.31)
2m 2
me0 b2 αK2 b2
where the following abbreviations are used:

%
λe0 λ
αK
4
= f 1 f 2 e02 and λK = me02 . (11.32a,b)
r0 α
For the moment of inertia of the single wheelset and the bogie frame, he uses the
approximations
me02 and F m F b 2 . (11.33a,b)
Three dimensionless parameters then remain in Eq. (11.31):

• the relationship μ between half the frame mass m F /2 and the wheelset mass m;
• the relationship β between the axle distance b and gauge width e0 ;
• a dimensionless parameter K b , which relates the bending stiffness cb to the contact
stiffnesses:
mF
= μ, (11.34a)
2m
b
= β and (11.34b)
e0
cb
= Kb . (11.34c)
αK2
Now the characteristic equation for the eigenvalues can be written

λ6K (1 + μ) 1 + β 2 (1 + μ) + λ4K K b (1 + μ) 1 + β 2 (1 + μ) +

+λ2K 1 + β 2 (1 + μ) + K b = 0 . (11.35)
Fig. 11.11 Critical speed as Kb

a function of the bending
10
stiffness parameter and the
relation b/e0 (according to Vcrit = 0.5
Keizer [7]) 8
6
Vcrit = 1.0
4
Vcrit
2
0 1 2 3 4 5 = b/e0
In addition, Keizer chooses μ = 1. The result is shown the diagram in Fig. 11.11,
taken from [7], with the critical speed as a function of K b and β.
At the border of the cross-hatched area, the critical speed becomes infinitely high.
For parameter values inside the area, a critical speed no longer exists. Further, the
diagram contains curves of constant critical speed, which, of course, do not result
from the characteristic equation (11.35) but from the characteristic equation of the
system from Eq. (11.29). Keizer also uses a dimensionless speed parameter
m
V 2 = v2 % . (11.36)
f 1 f 2 r0λe0
For a suitable choice of K b , the critical speed vanishes for the relationship
β 2.5. The corresponding axle distance, however, is very large. The dimensionless
bending stiffness K b can be recalculated as a longitudinal stiffness with the help of
Eqs. (11.34c) and (11.16a):
&
Kb λe03
cx = 2 f1 f2 . (11.37)
ex r0
If we now put 1.5 as the minimum value for the dimensionless bending stiffness K b
from Fig. 11.11 and apply all other values from Table 9.2, we get
cx 1.5 · 107 N/m .
This value is lower than the longitudinal stiffness of the Minden-Deutz link (cx =
5 · 107 N/m) but still relatively high.
To illustrate this in Fig. 11.12, a sketch of a bogie with the above parameters is
shown.
Fig. 11.12 Dimensions of a

bogie with infinitely high c y arbitrary
critical speed
c x = 0.75 . 10 N/m
1 7
2
b = 1.875 m
e x =0.95 m e0 =0.75 m
For interpretation of the diagram in Fig. 11.11, it is interesting to know what is

the real speed that belongs to the dimensionless speed parameter Vcrit = 1. We have
to solve Eq. (11.36) for v,
'
1 r0 e0
v crit
2
= 2
Vcrit f1 f2 , (11.38)
m λ
and obtain, with the corresponding values from Table 9.2,
m km
v crit = 156 = 560 .
s h
For realistic operation of rail vehicles, this speed is sufficient. The axle distance
could even be reduced, though the same minimum longitudinal stiffness must be
kept. The main problem is probably that infinite shear stiffness together with the
desired bending stiffness are impossible to realize.
11.2.5 Bogies with Infinite Bending Stiffness
The investigation of bogies with infinite bending stiffness is performed in a similar

way as for bogies with infinite shear stiffness. Again for the case of infinitely high
speed we get a dimensionless bicubic characteristic equation; cf. Keizer [7]. It turns
out that in this case, it is impossible to choose parameters b und K s in such a way that
only roots with negative real part arise. With infinite bending stiffness it is impossible
to achieve infinitely high critical speed. From solving the characteristic equation for
finite critical speeds v c , respectively the dimensionless speed Vc , we get the diagram
in Fig. 11.13.
Fig. 11.13 Line of equal Ks

critical speed for a bogie
10
with infinite bending
stiffness. Dependency of
critical speed on the shear 8 Vcrit = 0.5
stiffness parameter K s and
the relation b/e0 (according 6
to Keizer [7]) Vcrit = 1.0
4
0 1 2 3 4 5 = b /e 0
As above, we are interested in the axle distance and shear stiffness that have to be
chosen to achieve at least a dimensionless critical speed Vc = 1 (v = 156 ms ). From
Fig. 11.13, we can obtain β = 4.8, i.e., a very large axle distance of 7.20 m, as well
as K s = 4.8, i.e., a shear stiffness of cs = 0.32 · 107 m
N
.
Neither bogies with infinite bending stiffness nor bogies with infinite shear stiff-
ness are realistic from the design point of view. A comparison of the curves in
Figs. 11.11 and 11.13 shows, however, that it is better to increase the shear stiff-
ness and not the bending stiffness to achieve a high critical speed. One of the most
important reasons is that the necessary axle distances remain realistic. Further, a high
bending stiffness prevents good curving performance, cf. Chap. 14, while a high shear
stiffness influences the curving behavior very little.
From Figs. 11.11 and 11.13, we can also conclude that it is inadvisable to choose
both stiffnesses very high. If for infinite shear stiffness the bending stiffness is
increased or for infinite bending stiffness the shear stiffness is increased, the critical
speed first increases in both cases for constant b. For further increase of the stiffness
values, the critical speed again decreases.
11.2.6 Bogies with Finite Bending and Shear Stiffness
How bogies with finite bending and shear stiffness behave can be investigated only
numerically. A number of general and special conclusions can be found in the work
of Rocard, Wickens, and Scheffel [4, 5, 9–13]. Numerical investigations have been
performed, for example, by Kik [15].
v crit
cb cs
N/m
1 10 7
72 m/s 1 10 7 Nm / rad
2 10 7 2 10 7
3 10 7 3 10 7
4 10 7 4 10 7
5 10 7
Fig. 11.14 Qualitative lines of constant critical speed of a bogie as a function of cb and cs . Solutions
in the cross-hatched area are possible with conventional bogies (Calculation performed by Walter
Kik.)
The results can be illustrated in different ways. In Fig. 11.14, the critical speed is
shown as a function of bending stiffness cb (unit Nm/rad) and shear stiffness cs (unit
N/m). Both axes have linear scale.2
The calculation is done with some simplifying assumptions:
1. There is no secondary suspension, i.e., a single massless bogie or a two-axle
vehicle is investigated.
2. Wheelsets and bogie frame are regarded as massless.
3. Neither in the primary suspension nor in the cross-coupling is there damping.
4. The parameters for the contact geometry are the result of a quasilinearization.
As amplitude, 4 mm was chosen. S1002 wheel profiles and UIC 60 rail profiles
and normal gauge with inclination 1 in 40 was chosen. For the calculation, the
profiles were assumed to be elastic. For an axle load Q = 63.7 kN, this results
in the following quantities:
• equivalent conicity 0.161;
• roll angle parameter 0.0451;
• contact angle parameter 9.53.
5. For contact mechanics, the Cik -coefficients were not reduced.
2 Many thanks to Mr. Walter Kik, who prepared the results for Fig. 11.14 with the multibody simu-
lation program M EDYNA.
6. The lateral distance between contact points and the lateral distance between
attachment points of the primary longitudinal springs were equal:
2e0 = 2ex = 1.50 m .
These assumptions are justified because it is the only way to get general results.
In Fig. 11.14, one can see that the values of critical speed shown above the (cb , cs )-
plane form two mountain ridges almost parallel to the coordinate axes. For more
or less constant shear stiffness, one ridge extends into the area of infinite bending
stiffness; the other ridge has almost constant bending stiffness and extends into
areas with very high shear stiffness. The mountain can be characterized by its ridge
crown. The calculations show that the crown approaches a constant maximum for
increasing bending and shear stiffness. The maximum value for high shear stiffness
is only slightly larger than for high bending stiffness. In contrast to Fig. 11.11, for
increased shear stiffness the critical speed here is still finite. The reason for this is that
in the calculation, a value b/e0 = 0.751.35
was assumed, which is below the minimum
axle distance needed for infinitely high critical speed.
It can further be seen that the (cb , cs )-plane is divided into two parts by a dashed
line. The dashed line marks the maximum shear stiffness that is possible to achieve
with conventional bogies for a given bending stiffness. From Eqs. (11.16a) and
(11.16b) for this straight line we get the relationship (e0 ex )
cs = cb /b2 . (11.39)
Combinations of bending and shear stiffness in the not cross-hatched area are possible
only with cross-coupled bogies, i.e., bogies with a direct coupling between the two
wheelsets.
The stability mountain can vary depending on the size of the bogie (parameter
b/e0 ), though qualitatively, it always looks the same. Conventional and cross-coupled
bogies can thus be characterized by their cb -cs -values. For large cs -values, a some-
what higher critical speed is possible than for large cb -values. Looking at the bogies
designed for the German ICE 1 high-speed trains, one can conclude that they make
use of high cb -values. Bogies with high cs -values were, for example, developed by
Scheffel in South Africa [12]. Such bogies usually have better quasistatic curving
performance; cf. Chap. 14.
11.3.1 Equations of Motion of a Bogie
The bogie consists of three bodies. As long as all motions are considered and no
constraints exist, the displacement vectors can be defined as follows:
Frame: ( )
u F = u xF , u yF , u zF , ϕxF , ϕyF , ϕzF
k-th Wheelset: ( )
uk = u xk , u yk , u zk , ϕxk , ϕyk , ϕzk .
What degrees of freedom have to be introduced for a description of the lateral

running behavior of a bogie with two wheelsets if the constraints between wheel and
rail are considered? Show where there are nonzero elements in the mass, damping,
and stiffness matrix. In the displacement vector, place first the degrees of freedom of
the first wheelset, then those of the second wheelset, and finally those of the bogie
frame (index F).
11.3.2 Equations of Motion of a Freely Rolling

Wheelset at Low Speed
Analogous to the procedure for the wheelset, we shall investigate the natural oscil-
lations of a bogie without primary suspension at low speeds. Inertia effects can be
neglected. What do the oscillations look like if the terms from gravitational stiffness
and the C23 -term in the stiffness matrix are neglected? Give the equations of motion
of the case in which both effects are taken into account. Is the oscillation damped at
low speeds?
11.3.3 Equations for Bending and Shear Stiffness
For the cross-coupled bogie in Fig. 11.10, give the formulas for calculation of bending
and shear stiffness. The corresponding formulas without cross-coupling can be found
in Eqs. (11.16a) and (11.16b).
How are the equations changed if the cross-coupling includes a rotational stiffness
ĉ in addition to the lateral stiffness c?
References
1. T. Matsudaira, Dynamics of high speed rolling stock, in Quarterly Report of Railway Technical
Research Institute, JNR, Special Issue (1963), pp. 21–27
3. A.H. Wickens, The dynamic stability of simplified four-wheeled vehicles having conical
of lateral stability. Proc. Inst. Mech. Eng. 180(3F), 1–16 (1965)
6. R. Joly, Untersuchungen der Querstabiliät eines Eisenbahnfahrzeuges bei höheren
Geschwindigkeiten (Investigation of lateral stability of a railway vehicle at higher speeds).
Rail International - Schienen der Welt, 3, 168–204 (1972)
7. C.P. Keizer, Recent calculations on the hunting motions of railway vehicles running on 4
wheeled bogies. Int. J. Veh. Mech. Mobil. 4, 156–159 (1975)
8. C.P. Keizer, Some basic problems of bogies, an analytical approach. Veh. Syst. Dyn. 8, 359–406
(1979)
9. Y. Rocard, La Stabilité de Route des Locomotives, Première Partie, avec une note de M.R. Lévy
(Hermann & Cie, Paris, 1935) (Teil II siehe [10])
10. M. Julien, Y. Rocard, La stabilité de route des locomotives, deuxième partie (Hermann & Cie,
Paris, 1935)
11. A.H. Wickens, Steering and dynamics stability of railway vehicles. Veh. Syst. Dyn. 5, 15–46
(1975)
12. H. Scheffel, A new design approach for railway vehicle suspension. Rail Int. 10, 638–651
(1974)
13. H. Scheffel, The hunting stability and curving stability of railway vehicles. Rail Int. 5, 154–176
(1974)
14. H. Heumann, Lauf von Eisenbahnfahrzeugen mit zwei ohne Spiel gelagerten Radsätzen
beliebiger Belastung in der Geraden (Running of arbitrarily loaded railway vehicles on straight
track with two wheelsets supported without clearance). Organ Fortschr. Eisenb.-wes. 95, 43–54,
60–61 (1940)
15. W. Kik, Numerische Untersuchungen laufstabilisierender Maßnahmen an Eisenbahnfahrzeu-
gen (Numerical investigations of stabilizing measures for rail vehicles). ILR-Bericht 46, Institut
für Luft- und Raumfahrt, Berlin (1979)
Chapter 12
Lateral Eigenbehavior and Stability
of Bogie Vehicles
The investigation of the eigenbehavior and the stability of bogie vehicles, taking into
account all degrees of freedom, is performed with the same numerical procedures as
for a wheelset or a bogie. For presentation of the results, root loci curves and stability
cards are used, which are derived from the root loci curves. In the following, first a
characteristic result for a real vehicle is given. Thereafter, more general conclusions
based on the literature are discussed.
12.1 Stability of a Train with Two Vehicles
In 1975, at the TU Berlin Department of Aeronautics and Astronautics, the eigenbe-

havior of a metro train of the Hamburger Hochbahn (DT 2.5) was investigated [1].
The mechanical model is given in Fig. 12.1. It is cumbersome to derive the equations
of motion of such a system by hand, and so existing multibody simulation codes
were used. There are several packages that also include special wheel–rail modules.
Examples are Medyna and Simpack from Germany, Adams Rail from the USA
and Germany, Vampire from the UK, Nucars from the USA, Voco from France,
and Gensys from Sweden. For the following calculations, the program Linda [2],
a predecessor to Medyna, was used.
To investigate the lateral dynamics, the vehicle is modeled with 27 degrees of
freedom (seven for each bogie including the wheelsets and three each for u y , ϕx , ϕz
for the carbodies). The root loci curves calculated can be seen in Fig. 12.2. The area of
low frequencies and low damping is of special interest in a stability investigation. This
area was therefore magnified. One can observe that there are three root loci curves
that begin at the origin of the coordinate system at a speed of v = 0. These are the
curves that belong to the sinusoidal motions of the three bogies. All three curves
first increase damping with increased speed. The motions become more stable. Two
of the three root loci curves, however, change direction to lower damping values at
relatively low speeds, i.e., the stability decreases. At 16.1 and 17.1 m/s, the damping
is zero, and the running behavior becomes unstable. What happens?
DOI 10.1007/978-3-319-45376-7_12
226 12 Lateral Eigenbehaviour and Stability of Bogie Vehicles
101 102
x
601 509 511 512 602 510

533 103 534 537 105 538 536 104 535
301 302 303 306 305 304
508 516 515 507

528 532 531 552 551 543 544 540
504 527 520 519 539 503
501 526 x 517 518 538 502
525 529 530 549 553 554 550 542 541 537
505 513 514 506
y
105 rigid body element
301 wheelset element
512 spring element
suspension elements
602 damper element
Fig. 12.1 Mechanical model of a rail vehicle (DT 2.5 of Hamburger Hochbahn. Numbering of
elements separated into rigid-body elements, wheelset elements, spring and damper elements.)
2 eigenmodes
2 eigenmodes
2 eigenmodes
2 eigenmodes
Fig. 12.2 Root loci curves of the model in Fig. 12.1
To explain this phenomenon, we look at Fig. 12.3. On the right-hand side

(Fig. 12.3c), the magnified portion from Fig. 12.2 is shown again. On the left-hand
side, the results of two auxiliary calculations are given. In Fig. 12.3a, the carbodies
are fixed. They represent an inertial system that travels with speed v. In this case,
in the interesting part of the complex plane, only root loci curves for the sinusoidal
12.1 Stability of a Train with Two Vehicles 227
(a) [ Hz [ (b) [ Hz[ (c) [ Hz [

2.4 2 carbody 2.4 2.4
2 bogie
hunting modes 2.0 sway modes 2.0 2.0
1.6 1.6 1.6

1.2 1.2 1.2
0.8 0.8 0.8

intermediate
bogie 0.4 0.4 0.4
0 0 0
-1.4 -1.0 -0.6 -0.2 0.2 0.6 [ Hz[ -1.4 -1.0 -0.6 -0.2 0.2 [ H z [ -1.4 -1.0 -0.6 -0.2 0.2 0.6 [ H z[
carbodies fixed wheelsets fixed
Fig. 12.3 Root loci curves for submodels and comparison with curves from the whole vehicle
modes of the bogies can be observed. The curves for the leading and trailing bogies
are the same, since they are identical. In Fig. 12.3b, the wheelsets are fixed. For the
calculation of eigenvalues, only the degrees of freedom of the carbodies and the bogie
frames exist. The resulting root loci curves become isolated points in the complex
plane, since the vehicle speed no longer influences the result. Two of the dots are in
the parts of the diagram we can see in the figure. The corresponding mode shapes
belong to the carbodies.
Figure 12.3c again shows the superpositions that we get for the whole system.
Obviously, there is a coupling between carbody sway and the sinusoidal motions
of the leading and trailing bogies. These mode shapes are less damped than the
sinusoidal motions of the intermediate bogie. Altogether, we get four root loci curves
that represent characteristic motions of the leading and trailing bogies. These four
curves seem to merge. All four curves have intersections with the ω-axis, namely—as
a numerical calculation shows—at about 16 and 17 m/s as well as between 25 and
26 m/s. The stability limit therefore is at 16 m/s.
In Fig. 12.4, the mode shapes that belong to the critical eigenvalues at a speed of 18
m/s are shown. In the upper part, there is a snapshot of the first critical mode shapes,
while in the lower part, the motion of the carbodies is sketched for a sequence of
time steps. The carbody rotations about the longitudinal axis are indicated by means
of double arrows. The movements of the bogie are not shown for the sake of clarity.
The lateral displacements of the two bogies are more or less in phase with the lateral
displacements of the corresponding connection points at the carbodies.
Let us return to Fig. 12.3. From Fig. 12.3c, we conclude that the instability arises at
a frequency near the eigenfrequency of the carbody sway. If such motion patterns are
observed in reality, one is likely to interpret them as forced vibrations, whereby the
carbody sway is “excited” by bogie hunting. Since in reality, we observe a self-excited
vibration, this interpretation is wrong. In general, the qualitative results using this
explanation are correct, although a quantitative assessment is impossible. Since the
1. critical eigenmode
z
v x
y
1. critical eigenmode 2. critical eigenmode
0° 0°
60° 50°
90° 80°
140° 95°
190° 120°
250° 180°
270° 260°
320° 300°
Fig. 12.4 Mode shapes (carbody motions) corresponding to the two critical eigenvalues of the
model in Fig. 12.1 (the relative lengths of the arrows that symbolize the rotation of the bodies do
not match the size of the angles)
interpretation is very descriptive, it was, at least by the vehicle industry in Germany,

used to explain large lateral vibrations of rail vehicles. The interpretation dates back
to [3].
12.2 General Conclusions Regarding the Stability

of a Bogie Vehicle According to Matsudaira [4]
General conclusions for a bogie vehicle are of course difficult, due to the large num-
ber of parameters. One of the most promising attempts was made by the Japanese
researcher Matsudaira [4] in 1965. He made his conclusions in connection with
12.2 General Conclusions Regarding the Stability … 229
design studies on a bogie for a new Tokaido line. Another reason why the results
are interesting is that the concept of friction yaw damping was used. In the cited
publication, both tests on a roller rig (with a 1:5 scale model), on-track tests on a test
track, and simulation results are discussed. As a result of the tests, Matsudaira intro-
duced the terms carbody hunting or primary hunting and bogie hunting or secondary
hunting. The terms refer to the dominating motion. For the example in Fig. 12.4, the
separation is difficult; carbody and bogie hunting are coupled. Here we focus on the
theoretical part of the study. For details on the experimental parts, we refer to [4].
The following text is a shortened version of Masudaira’s publication. The theo-
retical background is the same as in Chaps. 3 and 8–11. Matsudaira also includes
nonlinear calculations with an analog simulation.
12.2.1 Theory and Simulation of Bogie Vehicle Hunting
A large number of stability simulations were performed since the start of the design
phase of a prototype bogie of a vehicle on the new Tokaido line (1957–1961). Some
of these calculations are briefly shown below.
As in this book, Matsudaira determines the eigenvalues λi = αi + iωi of the
equations of motion and investigates whether all real parts αi remain negative. The
imaginary part ωi again is the angular frequency, while the real part αi represents
the amount of damping of the oscillation. Damping can be negative or positive. The
stability of the system is known when the roots of the characteristic equation are
known.
To simplify the calculation, Matsudaira assumes that the wheelset axles are rigidly
connected to the bogie frame in both the longitudinal and lateral directions. The
system can then be described with seven degrees of freedom, i.e., lateral displacement
roll and yaw motion of the carbody and lateral displacement and yaw motion of the
bogies. The numerical parameters for the calculation were measured at or calculated
for a prototype bogie. The roots of the characteristic equation were determined for
each speed that results in a stability investigation of the vehicle.
The results are exemplified in Fig. 12.5. The values αi and ωi are shown depending
on the vehicle speed. Even though the characteristic equation has two times seven
roots, in Fig. 12.5 only five (complex conjugate roots) are shown. The others are
omitted because they have only negative real parts and thus do not influence the
eigenbehavior of the vehicle. The root loci curves for oscillation type 1 are shown,
even though α1 is negative for all investigated speeds, since there is a tendency for
instability.
Matsudaira uses the term “hunting” for self-excited vibrations for which the real
part of the eigenvalue is nonnegative. In this example, there are four different types of
hunting motions: In the area between 45 and 72 m/s, between 61 and 97 m/s, above 89
m/s, and above 97 m/s. The stability limit therefore is 45 m/s. Hunting motions 2 and
3 lose stability at 45 and 61 m/s respectively, although at higher speeds, they become
stable again (at 72 and 97 m/s). In the unstable area, i.e., between 61 and 97 m/s,
( rad /s ) d
/r a
f = 1680 t , c sx = 1000 N/ mm m
15 7
7,1 5
a L=
L,
( rad /s ) v/ 4
=
1.0 10
5 3
4 2
0.5 5 3 1
, 2
1 vcrit 4
vcrit 2 vcrit 3 vcrit 3 , vcrit 5
0 0
20 40 60 vcrit 2 80 100 120 140
a1
a2
-5
v (m/s)
-0.5
a3
-1.0 -10
a4
-1.5 -15
a5
-2.0 -20
Fig. 12.5 Typical example to illustrate the dependance of the roots λi = αi ± jωi (in the figure ai
instead of αi is given) on the vehicle speed; Matsudaira [4]
a linear investigation is no longer valid. Rather, a nonlinear stability investigation

should be carried out, which will be described in Chap. 13. Matsudaira stays with
the linear theory, since the ratio (α/ω) is moderate for hunting motions 2 and 3.
In the vicinity of vcrit,4 and vcrit,5 , qualitatively correct results can therefore still be
achieved.
The hunting motions with critical speeds vcrit,2 and vcrit,3 are of type 1 (carbody
hunting), while the hunting motions with vcrit,4 and vcrit,5 are of type 2 (bogie hunt-
ing).1 With the help of Fig. 12.5, it is not only the critical speed of each hunting
motion that can be calculated, but also the frequency and the amount of damping as
a function of vehicle speed.
As an example, in Fig. 12.6, the relationship between the longitudinal secondary
stiffness csx and the critical speed vcrit is shown. The two springs have a lateral distance
of 2830 mm and restrict the yaw motion of the bogie. The graphs in Fig. 12.6 form
1 The mode shape corresponding to each hunting motion can be derived by determining the relative
amplitude and phase for each system coordinate.
Fig. 12.6 Critical speed vc 120

as a function of secondary
longitudinal stiffness csx ; v crit 5
Matsudaira [4]. The 100
vcrit 4
secondary yaw stiffness can
be calculated by multiplying
2cxm by ex2 (lateral distance 80
between springs
vcrit (m/s)
2ex = 2830 mm)
vcrit 3
60
vcrit 2
40
vcrit 1
20
0 500 1000 1500 2000 2500 3000

c sx (N/mm)
a stability card. The cross-hatched areas are unstable. If the speed is increased, then
for csx = 1500 kN/m, at 53 m/s stability is lost but gained again at 73 m/s. At 104
m/s, another stability limit is reached.
Matsudaira’s results may be summarized as follows:
1. The same result as for model vehicles or real vehicles on the roller rig is obtained
in calculations: if the stiffness that prevents the bogie from yawing against the
carbody is small and if there is low damping in the system then two types of
hunting can be observed: the primary carbody hunting and the secondary bogie
hunting. Carbody hunting (1, 2, and 3) can again be divided into three types: First,
there is lower sway and upper sway, i.e., the roll center of the carbody is either
low or high. The third type is dominated by a carbody yaw motion. Bogie hunting
can be divided into two types. The bogies oscillate either in phase or out of phase.
(This classification is not always possible. Sometimes, it is impossible to identify
the modes explained above. This was the case in the example in Fig. 12.1.)
2. Each of these hunting motions appears only for speeds higher than the critical
speed and only in the unstable, cross-hatched areas in Fig. 12.6. The degree of
instability of the carbody hunting (expressed by α/ω) is relatively small; it reaches
its maximum somewhat above the critical speed, reduces with increased speed,
and becomes negative again, i.e., carbody hunting always becomes stable again.
In contrast, the degree of instability for bogie hunting increases dramatically
above the critical speed and is never reducing again. Therefore, bogie hunting is
much more critical. Nonlinear calculations prove that bogie hunting disappears
first when the vehicle speed is reduced far below the critical speed; see Chap. 13.
3. To avoid hunting, it is important to implement a resistance against bogie yaw
motions. If the secondary yaw stiffness is high enough, all hunting motions can
be suppressed, as can be seen in Fig. 12.6. This is true especially for carbody
hunting. For the design, it can be concluded that it is important to avoid play and
elastic deformations in friction yaw damping as long as possible if such elements
are used instead of viscous yaw dampers.
4. An efficient measure to suppress carbody hunting is to implement sufficient damp-
ing in the suspension system. Damping, however, has limited influence on bogie
hunting. Too much damping could instead significantly decrease the critical speed
of bogie hunting.
5. The higher the rail tread angle, the lower the critical speed. The reduction of the
critical speed is roughly proportional to the square root of the tread angle. When
the rail tread is worn, the equivalent conicity increases, and the vehicle is more
prone to hunting.
12.2.2 Bogie Hunting
For the calculations in Sect. 12.2.1, it was assumed that the wheelset is rigidly con-
nected to the bogie frame in both the longitudinal and lateral directions. In reality,
the coupling is always more or less flexible. This flexibility has a big influence on
bogie hunting stability. Therefore, the influence of the primary stiffness on the critical
speed was investigated with the model in Fig. 12.7.
The results are shown in Fig. 12.7. The figure illustrates the influence of the
secondary yaw stiffness csx (the lateral distance between them is 2830 mm) on the
critical speed vcrit . The longitudinal and lateral primary stiffnesses, cpx and cpy , are
chosen as parameters. From the figure, the following conclusions can be drawn:
1. A suitable choice of values for the primary stiffness is an efficient measure to
stabilize bogie hunting and in turn to increase the critical speed.
2. If no yaw stiffness exists between bogie and carbody, as is the case when the load of
the carbody is carried by a center plate, it turns out to be more efficient to increase
the lateral primary stiffness as much as possible and choose the longitudinal
primary stiffness as low as possible to increase the critical speed. This is at least
more favorable than choosing a low value for the lateral primary stiffness and a
high value for the longitudinal primary stiffness.
3. If the yaw stiffness is high enough, as is the case with elastic side-bearers, it is
more favorable to increase the longitudinal primary stiffness as much as possible
while choosing a soft lateral primary spring to increase the critical speed.
4. Nor is it advisable to let the lateral primary stiffness become zero. This means that
one should avoid lateral play in the axle bearing and in the arrangement carrying
the axle bearing.
Fig. 12.7 Critical speed of 300

bogie hunting as a function
of secondary yaw stiffness,
Matsudaira; parameters of
the different curves are the
values given for primary
px
stiffness [4]
0
500
200
py
c
0
vcrit (m/s)
00
30
0
00
0
30
00
50
100
1c 1c
2 py 2 px c sx
0 1000 2000 3000 4000 5000

c sx (N/mm)
12.2.3 Influence of Friction Yaw Damping on Bogie Hunting

(Nonlinear Stability Investigation)
Today, most passenger vehicles for higher speeds have viscous yaw dampers between
bogie and carbody to avoid bogie hunting. Another suspension element that was
used quite a lot in earlier high-speed bogies, for example in Germany, is friction
yaw damping. Friction yaw damping can, for example, be provided by suspended
side bearers on which either all or a part of the carbody weight rests. An alternative
is a center plate with a large diameter. In a real bogie, a certain amount of elastic
deformation is always present in such an arrangement, for example caused by rubber
bushings.
In all calculations above, linearity was postulated. Therefore, small amplitudes
were assumed. As mentioned several times, carbody hunting is characterized by
large oscillations of the carbody while the amplitudes of the bogie remain relatively
small. Observations of hunting motions of vehicles on roller rigs confirm that the
ϕz -amplitude of the bogie is small, at least at speeds somewhat below the critical
hunting speed. The results referring to primary hunting can therefore be regarded
csx F
100 v 100
z
F= 5000 N cs x (N/mm)
80 1c 1c 80
2 py 2 px c sx F
F=
150
00
vcrit (m/s)
vcrit (m/s)
60 60 N
10
00
0N
c= (N/mm)
40 40
500
0N
1000 N/mm
20 20
1500 N/mm
0 0.01 0.02 0 0.01 0.02

z,max z,max
Fig. 12.8 Critical speed of bogie hunting as a function of maximum bogie amplitude ϕzmax , for
constant friction force F (left) and constant longitudinal secondary stiffness csx (right), Matsudaira
[4]. The values marked with symbols were calculated with an analog computer
as valid. In contrast, the motion of the bogie for secondary hunting is high, and the
yaw amplitudes become so high that sliding in the side-bearers occurs. Therefore,
the results from a linear analysis with only elastic yaw moments are invalid for
this type of bogie. Hunting for the bogie in Fig. 12.8 was therefore investigated
with a nonlinear method, and the same example as given above was calculated. The
nonlinear calculation was carried out on an analog computer.
From the left- and right-hand diagrams of Fig. 12.8, the following conclusions
can be drawn:
1. When the yaw amplitude of the bogie exceeds a certain limit, sliding in the friction
surfaces of the side-bearers occurs. If the amplitude then is increased, the critical
bogie hunting speed decreases significantly.
2. If the yaw stiffness of the bogie is increased, the critical speed increases for small
amplitudes (see Sect. 12.2.2); for large amplitudes, however, the critical speed
decreases somewhat.
3. If the friction force is increased, the critical hunting speed remains unchanged for
small amplitudes. For large amplitudes, the critical speed increases.
In conclusion, the main possibility to stabilize bogie hunting and increase the crit-
ical speed is to increase the yaw stiffness of the bogie. Friction alone has no influence
on the control of the hunting motion. Friction plays a secondary roll, however, in
sustaining the effect of the spring to as high amplitudes as possible.
12.3 Suggestions for Further Work for This Chapter 235
12.3 Suggestions for Further Work for This Chapter
The following exercises are not exercises in the classical sense but can be understood
as topics for further study. Therefore, we call this section Suggestions for further
work.
12.3.1 Influence of Bending and Shear Stiffness

on the Stability of Bogie Vehicles
In Chap. 11, it is illustrated that the stiffness of the primary suspension can be
expressed with the help of bending and shear stiffness. How does the diagram from
Fig. 12.7 change if instead of the primary stiffnesses cpx and cpy , the bending and
shear stiffnesses are used?
12.3.2 Stability of a Vehicle with Independently Rotating

Wheels
Investigate the stability of a bogie vehicle with independently rotating wheels.
12.3.3 Friction and Viscous Yaw Damping
Compare the two yaw damper concepts in which either friction via suspended side-
bearers or viscous yaw damping is used.
References
1. K. Knothe, Die dynamische Analyse von Mehrkörpersystemen zur Stabilitätsuntersuchung von

Schienenfahrzeugen (The dynamic analysis of multibody systems for stability investigations of
railway vehicles), VDI-Berichte, vol. 269 (VDI-Verlag, Düsseldorf, 1976), pp. 77–86
2. K. Knothe, W. Kik, LINDA I – Ein Programmsystem zur Untersuchung des dynamischen Verhal-
tens von Schienenfahrzeugen (Linda I – A program system for the investigation of the dynamic
behavior of railway vehicles). ILR-Bericht 11, Institut für Luft- und Raumfahrt, Berlin (1976)
3. J. Klingel, Über den Lauf von Eisenbahnwagen auf gerader Bahn (On the running of railway
vehicles on straight track). Organ für die Fortschritte des Eisenbahnwesens, Neue Folge, vol.
20, Tafel XXI (1883), pp. 113–123
Chapter 13
Introduction to Non-linear
Stability Investigations
13.1 Preamble
The stability investigations for the wheelset in Chap. 10, the bogie in Chap. 11, and
for a bogie vehicle in Chap. 12 assume that the system can be described by linear
equations.
This might be possible for the vehicle if no gaps, no advanced suspension elements
such as air springs, and no friction elements are used. In the description of the wheel–
rail contact, however, nonlinearities are difficult to avoid, as described for the contact
geometry in Sects. 3.1 and 3.2 (especially Fig. 3.9), and for contact mechanics in
Sect. 3.4 and especially Fig. 3.20.
This has been well known for many decades. This fact had been detected in 1953,
in a tender in a competition of the ORE working group C9 regarding vehicle hunting
[1], even though the solutions submitted were all based on linear equations. One of
the side effects of the work in the ORE working group C9 was, however, that research
projects on the influence of nonlinearities on the running behavior of rail vehicles
were begun in several European countries, among others in Germany. In the Federal
Republic of Germany, on the initiative of Carl Theodor Müller in several dissertations
and research projects, nonlinear phenomena were studied; see, for example, [2–5]).
The simulations in the work supervised by C.Th. Müller were not done with the help
of digital computers but with analog computers. In Germany, Müller was 15 years
ahead of the development. First in 1982, at MAN Technology, simulations with an
analog computer were again carried out [6].
Unfortunately, for the high-speed test runs of the TGV in France in 1953 [7],
which almost ended in a catastrophe, no measurement results were available. It is
very likely that besides track properties, nonlinear processes in the wheel–rail contact
were also important.
A closed theory for nonlinear stability investigations in rail vehicle dynamics does
not exist. In the last 30 years, however, many studies that deal with this issue have
appeared.

DOI 10.1007/978-3-319-45376-7_13
238 13 Introduction to Non-linear Stability Investigations
1. De Pater dealt with nonlinear stability problems in 1961 [8]. In the ORE working
group C9, he inspired C.Th. Müller to work on such problems as well. Twenty
years later, another article with de Pater as coauthor was published [9].
2. Both theoretically and practically interesting investigations can be found in the
environment of Huilgol [10, 11].
3. In the USA, studies on nonlinear stability have been carried out by Cooperrider,
Hedrick, and Law [12]. A summary report by Hedrick [13] exists. To our knowl-
edge, the investigations were not continued, since there was no governmental
support in the area of rail vehicle dynamics as in Europe.
4. Finally, the school of True has to be mentioned. True devoted more than 20 years
to intense studies of different nonlinear phenomena in rail vehicle dynamics [14].
First, a well documented test on the roller rig in München–Freimann (1985)

will be discussed. Important nonlinear phenomena can be explained with the help
of the results (Sect. 13.2). Thereafter, one possibility to perform nonlinear stability
investigations will be described in detail. Basically, this method is based on Fourier
analysis (Sect. 13.3). In relation to this, a method known as “quasilinearization”
is mentioned (Sect. 13.4). Finally, in Sect. 13.5, the limits of Fourier analysis in
investigating nonlinear stability will be discussed.
13.2 Nonlinear Critical Speed
First, we look at the results from the roller rig in München–Freimann shown in
Fig. 13.1.
The lateral amplitude u y and the yaw angle ϕz of a wheelset were measured as
the speed was increased (see Fig. 13.1, lower figure). Up to a speed of v = 180 km h
,
the lateral displacements u y and the yaw angle ϕz remain small. They more or less
reflect the small excitation amplitude of the roller rig cylinders. At v = 180 km h
,a
sudden increase of lateral displacement and yaw angle can be observed. The speed
v = 180 km h
is the already known linear critical speed. In contrast to what is predicted
by the linear theory, the amplitudes do not increase to infinity. The amplitudes remain
at about 7 mm, even though the speed is further increased. This 7 mm represents the
gauge clearance between wheelset and rig cylinder; i.e., the result is expected.
Not expected, however, is the behavior when the speed is reduced. Even if the
speed is reduced below the linear critical speed v crit,lin = 180 kmh
, the lateral ampli-
tude of u y = 7 mm remains. Only if the speed is reduced to v = 135 km h
is the
wheelset amplitude again reduced to almost zero. The speed at which the amplitude
is reduced is 30 % lower than the linear critical speed. A similar phenomenon is
described by Matsudaira in [16]; see Sect. 12.2.1.
The second lower speed at which the amplitude decreases is called the nonlinear
critical speed v crit, nl .
If the vehicle is running with a speed between the linear and nonlinear critical
speeds, it is possible that a wheelset with amplitudes near zero is excited by a single
13.2 Nonlinear Critical Speed 239
uy
(mm)
10
7,5
2,5
0
t
-2,5
-5
-7,5
-10
z
(mrad)
20
10
0
t
-10
-20
v
(km/h)
200 180
150 130
100
50
0
t
Fig. 13.1 Limit cycle motions after exceeding the linear critical speed at the roller rig [15]
irregularity, so that the amplitude rises to 7 mm and remains at that level. This has to
be avoided in the daily operation of a vehicle. That, however, is possible only if the
maximum speed is below the nonlinear critical speed.
The existence of two critical speeds is a typical nonlinear phenomenon. As long
as the relations remain linear, only the linear critical speed exists. Below this speed,
the wheelset is runs stably. Only if the linear critical speed is exceeded do instability
effects arise.
13.3 Fourier Decomposition of Nonlinear Limit Cycle

Motions: The Method of Urabe and ReiTer
Nonlinear stability investigations can be carried out with a method introduced by

Urabe and Reiter [17, 18]. As can be seen from the titles of the work, it is a modifica-
tion of the Galerkin method. It is assumed that the solution is periodic. This periodic
solution is represented by a Fourier series.
The development of the procedure was performed by Moelle [19–21]. The bogie
investigated is shown in Fig. 13.2. The springs shown in the figure have to be inter-
preted as springs and dampers in parallel.
For more details, see [21]. In the following, the system of equations that describes
the motions is given only schematically,
1 1
M ü + D u̇ + S u + g NON = 0 . (13.1)
ωs
2 ωs
The solution u is assumed to be a Fourier series with the basic frequency ωs . As a

starting point, however, it is not the system of equations of motion (13.1) but the
m = 3588 kg
2
x = 2517 kg m
1 y = 9784 kg m
2
1 2 c zs
2 c zp z = 11170 kg m
2
zb
xb
1
2 c xp
=
1
2 c yp 1
2 c ys
m = 2256 kg
2
x = 1108 kg m
2
y = 193 kg m
x1 xb xT
2
z = 1108 kg m
y1 yb yT
Fig. 13.2 Bogie model for nonlinear stability investigation

13.3 Fourier Decomposition of Nonlinear Limit Cycle Motions: … 241
Fig. 13.3 Calculated limit uy c px = 2 E 7 N /m

cycle curves for the bogie [ mm [ c py = 2 E 7 N /m
from Fig. 13.2 with new UIC P = 289800 N
7
60 rail profile and S1002
wheel profile
6
unstable branch
5
bogie frame
4
leading wheelset
3
trailing wheelset
unstable branch
1
speed
0
30 40 50 60 70 80 v [m/sec]
corresponding version of the principle of virtual displacements that is chosen. By

this approach, it is assumed that the solution is periodic. The a priori unknown basic
frequency ωs is used in Eq. (13.1) as a normalization factor, which is introduced to
avoid time appearing explicitly. By this approach, the system of ordinary differential
equations is transformed into a system of algebraic equations that can be solved
iteratively.
The displacement vector u has the following components:
u = {u y1 , ϕz1 , ϕy1 , u x1 , u y2 , ϕz2 , ϕy2 , u x2 , u yb , ϕzb , ϕyb , u xb } .
The vector g N O N contains all nonlinear effects, especially the terms from nonlinear
contact geometry and contact mechanics. The vector also includes the vehicle speed
v and the basic frequency ωs . For more details, see [19, 21].
A characteristic result of the investigations is shown in Fig. 13.3. The maximum
lateral displacement (u y1,max ) as a function of the vehicle speed is given for a new
UIC 60 rail profile and S1002 wheel profiles. The curves of Fig. 13.3 are called
limit cycles. The limit cycles for the leading and trailing wheelset are qualitatively
the same. The limit cycles consist of stable sections and unstable sections. Sections
where the speed decreases with increasing amplitude u y are unstable. If a motion
state near a stable branch is given, after a while this stable solution will be observed.
In nonlinear dynamics, such stable branches are called attractors. Motion states on
unstable branches do not stay there, but decrease or increase in amplitude until the
solution has reached a stable branch.
uy
[ mm [
7
3
worn rail
2
new rail
1 slightly worn
new rail
1 rail
0
1 0.5 0
speed
rL- rR [ mm [ 10 0 30 40 50 60 70 80 v [m/sec]
Fig. 13.4 Calculated bifurcation diagrams for the bogie in Fig. 13.2 on new and worn rails (wheel
profile S 1002)
A speed of v = 73 ms belongs to a lateral displacement amplitude of u y,max ≤

1.5mm. This is exactly the linear critical speed v crit,lin of the system. For slightly
higher lateral displacement amplitude for a new rail profile, the critical speed
decreases to 33 ms . For a displacement amplitude of 4.8 mm, the critical speed
increases again to 65 ms . For a further increase of the lateral displacement ampli-
tude of the leading wheelset, the speed decreases again to about 60 ms . A further
increase of the amplitude implies that the contact point is in the flange region and
remains there; the critical speed increases significantly. In principle, the leading and
trailing wheelsets exhibit the same behavior.
The behavior does not change significantly when the new rail profiles are replaced
with worn profiles. The wear profile was generated artificially, by approximating the
tread area of the rail profile with an ellipse. For the wheels, new S1002 profiles were
kept. Qualitatively, the limit cycle curve remains the same (see Fig. 13.4). Only for
small lateral displacements can differences be seen. For displacements of 3.5 mm
or more, the limit cycles are the same. To visualize the influence of the profile, the
r-function is given in the left part of the figure.
On the basis of such limit cycle or bifurcation diagrams, the results from the
roller rig tests can also be interpreted. In Fig. 13.5, the bifurcation diagram for a
profile combination of new S1002 wheels on slightly worn rail profiles, a gauge of
1435 mm and a rail cant 1 in 40 is shown. The wheelset on which the measurements
were carried out is first in a centered position. The vehicle speed is increased until
the linear critical speed of 72 ms is reached. At this speed, the zero attractor loses
13.3 Fourier Decomposition of Nonlinear Limit Cycle Motions: … 243
uy lateral
[ mm [ amplitude u y1
7
u z1
6
5
u y1
4
3
track gauge 1435 mm
2 rail cant 1: 40
1
speed
0
30 40 50 60 70 80 90 v [m/sec]
58 72
Fig. 13.5 Interpretation of the results shown in Fig. 13.1 by calculation with the method of Urabe
and Reiter
its stability, and the motion state of the wheelset approaches the stable attractor
with about 6.5 mm amplitude. This is at the same time the gauge clearance for this
wheelset. If now the speed is decreased, the amplitude is first significantly decreased
at 60 ms . At 58 ms , the amplitude is zero again. The leading wheelset is in a centered
position again.
The speed 72 ms is in this case the linear critical speed, while 58 ms is the above
introduced nonlinear critical speed. The results are not exactly the same as for the
tests on the roller rig. The tendency, however, is the same. Therefore, it can be
assumed that the phenomena observed on the roller rig can be explained with theses
simulation results.
13.4 Simplified Investigation of Nonlinear Stability

with the Method of Quasilinearization
A simplified nonlinear method for stability investigations called quasilinearization or

describing function technique was developed especially in the USA. In the following,
we look at only the simplest case, in which only the nonlinear wheel–rail geometry as
r = r (u y ) is taken into account. For further studies, we recommend [13]. Details
regarding rail vehicle dynamics can be found in [12, 22]. We further assume that
the motion is a harmonic one, i.e., the method of quasilinearization is a one-periodic
version of the method of Urabe–Reiter:
ũ y = û y eiωt .
One reason for the restriction to nonlinearity in the contact geometry is that we can use
the quasilinearized nonlinear rolling radius difference; see Sect. 3.2.3. According to
Eq. (3.10), the nonlinear rolling radius difference is replaced by an equivalent linear
relationship
r (u y ) = 2λe u y , (13.2)
with the equivalent conicity λe ,
π
1
λe = [rL (u y0 sin τ ) − rR (u y0 sin τ )] sin τ dτ . (13.3)
πu y0
0
Starting from Eq. (9.13), the quasilinearized equations of motion for the wheelset
can be written as

m 0 ü y 2Qζ 0 uy
+ +

dy 0 u̇ y cy uy
+ +
0 ex2 d ϕ̇z 0 ex2 cx ϕz

1 2GabC22 0 u̇ y
+
v0 0 2GabC11 e02 ϕ̇z

0 −2GabC22 uy 0
e0 λ(û y ) = ,
2 r0 GabC11 0 ϕz 0
where as an insignificant simplification, the terms proportional to C23 are neglected.

The analysis of the homogeneous equations of motion is performed as above.
The difference is that û y is given, and thus the eigenvalues depend on û y . After the
eigenvectors for the u y -component are calculated, the chosen û y is introduced. With
this, the yaw angle ϕ̂z is also known.
Since v crit in this case can be determined analytically (10.39), we can directly use
the relation
2 e0 r 0 1
v crit
2
= ωcrit . (13.4)
λ(û y ) 1 − k(û y )
What remains is to replace ωcrit

2
from Eq. (10.40) and k from Eq. (10.41). It is essential
that v crit depend on the lateral displacement amplitude û y .
2
In looking at a bogie with two wheelsets, we see that the lateral displacement
amplitude will not be the same for both wheelsets. In that case, the solution has to
be found iteratively. One of the two amplitudes is chosen. From the beginning, the
other one can, for example, be determined from a linear eigenvalue analysis. With
this, for both wheelsets an equivalent conicity (not the same) is known. The next
13.4 Simplified Investigation of Nonlinear Stability … 245
iteration step gives an improved second amplitude. The iteration procedure can be
interrupted when the ratio û y1 /û y2 saturates at a desired accuracy interval.
As long as the contact point stays on the running tread, only small differences to the
method of Urabe–Reiter can be found. The reason is that in this case, the creepages
are small, and nonlinearities in the contact mechanics are negligible. If nonlinearities
depend on different variables (for example in nonlinear contact theory), the method
of quasilinearization becomes much more complicated. Further nonlinearities can in
principle be handled without difficulty, as long as they depend on one variable.
13.5 Limits of Fourier Decomposition
Many problems have been handled with the method of Urabe–Reiter [20]. However,
the methodology has its limitations. Difficulties can arise when whole vehicles are
investigated. The difficulties have partly mathematical and partly mechanical reasons.
The mathematical reasons refer to the nonlinear iteration procedure (Newton) that is
used to find the solution of the system of nonlinear algebraic equations. The iteration
procedure does not always yield a solution.
The real reasons, however, are often mechanical. Let us look at the vehicle that
was investigated in Chap. 12. Two of the root loci curves of this vehicle travel at
almost exactly the same frequency from the negative to the positive half-plane; see
Fig. 12.2. One could say that the nonlinear iteration procedure in that case would
have problems to decide in favor of one of the motion states shown in Fig. 12.4.
It is also possible that the basic assumption of a periodic solution is wrong. The
solution can be doubly periodic or even chaotic. If no periodic solution exists, other
procedures have to be used. Such methods have mainly been developed by True
[23–25].
13.6 Nonlinear Stability Investigations in the Time Domain
In existing commercial multibody simulation tools, the simulations are today mainly
carried out in the time domain. In this type of simulation, it is impossible to calculate
the nonlinear critical speed exactly, but just an estimate. From the engineering point
of view, this is sufficient, since the simulations include a number of uncertainties
regarding input parameters and model. Even a mathematically exact critical speed
would not be exact in reality. The advantage of time-domain simulations is that all
nonlinearities in the mechanical system can be taken into account.
Usually, the vehicle is excited by an initial disturbance, and it is tested whether the
triggered oscillation disappears. For a number of relevant wheel–rail profile com-
binations, it is tested at what speed the oscillation no longer disappears. Another
possibility is to carry out the simulations equivalent to the stability test on the roller
rig; cf. Fig. 13.1. The simulations are begun on ideal track (no track irregularities)
at a high speed, so that the vehicle runs on a limit cycle. The speed is then con-
tinuously reduced until the oscillation disappears. A good description of different
methodologies to determine the critical speed can be found in [26]. A similar method
to determine the nonlinear critical speed is suggested in [27]. The difference between
this and the method described above is that the speed is not reduced continuously
but in discrete steps, as suggested by [14].
13.7 Ideas for Further Study in this Chapter
13.7.1 Stability Investigation for the Boedecker Vehicle
In 1887, Boedecker described the instability of two-axle vehicles in which the

wheelsets are rigidly connected to the vehicle. The most important detail in his
derivation was that he used a local Coulomb law and not a linear creepage–creep
force law, as in our stability investigations.
Try to redo the calculation from Boedecker’s book for a bogie with the method
of quasilinearization. Use the following model (Fig. 11.9):
• Rigid bogie frame;

• two rigid wheelsets that are rigidly connected to the bogie frame;
• inertially guided carbody, to which the bogie frame is connected by lateral and
yaw stiffness (see Fig. 11.9);
• nonlinear creepage–creep force law based on a local formulation of the Coulomb
law (friction coefficient equal in both directions), limited to longitudinal and lateral
creep (no spin creep);
• linear contact kinematics with double cone (see Fig. 3.7). Use a cone angle 1 : 20.
The only nonlinearity is the creepage–creep force law. For this relation, first a linear
damping law has to be found with quasilinearization. In the next step, the quasilin-
ear equations of motion have to derived and tested for stability. For the numerical
simulation, the secondary yaw stiffness can be neglected, and the secondary lateral
stiffness might be varied.
References
1. ORE. Wechselwirkung zwischen Fahrzeug und Gleis. Bericht über das Preisausschreiben zur
Lösung des Schlingerproblemes (Interaction between vehicle and track. Report on the com-
petition to solve the hunting problem.). ORE-report to question C 9. Work result Nr. 2 Vol. 1.
Utrecht, UIC, ORE (1957)
2. G.-P. Bracker, Einfluß der Gehänge- und Tragfederkräfte auf das Laufverhalten freier Lenkrad-
sätze im geraden Gleis. Dissertation, TH München (1966)
13.4 Simplified Investigation of Nonlinear Stability … 247
3. H. Dillmann, Einfluß der Schlupfabhängigkeit der Reibung auf den Bogenlauf zweiachsiger
Schienenfahrzeuge mit freien Lenkachsen - Auszug aus einer Dissertation (Influence of creep
dependency of friction on the curving behavior of two-axled railway vehicles with free steering
axles). Glas. Ann. 83(12), 420–433 (1959)
4. K. Inderst, C.T. Müller, Energiefluß beim Schlingern von Schienenfahrzeugen (Flow of energy
during hunting of railway vehicles). Research report, laboratory Minden of DB, o. J. (1967).
(not published; from the estate of C. Th. Müller, O. Krettek)
5. K. Kämpfe, Schwingungsverhalten eines zweiachsigen Eisenbahnfahrzeugs mit freien
Lenkradsätzen und reibungsbehafteten Kopplungen beim Lauf im geraden Gleis (Dynamic
behaviour of a two-axled railway vehicle with free steering wheelsets and friction couplings
running on straight track). PhD thesis, TH München (1961). (Text, pictures and equations)
6. A. Schmidt, D. Moelle. Ermittlung der Grenzzykelbewegungen der nichtlinearen Drehgestell-
dynamik, Vergleich zwischen Hybridsimulation und digitalen Grenzzykelberechnungen
(Determination of limit cycle motions in nonlinear bogie dynamics. Comparison between
hybrid simulation and digital limit cycle analysis). Report Nr. B 099006 EDS 019, MAN
München, MAN Neue Technologie München and TU Berlin, ILR (1982)
7. J. Salin, Regards en arrière. Revue Générale des Chemins de Fer 100, 216–221 (1981)
8. A.D. de Pater, The approximate determination of the hunting movement of a railway vehicle
by aid of the method of Krylov and Bogoljubov. Appl. Sci. Res. 10, 205–228 (1961)
9. J.P. Meijaard, A.D. DePater, Railway vehicle system dynamics an chaotic vibration. Int. J.
Nonlinear Mech. 24, 1–17 (1989)
10. R.R. Huilgol, Hopf-Friedrichs bifurcation and the hunting of a railway axle. Q. J. Appl. Mech.
36, 85–94 (1978)
11. X. He, R.R. Huilgol, Application of Hopf bifurcation at infinity to hunting vibrations of rail
vehicle trucks, in Dynamics of Vehicles on Roads and Tracks. Proceedings of the 12th IAVSD
Symposium held in Lyon, France, August 1991. Supplement to Vehicle System Dynamics, vol.
20, ed. by G. Sauvage (Swets & Zeitlinger, Amsterdam/Lisse, 1992), pp. 240–253
12. N.K. Cooperrider, J.K. Hedrick, E.H. Law, C.W. Malstrom, The application of quasilineariza-
tion techniques to the prediction of nonlinear railway vehicle response, in The Dynamics of
Vehicles on Roads and on Tracks. Proceedings of the IUTAM Symposium held at Delft, The
Netherlands, August 1975, ed. by H.B. Pacejka (Swets & Zeitlinger, Amsterdam, 1976), pp.
314–325
13. J.K. Hedrick, Nonlinear system response: quasi-linearization methods, in Nonlinear System
Analysis and Synthesis: Volume 1 - Fundamental Principles. Presented at the Winter Annual
Meeting of the American Society of Mechanical Engineers; New York, December 5–10, 1976,
ed. by J.K. Hedrick, H.M. Paynter (1978), pp. 97–124
14. H. True, On the theory of nonlinear dynamics and its application in vehicle system dynamics.
Veh. Syst. Dyn. 31(5–6), 393–421 (1999)
15. A. Schmidt, L. Moelle, Durchführung Rollprüfstandsversuche, MAN-Anteil. Rad/Schiene
Forschung, BMFT VorhabenTV79604 (Implementation of roller rig tests). Final report Nr.
K 096 991 - EDS, MAN Neue Technologie Munich (1982)
17. M. Urabe, Galerkin’s procedure for nonlinear periodic systems. Arch. Ration. Mech. Anal. 20,
120–152 (1965)
18. M. Urabe, A. Reiter, Numerical computation of nonlinear forced oscillations by Galerkin’s
procedure. Arch. Math. Anal. Appl. 14, 107–140 (1966)
19. R. Gasch, D. Moelle, Nonlinear bogie hunting, in The Dynamics of Vehicles on Roads and on
Tracks. Proceedings of the 7th IAVSD-Symposium held at Cambridge, UK, August 1981, ed.
by A.A. Wickens (Swets & Zeitlinger, Lisse/Amsterdam, 1982), pp. 455–467
20. R. Gasch, D. Moelle, K. Knothe, The effects of nonlinearities on the limit-cycles of rail-
way vehicles, in The Dynamics of Vehicles on Roads and on Tracks. Proceedings of the 8th
IAVSD-Symposium held at Cambridge, Mass./USA, August 1983, ed. by J.K. Hedrick (Swets
& Zeitlinger, Lisse/Amsterdam, 1984), pp. 207–224
21. D. Moelle, Digitale Grenzzykelrechnung zur Untersuchung der Stabilität von Eisenbahn-
drehgestellen unter dem Einfluß von Nichtlinearitäten (Digital limit-cycle analysis for the
investigation of the stability of railway vehicle bogies under the influence of nonlinearities.).
PhD thesis, Technical University Berlin (1990)
22. N.K. Cooperrider, The hunting behaviour of conventional railway trucks. J. Eng. Ind. 94, 752–
762 (1976)
23. C. Knudsen, R. Feldberg, H. True, Bifurcations and chaos in a model of a rolling railway
wheelset. Philos. Trans. R. Soc. Lond. A 338, 455–469 (1992)
24. H. True, Railway vehicle chaos and asymmetric hunting, in The Dynamics of Vehicles on
Roads and on Tracks. Proceedings of the 12th IAVSD Symposium held at Lyon/France, August
1991. Supplement to Vehicle System Dynamics, vol. 20, ed. by G. Sauvage (Swets & Zeitlinger,
Amsterdam/Lisse, 1992), pp. 625–637
25. H. True, Does a critical speed for railroad vehicles exist?, in Proceedings of the 1994 ASME/IEE
Joint Railroad Conference held in Chicago, Illinois, March 1994, ed. by K.L. Hawthorne, R.J.
Hill (1994), pp. 125–131
26. O. Polach, On nonlinear methods of bogie stability assessment using computer simulations.
Proc. Inst. Mech. Eng. Part F: J. Rail Rapid Trans. 220, 13–27 (2006)
27. S. Stichel, Limit cycle behaviour and chaotic motions of two-axle freight wagons with friction
damping. Multibody Syst. Dyn. 8(3), 243–255 (2002)
Chapter 14
Quasistatic Curving Behavior
14.1 Historical Introduction
Interest in curving behavior of rail vehicles began earlier than interest in stability.
In Germany, Redtenbacher is probably the first to have dealt intensely with curving
behavior [1]. In the monograph by Boedecker [2], curving behavior is also the focus.
Although the question of running on a straight line (and so the question of stability) is
mentioned as a problem, the issue is, however, regarded as secondary in Boedecker’s
work.
Around 1900, Uebelacker was the first to intensely discuss curving behavior
[3]. Among Redtenbacher, Boedecker, and Uebelacker, quasistatic curving behavior
dominated. The time after 1910 was, regarding quasistatic railway behavior, domi-
nated by Heumann, whose first work was published in 1913 [4].
In 1941, the German ministry of railways put together a working group called the
“Dauner–Hiller–Reck–Kompendium,” with the instruction to work out preliminary
guidelines for vehicle manufacturing to achieve good guidance of vehicles on the
track [5, 6]. Heumann played an essential part in this working group. However, he
did not summarize his findings regarding curving behavior before the 1950s. First
he published a number of articles and then a special issue in 1954 [7]. Over the next
30 years, this special issue by Heumann was the bible for the design of rail vehicles.
Heumann’s theory of curving behavior is particularly valid regarding curves with a
small radius.
The historical development in England is portrayed by Gilchrist [8] and Wickens
[9]. After Gilchrist, Mackenzie was the first to deal intensely with curving behavior
[10]. In Germany, Mackenzie’s results where picked up and further developed by
Boedecker [2], Helmholtz [11], and Uebelacker [3]. In England, this school was
pursued by Porter in 1934/1935 [12–15], who sadly died of pneumonia at only 28
years old.

DOI 10.1007/978-3-319-45376-7_14
250 14 Quasistatic Curving Behavior
14.2 General Remarks
In modern simulation tools, it is standard to define transition curve, circular curve,

and superelevation (cant) as the design geometry. Track irregularities are regarded
as small deviations of the design (ideal) geometry. The solution is usually calculated
in the time domain.
One of the primary goals of curving calculations is usually to derive the position
of the running gear and the wheelset relative to the track when initial disturbances
are damped out. This motion condition, which is in between a static and a real
dynamic condition, is usually called quasistatic curving. The relative position of the
running gear determines the average level of creepages and creep forces in the wheel–
rail contact. We begin with a single wheelset (Sects. 14.3 and 14.4). Thereafter, the
behavior of a whole bogie in a curve is discussed (Sect. 14.5). The higher the average
level of creepages and creep forces, the higher the wear and the risk of rolling contact
fatigue (RCF) on wheel and rail; cf. Sect. 14.6. Prerequisites for a quasistatic curving
condition are
• constant speed v0 ,
• constant curve radius,
• constant superelevation,
• ideal track, i.e., no track irregularities.
There are two main possibilities for investigating quasistatic curving behavior:
1. One possibility is to calculate in the time domain. Ideal track is assumed, and the
time step integration is performed until initial disturbances are damped out and
a quasistatic solution is achieved. Often, the simulation starts with a vehicle on
straight track. The simulation is then carried out until the vehicle has negotiated
the transition curve and reached the circular curve.
2. The other possibility is to let the time derivatives in the equation be zero. In this
case, we no longer have differential equations but a system of nonlinear equations.
From solving the nonlinear equations, the relative position of the vehicle on the
track and the corresponding forces are found.
Both for quasistatic and dynamic solutions, linear equations can be assumed. This
approach is described by Boocock [16]. It will, however, not be further discussed
here, since a nonlinear approach is more appropriate in our opinion.
We are going to discuss the nonlinear quasistatic calculation in more detail. Strictly
speaking, this methodology has nothing to do with rail vehicle dynamics. However,
it enables a basic understanding of the issues of rail vehicles negotiating curves.
Further, an optimal design for good curving behavior and optimal design for high
speeds on straight track are in general contradictory. Therefore, both aspects have
always to be taken into account. In this respect, quasistatic curving also influences
the dynamic behavior of a rail vehicle and should be part of this book.
Just as when we derived the equations that describe the stability behavior of a rail
vehicle, we will also start with a single wheelset for the curving behavior.
14.3 A Single Wheelset in a Curve 251
14.3 A Single Wheelset in a Curve
14.3.1 Free Wheelset in a Curve (Kinematic Curving)
Let us look at a free (not suspended) wheelset with conical profiles. We assume pure
rolling, i.e., no creepages occur at the contact point. Further, no centrifugal forces
act on the wheelset.
To achieve radial steering, i.e., the wheelset axis points to the center of the curve,
the outer wheel has to roll a longer distance than the inner wheel. This is possible if
the wheelset is displaced toward the outside of the curve and in turn, the outer wheel
rolls on a larger radius than the inner wheel; cf. Fig. 14.1. The ratio needed between
outer and inner curve radii is here called the rolling line (in German: Rolllinie) and
can be calculated as
rout R + e0
= , (14.1)
rin R − e0
where it is assumed that ea = ei = e0 . This relationship was derived by Redtenbacher

[1]. For conical profiles,
rout = r0 + δ0 u y , (14.2a)
rin = r0 − δ0 u y . (14.2b)
r out
uy
R e0
r in
e0
Fig. 14.1 Radial steering of a free wheelset with conical profiles in a curve according to [17]
underradial radial overradial

position position position
z>
0 z
=0 z>
0
z z
v0 v0 v0
Fig. 14.2 Possible wheelset positions on the track according to [17]
Equations (14.2a) and (14.2b) substituted into Eq. (14.1) results in the necessary
lateral wheelset displacement
r0 e0
uy = . (14.3)
δ0 R
If this lateral displacement is achieved exactly, a free wheelset is rolling on its rolling
line and will negotiate a curve with a radial position.
The lateral displacement needed for radial steering is reduced with increased curve
radius R and higher conicity δ0 . Already here a conflict to running on straight track
is indicated. For curving according to Eq. (14.3), a high conicity δ0 is an advantage.
For running stability on straight track, in contrast, low conicity is preferred; see
Eq. (10.39). If the required rolling radius difference cannot be achieved, e.g., because
the wheelset is suspended in a bogie, the wheelset assumes either an overradial or
underradial position, as can be seen in Fig. 14.2.
14.4 Wheelset Guided in Track Following Frame
The wheelset is now assumed to be suspended in a guided frame that follows the
track with speed v0 . The frame does not move relative to the track center line and
also follows the superelevation in curves. The wheelset can move and rotate relative
to the guided frame. The displacements and rotations of the wheelset are given in a
track-following coordinate system (speed v0 ); cf. Fig. 14.3. For a superelevation h 0
of the outer wheel in a curve, the (x, y)-plane of the coordinate system is rotated
with
h0
sin ϕx = (14.4)
2e0
14.4 Wheelset Guided in Track Following Frame 253
Fig. 14.3 Free-body ex

diagram of wheelset with (a)
acting forces
z
cy y cy
uy
FzR FzL
z
Hy c y uy c y uy m r a y
y uy
T R L
R
NR NL
T L
M R M L
(b)
cx cx
z
cy y cy
x
uy
c x ex z c x ex z
Hy c y uy z uy c y uy m a y
y
T R T L
relative to the (x, y)-plane on straight track. The maximum cant h 0 usually is 150–
180 mm. This means that the relationship h 0 /2e0 is at most about 1/10. Therefore,
we can simplify to obtain
cos ϕx ≈ 1 .
In the following, we will derive the equations needed to determine the position
of the wheelset and the acting forces in case of quasistatic curving. The wheelset is
still considered to be rigid; the wheel–rail profile combination, however, is arbitrary.
Further, we consider a right-hand curve; i.e., the rotation ϕ̇z,track of the track around
the z-axis is positive.
To begin with, the creepage equations derived in Sect. 3.2.6 have to be extended.
Another relative rotational speed arises in the contact points due to the curve radius.
The total rotational speed ϕ̇z,tot of the wheelset relative to the rail can be written as
v0
ϕ̇z,tot = ϕ̇z,track + ϕ̇z ≈ + ϕ̇z . (14.5)
R
There will also be a nonnegligible longitudinal displacement of the contact point
when the contact point of the outer wheel is in the flange or the gauge corner. For the
derivation of the displacement of the contact point in the longitudinal direction in
relation to the attack angle ϕz , see Fig. 14.4a, b, which can be found in Heumann’s
work [7]. Figure 14.4a shows a horizontal cut of the outer, in this case left, wheel
through the contact point. The vertical cut is shown in Fig. 14.4b. The (F, F)-plane
is rotated to the plane (E , D), whose normal direction is the rolling direction, with
the angle ϕz . In the (F, F)-plane, the normal force P also acts at the contact point.
The introduced point B is situated vertically above the contact point. The distance
from the contact point equals the rolling radius rL . From Fig. 14.4b, one can see that
B E = rL tan δL , (14.6)
and from Fig. 14.4a, that

ξKL = B E sin ϕz . (14.7)
Equation (14.6) substituted into Eq. (14.7) with sin ϕz ≈ ϕz finally results in the
relation we have looked for:
(a) (b)
KL F
P
A
z rL
D B L
A B
B Au
main radius
z
B
z
F E
Fig. 14.4 Longitudinal displacement of contact point in a curve according to [7], p. 88

ξKL = rL tan δL · ϕz , (14.8a)

ξKR = −rR tan δR · ϕz . (14.8b)
We also have to free ourselves from the linearized considerations in Chap. 3. This
means that we cannot use the geometric parameters from Sect. 3.2. When curving,
the contact point of the wheel often moves to the flange or even to the gauge corner.
This makes it impossible to linearize the wheel–rail geometry. The assumption of a
quasistatic condition does, however, allow the following simplifications:
u̇ x = u̇ y = ϕ̇x = ϕ̇y = ϕ̇z = 0 .
The equations for the creep velocities (3.23a)–(3.23c) with the extensions from
Eqs. (14.5), (14.8a) and (14.8b) then read as follows:
eL,R v0
vξL,R = v0 − 0 r0 ∓ 0 · rL,R ± , (14.9a)
R
v0 + 0 r0
vηL,R = − ϕz · cos δL,R ∓ ξKL,R 0 sin δL,R , (14.9b)
2
cos δL,R v0
ωζL,R = −0 sin δL,R − . (14.9c)
R
As in Chap. 3, the reference velocity
v0 + 0 r0
vm =
2
is introduced so that the creepages can be written
0 · rL,R eL,R
νξL,R = νξ0 ∓ ± , (14.10a)
vm R
ξKL,R 0
νηL,R = −ϕz · cos δL,R ∓ sin δL,R , (14.10b)
vm
0 cos δL,R
νζL,R =− sin δL,R − (14.10c)
vm R
using the same abbreviations for νξ0 as in Chap. 3. The difference between v0 and
vm has been neglected in the terms using the curve radius R.
As with the derivation of the equations of motion for the lateral dynamics, we
use a free-body diagram and establish force equilibria and moment equilibria for
the wheelset. For a rigid wheelset, there are three force equilibria and three moment
equilibria. In Fig. 14.3, the wheelset is shown with the acting forces. The creep
forces, which are assumed to be positive for the time being, are drawn in the negative
coordinate direction, showing their effect on the wheelset. Additionally, the following
assumptions apply:
• As in Chap. 8, there are no displacements u x or rotations ϕy .

• The difference between the x- and ξ-directions can be neglected. This means that
there will be no distinction between the (y, z)-plane and the (η, ζ)-plane.
• The contribution of the lateral creep forces (caused by the displacement of the
contact point) to the moment equilibrium about the z-axis is also ignored.
• The vehicle moves with a so-called cant deficiency, i.e., the superelevation of the
rail is not able to fully compensate for the centrifugal acceleration v 2 /R.
With these assumptions, the equilibrium equations become

Fx = 0 = − TξL − TξR + cx ex ϕz − cx ex ϕz = −TξL − TξR (14.11a)

Fy = 0 = − TηL cos δL − NL sin δL − TηR cos δR + NR sin δR
+ Hy − 2cy u y + m r ay , (14.11b)

Fz = 0 = − TηL sin δL + NL cos δL + TηR sin δR + NR cos δR
+ FzL + FzR − m r g , (14.11c)

Mx = 0 =(−TηL cos δL − NL sin δL )rL
− (TηR cos δR − NR sin δR )rR
− (TηL sin δL − NL cos δL )eL
− (TηR sin δR + NR cos δR )eR + FzL ex − FzR ex , (14.11d)

My = 0 = + TξL rL + TξR rR + MζL sin δL − MζR sin δR , (14.11e)

Mz = 0 = + TξL eL − TξR eR − 2cx ex2 ϕz − MζL cos δL
− MζR cos δR . (14.11f)
Here Hy is the force that acts on the wheelset because of the centrifugal force; ay is the
so-called track plane acceleration. The acceleration is given in the canted coordinate
system.
In this system of equations, the normal forces NL and NR are unknown, as are also
the lateral displacement u y and the yaw angle or attack angle ϕz . The contact angles
are functions of u y . The creep forces can be determined from the normal forces and
u y and ϕz . Hence in total, there are only four unknowns for six equations so far.
In further considerations, we will neglect the spin moments MζL and MζR . In
looking at the force balance and the moment balance with respect to the y-axis, it
is discovered that they can be achieved at the same time only under certain circum-
stances: the longitudinal creep forces need to be the same in magnitude but oppositely
oriented, and the rolling radii rL and rR must be the same. In curving, however, this
will almost never be the case. This means that a driving moment has to be introduced
so that Eq. (14.11e) with MζL = 0 and MζR = 0 can be written as
Ma = −TξL rL − TξR rR .
Because of the driving moment, a residual longitudinal creep force, which we call
Tξa , remains. Equation (14.11a) then becomes
Tξa = −TξL − TξR .
Hence, the missing unknowns in our system of equations are a driving moment
Ma and the longitudinal creep force Tξa . If the driving moment compensates the
longitudinal creep force Tξa created by the curve, then the creep νξ0 equals zero,
despite the driving moment. Then
v0 = 0 r0 .
The Eq. (14.10) for the longitudinal creepages is then simplified (vm ≈ v0 ) to
rL,R eL,R
νξL,R = ∓ ± . (14.12)
r0 R
Figure 14.5 shows an example of a wheelset with underradial position in a curve.

The wheelset is displaced outward beyond the rolling line. The creepages in the
longitudinal and lateral directions are given by the Eqs. (14.12) and (14.10b). On
both wheels, the lateral creepages are negative: on the left wheel, both terms from
Eq. (14.10b) are negative. On the left wheel, the term that arises from the displacement
Fig. 14.5 Forces acting on a z

wheelset in an underradial
position in the curve. The
wheelset is displaced
outward beyond the rolling
line. The solid line
longitudinal creep force TL
represents a nondriven state,
the dashed lines a driven
state T L
v0
T R
T L
Fig. 14.6 Needed relative

angle between wheelset and
bogie frame for radial
steering in a curve
b
v0
e0 e
x
zl
R
zt
ex
of the contact point can be set to zero, since ξKR ≈ 0. The result is that both lateral
creep forces acting on the wheelset point outward in the curve. This is the opposite
of what would be needed to help the wheelset to negotiate the curve.
When looking at the longitudinal creep forces, we begin by considering a driven
state, marked by dotted arrows. Because of the assumptions made (wheelset displaced
outward beyond the rolling line), the creep resulting from the rolling radius difference
on the outer wheel is larger then the creep from the curvature. The creep is negative,
but the creep force acting on the wheel is positive. On the inner wheel, the rolling
radius difference is negligible, since the contact point is normally on the tread area.
The creep from the curvature results in a positive creep force on the inner wheel
as well. The sum of the creep forces is not equal to zero and, as discussed above,
a driving moment is needed to fulfill the moment equilibrium with respect to the
y-axis.
In the nondriven state, the longitudinal creep forces have to balance each other.
The longitudinal creep on the inner wheel can only be positive, i.e., the creep force
on the wheel point in the negative x-direction if the creep νξ0 in Eq. (14.10) is greater
than zero. This means that for curving without driving moment, a longitudinal creep
always arises that can be interpreted as braking of the wheelset (cf. Fig. 3.17).
Let us now look at the moment equilibrium about the vertical axis Eq. (14.11f)
in more detail. The larger the relative angle ϕz between wheelset and bogie frame,
the larger the restoring moment from the longitudinal spring forces. According to
Fig. 14.6, the relative angle needed for a radial position can be written as
b
ϕz = . (14.13)
R
The force moment to obtain this rotation can be built up only by the longitudinal
creep forces. To achieve radial steering, we must have
Table 14.1 Optimal choice of some key parameters for running on straight track and in curves
Parameter Stability on straight track Curving
Conicity λ Low High
Longitudinal primary stiffness cx High Low
Wheel distance in bogie 2b Long Short
2cx ex2 b
eL TξL − eR TξR = . (14.14)
R
In Chap. 11, the term cx ex2 was introduced as the bending stiffness of the bogie.
The stiffer the longitudinal wheelset guidance (bending stiffness), the greater
must be the longitudinal creep forces for radial steering. This, in turn, means that the
wheelset has to move farther outward in the curve than for ideal curving of a free
wheelset. The larger rolling radius difference results in a force moment balancing the
moment from the spring forces. Hence for good curving behavior of a rail vehicle, the
bending stiffness should be as low as possible. Also, a low lateral stiffness enabling
lateral displacement of the wheelset is beneficial. Both, however, are contradictory
to the demands for high critical speed on straight track. The suspension design of
a vehicle is therefore always a compromise between stability on straight track and
good curving performance, i.e., a curving performance resulting in low wear rates.
Table 14.1 summarizes how the most important design parameters should be chosen
for running on straight track and in curves. The authors which to point out, however,
that in reality, the phenomena are more complex than described here. For example,
for very low conicities, significant yaw motions of the carbody can also arise if the
excitation frequency according to Klingel coincides with the yaw eigenfrequency of
the carbody.
14.5 Curving of Bogies and Entire Vehicles
In this section, the curving behavior of bogies and two-axle vehicles will be discussed.
The governing equations are similar to those derived above for a single wheelset.
Calculations without computer are possible only with a large number of simplifi-
cations. To determine the quasistatic equilibrium is quite elaborate because of the
nonlinearities. It has to be calculated numerically on a computer. Most methods used
today are based on finding the state with minimum energy. An iterative solution is
necessary. Multiple local minima are possible, which implies that there is not always
a unique solution. Numerical aspects will not be discussed further, however. Details
can be found in, e.g., [18–21].
14.5.1 Curving According to Uebelacker and Heumann
Before powerful computers were available, the calculation method developed by

Uebelacker [3] and Heumann [5, 7, 22, 23] was used. In the following, the basics
of the method will be presented. A more extensive description can be found in
Bussmann’s doctoral thesis [24]. In the thesis can also be found comparisons to
results produced with modern numerical methods.
To allow for hand calculations, besides the assumptions made in Sect. 14.4, a
number of other simplifying assumptions are necessary:
1. The vehicle or bogie has a rigid frame. The wheelsets are guided in the longitudinal
direction in this frame, i.e., cx = ∞.
2. Wheel treads are cylindrical, i.e., δ0 = 0.
3. The load-carrying and guiding functions are fully separated. At the wheels where
flange contact occurs, a second contact point arises with a normal direction parallel
to the track plane.
4. There are no driving, braking, or coupler forces.
5. The resulting creep force at each wheel has reached saturation (μQ).1
Heumann’s theory allows us to neglect some of the assumptions. The calculations,
however, become very elaborate, so that it is easier in that case to use computer
simulation instead.
With the assumptions made in Heumann’s theory, the motion of the bogie is
described by a roll motion of the wheelset in the direction of the longitudinal axis of
the vehicle and a rotation about the vertical axis. The normal from the center of the
curve to the longitudinal axis of the vehicle is called the friction center (in German:
Reibungsmittelpunkt). A rolling motion of cylindrical wheels in the longitudinal
direction does not induce curving forces. To determine the forces, it is therefore
sufficient to look at the rotation of the vehicle or bogie around the friction center. Its
longitudinal distance lRM from the leading wheelset is unknown a priori. Depending
on bogie design and curve, the friction center can be situated in front of or behind the
trailing wheelset; cf. Fig. 14.7. In large-radius curves, it is always behind the trailing
wheelset, and the inner wheel of the trailing wheelset does not have flange contact.
This situation is called free curving (in German: Freilauf). In narrow curves, the inner
wheel of the trailing wheelset has flange contact, since the lateral clearance limits
the lateral wheelset movement. This position of the wheelset, where the flanges of
the outer wheel of the leading axle and the inner wheel of the trailing axle are in
contact, is called constrained curving (in German: Spießgang).
For constrained curving, the position of the friction center is determined geometri-
cally by the wheelset distance, the lateral clearance, and the curve radius. According
to Fig. 14.8,
s lRM − b
tan ϕz,Ra ≈ = . (14.15)
2b R
1Q is the wheel–rail force perpendicular to the track plane. For cylindrical wheels, Q = N .
14.5 Curving of Bogies and Entire Vehicles 261
Fig. 14.7 Position of

friction center (•) for running direction
different curve radii and
different bogie positions
according to [24] 2b
lRM
S
free
curving
constrained
curving
friction centre
After the equation has been rearranged, the position of the friction center can be
found as
sR
lRM = b + . (14.16)
2b
The longitudinal and lateral creep forces can be calculated directly for constrained
curving. With the assumed sliding regime and Heumann’s assumption that the result-
ing creep forces are perpendicular to a line from the friction center to the contact
Fig. 14.8 Geometrically b

determined friction center
for constrained curving. (The z, Ra
z,Ra
vehicle is reduced to a line;
the distance between the two
rails is the lateral clearance.)
According to [25]
~s
l RM
RM
~R
z, Ra
z,Ra
point at each wheel, the resulting creep force at each wheel is (cf. Fig. 14.9)
TiL,R = μQ iL,R i = 1, 2 . (14.17)
The longitudinal and lateral creep forces can be written as
TξiL,R = μQ iL,R sin ξiL,R , (14.18a)

TηiL,R = μQ iL,R cos ξiL,R , (14.18b)
with
e0 lRM 2b − lRM
sin ξiL,R = , cos ξ1L,R = , cos ξ2L,R = .
qiL,R q1L,R q2L,R
Unknown are the two guiding forces H1 and H2 . They are calculated from the force
equilibrium in the y-direction and the moment equilibrium about the vertical axis:
H2 − H1 = Tη2L + Tη2R − Tη1L − Tη1R

ay
− (Q 1L + Q 1R + Q 2L + Q 2R ) (14.19a)
g
H1
1L
Q 1L Q 1L cos 1L
Q 2L q1L
q2L
2L RM
1L
l RM
Q 1R
q2R
q1R
Q 2R
whelset 2 1
H2 rotation
Fig. 14.9 Lever arms, creep forces, and guiding forces for constrained curving, according to [25]
H1 2b = (Tη1L + Tη1R )2b + (Tξ1L + Tξ1R + Tξ2L + Tξ2R )e0

2bay
+ (Q 1L + Q 1R ) . (14.19b)
g
Here ay is the track plane acceleration in the curve. The force equilibrium in the x-
direction for nondriven (nonbraked) bogies is always fulfilled because of symmetry,
since the friction center is always at the center of the vehicle in the lateral direction.
If Eqs. (14.19a) and (14.19b) result in a guiding force in the opposite direction to
that assumed, then the assumption of constrained curving was wrong. The guiding
forces can only be compressive forces. Instead, free curving will occur.
For free curving, only the guiding force at the leading wheelset is present. The
second unknown is the position of the friction center. The creep forces can no longer
be determined as in constrained curving, since the friction center is unknown. The
solution, therefore, has to be found iteratively. The minimum principle, established
by Heumann, assumes that the guiding force is minimized. The task is therefore to
find the friction center in a way that
d Hy1
= 0.
dlRM
The forces acting for free curving are indicated in Fig. 14.10.
1L
H1
Q 2L Q 1L Q 1L cos 1L
q1L
q2L
2L
RM 1L
l RM Q 1R
q2R Q 2R q1R
wheelset 2 1
rotation
Fig. 14.10 Lever arms, creep forces, and guiding forces for free curving according to [25]
14.5.2 Curving of Bogies with Suspension
The simplified curving theory in the section above leads to realistic results only in
tight curves. Especially two of the assumptions above (cf. p. 249) strongly influence
the results: the assumption of cylindrical wheels and the totally rigid wheelset sus-
pension. In Fig. 14.11, therefore, for a typical case, the real directions of the creep
forces as they act on the wheelset are shown. The forces are shown in a wheelset-fixed
coordinate system for a bogie negotiating a curve of ca. 1000 m radius. The vehicle
is assumed to run at cant deficiency, i.e., in the wheel–rail contact, centrifugal forces
also have to be transferred. The longitudinal suspension between wheelset and bogie
is assumed to be relatively stiff. In Fig. 14.11, this is illustrated by a rigid connection.
In the given example, only one contact point exists at each wheel. This is not always
the case. As in the theory of Heumann, two-point contact can occur.2 At the outer
wheel of the leading wheelset, the contact point is assumed to be in the flange root.
Also for large cant deficiencies, the contact point at the outer wheel of the trailing
wheelset will move to the flange root to balance the centrifugal force. As can be seen
in comparison to Fig. 14.10, the results differ from those of Heumann’s theory for
free curving. The main reason is that Heumann uses cylindrical wheels.
The first deviation from Heumann’s theory for the creep force for free curving is
that the lateral creep forces at the trailing wheelset point to the inner side of the curve.
2 Two- or even three-point contact is most common at the outer wheel of the leading wheelset.
Fig. 14.11 Sketch of resulting creep forces for a bogie with relatively stiff longitudinal wheelset
guidance according to [17] (Creep forces acting on the wheelset in a wheelset-fixed coordinate
system are shown)
They are quite large, since they have to take up a part of the centrifugal forces. Further,
in moderate curves, as in the example with a curve radius of 1000 m, the wheelset can
build up a sufficient rolling radius difference inducing a pair of longitudinal creep
forces that rotates the wheelset around the negative z-axis. With cylindrical wheels
this is impossible. At the trailing wheelset, a relatively small moment rotating in the
opposite direction arises.
Ideally, each wheelset takes up an equally large part of the centrifugal force. The
forces at the trailing wheelset, however, have to balance the force moment from the
longitudinal creep forces at the leading wheelset. Further, the lateral creep forces at
the leading wheelset also point outward in the curve and can therefore not balance
the centrifugal force. The only possibility for the leading wheelset to take up parts of
the centrifugal force is the lateral component of the normal force in the flange root.
Therefore, the sum of lateral forces3 at the trailing wheelset will be significantly
larger than at the leading wheelset.
The discussion above illustrates the advantages of bogies with soft longitudinal
wheelset guidance in curves: The longitudinal creep forces needed for radial steering
are much lower. The wheelset can even take a slightly overradial position, which
makes the lateral creep forces point to the inside of the curve, so that they can help
to compensate for the centrifugal force. Both of these factors result in a more equal
distribution of the track shift forces. In consequence, the lateral creep forces at the
trailing bogie are lower. An example for typical creep forces for such a bogie in the
same curve as above is given in Fig. 14.12.
3 The sum of all force components in the lateral direction (track plane) is called the track shift force

or Y -force.
Fig. 14.12 Sketch of resulting creep forces for a bogie with soft longitudinal wheelset guidance
according to [17] (Creep forces acting on the wheelset in a wheelset-fixed coordinate system are
shown)
14.6 Wear Calculation in the Wheel–Rail Contact
As already mentioned in the introduction to this chapter, the estimation of the

expected wear on wheel and rail is an important criterion for the design of the vehi-
cle. Wear here means only abrasive wear; plastic deformation or work hardening
phenomena are not considered. The exact determination of wear is not easy, since it
depends on a number of parameters.
The usual criteria are based on the determination of the work performed in the
contact area. The material removal is assumed to be proportional to the work in the
contact area. The work performed is the sum of the products from creep forces and
sliding distances:
W = Tξ dsξ + Tη dsη + Mζ dϕζ [Nm] . (14.20)
Work per time unit is

dW
P [Nm/s] = ,
dt
i.e., the sum of the products of creep forces and sliding speeds. The quotient P
divided by the vehicle speed results in the work per meter running distance or energy
dissipation in the contact, which is the sum of the products of creep forces and
creepages,
P W
= = Tξ νξ + Tη νη + Mζ νζ [Nm/m] . (14.21)
v l
W
The quantity is easily determined in multibody simulations. Sometimes, the
l
specific energy dissipation normalized with the size of the contact patch is also
determined:
P W W
PA = or = .
A l A lA
14.6 Wear Calculation in the Wheel–Rail Contact 267
(a) 25 (b)
1200
20 1000 two-axle wagon

[ m rad [
two-axle wagon
W [ Nm/ m [
9m axle distance
9m axle distance 800
15 Y25 bogie
Y25 bogie 600
z
10 link bogie
l
link bogie 400
5
200
0 0
100 200 300 400 500 600 700 100 200 300 400 500 600 700
curve radius R [m] curve radius R [m]
Fig. 14.13 Comparison between different freight wagon running gear a attack angle of leading
wheelset, b energy dissipation per meter of outer wheel of the leading wheelset. Friction coefficient
μ = 0.3, track plane acceleration ay = 0.6 sm2 . According to [26]
The energy dissipation in the contact, and in turn the wear, depend strongly on the
curve radius and on the type of running gear. In Fig. 14.13, for a freight wagon with
three different types of running gear, the attack angle for the leading wheelset and
the energy dissipation for the outer wheel of that wheelset are shown [26]. It can be
seen that the two-axle vehicle has the highest values, since the axle distance of 9 m
is much longer than that in the bogies of the other two vehicles (1.8 m). The Y25
bogie is worse than the link bogie, since the wheelset guidance is much stiffer.
The criteria to judge wear introduced above have two disadvantages:

• The absolute wear cannot be predicted; only comparisons between different cases
are possible.
• It is impossible to predict where in the contact patch wear occurs, since only global
values for sliding speeds, creepages, and creep forces are considered.
With the help of proportionality factors derived from measurements, however, the
relationship between energy dissipation and material removal can be established
[27–32]. From laboratory tests, the proportionality factors given in Fig. 14.14 have
been derived for UIC90A rail steel [29, 33, 34]. Two different wear regimes can be
seen: for low values of energy dissipation, mild oxidative wear is observed, while
large-energy dissipation leads to severe metallic wear. The proportionality factors
for the two areas differ by about a factor of 10. It has to be said, however, that it is not
easy to derive the proportionality factors. Therefore, different values can be found
in the literature.
If the contact area is divided into elements, the material removal can be calculated
for each element where sliding occurs. In the adhesion zone of the contact patch, no
wear will occur. This can be used to try to predict the long-term changes of wheel
and rail profiles with the help of simulations. A possible methodology is described
in Linder’s thesis [33].
[10-6 g / Nm ] material removal

per frictional work k v,severe= 7.8 x10-6
8
6
p v,limit
4
2 k v,mild = 7.8 x10-7

frictional work
per area
0 2 4 6 8 [ W/mm2 ]
Fig. 14.14 Quantitative wear calculation law for the example of the UIC90A wheel steel. According
to [34]
Fig. 14.15 Wear map for steel grade UIC900A under dry conditions [36]. Contact pressure p and
sliding velocity vsli p . Material hardness H
Another wear model is based on Archard [35] and was implemented, for example,
by Jendel [36] and Enblom [37]. The wear depth is calculated as a function of contact
pressure, sliding distance, and surface hardness:
pz · ds
dz = ki · . (14.22)
H
14.6 Wear Calculation in the Wheel–Rail Contact 269
The wear coefficients are to be determined by testing, replicating the relevant wear
mechanisms in the laboratory or identifying them in the field. The Archard wear
coefficients are presented in two-dimensional maps as a function of contact pressure
and sliding velocity; see Fig. 14.15.
14.7.1 Directions of Creep Forces for Different Wheelset

Positions
In what direction do the creep forces point if the wheelset in Fig. 14.5 cannot achieve
sufficient rolling radius difference?
In what direction do the creep forces point if the wheelset is displaced more to
the outside in the curve than would be needed for ideal curving of a free wheelset?
Two-point contact at the outer wheel is assumed, i.e., there is one contact point on
the wheel tread and another contact point at the wheel flange.
Fig. 14.16 Bogie with

z
mounting errors of the
wheelsets
uy
14.7.2 Deviation of Angle of Attack or Lateral Displacement

of Wheelsets
The influence of a deviation in the wheelset mounting in a bogie (Fig. 14.16) can be
described with the equations for quasistatic curving as well. Find a simple motivation
for why this is the case.
Qualitatively, what amounts of creepages and creep forces arise in both cases if
the center of gravity of the bogie frame moves in the x-direction with constant speed?
14.7.3 Curving of Single Wheelset
Develop an algorithm that solves Eqs. (14.11a)–(14.11f). Assume conical wheel pro-
files and the vehicle data from Table 9.2. The creep forces can be calculated with the
simplified theory of Shen–Hedrick–Elkins. Determine the wheelset position in the
curve for a number of curve radii. Also vary vehicle parameters and conicity.
References
1. F.J. Redtenbacher, Die Gesetze des Locomotiv–Baues, (The Laws of Design of Locomotives)
(Bassermann, Mannheim, 1855)
2. C. Boedecker, Die Wirkungen zwischen Rad und Schiene und ihre Einflüsse auf den Lauf und
and rail and their influences on the running behavior and the resistance of vehicles in railway
trains). (Hahn’sche Buchhandlung, Hannover, 1887)
3. H. Uebelacker, Untersuchung über die Bewegung von Lokomotiven mit Drehgestellen in
Bahnkrümmungen (Investigation on the motion of locomotives with bogies in curves). Organ
für die Fortschritte im Eisenbahnwesen, 58: Beilage (1903), pp. 158–162
4. H. Heumann, Zum Verhalten von Eisenbahnfahrzeugen in Gleisbögen (The behavior of railway
vehicles in curves). Organ Fortschr. Eisenb.-wes., 68 (1913), pp. 104–108, 218–121, 136–140,
158–163
5. N.N. Dauner, (Vorsitzender). Vorläufige Richtlinien für den Fahrzeugbau zur Erzielung guter
Führung der Fahrzeuge im Gleis, aufgestellt von der Arbeitsgemeinschaft für die Untersuchung
der Fahrzeugführung im Gleis im Auftrag des Reichsverkehrsministeriums (Preliminary guide-
lines for vehicle design in order to get good vehicle behavior on track). Technischer Bericht,
Deutsche Reichsbahn, Stuttgart, 1944. Mitglieder der Arbeitsgemeinschaft waren: Abt.Präs.
Dauner (Vorsitzer), OR. Dr. Ing. Bäseler, OR. Dr. Ing. Bingmann, Prof. Dr. Ing. Heumann
Technische Hochschule Aachen, OR Hiller (Schriftführer), RDir Hörmann ab Jan. 1942, RDir.
Jaehn, OR. Krauß vom Juli 1942 bis Febr. 1943, Abt.Präs. Dr. Ing. Meier ab Febr. 1943, OR
Dr. Ing. Mielich, Prof. Dr. Ing Pflanz Technische Hochschule Prag, OR Schmidt-Kleewitz bis
Mai 1942, OR. Dr. Ing. Schramm vom Mai 1942 bis Juli 1942, RDir. Dr. Ing. Troitzsch, Dr.
Ing. Vogel Gesellschaft für Oberbauforschung Berlin, Abt. Präs. Dr. Ing. e.h. Wagner bis Nov.
1942, RDir. Witten ab Nov. 1942
6. W. Dauner, E. Hiller, W. Reck, Einfürung in die Spurführungsmechanik der Schienenfahrzeuge
(Introduction to track guidance behavior of railway vehicles). Archiv für Eisenbahntechnik,
Beiheft zu Eisenbahntechnische Rundschau, Folge 2, S (1953), pp. 1–26
References 271
7. H. Heumann. Grundzüge der Führung der Schienenfahrzeuge. Elektrische Bahnen, 49–

52:Sonderdruck von Arbeiten aus den Jahren 1950–1953 (Principles of guidance of railway
vehicles. Reprint of publications from 1950–1953 and 1954)
8. A.O. Gilchrist, The long road to solution of the railway hunting and curving problem. Proc.
Inst. Mech. Eng. 212, 219–226 (1998)
9. A.H. Wickens, The dynamics of railway vehicles - from stephenson to carter. Proc. Inst. Mech.
Eng. Part F 212, 209–217 (1998)
10. J. Mackenzie, Resistance on railway curves as an element of danger. Proc. Inst. Civ. Eng. 74,
1–83 (1883)
11. R. von Helmholtz, The causes of wear of tyre and rail in track curves and design possibilities
for their avoidance. Z. VDI 32(330–335), 353–358 (1888)
12. S.R.M. Porter, The mechanics of a locomotive on curved track. Proc. Inst. Mech. Eng. 126,
457–461 (1934)
13. S.R.M. Porter, The mechanics of a locomotive on curved track. Railway Eng., 55 (1934/1935),
pp. 205–206, 255–257, 282–287, 318–330, 384–386, 424–428
14. S.R.M. Porter, The Mechanics of a Locomotive on Curved Track (Railway Gaz, London, 1935)
15. S.R.M. Porter, The mechanics of a locomotive on curved track. Railw. Gaz. 91(232–238),
432–434 (1935)
16. D. Boocock, The steady state motion of railway vehicles on curved track. J. Mech. Eng. Sci.
11(6), 556–566 (1969)
17. E. Andersson, M. Berg, S. Stichel, Rail Vehicle Dynamics, Rail Vehicle Dynamics (KTH Royal
Institute of Technology, Stockholm, 2014). ISBN 978-91-7595-420-2
18. W. Kik, K. Knothe, H. Steinborn, Theory and numerical results of a general quasi-static curving
algorithm, In Wickens, A., ed. Proceedings 7th IAVSD Symposium on Dynamics of Vehicles
on Roads and on Tracks, Cambridge, Sept 1981. (Swets & Zeitlinger, Amsterdam, 1982), pp.
427–440
19. W. Kik, H. Steinborn, Führ- und Störverhalten - Ermittlung statischer und quasistatischer Gle-
ichgewichtslagen, (Guiding and disturbance behavior - Determination of static and quasistatic
positions of equilibrium). VDI-Berichte 510, 275–284 (1984)
20. K. Knothe und, L. Mauer, Inkrementelle Formulierung eines Algorithmus für Mehrkörpersys-
teme zur Untersuchung der stationären Gleichgewichtslage von Eisenbahnfahrzeugen (Incre-
mental formulation of an algorithm for multibody systems in order to investigate the stationary
equilibrium of railway vehicles). ILR-Mitteilung 59. (Institut für Luft- und Raumfahrt, TU
Berlin, 1979)
21. K. Knothe und L. Maurer., Inkrementelle Formulierung eines Algorithmus für Mehrkörper-
systeme mit Anfangslasten und nichtlinearen Zwangsbedingungen (Incremental formulation
of an algorithm for multibody systems with initial loads and nonlinear constraints). Z. Angew.
Math. Mech. 60, T42–T44 (1980)
22. H. Heumann, Grundzüge des Bogenlaufs von Eisenbahnfahrzeugen (Principles of curving of
railway vehicles). Die Lokomotive 39, (1942)
23. H. Weber, Prof. Heumanns Arbeiten auf dem Gebiet der Spurführung im Zeichen der heutigen
Rad/Schiene-Technik (Prof. Heumann’s papers in the area of track guidance under the influence
of today’s wheel-rail technology). ZEV Glasers. Annalen 102(7/8), 201–213 (1978)
tberichte VDI, Reihe 12, Nr. 338 (also dissertation TU Berlin). VDI-Verlag, Düsseldorf, 1997
26. S. Stichel, Modellierung und Parameterstudien zum Fahrverhalten von Güterwagen mit UIC-
Fahrwerken. ZEV+DET Glas Ann. 123(7/8), 289–296 (1999)
27. T.M. Beagly, Severe wear of rolling/sliding contacts. Wear 36, 317–335 (1976)
28. P. Bolton, P. Clayton, Rolling-sliding wear damage in rail and tyre steels. Wear 93, 145–165
(1984)
29. H. Krause, G. Poll, Wear of wheel-rail surfaces. Wear 113(1), 103–122 (1986)
30. H. Krause, H. Lehna, Investigation of tribological characteristics of rolling-sliding friction

systems by means of systematic wear experiments under well-defined conditions. Wear 119,
153–174 (1987)
31. O. Pigors, Verschleißuntersuchungen an Radwerkstoffen im Labor (Wear investigations for
wheel materials in the laboratory). DET - Die Eisenb. 23, 359–361 (1975)
32. H. So, The mechanism of oxidational wear. Wear 184, 161–164 (1995)
33. C. Linder, Verschleiß von Eisenbahnrädern mit Unrundheiten (Wear of out-of-round wheels).
Ph.D. thesis, ETH Zürich (1997)
34. W. Specht, Beitrag zur rechnerischen Bestimmung des Rad- und Schienenverschleißes durch
Güterwagendrehgestelle (Contribution to the arithmetic determination of wheel and rail wear
by freight wagon bogies, Ph.D. thesis, RWTH Aachen (1985)
35. J.F. Archard, Contact and rubbing of flat surfaces. J. Appl. Phys. 24, 981–988 (1953)
36. T. Jendel, Prediction of wheel profile wear - methodology and verification. Licentiate Thesis
TRITA-FKT 2000:49, Royal Institute of Technology, Department of Vehicle Engineering,
Railway Technology, Stockholm (2000)
37. R. Enblom. On simulation of uniform wear and profile evolution in the wheel–rail contact. Ph.D.
thesis TRITA-FKT 2006:83, Royal Institute of Technology, Department of Vehicle Engineer-
ing, Railway Technology, Stockholm (2006)
Chapter 15
Determination of Load Collectives
for Vehicle Components
15.1 Introduction
For most components of rail vehicles, including bogie frames and wheelsets, dynamic
loads are dimensioning loads. The dynamic loads are superimposed on static loads
stemming from the weight of the vehicle. This is especially true for high-speed
vehicles. For trams, metros, and freight wagons, the maximum static load can be
decisive for dimensioning components. For optimal dimensioning of a component,
it is therefore important to know the dynamic loads exerted on a vehicle throughout
its lifetime as exactly as possible.
Dynamic loads, however, are today often taken into account via dynamic factors
(e.g., [1]). The loads on the considered component have to be below the endurance
limit according to the concept of Wöhler. Such a calculation can, for example, be
found in [2] for a wheelset with rubber suspended wheel rim. The failure of such a
wheel caused a disastrous accident in Eschede in 1998. In [3], the dynamic maximum
loads where not taken into account by load factors but by a calculation in the frequency
domain. The general concept was the endurance limit, however.
More realistic is a calculation of a component based on a fatigue damage spectrum
with variable amplitude fatigue calculation. A number of excellent textbooks on this
topic exist. For further studies, [4–7] are mentioned here as examples. As the name
indicates, to perform such a calculation, the load spectra for the component have
to be known. In some cases, for dimensioning a bogie frame, measured forces in
the primary suspension have been used [8, 9]. This is, of course, possible only if
the vehicle in question is similar to the vehicle on which the measurements where
carried out.
In the authors’ opinion, multibody simulation provides a heretofore seldom used
opportunity to gain knowledge about loads on a rail vehicle component. For other
vehicle dynamics investigations, detailed vehicle models are built up. Also, the forces
in the connection elements, i.e., the springs and dampers, are results of this type of
simulation and could be used to generate load spectra. However, the representation
of flexible bodies in multibody simulation models has to be improved. Other modes

DOI 10.1007/978-3-319-45376-7_15
274 15 Determination of Load Collectives for Vehicle Components
than those typically used for ride comfort evaluation probably have to be used to
enable correct determination of stresses in the components. Procedures to do this
exist.
One of the main problems is probably assumptions regarding the operational
conditions of the vehicles. Among these is information about track irregularities, the
distribution of speeds, and curve radii. Also, information about single events such
as running on switches and crossings and their frequency on the network can be
important. Further, rail vehicles have a long life, 30 years or more. It is a big task
to predict all operational conditions for a vehicle over such a long time. This is one
of the reasons while simplified methods are still frequently used. The authors are
convinced, however, that operators in the future will order vehicles based on certain
fatigue load spectra. The infrastructure owner then has to ensure that the network
is maintained according to defined standards. Therefore, the task is easier for rail
vehicles than for road vehicles.
In the following section, methods to determine fatigue load spectra for rail vehicles
are presented. A concept like the one described here will sooner or later become a
standard procedure for the fatigue design of rail vehicle components, like ride comfort
and running stability calculations today.
15.2 General Procedure
The general procedure is sketched in Fig. 15.1. The fatigue calculation is performed
in several steps:
• Starting from a multibody simulation (1a), a stress simulation (1b) is carried out.
• Based on this a stress, collective (2) has to be determined.
• In a final step, a damage hypothesis has to be employed and the proof of strength
has to be performed (3).
As indicated above, first the stresses in a component in selected cross sections have
to be determined (Sect. 15.3). For this task, the forces in the coupling elements
(springs and dampers) that act on the component and the inertial forces within the
component have to be known. This is a result of the multibody simulation. Further,
the relationship between loads and stresses has to be known. This relationship can
be determined with a finite element model of the component.
In the next step, the stresses have to be evaluated (Sect. 15.4). For the procedure
presented here, load collectives have to be generated. The load collectives contain
information about what stress amplitudes can be expected with what frequency during
the lifetime of the component.
The third and last step is the proof of strength itself; see Sect. 15.5. This is per-
formed with the help of a damage hypothesis. It is assumed that every load cycle
causes a certain amount of damage. Such damage accumulates to a total degree of
15.2 General Procedure 275
load case 1 FE-model stresses in cross section

of component
vehicle response
z z
y cut
y z x
mbs- Fx x
transformation Fy
calculation = B uu Fz y x
Iy ,Iz ,A z
cut
(1b)
cycle
(1a)
counting
stress collective
(n n ) i
summation
max ni
ai
other load cases

n tot
m
0 2
10 10 10 4 10
6
10 8
log H
(2)
proof of strength
damage accumulation
D<1?
(3)
Fig. 15.1 General procedure of proof of strength
damage that must not exceed a certain level. In most cases, the damage is normalized
with the number of cycles to failure to be between 0 and 1.
These three steps are described in detail in Sects. 15.3–15.5.
15.3 Stress Calculation in Components
Stresses in a component result from forces that are introduced via the coupling
elements or from inertia forces in the component itself. How to determine spring
forces, damper forces, or inertia forces from a multibody simulation is described
in other parts of this book. For the determination of stresses, there are different
possibilities.
15.3.1 Finite Element Calculation in Each Time Step
The first possibility is to perform the stress calculation within a finite element (FE)
program. We assume that an FE model of the investigated component, for example a
bogie frame, exists. The time histories of the vehicle reactions are determined with
the multibody simulation tool. In each time step, the forces acting on the bogie frame
and the rigid-body accelerations are transferred to the FE program. Mass, center of
gravity, and moments of inertia of the multibody model must have been calculated
from the mass matrix in the FE program. In commercial FE programs this is a standard
calculation. If this is not the case, the following must hold:
⎡ ⎤
ρ dV 0 ⎡ T⎤ ⎡ ⎤
⎢V ⎥ ex
⎢ 0 ρ dV ⎥ ⎢ ⎥
0 ⎥ = ⎣ eT ⎦ ⎣ M
⎢ y FEM ⎦ ex ey ez , (15.1)
⎣ V ⎦
0 0 ρ dV eTz
V
where the vectors ex , ey , and ez represent rigid-body motions in the x-, y-, and and
z-directions. The reader is asked to derive the equations for determination of center
of gravity and inertia matrix on his own.
The result of the FE calculation is the time histories of stresses in each element
of the model. The numerical effort of this procedure, however, is immense, since
the system of linear equations of the FE program, typically consisting of several
thousand unknowns, has to be solved in each time step. Therefore, this method is not
optimal.
15.3.2 Stress Calculation with the Help of Transformation

Matrices
In the second methodology, the stresses are calculated only at a limited number of
cross sections in the bogie frame. With the FE program it is possible to calculate a
transformation matrix between the external forces acting on the component and the
stresses in certain cross sections. This assumes that the FE model is linear elastic and
that the contributions from the different forces can be superimposed on the complete
stress state. Forces in connection elements are a standard output in each multibody
simulation program. In general, the relation
⎧ ⎫ ⎡ ⎤⎧ ⎫
⎪
⎪ σ1 ⎪⎪ b11 b12 ··· b1n ⎪
⎪ f1 ⎪
⎪
⎨ σ2 ⎪
⎪ ⎬ ⎢ b12 ⎥⎪⎨ ⎪
⎢ b22 ··· b2n ⎥ f2 ⎬
.. =⎢ . .. .. .. ⎥ .. , (15.2a)
⎪
⎪
⎪ . ⎪ ⎣ ..
⎪ . . . ⎦⎪⎪ . ⎪
⎪
⎩ ⎪ ⎭ ⎪
⎩ ⎪
⎭
σm bm1 bm2 · · · bmn fn
15.3 Stress Calculation in Components 277
is obtained, or more briefly,
σ = B σf f . (15.2b)
With the help of a transformation matrix between the outputs of the multibody sim-
ulation program and the forces in the connection elements, a transformation matrix
between displacements, velocities, and accelerations of the degrees of freedom and
the stresses can finally be determined:
σ = B σf B fu u = B σu u . (15.3)
In that case, the stress calculation is reduced to a matrix multiplication in each time
step. Below, the methodology is explained with the help of a simple example.
Example for a Rigid Body
As an example, we take the longitudinal beam of a bogie frame. The forces in
Fig. 15.2 that are introduced in the structure via the primary suspension are results of a
multibody simulation. Additional outputs are the accelerations. With these measures,
the internal forces in the longitudinal beam can be calculated. The equation for the
tensile stress can be written as
My (x) z Mz (x) y N (x)
σx = + + . (15.4)
Iy (x) Iz (x) A
Below, we will look at the calculation of the bending moment My (x) in more detail.
According to Fig. 15.3, the internal moment can be calculated as My (x = l) with the
vertical force Fz and the inertia forces
z
cut
y
z z
x
x
Fx
Fy y x
Fz
I y ,I z , A
cut
Fig. 15.2 Forces on the longitudinal beam of a bogie frame

l
z, u z (x) lf z z
l yl z x
..
u z (x)
y x
Fz
x
lcg
Fig. 15.3 Stresses in the longitudinal beam of the bogie frame resulting from the force in the
primary spring Fz and the inertia forces
l
My (x = l) = Fz lF − μü z (x) · (l − x) dx . (15.5)
0
In case of rigid bodies,
ü z (x) = ü z,rigid − ϕ̈y,rigid · (lcg − x) , (15.6)
where lcg is the distance from the origin of the coordinate system to the center of
gravity of the bogie frame. The rigid-body accelerations can be extracted from the
integral
My (x = l) =
l l
Fz lF − μ(l − x) dx ü z,rigid + μ(l − x)(lcg − x) dx ϕ̈y,rigid . (15.7)
0 0
The integrals in Eq. (15.7) can be determined in advance. Letting
μ l = m LB , and lF = lcg = l ,
Equation (15.7) can be written as
l l2
My (x = l) = Fz lF − m LB ü z,rigid + m LB ϕ̈y,rigid . (15.8)
2 3
In multibody simulation programs, this expression can be defined as output, since it
can be represented as a linear combination of displacements and accelerations. In a
similar way, the internal forces N (x) and moments Mz (x) can be determined.
15.3 Stress Calculation in Components 279
With Eq. (15.4), the stresses σx (x = l) can be defined directly, since they are
linear combinations of outputs from the simulation as well.
Procedure in Case of Flexible Bodies
In the case of flexible bodies, the procedure is similar. The displacements in the
flexible body are usually approximated with a Ritz approach. In the bodies with a
flexible representation, the rigid-body degrees of freedom are replaced by generalized
degrees of freedom. The set of modes that are used in the Ritz approach have to be
able to represent kinks in the internal moments and jumps in the internal forces.
Detailed descriptions of this topic can be found in Dietz [10] and in Schwertassek
and Wallrapp [11].
Implementation of the Procedure for Different Analysis Methods in Multibody
System Codes
The methodology described above can be applied both in the time domain (time step
integration) and in the frequency domain (spectral analysis). In the time domain, with
Eq. (15.3) the solution is already given. The transformation is performed in each time
step. The result is the stress–time history.
If a spectral analysis is performed, the matrix of the power spectral densities Su
of the output quantities have to be multiplied by the transformation matrix B σu from
both sides,
Sσ () = B σu ()Su ()B σu ()T . (15.9)
The results are spectral densities of stresses at some selected cross sections.
15.4 Determination of Load Collectives
The result of Sect. 15.3 is stresses—either as a function of time or as spectral

densities—at a number of cross sections in the structure. For the strength verifi-
cation, these have to be further evaluated. This is the second step in the calculation
scheme from Sect. 15.2. To understand the evaluation methods presented below, it is
useful to think about how to determine sustainable stresses. The proof of strength in
a structure then is the comparison between actual stresses and sustainable stresses.
15.4.1 Determination of Sustainable Stresses
The appropriate method to calculate sustainable stresses depends on the type of

loading of the structure. The most important distinction is between static and dynamic
loading. For rail vehicles, dynamic loads are very important.1 Therefore, the fatigue
1 Static loads, for example from lifting a carbody in the workshop, are not discussed here.
vibration resistance
low cycle fatigue high cycle fatigue fatigue strength variable amplitude
cycle fatigue
load cycle
a> D
a
D
m m m
0 0 0
t t t
a
number of cycles N : < 5 10 4 5 10 4 < N < 2 106 > 2 10 6 N = 10 4 - 106

W
a/ D
a/ D
a/ D
öh W fat
a/ D
W W
ler öh öh öh igu
cu le le le el
rv rc rc rc ine
e ur ur
log
log
log
ur
log
ve ve ve
1 1 1 1
log N 2 10 6 log N 2 10 6 log N 2 10 6 log N

frequency distribution failure area
Fig. 15.4 Classification of vibration resistance (according to [4]); σa = stress amplitude, σ D =

endurance limit σ̄a = maximal stress amplitude, σm = mean stress of one-stage test, σ̄a = maximal
stress amplitude of stochastic stress history σ̄m = mean stress of stochastic stress history
strength is important when a structure is being dimensioned. On the one hand, the
number of load cycles must be known. Further, an important distinction is whether
the amplitudes and mean values of the load cycles are constant.
Most information regarding fatigue strength of materials stems from tests con-
ducted with loads oscillating with a sinusoidal shape with constant amplitude and
mean value. A reason is, on the one hand, the available test equipment. On the other
hand, tests with random loads tend to exceed a reasonable effort. In most cases,
the real loads have varying amplitudes. Because of a lack of characteristic values,
the Wöhler diagrams that are results of so-called single-stage tests are still used for
strength verification. The sustainable stress is given as a function of mean stress, as
can be seen in Fig. 15.4.
Low cycle fatigue usually covers the range up to 5 · 104 load cycles. Since all
structures in rail vehicles are exposed to more load cycles, low cycle fatigue is not
described here.
High cycle fatigue occurs between 5 · 104 and 2 · 106 load cycles. As can be seen
in Fig. 15.4, high cycle fatigue is characterized by a dependence of the sustainable
stress on the number of load cycles. In traditional fatigue calculation of rail vehicles,
high cycle fatigue is not important either, since all relevant dynamic load cases
reach numbers of load cycles that are above 2 · 106 . For the assessment of variable
amplitude fatigue strength as described below, however, this range of load cycles is
again interesting.
15.4 Determination of Load Collectives 281
For a number of load cycles above 2 · 106 , we talk about the endurance limit. In
Fig. 15.4, from the horizontal line it can be concluded that the sustainable stress is
independent of the number of load cycles. According to Wöhler, for a given mean
stress, there is a stress amplitude that can be repeated with an infinite number of cycles
without causing fatigue damage. Since nobody knows whether an infinite number of
load cycles is really allowed, sometimes the term technical endurance limit is used.
In the damage accumulation hypotheses used for assessment of variable amplitude
fatigue (Sect. 15.5), however, there are also other theories, where stress amplitudes
below the endurance limit also contribute to the total damage.
As mentioned above, most structures are exposed to loads with varying ampli-
tudes, as in Fig. 15.4 (right).
The maximal load occurs very seldom in a stochastic stress history. To require
that this stress be sustained for an infinite number of load cycles is too conservative.
Therefore, the strength is assessed with variable amplitude fatigue. In this approach,
one tries to evaluate what amplitudes occur with what frequencies in the stochas-
tic process (cf. Sect. 15.4.2). In Fig. 15.5, it can be seen that for a given maximal
amplitude, the number of tolerable load cycles depends on the shape of the load col-
lective. Or in other words, for a given number of load cycles, the sustainable stress
level depends on the shape of the load collective. The slimmer the load collective
becomes, i.e., the more seldom high amplitudes occur in relation to the total number
of load cycles, the lighter the structure can be.
shape of collective
N/mm 2
300
a
max. stress amplitude
200
PÜ= 90%
R = -1
100
F F
weld sample St 37/52
50
10 4 10 5 10 6 10 7 10 8
number of cycles to failure N
Fig. 15.5 Sustainable stress amplitude in relation to the shape of the collective (from [4])
15.4.2 Cycle Counting Methods to Determine Load

Collectives
From Sect. 15.4.1, it is clear that it is necessary to evaluate a stochastic process in

a way that it becomes possible to assess the strength of the structure. This is done
with cycle counting methods. The result is load collectives. Load collectives contain
information as to the number of load cycles with certain features such as amplitude
for which a mean value can be sustained.2 Here we give only some general remarks
regarding the variety of existing cycle counting methods. More thorough descriptions
can be found in Buxbaum [4], Hänel [12], Haibach [6], Liersch [13], and Naubereit
and Weihert [7].
The evaluated process should possess as many properties of the original stochastic
process as possible. At least it should be statistically equivalent. With usual cycle
counting methods, however, the influence of order and vibration frequency of load
cycles is lost. The evaluation can be performed in the time domain or in the frequency
domain.
Time Domain
If the load or stress history is a result of measurements or from time step integration
in a simulation and available as stress–time history, then the evaluation is performed
by real counting of each load cycle. The load cycles are then sorted into predefined
classes xi .
There are counting methods that register one parameter and methods that store
two parameters. Many different variants exist that are defined, e.g., in DIN 45667
[14]. The result of a one-parameter counting method is presented as, for example,
the amplitude of a stress history as a function of the frequency of occurrence.
The accuracy of the evaluation can be increased by evaluating two parameters of
the process. Usually, this means amplitude σa and mean value σm of the stress history.
The result is a two-parameter load collective. Basically, there are two possibilities
for two-parameter classification.
The first possibility is two-parameter peak value counting, whereby the cycles are
taken from the load history in their natural order (Fig. 15.6). Two adjacent extreme
values and the corresponding mean value are counted as one cycle. The second
possibility is the rain-flow counting method. In this methodology, first the cycles
with the smallest amplitudes are extracted. Thereafter, cycles with larger and larger
amplitudes are continuously extracted (Fig. 15.7). Only complete cycles are extracted
from the process. To judge the residuum, i.e., the rest of the process that no longer
contains any complete cycles, there are also different possibilities [13, 15]. The
simplest possibility is simply to neglect the residuum. The rain-flow cycle counting
method seems to emerge as the standard counting method, since according to today’s
knowledge, it reflects the material behavior in the best way [15].
2 The transformation from forces into stresses could be performed after the determination of collec-
tives.
8
7
6
class 5
4
3
2
1
class of maxima
1 2 3 4 5 6 7 8
1 1
2 1
class of minima
3 2
4
5 2
6 1 1
7
8
Fig. 15.6 Two-parameter peak value counting (according to [12])
8
class
6 class of maxima
4
1 2 3 4 5 6 7 8
2
1 1
class of minima 2
8 1 1
3
class
6 4
4 5 2
2 6 2
7
8 8
class
6
4
2
Fig. 15.7 Rain-flow classification (according to [12])
In Fig. 15.8, the difference between the two above-mentioned counting methods
is illustrated with the help of a short part of the load history. Rain-flow counting
leads to higher extracted amplitudes. It will always give a more conservative load
collective.
One problem that usually arises in calculating in the time domain is that only
very short parts of the usual operation of a rail vehicle can be simulated, due to long
simulation times. This means that the load collectives have to be extrapolated. The
extrapolation is not trivial. It is not enough to just multiply the classified cycles by
a certain factor. Statistically, the amplitudes that arise for a one-million-kilometer
running distance are higher than those for a one-kilometer running distance. This
has to be taken into account in the extrapolation. This will not be discussed here.
Possible solutions can be found, for example, in [4, 16] or [17].
Frequency Domain
Load collectives can also be determined if the stresses are calculated as spectral den-
sities. We assume that the track excitation is a stationary random process with normal
(a) (b)
a1 m1
a1 m1
a2 m2
a2 m2
original process replaced process t original process replaced process t
Fig. 15.8 Comparison between two-parameter peak counting (left) and rain-flow counting (right)
(according to [12])
distribution (cf. Sect. 6.3). For linear models, in this case the vehicle response is also
a stationary normal distributed process. The load collective can then be determined
by characteristic values of the random process. In [18], these are introduced in a very
illustrative way. In the following, the parameters needed for evaluation are briefly
mentioned.
Parameters of a Stationary Random Process with Normal Distribution
The spectral density of a stationary random process with normal distribution has the
three parameters (moments) below:
∞
s02 = () d , (15.10a)
0
∞
s22 = 2 () d , (15.10b)
0
∞
s42 = 4 () d . (15.10c)
0
The measures s0 , s2 , and s4 can be interpreted as the quadratic mean values of the
process and its first and second derivatives. The parameter s0 has here as a standard
deviation the unit of a stress.
The mean values of the number of rising zero crossings per second N0 and of the
maxima N1 are defined as
N0 = s2 /(2πs0 ) and N1 = s4 /(2πs2 ) , (15.11a, b)
where N0 is the “effective frequency” of the random process.

The relationship between the number of zero crossings and the number of maxima
is the coefficient of irregularity i of the process:
i = N0 /N1 . (15.12)
From i, the bandwidth of the stochastic process can be calculated:

ν= 1 − i2 ; 1 ≥ ν ≥ 0 . (15.13)
The coefficient of irregularity contains information about the “benignity” of the

process with respect to the sensitivity regarding the choice of cycle counting method.
As with classification in the time domain, in the frequency domain there are also
different possibilities to derive a load collective. Both analytic approximations exist
as empirical expressions. All, however, use the above-mentioned characteristic para-
meters. As in the previous section, we jump directly to the two-parameter counting
methods.
At the beginning of the 1960s, Kowalewski [19] and Sjöström [20] indepen-
dently derived an approximative solution for the two-dimensional density function
of two consecutive extreme values with the corresponding mean values of a Gaussian
process. If it is assumed that the two parameters are statistically independent, they
can be combined by multiplication:
σa −σa2 /(2s02 i 2 ) 1
e−σm /(2so ν ) . (15.14)
2 2 2
f (σa , σm ) = f a (σa ) · f m (σm ) = e ·√
s02 i 2 2π s0 ν
The density function f (σa , σm ) has the unit 1/stress2 . The amplitudes σa are
Rayleigh distributed; the instantaneous mean values σm have a normal distribution.
The stress amplitudes in Fig. 15.9 are normalized. As in the time domain, a two-
dimensional load collective can be derived that is equivalent to a two-parameter
0.07
0.06
0.05
m, a )
0.04
f(
0.03
0.02
4
0.01 3.5
3
0 2.5
-5 -4 2
-3 -2 1.5
-1 0 1
a
1 2 3 4 0.5
m 5 0
Fig. 15.9 Normalized two-dimensional density function of amplitudes and mean values
peak counting method (Fig. 15.6). The total number of load cycles per second with
the current mean value in the interval σm,i −σm,i+1 and with amplitudes in the interval
σa,i − σa,i+1 is calculated from the double integral
σm,i+1 σa,i+1
N = N1 f (σm , σa ) dσm dσa , (15.15)

σm,i σa,i
with the effective frequency from Eq. (15.11) and the density function from
Eq. (15.14). The results become better, the nearer the coefficient of irregularity is
to 1. With the help of a suitable transformation of the two-parameter collective into
a single-parameter one, as described in Sect. 15.4.3, for processes with large band-
width, results can be also obtained that are comparable to collectives from test results.
The derived collectives are equivalent to those determined in the time domain.
The proof of fatigue strength from here on is analogous in the time and the frequency
domains.
Since the rain-flow counting method has evolved as the method that represents
the fatigue behavior of a material in the best way, researchers have tried to develop
an equivalent method in the frequency domain. Up to now, however, there is no exact
theoretical solution in the frequency domain. Further, for all existing methods in the
frequency domain, the random process has to be a stationary Gaussian process.
The usual way to get an empirical approximation is to compare it with a real
rain-flow counting in the time domain. Dirlik [21] has developed a closed solution
for a rain-flow collective derived from a spectral density with the help of extensive
simulations. For the probability density of the distribution of stress cycles, i.e., double
amplitudes, he writes
2
D1 e −Z
Q +
D2 Z e −Z −Z 2
2R 2 + D Z e 2
Q 2 3
p(2σa ) = R , (15.16)
2s0
σa
with Z = and the auxiliary variables
so
s12 i − σm − D12
D1 = 2(σm −2i ) ;
2
σm = ; R=
;
2
s0 N 1 1+i 1 − i − D1 + D12
1 − i − D1 + D12
D2 = 1− R
D3 − D2 R) .
; D3 = 1 − D1 − D2 ; Q = 1.25(i − D
1
A new definition of the rain-flow method by Rychlik [22] opened the way to
a theoretical solution to derive rain-flow load collectives from spectral densities.
Bishop and Sherrat [23] picked up the new formulation and refined it. The basic
idea is to calculate the probability density for nonadjacent maxima and minima from
the probability density of adjacent maxima and minima. Hereby the assumption
is made that the process is defined as a Markov chain, i.e., at a certain number
of points. Since for the probability density of two adjacent maxima and minima,
only the approximation from Eq. (15.14) is available, the final result can only be
an approximation as well. In [23], this solution was compared with an empirically
obtained solution. It turns out that the method gives the closest results to a rain-flow
counting in the time domain. The results, however, are only marginally better than
those obtained with the approximation from Dirlik.
15.4.3 Conversion from a Two-Parameter to a One

Parameter Collective
The result from a two-dimensional classification method is, as described above,

a two-dimensional matrix with amplitudes and the corresponding mean values of
a stochastic process. The variable-amplitude fatigue strength proof with the help
of damage accumulation methods is defined only for one-dimensional collectives.
Therefore, one has to try to converge the two-dimensional collective into a one-
dimensional one without losing the information about the mean values. For rail
vehicles, e.g. in [24–26], the method of extended amplitudes described by Hänel [27]
is suggested. It can be applied to results from time domain and frequency domain
calculations. The extended amplitude is calculated in relation to the mean value of
the load cycles,
σae = σa + ψ|σm − σ̄m | , (15.17)
where σ̄m is the global mean value of the process. The mean stress sensitivity ψ
represents the reduction of sustainable stress, as can be seen, for example, from the
Smith diagram (Fig. 15.10). For one-stage tests, ψ can be determined directly from
the Smith diagram, independent of whether ψ is constant or weakly depending on
σm . The following relation is given:
σai = σw − ψσmi . (15.18)
For stochastic loads, ψ has to be determined experimentally. In Fig. 15.11, a load cycle
transformed to the global mean value is shown. The amplitude of the transformed
load cycle is larger than the original one. Figure 15.12 gives an example of a one-
parameter collective.
In [27], as a result of random tests for ψ values between 0.2 and 0.8 for the rain-flow
method (ψ = 0.3 is recommended) and between 0.6 and 0.8 for the peak counting
method are determined. Since rain-flow counting results in a more conservative
collective than peak counting (cf. Sect. 15.4.2), for the latter, higher values should
be used to get one-parameter collectives causing the same amount of damage in the
material.
Haibach [6] suggests the procedure of amplitude transformation for the reduction
to one-parameter collectives. In this method, the reduction is performed already
R eH
m2 a2
m1
w
a1
m1 m2 m
Fig. 15.10 Determination of mean stress sensitivity ψ from the Smith diagram
ae
m
m m
original process t replaced process t
Fig. 15.11 Rain flow classification and reduction to a one-parameter collective (from Hänel [12])
during the classification of the stochastic process. The one-parameter collective is

derived directly. The motivation from material physics is the same as above, namely
that the damage caused by a load cycle with a certain amplitude increases with
increased mean value of the load cycle. Amplitude transformation, however, is not
possible in the frequency domain.
Fig. 15.12 One-parameter

load collective; H = ( n )i
cumulative frequency of
occurrence, n i = number of max ni
load cycles in the ith class,
ai
n tot = total number of load
cycles in collective, σai =
stress amplitude in the ith
class, σmax = maximal stress n tot
in collective m
0 2 4 6 8
10 10 10 10 10
log H
Table 15.1 Simplified list of operation cases for Berlin S-Bahn with 11 different load cases,
percentage in relation to total running distance
v (km/h) Pay load (%) Straight track Curve Switch
(%)
R = 250 m R = 500 m
0–20 20 3.45
55
90
20–40 20 8.81
55
90
40–60 20 26.16 9.58 %
55 5.99
90 0.70
60–80 20 30.79 5.76 % 0.66 %
55 7.06
90 0.86
15.4.4 Superposition to Total Load Collective
The calculation of dynamic loads in vehicle components is usually divided into sep-
arate parts. Simulations with empty or with loaded vehicle, or at different speeds, are
usually performed separately. In Table 15.1, a matrix with 11 load cases is given that
represent the different operational conditions (in German called “Einsatzspiegel”
[28]) of a commuter train vehicle (S-Bahn) in Berlin. Each operation case is given
with its percentage of the total running distance. The matrix could have more ele-
ments; for example, different track qualities could also be introduced.
For each operation case, a load collective at the chosen spots in the structure is
derived as explained in previous sections. If the calculations are performed in the time
domain, extrapolation is needed in each element of the matrix. Since the total running
distance throughout the life of this type of vehicle is about 4.5 million kilometers, the
average running distance in each load case is about 400,000 km. It is impossible to
simulate such a running distance. The extrapolation implies the problems that were
discussed in Sect. 15.4.2.
If the calculation is performed with spectral analysis, then the number of load
cycles per second in each class is determined with Eq. (15.15). This number has
to be multiplied by the number of seconds traveled per load case, which in turn is
calculated from the running distance and the running speed.
The superposition of the different cases, in our example 11 load cases, is simple.
The numbers of load cycles in each class are just added. One has to be careful,
however, in choosing the same classes for each load case.
In his dissertation, Flach [16] points out that when using this procedure, the
changes of maximal stresses must not be neglected. These are the load cycles that
arise from the transition from one load case to another, for example changing from
high negative additional stresses in a right-hand curve to high positive stresses in a
left-hand curve. This implies that not only the frequency of occurrence of a load case
has to be known, but also the order along the line.
In a British study that exemplifies a fatigue calculation for a running gear on a
vehicle of the London metro [29], instead of a matrix with load cases, a representative
part of the line is chosen. For this track, a load collective is determined, which then
is extrapolated to get the load collective representing the whole life of the vehicle.
Again it has to be pointed out that it is not easy to extrapolate a 10 km simulated track
to several millions of kilometers and to represent all loads with the right number of
cycles.
15.5 Damage Accumulation: Proof of Strength
The last step in our calculation scheme is the proof of strength of the investigated
structure. This is achieved by evaluating the total load collective determined above.
The variable-amplitude fatigue strength proof is carried out using a damage accu-
mulation hypothesis.
15.5.1 Damage Accumulation Hypotheses
In 1945, Miner published his methodology of linear damage accumulation [30].

Already in 1924, in Sweden, Palmgren published a study in which the same procedure
is used to predict the fatigue life of rolling bearings [31]. Today, this linear damage
accumulation procedure is called the Palmgren/Miner hypothesis.
The Palmgren/Miner hypothesis is the best-known method to proof fatigue
strength on structures exposed to loads with variable amplitudes. Two assumptions
form the basis of the hypothesis:
15.5 Damage Accumulation: Proof of Strength 291
Wöhler curve
original process replaced process
N1 N2 N
N1 N2
Fig. 15.13 Linear damage accumulation hypothesis (from [4])
• It is assumed that the damage D increases linearly in a test with constant amplitude.
• Further, it is assumed that the damages from different load cases i can be super-
imposed.
This results in an equation for the total damage,
ni
D= Di = , (15.19)
i i
Ni
where n i is the number of load cycles in a load level i, and Ni the maximum permitted
number of load cycles derived from a Wöhler line. Failure of the component occurs
if the damage sum is equal to 1. Therefore, the following must hold:

Di ≤ 1 . (15.20)
i
The calculation procedure is illustrated in Fig. 15.13. Palmgren and Miner were
criticized because of the simplified assumptions in their method. Some of the main
issues are:
• The assumption of linear damage accumulation does not have a physical back-
ground.
• Stresses below the endurance limit do not contribute to the damage.
• The order of the loads is disregarded.
• There is no distinction between the damage during crack initiation and during
crack propagation, even though these processes might have different mechanisms.
Nevertheless, the Palmgren/Miner method is frequently used today. The main reason
is probably that the well-known Wöhler curves can be used, and no tests determining
new characteristic curves are needed.
Most of the other known damage accumulation hypotheses are based on Palm-
gren/Miner. Good overviews can be found in Haibach [6] and Buxbaum [4]. The most
significant difference is whether load cycles with amplitudes below the endurance
limit are taken into account. For fatigue life calculations of rail vehicles, the damage
accumulation hypothesis from Corten and Dolan [32] is sometimes recommended.
As a basis, a system of Wöhler curves with different notch types is used. The char-
acteristic property of the procedure proposed by Corten and Dolan is that the line
of high-cycle fatigue (in a double logarithmic scale) continues with the same slope
below the endurance limit. Further, the damage accumulation is not linear, i.e., dam-
ages can no longer just be added.
15.5.2 Concepts for Variable-Amplitude Fatigue Strength

Proof for Rail Vehicles
For the proof of fatigue strength with the method presented in this chapter, some
problems usually arise. Wöhler curves often exist only for samples without notches.
The effects of size, surface quality, and notches have to be estimated. Finally, as
mentioned before, the Wöhler curves are based on single-amplitude tests that do
not reflect the real loading. The optimum would be fatigue calculation of a vehicle
component by tests with variable amplitudes. Due to economic restrictions, this is
usually not possible.
As mentioned above, most regulations for fatigue life calculations of rail
vehicles are based on the endurance strength. On the basis of a standard from
the former GDR TGL 28875 [33] (Schienenfahrzeugbau-Berechnungsgrundlagen-
Festigkeitsnachweis), however, a variable-amplitude fatigue strength proof is possi-
ble. First, the proof of strength is performed separately for each stress component
and for each load collective. The following must hold:
max σ̂ ≤ lim σbe , (15.21)
where max σ̂ is the maximum value of the collective as a maximum stress (not stress
amplitude); lim σbe is the permissible maximum level of the collective based on
the sustainable stress level from a single-amplitude test depending on the relation
between maximum and minimum stress, the shape of the collective, and the notch
type. A detailed description of the procedure and the suggested superposition of
different stress components can be found, for example, in [26].
Also, in a draft version of a European standard for calculation of fatigue strength of
carbodies of rail vehicles [34], the possibility of proof of strength in case of variable
amplitude loads is given.
Most concepts for variable-amplitude fatigue strength (also the TGL mentioned
above) are based on nominal stresses. The calculation of nominal stresses is compar-
atively easy, and a large number of Wöhler curves for nominal stresses are available,
especially for welded structures. The use of nominal stresses, however, is also criti-
cized, since the mechanisms around notches are not taken into account in an adequate
way.
15.5 Damage Accumulation: Proof of Strength 293
Buse and Voss [24] proposed a procedure for variable-amplitude fatigue strength
proof on bogies of high-speed vehicles based on a notch base concept. Since peak
stresses in the notch base might exceed the elasticity limit, the linear relationship
between load and stress is lost. With the Neuber rule, the elastoplastic stress hysteresis
in the notch base can be derived from the elastic stress if the cyclic stress–strain curve
of the material is known; see, for example, [6].
A proof of fatigue life as proposed here will probably not be the standard method
in the coming years. The possibilities of multibody simulations today and the possi-
ble coupling between multibody simulations and finite element calculations provides
excellent possibilities, however, to take dynamic loads into account in a more exact
way than today. We know that such concepts are used more and more within com-
panies today.
15.6.1 Transformation Matrix Between Degrees of Freedom

in a Multibody Simulation Model and Stresses
in the Bogie Frame
Give the transformation matrices Bσf and Bσu for the stress σx (x = l) in a bogie frame
according to the example in Sect. 15.3.2. Assume a model for the lateral dynamics
of a bogie like that in Chap. 11.
15.6.2 Determination of Collective of Spring Forces

with Spectral Analysis
Look at the vehicle model from Sect. 7.5.1. Calculate the two-dimensional load col-
lective for the primary and secondary spring forces for the same stochastic excitation
and for a two-million-kilometer running distance.
References
1. Leichtbau der Verkehrsfahrzeuge (Lightweight design of transport vehicles). Lastannahmen

und Sicherheiten für Schienenfahrzeuge (Load assumptions and safety of rail vehicles).
Leaflet steel 415, 3rd edn, Forschungskreis 1 der Studiengesellschaft “Leichtbau der Verkehrs-
fahrzeuge”, o. O. (1970)
2. K. Knothe, A. Böhmer, Eschede - Einige Bemerkungen zur Dauerfestigkeit von Eisenbahn-
radsätzen (Eschede - Some remarks on the fatigue strength of railway wheelsets). Int. Verkehr-
swesen 50(11), 542–546 (1998)
3. A. Böhmer, T. Klimpel, K. Knothe, Dynamik und Festigkeit von gummigefederten Radreifen

(Dynamics and strength of rubber suspended wheel rims). ZEV+DET Glasers. Annalen 124(3),
223–230 (2000)
4. O. Buxbaum, Betriebsfestigkeit. Sichere und wirtschaftliche Bemessung schwingbruchge-
fährdeter Bauteile. 2. Auflage (Variable-amplitude fatigue analysis. Safe and economic design
of devices with risk of fatigue failure. 2nd edn). Verlag Stahleisen mbH, Düsseldorf (1992)
5. H. Gudehus, H. Zenner, Leitfaden für eine Betriebsfestigkeitsrechnung - Empfehlung zur
Lebensdauerabschätzung von Maschinenbauteilen (Guide for variable-amplitude fatigue
analysis - Recommendation for the life span appraisal of components of machines) (Verlag
Stahleisen mbH, Düsseldorf, 1995)
6. E. Haibach, Betriebsfestigkeit: Verfahren und Daten zur Bauteilberechnung, (Fatigue. Process
and data for the component analysis) (VDI-Verlag, Düsseldorf, 1989)
7. H. Naubereit, J. Weihert, Einführung in die Ermüdungsfestigkeit (Introduction into fatigue
strength) (Hanser, München und Wien, 1999)
8. G. Voß, Bedeutung der Lastannahmen für die Bemessung von Schienenfahrzeugen (Importance
of design loads for the design of railway vehicles). ZEV - Glas. Ann. 107(2), 33–36 (1983)
9. G. Voß, Ermittlung von Betriebslastkollektiven für Hochgeschwindigkeitsdrehgestelle (Deter-
mination of operation load collectives for high-speed bogies). ZEV - Glas. Ann. 111(6), 180–
187 (1987)
10. St. Dietz. Vibration and Fatigue Analysis of Vehicle Systems Using Component Modes.
Fortschritt-Berichte VDI (also Ph.D. thesis TU Berlin), Reihe 12, Nr. 401. VDI–Verlag, Düs-
seldorf (1999)
11. R. Schwertassek, O. Wallrapp, Dynamik flexibler Mehrkörpersysteme (Dynamics of flexible
multibody systems) (Vieweg, Braunschweig, Wiesbaden, 1998)
12. B. Hänel, Über die Ableitung von Beanspruchungskollektiven aus dem Beanspruchungsprozeß
(On the derivation of demand collectives based on the demand process). IfL-Mitteilung 27, 49–
62 (1988)
13. J. Liersch, Klassieren von stochastischen Beanspruchungen mit Hilfe eines Rain-Flow-
Zählverfahrens (Classification of stochastic demands using a rainflow counting procedure).
Diploma thesis, TU Berlin, Institute for aeronautics and astronautics (1994)
14. DIN. Klassierverfahren für das Erfassen regelloser Schwingungen (Cycle counting methods
for random vibrations). DIN 45667, Deutsches Institut für Normung e.V., Beuth Verlag, Berlin
(1969)
15. U.H. Clormann, T. Seeger, RAINFLOW-HCM Ein Zählverfahren für Betriebsfestigkeitsnach-
weise auf werkstoffmechanischer Grundlage (RAINFLOW-HCM A numerical procedure for
the verification of operational stability based on material mechanics). Stahlbau 55, 65–71
(1986)
16. M. Flach, Rechnerische Lebensdauerberechnung für stochastische Lasten im Schienen-
fahrzeugbau (Lifespan analysis for stochastic loads at railway vehicle construction). TIM-
Research reports, FOMAAS, Universität-GH Siegen (1999)
17. E.J. Gumbel, Statistics of Extremes (Columbia University Press, New York, 1958)
18. B. Hänel, Betriebsfestigkeitsnachweis auf der Grundlage der Spektraldichte des regellosen
Beanspruchungsprozeßes (Variable-amplitude fatigue proof based on the spectral density of
random demand). IfL-Mitteilung 15, 252–260 (1976)
19. J. Kowalewski. Über die Beziehungen zwischen der Lebensdauer von Bauteilen bei
unregelmäßig schwankenden und bei geordneten Belastungsfolgen (On the relation between
life of components for irregularly fluctuating and regular load sequences). Bericht Nr. 249,
Deutsche Versuchsanstalt für Luft- und Raumfahrt (1963)
20. Sjöström, S, On Random Load Analysis. Technical report 181, Transactions of the Royal
Institute of Technology, Stockholm, Sweden (1961)
21. T. Dirlik, Application of Computers in Fatigue Analysis Thesis (University of Warwick, Coven-
try, 1985)
22. G. Lindgren, I. Rychlik, Rain flow cycle distributions for fatigue life prediction under gaussian
load processes. Fatigue Fract. Eng. Mater. Struct. 10(3), 251–260 (1987)
References 295
23. N.W.M. Bishop, F. Sherrat, A theoretical solution for the estimation of rainflow ranges from
power spectral density data. Fatigue Fract. Eng. Mater. Struct. 13(4), 311–326 (1990)
24. H. Buse, G. Voss, Lebensdauersimulation durch Betriebsfestigkeitsrechnung an Drehgestellen
des Hochgeschwindigkeitsverkehrs (Life cycle simulation by analysis of variable-amplitude
fatigue analysis for bogies of high-speed vehicles), vol. 1219, pp. 543–563. VDI-Verlag, Düs-
seldorf (1995)
25. C. Dorn, Beitrag zur Synthetisierung von Rainflow-Kollektiven für die betriebsfeste Dimen-
sionierung von Schienenfahrzeugen (Contribution to the synthesis of rainflow collectives for
the dimensioning of railway vehicles). Dissertation, Universität Hannover (1993)
26. S. Stichel, Betriebsfestigkeitsberechnung bei Schienenfahrzeugen anhand von Simulation-
srechnungen (Variable-amplitude fatigue analysis of railway vehicles based on simulations).
VDI Fortschritt–Berichte (also Ph.D. thesis TU Berlin), Reihe 12, Nr. 288. VDI–Verlag, Düs-
seldorf(1996)
27. B. Hänel, Zur Mittelspannungsempfindlichkeit bei regelloser Beanspruchung (Mean stress
sensitivity for random demand). IfL-Mitteilung 26, 174–181 (1987)
28. M. Ehinger, T. Fretwurst, Betriebsfestigkeit im Schienenfahrzeugbau - Stand und Anwend-
barkeit (Variable-amplitude fatigue analysis of railway vehicles. Existence and applicability).
Wiss. Z. Hochschule für Verkehrswesen Dresden 37(5), 845–875 (1990)
29. R.K. Luo, B.L. Gabittas, B.V. Brickle, An integrated dynamic simulation of metro vehicles
in a real operating environment, in The Dynamics of Vehicles on Roads and on Tracks, vol.
23, Proceedings of the 13th IAVSD Symposium held in Chengdu/China in August 1993; Sup-
plement to Vehicle System Dynamics, ed. by Z. Shen (Swets & Zeitlinger, Amsterdam/Lisse,
1994), pp. 334–345
30. M.A. Miner, Cumulative fatigue damage. J. Appl. Mech. 12(3), 159–164 (1945)
31. A. Palmgren, Die Lebensdauer von Kugellagern (Life of ball-bearings). VDI-Z. 58, 339–341
(1924)
32. H.T. Corten, T.J. Dolan, Cumulative Fatigue Damage (Institution of Mechanical Engineers,
London, 1958)
33. T.G.L. Schienenfahrzeugbau, Berechnungsgrundlagen, Festigkeitsnachweis (TGL Railway
vehicle design, basis for calculations, fatigue assessment). Departemental standard TGL
33398/01-22, DDR (1986)
34. CEN. Railway applications. Structural requirements of railway bodies. European standard,
Brussels, 2010. Amendment (2015)
Chapter 16
Appendix
16.1 List of Symbols
Skalars, Latin
a m Semiaxis of contact ellipse

a m Coefficient for actuator force in Eq. (5.37)
a0∗ m Semiaxis of adhesion zone of contact patch
an – Fourier coefficients
ay m/s2 Track plane acceleration in curve
A m2 Cross-sectional area
b m Semiaxis of contact ellipse
b Ns/m Coefficient for actuator force in Eq. (5.37)
2b m Longitudinal wheelset distance in a bogie
bn – Fourier coefficients
B(e) – Combination of elliptic integrals (Hertz theory)
Bl,v ( f ) – Frequency weighting function lateral/vertical
BGl,v ( f ) – Frequency weighting function lateral/vertical for feight traffic
c N/m Spring stiffness
cH N/m Hertzian spring stiffness
ctors Nms/rad Torsional spring stiffness
cx N/m Longitudinal wheelset stiffness (x-direction)
cy N/m Lateral wheelset stiffness (y-direction)
cb Nm/rad Bending stiffness of a bogie
cs N/m Shear stiffness of a bogie
c m Contact radius for point contact in Eq. (3.30)
Ci j – Creep coefficients (Kalker-coefficients)
Ck m Fourier coefficients
d Ns/m Damping constant
ds m Sliding distance in wear calculation
dz m Wear depth in Archard’s wear law
D – Relative damping (damping factor)
D(e) – Combination of elliptic integrals (Hertz theory)
e m Eccentricity of contact ellipse in Eq. (3.34)
2ex m Lateral distance between points of attack of longitudinal primary springs cx

DOI 10.1007/978-3-319-45376-7_16
298 16 Appendix
e0 m Half lateral distance between contact points for centered wheelset

position
ea , (ei ) m Lateral distance from track center line to outer (inner) wheel
contact point
E N/mm2 Modulus of elasticity
E – Comfort value in Table 7.1
E∗ N/mm2 Equivalent modulus of elasticity (Hertz theory)
f 1/s Frequency
fξ , f1 N Adhesion gradient in longitudinal direction
fη , f2 N Adhesion gradient in lateral direction
F N Force
F( f ) – Frequency weighting function in Eq. (7.4)
Fc N Spring force
Fd N Damping force
F1 , F2 – Help variables in Johnson’s contact theory based on Hertz
g m/s2 Gravity coefficient
gR , g S – Auxiliary variables in Eqs. (3.2) and (3.3)
hR , hS – Auxiliary variables in Eqs. (3.2) and (3.3)
G N/mm2 Shear modulus
h m Amplitude of a general periodic excitation
h m Track superelevation, height difference between outer and inner
rail
H (i) – Transfer function
H N Horizontal force
H N/m2 surface hardness
Hy N Lateral force between vehicle and wheelset due do centrifu-
gal force in curve, guiding force in curving theory according
to Heumann,
Iy , Iz m4 Area moments of inertia
J kg m2 Moment of inertia
ki – Wear coefficient in Archard’s wear law
kv kg/Nm Proportionality factor between energy dissipation and material
removal
Kb – Dimensionless bending stiffness in Eq. (11.34c)
l m length of pendulum
lRM m Distance between leading wheelset in bogie and friction center
L m Wavelength
L , Lξ, Lη m/N Flexibility in simplified creepage–creep force relation
m kg Mass
mW kg Wheelset mass
m CB kg Carbody mass
M Nm Moment
Ma Nm Wheelset driving moment
Mζ Nm Spin moment
N N Normal force in contact point
NL N/m Normal line load for contact of two cylinders
NMV – Ride comfort values according to CEN (UIC 513)
p0 N/mm2 Maximum surface pressure (hertz)
pz N/mm2 Normal pressure in the contact area
P Nm/s Frictional power (work per time unit)
PV Nm/m Frictional work per meter
PA Nm/sm2 Specific energy dissipation per area
PCT – Percentage disturbed by curve transitions
16.1 List of Symbols 299
PDE – Percentage disturbed from discrete events

qi Generalized displacement
qi m Lever arm from friction center to wheel contact point in
Heumann’s curving theory
q Amplitude of eigenmode
q0 N/m Shear flow
q N/m3 Tangential stress in the contact of two cylinders (line contact)
qξ , qη N/mm2 Tangential stresses in the contact
2Q N Axle load
Q = 21 mg N Wheel load (also vertical force at wheel contact point)
r0 m Rolling radius in centered wheelset position
rL m Rolling radius on left wheelset in noncentered position
rR m Rolling radius on right wheelset in noncentered position
R m Radius, especially radius of curvature on wheel profile
Ra m Main radius of curvature of elliptic stamp
Rb m Main radius of curvature of elliptic stamp
Rξ1 m Longitudinal wheel radius of curvature
Rη1 m Lateral wheel radius of curvature
Rξ2 m Longitudinal rail radius of curvature
Rη2 m Lateral rail radius of curvature
Rξ∗ m Equivalent longitudinal radius of curvature
Rη∗ m Equivalent lateral radius of curvature
Rm∗ m Equivalent mean radius of curvature
R , (R ) m Largest (smallest) equivalent radius of curvature
RW m Wheel radius of curvature for circular profiles
RR m Rail radius of curvature for circular profiles
s m Lateral clearance between wheel and rail
s, sξ , sη – Local (true) creepage
S – Centre of gravity
S() – Two-sided power spectral density
Sa () (m/s2 )2 Power spectrum of vehicle response (acceleration)
t s Time
T s Oscillation period
T, Tν N Resulting creep force, tangential force
Tξ , Tη N Longitudinal respectively lateral creep force
Tξa N Resulting creep force in longitudinal direction due to a traction
moment
ux m Longitudinal displacement
uy m Horizontal displacement, lateral displacement
uz m Vertical displacement
uξ , uη m Displacement difference between wheel and rail in contact patch
...
uy m/s3 3rd time. derivative of lateral displacement (jerk)
U Nm Potential energy
v, v0 m/s vehicle speed
vcrit m/s critical speed
vξ , vη m/s Wheel velocities in contact point
vξ , vη m/s Rail velocities in contact point
V () – Transfer function
W Nm Frictional work
W/L Nm/m Frictional work per length unit
Wz – Ride index (German: Wertungsziffer)
x, y, z m inertial coordinates
yt (z t ) m Track excitation from lateral alignment irregularity (vertical
alignment)
300 16 Appendix
Scalars, Greek
α – Dimensionless variable in transformation equation (11.13)

α – Phase angle caused by base point excitation
α rad/s Real part of eigenvalues (damping gets negative)
αa αδ – Auxiliary variables for Hertz theory
β – Dimensionless variable in transformation equation (11.13)
β – Dimensionless axle distance in Keizer’s work, Eq. (11.34b)
β – Phase angle of excitation
χ m Coefficient of gravitational stiffness of yaw angle
δ – Wheel–rail contact angle
δ m Elastic deformation
δ – Real part of eigenvalue (positive as damping)
δ0 – Wheel–rail contact angle in centered wheelset position
δR (δL ) – Wheel–rail contact angle at right (left) contact point
δu – Virtual displacement
– Determinant
l m Extension (positive) or compression (negative) of spring
– Difference
r m Rolling radius difference
δ – Contact angle difference
ε – Normalized coefficient of contact angle difference
ε – Strain
εe – Normalized coefficient of equivalent contact angle difference
γ – System phase angle
λ – Conicity
λe – Equivalent conicity
μ – Dimensionless mass in Keizer’s work, Eq. (11.34a)
μ kg/m Mass density of beam (Chap. 15)
ν – Rigid-body creepage
νξ , νη – Longitudinal and lateral rigid-body creepage
νζ 1/m Spin creepage
ν – Poisson ratio
σ – Normalized roll angle coefficient
σe – Normalized equivalent roll angle coefficient
σ N/mm2 Normal stress
σY N/mm2 Yield stress
σv N/mm2 Equivalent stress
τ N/mm2 Shear stress, normalized shear stress
ϕ rad Angle, angle in complex notation
ϕk rad Phase angle
ϕ rad Rotational degree of freedom
ϕx – Roll angle
ϕy – Rotation about y-axis
ϕz – Yaw angle
ϕdr rad Relative angle between wheelset and bogie with respect to ver-
tical axis
ϕ̇z,track rad Wheelset rotational speed with respect to vertical axis due to
track curvature
ϕ̇z,tot = rad Total wheelset rotational speed
ϕ̇z,track + ϕ̇z relative to track in the curve
() – Single sided PSD
16.1 List of Symbols 301
– Help variable for Johnson–Vermeulen solution of tangential contact

problem
z = x kg/m2 Wheelset moment of inertia about vertical axis
xF kg/m2 Bogie frame moment of inertia with respect to longitudinal axis
yF kg/m2 Bogie frame moment of inertia with respect to lateral axis
zF kg/m2 Bogie frame moment of inertia with respect to vertical axis
ξK L Longitudinal displacement of left contact point in curve (German:
Kurve)
ξK R Longitudinal displacement of right contact point in curve (German:
Kurve)
ζ 1/m Coefficient of gravitational stiffness of lateral displacement
ω rad/s Eigenfrequency
ωζ rad/s Angular velocity in wheel–rail contact point
rad/s Excitation frequency
0 rad/s Wheelset angular velocity, especially angular velocity for centered
wheelset position
ξ, η, ζ m coordinate system, fixed to contact point
η – Dimensionless excitation frequency
Vectors and Matrices

A – State space matrix
C – Damping matrix
I – Unit matrix
f˚ – Creep force vector
fQ – Gravitational force vector
fF – Suspension force vector
K – Stiffness matrix
M – Mass matrix
p0 , pQ – Vector with gravitational forces
p(t) – Excitation vector
T – Transformation matrix
y(t) – Vector with degrees of freedom of second-order differential equation
Indices
B Bogie (frame and wheelsets)
c Cosine term
C Creepage term
CB Carbody
c Spring term
d Damping term
e Equivalent
F Suspension term
F Bogie frame
eff Effective value
f Free
i See a
302 16 Appendix
in Inner (in curve)

l, L Left
l Leading axle
le Leading edge of contact patch
out Outer (in curve)
p Primary
q Quadratic mean value
Q Gravitational force term
red Reduced
rel Relative
r, R Right
RM Friction center (Im German: Reibungsmittelpunkt)
s Sine term
s Secondary
sat Saturated
stat Static
t Trailing axle
t(rack) Track
W Wheelset
x, y, z Coordinate directions
x’, y’, z’ Coordinate directions of a track following body fixed coordinate
system
ξ, η, ζ Coordinate directions in contact point coordinate system
Symbols and Labels
û Excitation amplitude
ū Conjugated complex quantity
ū Mean value
ũ Time dependent
˙ ũ¨
ũ, First and second time derivative
u Derivative with respect to distance
[]T Transposed matrix, transposed vector
16.2 Coordinate Systems
During the development of multibody system tools, the introduction of different coor-
dinate systems is necessary, especially if large displacements and rotations are taken
into account. But also in most cases, it is of great help in our linearized equations.
The definition of such coordinate systems is unfortunately not standardized. For
example, the contact mechanics code Contact uses a different coordinate system
from that of the the MBS code Medyna. Therefore, one has to be very careful when
developing interfaces between different simulation tools.
16.2 Coordinate Systems 303
Here we restrict ourselves to the case that the track (rail and sleeper) is rigid and
in all respects, including rail profiles, symmetric to the track center line. This means
that we exclude the ability of the rail to rotate on the sleepers or on slab track, and
we also exclude asymmetric profiles in, e.g., curves.
To characterize a coordinate system, the origin O and the unit vector in the direc-
−
→ −
→ − →
tion of the coordinate axes are needed ( i , j , k ).
Inertial Coordinate System
−
→− →− →
In the inertial coordinate system O I i I jI kI -, the absolute movements of the vehicle
are determined. It is defined in a way that it coincides with the reference coordinate
system at t = 0.
Reference Coordinate System

The ideal track geometry, i.e., curves and superelevation, are captured with the ref-
erence coordinate system.
−
→− →− →
The reference coordinate system O0 i 0 j0 k0 -moves with the vehicle speed v. The
−
→− →
O0 i 0 j0 -plane is parallel to the top of the rail plan of the undisturbed track. If only
one wheelset is present, it is convenient to put the origin of the coordinate system
at the center of gravity of the wheelset (in centered position on straight track). The
−
→ −
→
i 0 -axis points in the direction of travel of the vehicle; the j0 -axis coincides with the
−
→
wheelset axle and points to the left; the k0 -axis completes the right-hand coordinate
system.
Track Coordinate System

The track coordinate system captures the track irregularities. They are measured in
the reference coordinate system.
−
→−→− →
Also, the track coordinate system Ot i t jt kt moves with speed v. For the defini-
tion of the track coordinate system, the introduction of the rolling line is convenient.
A symmetric wheelset is displaced laterally on a disturbed track until left and right
rolling radii are similar (rL = rR = r0 ). The origin Ot is at the center of gravity
−
→
of this wheelset; the i t -axle points into the direction of travel of the vehicle; the
−
→ −
→
jt -axis points to the left; the kt -axis completes the right-hand coordinate system.
Wheelset Coordinate System
Wheelset displacements are expressed as motions of a wheelset coordinate system
relative to the reference coordinate system.
The origin of the wheelset coordinate system is at the center of gravity of the
−
→ −
→
displaced wheels. Its i W -axis is equal to the direction of travel, the jW -axis coincides
−
→
with the wheelset axle and points to the left and the jW -axis completes the right-hand
−
→
coordinate system. Despite the rotation with respect to the jW -axis, the coordinate
system moves together with the wheelset.
For the above-described choice of the coordinate system, the wheelset behaves
like a gyro, i.e., gyroscopic matrices have to be taken into account.
304 16 Appendix
Nominal Configuration
The static state of the vehicle, i.e., the state in which only gravitational forces or other
time-independent forces (for example centrifugal force in a curve with constant radius
and superelevation) exist is called the nominal configuration. On straight track, the
determination of the nominal configuration is trivial. Only in the vertical direction
are the primary and secondary springs compressed.
For quasistatic curving (no track irregularities and no transient processes), how-
ever, the determination of the nominal configuration is not easy [1–4].
The displacements of the different bodies in the vehicle can be divided into the
displacement in the nominal configuration and the excitation state.
Contact Coordinate System
−→→− →
In the contact coordinate system OK ξ − η ζ , the contact mechanical processes (creep
force, creep relations) are described.
The origin OK is the center of the contact ellipse as long as the contact patch
−
→→ −
→
is elliptical. The OK ξ − η -plane is the tangential contact plane; ξ points in the
−
→
direction of travel, ζ is the outer normal of the rail, and −→
η completes the right-
hand coordinate system.
16.3 Fundamentals of Contact Mechanics
16.3.1 Hertz Contact Mechanics
Initial Comment
In the following, a short summary of Hertz contact mechanics will be given. The main
purpose is not to provide a complete mathematical derivation with elliptic integrals.
The aim is rather to give an impression of the results that can be expected from using
Hertz contact mechanics. The presentation is based on the textbook of Johnson [5].
The notation, however, is adapted to mesh with our notation.
Contact Condition
Consider two bodies with second-order surfaces. At the outset, the two bodies just
touch. They then approach each other with a distance δ; see Fig. 16.1. The total
approach is the sum of the deformations of bodies 1 and 2, i.e., δ1 plus δ2 . Now the
new distance of the points A1 and A2 can be expressed with the same coordinate
(x, y):
h (x, y) = h(x, y) − (δ1 + δ2 ) + (w1 (x, y) − w2 (x, y)) . (16.1)
The two δ-terms are compressions with the same sign, which have to be added. The
two displacements w1 and w2 are positive if the respective point is displaced in the
16.3 Fundamentals of Contact Mechanics 305
Fig. 16.1 Approach of the z

two bodies 1 and 2 . The M1
y-axis, which is not shown, 1
points into the figure. It is
1
possible to see Fig. 16.1 in
accordance with Fig. 3.1 if
we assume that the
(OK , ξ, η, ζ)-coordinate
system in Fig. 3.1 coincides
with the
(O, x, y, z)-coordinate
system in Fig. 16.1 R x1
2 w2
A2
1
0 w1 x
A1
a a
R x2
2
2
M2
positive z-direction (cf. Fig. 16.1). This is the case for A1 but not for A2 . Therefore,
the sign of w2 is negative.
If the points A1 and A2 are in the contact patch, the distance h (x, y) equals zero,
and we get
w1 (x, y) − w2 (x, y) = (δ1 + δ2 ) − h(x, y) . (16.2)
We now add the two compressions, i.e., δ = (δ1 + δ2 ), and replace the expression
we get for h(x, y) if the contact partners are parabolic with the same directions of
the main axes:
x2 y2
w1 (x, y) − w2 (x, y) = δ − − , (16.3)
2 Rx∗ 2 Ry∗
with
1 1 1
∗
= + ,
Rx Rx1 Rx2
1 1 1
∗
= + .
Ry Ry1 Ry2
306 16 Appendix
Fig. 16.2 Displacement P

distribution of a half-space r
with single load
w (r )
The curvature is positive if the center of the curve is inside the body.
If the two points A1 and A2 are outside the contact patch, the distance is greater
than zero, h (x, y) > 0, which means that
x2 y2
w1 (x, y) − w2 (x, y) > δ − − . (16.4)
2 Rx∗ 2 Ry∗
Constitutive Equations
During the approach of the two bodies, contact stresses arise. With symmetry con-
siderations and with the assumptions in Chap. 3, it can be shown that normal loads
induce only relative normal displacements w1 (x, y) − w2 (x, y) and no relative tan-
gential displacements u 1 (x, y) − u 2 (x, y) and v1 (x, y) − v2 (x, y) (and conversely).
Therefore, in the normal contact problem for a half-space from the same material,
only normal stresses arise in the contact patch.
The problem is to determine the pressure distribution in the contact patch that
leads to the fulfillment of the contact condition. The starting point is the displacement
distribution of a single load on the half-space (Fig. 16.2),
1 − ν 2 2Q
w(r ) = , (16.5)
πE r
with modulus of elasticity E and Poisson ratio ν. For a distributed load p(x, y)
(Fig. 16.3), we get

1 − ν2 p(ξ, η)
w(x, y) = dξdη. (16.6)
πE (x − ξ)2 + (y − η)2
16.3.2 Contact Equation
The load p(x, y) now acts on both body 1 and body 2. For the resulting vertical
displacements, one must have
Fig. 16.3 Notation for a P

distributed load on a contact Pz ( , )
patch
B A
w (x,y)
y
contact -
area B( , )
r
A (x,y)

1 1 − ν12 1 − ν22 p(ξ, η) dξdη
w1 (x, y) − w2 (x, y) = + .
π E1 E2 (x − ξ)2 + (y − η)2
(16.7)
If we now demand that Eq. (16.4) be equal to Eq. (16.7), we arrive at the contact
equation

1 1 − ν12 1 − ν22 p(ξ, η)
+ dξdη
π E1 E2 (x − ξ)2 + (y − η)2
2 2
x y
= δ − ∗
− . (16.8)
2 Rx 2y
This is a singular integral equation for the area pressure p(x, y). The solution is
a half-ellipsoidal pressure distribution:

x2 y2
p(x, y) = p0 1 − 2
− 2. (16.9)
a b
Strictly speaking, the equations hold only for like materials. But the Hertz solution
is also a good approximation for different materials. The abbreviation
1 1 − ν12 1 − ν22
= + (16.10)
E∗ E1 E2
is introduced. For bodies from the same material, the equation is simplified to
1 2(1 − ν 2 )
∗
= . (16.11)
E E
308 16 Appendix
For the solution of the integral equation (16.8), the maximum area pressure p0
as well as the two radii a and b of the contact ellipse are unknown. From purely
geometric considerations, it can be shown that

a Rx∗
= . (16.12)
b Ry∗
The question is then the values of the radius a of the contact pressure and the
maximal area pressure of the corresponding load,

2π
P= p(x, y)d xd y = abp0 . (16.13)
3
Since usually 2Q is given, the question now becomes the size of the radius a and the
elastic approach δ.
16.3.3 Basic Equations for the Tangential Contact Problem

According to Carter
As in the three-dimensional contact problem, in the two-dimensional case there are

also three basic types of equations:
• constitutive equations;
• kinematic equations;
• contact conditions.
In the Carter solution, only one resulting quantity is calculated by integrating stresses,
the longitudinal creep force Tx .
Constitutive Equations
In the two-dimensional case, the displacement vector and the vector with the tangen-
tial forces can be written as
u = {u x , u z }T = {u x1 − u x2 , u z1 − u z2 }
and
q = {qx , p}T .
From these, the constitutive equation is found:

u(x) = G(x − ξ) q(ξ) dξ + C , (16.14)

where C = {C1 , C2 }T is a vector comprising two constants, and G(x − ξ) is again

an influence matrix:

4(1 − ν 2 ) ln(x − ξ) 0
G(x − ξ) = . (16.15)
πE 0 ln(x − ξ)
It is impossible to make the two constants C1 and C2 disappear. Because of their exis-
tence, the elastic approach of the normal contact problem and the relative displace-
ment of the tangential contact problem cannot be calculated. The stress calculation
is possible, however.
We find again that the normal and tangential contact problems are not coupled
by the constitutive equations; see Eq. (16.14). The decoupling is valid only if the
constitutive equations are given with respect to the differential displacements.
Kinematic Equations
Assume two cylinders rolling on each other (Fig. 16.1). The creep is defined as the
speed difference of body 1 with respect to 2 , divided by a mean speed vm :
v1 − v2 Rx2 2 − Rx1 1
νξ = = , (16.16)
vm vm
with
Rx1 1 + Rx2 2
vm = .
2
If the cylinder is rolling on a plane, the creep
2(v0 − r0 )
νx = (16.17)
v0 + r0
results. The local creep s(x) in the contact patch can be written as
∂u x (x, t) 1 ∂u x (x, t)
s(x, t) = νx (t) + − . (16.18)
∂x vm ∂t
Contact Conditions
The contact conditions become much simpler in the two-dimensional case. Further,
we consider only the stationary case:

sx (x) = 0
(x, y) ∈ adh ⇐⇒ (16.19)
|qx (x)| ≤ μ p(x)
sx (x) = 0
(x, y) ∈ slide ⇐⇒ (16.20)
qx (x) = + |ssxx (x)
(x)|
μ p(x)
310 16 Appendix
Creep Forces
In the case regarded here, there is only a longitudinal creep force

Tx (x, t) = qx (x, t)d x . (16.21)

Carter Solution
With the help of the second constitutive equation (16.14), the tangential displace-
ments of the material in the contact patch due to the acting tangential stresses can be
calculated:
a0
4(1 − ν)2
u x (x, t) = qx (ξ) ln(x − ξ)dξ + C2 (t) . (16.22)
πE
−a0
As for the normal contact problem, the constant C2 depends on the choice of the
reference point u x . In the stationary case, this is unimportant, since the constant
disappears after partial derivation ∂u
∂x
x
. The remaining expression is
a0
∂u x (x) 4(1 − ν 2 ) 1
= qx (x, ξ)dξ . (16.23)
∂x πE x −ξ
−a0
The stationary contact equations for the stationary case are given below. The aim is
to derive the tangential stresses qx (x) and the creep force Tx with a given rigid body
creep νx .
In the adhesion area adh , we get
a0
4(1 − ν 2 ) 1
qx (ξ)dξ = −νx ∀ x ∈ adh , (16.24)
πE x −ξ
−a0
with the constraint

|qx (x)| ≤ μ p(x) . (16.25)
In the sliding area slide , we get
∂u x (x)
νx + ∂x
qx (x) = ∂u x (x)
μ p(x) ∀ x ∈ slide . (16.26)
|νx + ∂x
|
Outside the contact area, the following must hold:
qx (x) = 0 ∀ x ∈
/ . (16.27)
Again, a singular integral equation, Eq. (16.24), has to be solved, and a constraint
has to be satisfied.
16.4 Function for the Vermeulen–Johnson Solution
Equation 3.55 contains a function , which is a combination of complete elliptic

integrals and is taken directly from the original work of Vermeulen and Johnson [6],
together with a function 1 , which is needed for pure lateral creepage νη . We can
write

νξ 3
Tξ = μ N 1 − 1 − , (16.28a)
νξsat

νη 3
Tη = μ N 1 − 1 − , (16.28b)
νηsat
with
3μ N
νξsat = , (16.29a)
Gabπ
3 μ N 1
νηsat = . (16.29b)
Gabπ
For the two functions and 1 , the following relations can be found:
⎧
⎪
⎪ B − ν(D − C) a < b,
⎪
⎪
⎪
⎪
a ⎪
⎨ π(4 − 3ν)
= a = b, (16.30a)
⎪ 16
b ⎪
⎪
⎪
⎪
⎪
⎪ b
⎩ [D − ν(D − C)] a > b ;
a
⎧
⎪ a2
⎪
⎪ B − C a < b,
⎪
⎪ b2
⎪
⎪
a ⎪
⎨
π(4 − ν)
1 = a = b, (16.30b)
b ⎪
⎪ 16
⎪
⎪
⎪
⎪
⎪
⎪
⎩ b [D − ν(C)] a > b.
a

2
Here B, C, and D are complete elliptic integrals as functions of k = 1 − ab2 for

2
a < b and k = 1 − ab2 for b > a. There exist algorithms to solve them, which can
also be found in [7].
312 16 Appendix
16.5 Basic Equations of the Simplified Theory

of Rolling Contact
Adhesion area Sliding area

X2 + Y 2 ≤ μp (16.31a) X2 + Y 2 = μp (16.31b)
Abbreviations
Vξ1 − Vξ2
νξ = , (16.32a)
vm
Vη1 − Vη2
νη = , (16.32b)
vm
ζ1 − ζ2
νζ = , (16.32c)
vm
Vξ1 + Vξ2
vm = − , (16.33)
2
u ξ = u ξ2 − u ξ1 , (16.34a)
u η = u η2 − u η1 . (16.34b)
Kinematic relations
∂u ξ
sξ = (νξ − νζ η) + (16.35a)
∂ξ
∂u η
sη = (νη + νζ ξ) + (16.35b)
∂ξ
Constitutive equations
uξ
X= (16.36a)
Lξ
uη
Y = (16.36b)
Lη
Contact conditions
Adhesion area Sliding area

sξ = 0 sη = 0 (16.37a, b) sξ = λX , (16.38a)
sη = λY , (16.38b)
with λ ≤ 0.
16.6 Stability Criteria of Characteristic Equations with Hurwitz Criterion 313
16.6 Stability Criteria of Characteristic Equations

with Hurwitz Criterion
A systematic way to derive the stability criteria was given by Hurwitz [8]. Hurwitz
investigates the conditions under which an algebraic equation with real-valued coeffi-
cients in the shape of Eq. (10.25) has roots with only negative real parts. The Hurwitz
matrix plays an important role in deriving the stability condition. The mathematical
proof from [8] will not be repeated here. We shall, however, illustrate how to arrive
at the Hurwitz matrix. We begin with equations (10.26a, b). From these, a system
of equations for the determination of the powers (−ωcrit 2 k
) for k = 0 to n can be
built. To do this, the equations (10.26a, b) have to be multiplied by powers ωcrit
2
. The
system of equations can be built up from the resulting equations:
⎡ ⎤⎡ ⎤ ⎧ ⎤
a1 a3 a5 a7 a9 a11 a13 a15 ... (−ω 2 )0 ⎪
⎪0
⎢ a0 ...⎥ ⎢ (−ω 2 )1 ⎥ ⎪⎪0⎥
⎢ a2 a4 a6 a8 a10 a12 a14 ⎥⎢ ⎥ ⎪⎪ ⎥
⎢0 a1 a3 a5 a7 a9 a11 a13 ...⎥ ⎢ (−ω 2 )2 ⎥ ⎪⎪
⎪ ⎥
⎢ ⎥⎢ ⎥ ⎪0⎥
⎢0 ...⎥ ⎢ (−ω 2 )3 ⎥ ⎪⎪ ⎥
⎢ a0 a2 a4 a6 a8 a10 a12 ⎥⎢ ⎥ ⎪ 0
⎨ ⎥
⎢0 0 a1 a3 a5 a7 a9 a11 ...⎥ ⎢ (−ω 2 )4 ⎥ 0 ⎥
⎢ ⎥⎢ ⎥= ⎥ . (16.39)
⎢0 0 a0 a2 a4 a6 a8 a10 ...⎥ ⎢ (−ω 2 )5 ⎥ ⎪⎪ ⎥
0
⎢ ⎥⎢ ⎥ ⎪
⎪ ⎥
⎢0 ...⎥ ⎢ 2 6⎥
⎪0⎥
⎪
⎢ 0 0 a1 a3 a5 a7 a9 ⎥ ⎢ (−ω 2 )7 ⎥ ⎪ ⎥
⎢0 ...⎥ ⎢ ⎥ ⎪⎪0⎥
⎪
⎣ 0 0 a0 a2 a4 a6 a8 ⎦ ⎣ (−ω ) ⎦ ⎪ ⎪ ⎦
.. .. .. .. .. .. .. .. .. .. ⎪
⎩ ..
. . . . . . . . . . .
The matrix of coefficients is the Hurwitz matrix. According to Hurwitz, the following
stability criteria are satisfied:
The characteristic polynomial has roots with only negative real parts if and only if
1. all coefficients ai are different from zero and positive, and

2. the subdeterminants derived from the Hurwitz matrix assume only positive values.
As illustration, the stability criterion for the fourth-order characteristic equation
a 0 + a 1 λ + a 2 λ2 + a 3 λ3 + a 4 λ4 = 0 (16.40)
is given. First, all coefficients from Eq. (16.40) have to be positive:
ak > 0 for k = 0, . . . , 4 . (16.41a)
The Hurwitz matrix for the fourth-order polynomial looks like this:
⎡ ⎤
a1 a3 0 0
⎢ a0 a2 a4 0⎥
⎢ ⎥ .
⎣0 a1 a3 0⎦
0 a0 a2 a4
314 16 Appendix
Table 16.1 Stability criteria for characteristical polynomials of the orders 2–6
Order of Additional stability criterion Limit frequency
polynomial of stability ωcrit
2
a0
2 a1 ≥ 0 a2
a1
3 a1 a2 − a0 a3 ≥ 0 a3
a1
4 a1 a2 a3 − a0 a32 − a4 a12 ≥ 0 a3

a3 ± a32 −4a1 a5
5 (a3 a4 − a2 a5 )(a1 a2 − a0 a3 ) − (a1 a4 − a0 a5 )2 ≥0 2a5

a3 ± a32 −4a1 a5
6 [a5 (a3 a4 − a2 a5 ) − a6 (a32 − 2a1 a5 )](a1 a2 − a0 a3 ) 2a5
−a5 (a1 a4 + a0 a5 )2 − a1 a3 a6 (a0 a5 − a1 a4 )2 ≥ 0
The determinant criteria can now be written as
k > 0 for k = 1, . . . , 3 , (16.42a)
with
1 = a1 , 2 = a1 a2 − a0 a3 and 3 = a1 a2 a3 − a0 a32 − a4 a12 . (16.43a)
In Table 16.1, all stability criteria for polynomials up to order 6 are gathered. For
polynomials of higher order, it is no longer possible to give an expression for the limit
frequency for stability in closed form, since cubic and even higher-order equations
have to be solved.
16.7 Critical Speed of Single Wheelset Taking Nondiagonal

Elements of Creepage Damping Matrix into Account
For the sake of completeness, we give the relations taking the d12C -terms into account.
Instead of (10.36) and (10.37), we get
d11C s22 + d22C s11 − d12C s21 − d21C s21
ωcrit
2
= (16.44)
d11C m 22 + d22C m 11
and

d11C d22C m 22 s11 − m 11 s22 2
k=
−s12 s21 m 11 d22C + m 22 d11C

d12C s21 + d21C s12 m 211 s22 d22C + m 222 s11 d11C
−
d11C d22C m 11 d22C + m 22 d11C
16.7 Critical Speed of Single Wheelset Taking Nondiagonal … 315

m 11 m 22 (d11C s22 + s11 d22C − d12C s21 − d21 s12 )
− .
m 11 d22C + m 22 d11C
For the critical speed, we finally get
d11C d22C − d12C d21C 1

vcrit
2
= ωcrit
2
. (16.45)
(−s12 s21 ) 1−k
If numerical values are inserted in these equations and they are compared with
(10.36)–(10.38), it is observed that it is justified to neglect the d12C -terms.
References
tberichte VDI, Reihe 12, No. 338 (also dissertation TU Berlin). VDI-Verlag, Düsseldorf (1997)
2. W. Kik, H. Steinborn, Führ- und Störverhalten - Ermittlung statischer und quasistatischer Gle-
ichgewichtslagen (Guiding and disturbance behavior - Determination of static and quasistatic
positions of equilibrium). VDI-Berichte 510, 275–284 (1984)
3. W. Kik, H. Steinborn, Wheel/rail connection element for use in a multi-body-algorithm, in The
Dynamics of Vehicles on Roads and on Tracks. Proceedings 8th IAVSD-Symposium. Cambridge,
Mass. August 1983., ed. by J.K. Hedrick (Swets & Zeitlinger, Lisse, 1984), pp. 303–316
4. W. Kik, H. Steinborn, A nonlinear wheel–rail connection element and its application for the
analysis of quasi-static curving behaviour, in Advanced Railway Vehicle System Dynamics,
Chapter 8, ed. by J. Kisilowski, K. Knothe, Wydawnicta Naukowo-Techniczne, (1991), pp.
243–271
5. K.L. Johnson, Introduction to contact mechanics - a summary of the principle formulae, in
Fundamentals of Friction: Macroscopic and Microscopic Processes. Proceedings of the NATO
Advanced Study Institute on Fundamentals of Friction. Braunlage/Harz, Germany, July 29 -
August 9, 1991 (Kluwer Academic Publishers, Dordrecht, Boston, London, 1992), pp. 589–603
6. J. Vermeulen, K.L. Johnson, Contact of nonspherical elastic bodies transmitting tangential forces.
J. Appl. Mech. 31, 338–340 (1964)
7. E. Jahnke, F. Emde, F. Lösch, Tafeln höherer Funktionen (Tables of higher functions), vol. 7
(Teubner, Stuttgart, 1966), durchges. u. erw. Aufl
8. A. Hurwitz, Ueber die Bedingungen, unter welchen eine Gleichung nur Wurzeln mit negativen
realen Teilen besitzt (On the conditions under which an equation has only roots with negative
real parts). Math. Ann. 46, 273–284 (1895)
Index
B arbitrary profiles, 43–45

Bearing task conical and circular profiles, 37–43
comfort, 2 creep calculation, 50
Bending stiffness, see bogie equivalent circular profiles, 47, 48
Bogie quasilinearization, 45–47
bending stiffness, 207 with track irregularities, 48, 49
friction yaw damper, 234 Creepage
friction yaw damping, 232–235 in curve, 255
running gear design, 23 on straight track, 50–51
Minden–Deutz, 18–21 Creep velocity
overview, 23 curve, 255
SGP-ICE-bogie, 21 Critical speed
Y25, 21 bogie, see stability of bogie hunting
shear stiffness, 207 linear, 192, 225, 227
yaw damper, 235 linear (single wheelset), 190
yaw friction damping, 18, 20 nonlinear, 238–239
Bogie hunting single wheelset (linear), 189
wavelength, 212 Curving
Bogie stability, 195 bogie positions, 261
constrained curving, 260
free curving, 263
C bogies
Cant deficiency, 256 with primary suspensionn, 264
Circular profiles bogies and vehicles, 259–265
equivalent, 47–48 creepages, 257
Comfort, 1 dynamic, 2
Complex syntax, 108–123 kinematic, 251
calculation rules, 108–111 longitudinal creep forces, 257
Concentration processes in railway industry, optimal solution
11 not unique, 259
Conicity, 159 quasistatic, 250
effective, 45 quasistatic solution, 2
efficient, 47 assumptions, 250
equivalent, 45–167 equation system, 257
experimental determination, 167 nonlinear, 250
Contact kinematics for the wheel/rail con- radial position, 258
tact, 36–51 remaining creep force, 257
DOI 10.1007/978-3-319-45376-7
318 Index
simplifications, 260 Friction yaw damping, see bogie, friction

time domain, 250 yaw damping
Uebelacker and Heumann, 260 Function of wheel–rail system
wear, 266–269 ride comfort, 141–156
wheelset positions, 252
G
D Guiding task, 1
Damage, see long term behaviour stability, 2
Damage hypothesis, 274
Definition of forces and creepage in the
wheel/rail contact, 33 H
Description of type of motion, 18 History of railway technology
curving, 249–250
Boedecker, 249
E Dauner–Hiller–Reck, 249
Effective value, 127, 128 Heumann, 249
determination of effective value, 129 Mackenzie, 249
Eigenmode Porter, 249
of wheelset, 198 Redtenbacher, 249, 251
Eigenvalue calculation, see stability Übelacker, 249
numerical, 195 Uebelacker, 249
Equivalent circular profiles, 47 History of railway technology research, 4–9,
Equivalent coefficient of contact angle dif- 60
ference, 46 Boedecker, 5
Equivalent conicity, 45, 47 curving
Equivalent rail profile radius, 48 Redtenbacher, 4
Equivalent running parameter, 46 Klingel, 4
Equivalent wheel profile radius, 47 Krugmann, 9
Evaluation criteria, 2–3 Matsudaira, 7
comfort, 2, 3 Müller, C.Th., 9
economy, 2 nonlinear stability, 238
external noise, 3 ORE-committee for hunting, 8
life cycle costs, 2 SNCF-tests1955, 7
limit values, 2 Stephenson, 4
safety, 2, 3 Tokaido Line, 9
traction, 3 wheel/rail contact, 60
Bufler, 60
Carter, 6, 60
F Fromm, 7, 60
Fatigue damage spectra Heinrich, 60
general procedure Johnson, 60
load collectives, 274 Kalker, 60
Filter Reynolds, 60
bandpass characteristics, 134 Rocard, 7
filtering, 141 Wickens, 9
selective, 133 Hunting
time domain, 134 bogie vehicle, 229–232
Flexible carbody, 23–26 of bogie, see bogie stability
Fourier representation, 120 of bogie with secondary stiffness
Fourier-series presentation, 119 frequency, 214
Frequency domain of the rigid bogie
vehicle dynamics, 17 critical speed, 214
Frequency-wavelength-speed-diagram, 112 of wheelset, 166
Index 319
assumptions for Klingel equation, track, 26–27

165–166 model parameter, 27
derivation of Klingel equation, 162, vehicle, 17–26
165 coupling elements, 24
Klingel, 161 flexible models, see flexible carbody
summary, 167 rigid body models, 17, 23
wavelength, 165 suspension elements, 23
L N
Lateral dynamics Normal contact in the wheel/rail contact, 51–
introduction, 159–168 59
sinusoidal motion, 160 assumptions, 52–53
Load assumptions elliptical contact, 57–58
for variable amplitude fatigue calcula- equivalent contact measures of the Hertz
tion, 274 theory, 54–56
Load damage spectra Hertz contact stiffness, 59
general procedure Hertz theory, 53–59
damage hypothesis, 274 line contact, 58, 59
stress calculation, 275 nonelliptic contact patches, 53
Loads on bogie frames , see variable ampli- spherical contact, 56–57
tude fatigue
Longitudinal displacement, 255
Long term behaviour, 3 O
track, 3 Overview of contents, 11–13
vehicle, 3, 273
P
M Power spectral density, 133–140
Mbs-program analog derivation, 133–134
Adams Rail, 17, 37, 225 calculation for ICE 2.2, 136–138
Gensys, 17, 225 derivation from correlation function, 133
Linda, 225 derivation from time series, 133
Medyna, 10, 17, 37, 221, 225 digital derivation, 134
Nucars, 225 digital determination, 134
Simpack, 10, 17, 37, 225 in DB network
Vampire, 17, 225 analytical expressions, 136
Voco, 17, 225 in DB network for ICE 1, 135–136
Mean value relationship between spatial and angular
arithmetic, 127 frequency, 139
effective, 127 Power spectrum
quadratic, 127 of track irregularities, 133
Mode shapes Power spectrum of vehicle response
vehicle, 227 determination of
Modeling, 17–31 for periodic excitation, 130
coordinate systems, 18 for stochastic excitation, 131, 132
excitation, 28–31 Primary stiffness
moving irregularity, 31 optimal parameters
moving track irregularity, 30 stability and curving, 259
out-of-round wheels, 28 Principle of impulse and momentum, see
track irregularities, 29, 30 vertical dynamics for a two-axle vehi-
program systems, 17 cle
running gear, see bogie, running gear Principle of virtual displacements for rigid-
design body systems, 91–93
320 Index
Profile geometry in the wheel/rail contact, finite bending and shear stiffness,
33–36 220–222
finite bending and shear stiffness
(diagram), 221
R infinite bending stiffness, 219
Random vibrations, 127–140 infinite shear stiffness, 219, 220
Research and development in Germany, 10– of wheelset
11 Boedecker, 161
BMFT-funding, 10 Stability of bogie hunting
new system boundaries in 1990, 10–11 analytic, 201
Response spectrum coordinate transformation, 211
interpretation, 139–140 infinite shear stiffness, 216
Ride comfort, 141–156 infinitely high bending and shear stiff-
comfort criteria, 141–156 ness, 211
CEN discrete events, 153–155 Stability of four-axle vehicle, see bogie sta-
CEN mean comfort, 151–152 bility
CEN transition curves, 153 Stability of wheelset hunting
CEN transitions, 153 coordinate transformation, 203
comfort values, 142–144, 148 Standard deviation, 127
EN 12299, 151 Standards
general periodic, 145–146 CEN WG10, 2
ISO 2631, 148–151 DIN 13647 (rail profile), 33
lateral-vertical, 143 DIN 5573:95 (rail profile), 33
random, 147–148 EN 12299, 151, 153–155
transfer function, 150–151 EN:14363, 132
weighting function, 144 ENV 12299, 149, 151
Wertungsziffern, 142 ORE 116DT81 (wheel profile), 35
duration of exposure, 156 ORE S1002 (wheel profile), 35, 44
measurement vs. simulation, 155–156 UIC
passenger, 141–156 90A, 267
running behaviour UIC 60 (rail profile), 35, 36, 44
freight wagon, 144 UIC 513, 149, 151
system property, 156 UIC 518, 2, 132
RMS-Value, see root mean square Static condensation
Rolling line, 251 bogie, 205
Root loci curves, 197–198 Symmetry, 17
bogie, 197 System view, 3
characterization, 198–200
Root mean square (RMS-Value), 127
T
Tangential contact for wheel and rail, 60–77
S adaption of theory to reality, 76–77
Shear stiffness, see bogie additional assumptions, 62–63
Short-term dynamics, 3 approximate solution John-
Spectrum son/Vermeulen, 67–70
amplitude spectrum, 120, 121 approximate solution Shen–Hedrick–
phase spectrum, 120 Elkins, 70–71
power spectrum, 130 friction phenomena, 61–62
spectrum of squares of amplitudes, 130 sign definition, 60–61
Speed, critical, see critical speed simplified theory, 71–76
Stability, 1 solution for elliptical contact (Kalker),
of bogie hunting, 222 66–67
analytical, 215 solution for line contact, 63–65
Index 321
Tasks of the wheel/rail system, 1–2 free system, 87–89

bearing task, 1 spring forces, 85–87
guiding task, 1 with principle of linear and angular
traction task, 1, 2 momentum
Track excitation free system, 88
harmonic, 127, 128 with principle of virtual displacements,
periodic, 128 93–97
stochastic, 128 constraints, 94
Track irregularities, 1, 29 equations of motion in matrix syntax,
Track irregularity (random) 98–99
Gaussian distribution, 132 formalized procedure, 95–97
Track plane acceleration, 256, 263 formalized procedure: external and
Transfer function constraint forces, 97
acceleration, 116 formalized procedure: mass matrix,
elastic carbody, 123 97
displacement, 116, 122 formalized procedure: virtual energy,
graphical representation, 118 96–97
Vertical-lateral dynamics
separation, 2, 17
V Vertical level
Variable amplitude fatigue, 273–293 time series from measurement
multibody dynamics DB network, 137
elastic bodies, 273 Vertical vibrations
principal procedure, 274 base-point excitation, 111–115
Vertical dynamics for a two-axle vehicle, 81– general periodic track irregularities,
102 119–124
assumptions, 81–82 elastic carbody, 123–124
eigenvalues and eigenmodes, 102 harmonic and periodic track irregulari-
equations of motion, 90–91 ties, 107–124
equations of motion for elastic carbodies, harmonic track irregularities, 111–119
99–100 interpretation of solution, 116–119
solution for free vibrations, 101–102 solution for bounce, 112–115
with principle of impulse and momen-
tum, 82–91
calculation scheme, 82–83 W
constraints, 83–84 Wavelength, see hunting
displacement degrees of freedom, 83 Wear calculation, 266–269
elimination of constraint forces, 89– Wheel/rail contact, 1, 33–77
91 contact kinematics, 51

Book - Rail Vehicle Dynamics

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Book - Rail Vehicle Dynamics

Uploaded by

Copyright:

Available Formats

Klaus

Rail Vehicle Dynamics

ISBN 978-3-319-45374-3 ISBN 978-3-319-45376-7 (eBook)

Original version published by SpringerVieweg 2003

Printed on acid-free paper

This Springer imprint is published by Springer Nature

Berlin, Germany Klaus Knothe

3 Modeling of Wheel/Rail Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

4.2.5 Principle of Linear and Angular Momentum

5.5 Exercises for This Chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

9 Derivation of Equations of Motion for Lateral Dynamics . . . . . . . . 169

11.2.3 Realization of Inﬁnitely High Bending Stiffness

14.4 Wheelset Guided in Track Following Frame . . . . . . . . . . . . . . . 252

16.3 Fundamentals of Contact Mechanics . . . . . . . . . . . . . . . . . . .... 304

1.1 The Basic Function of the Wheel/rail System

© Springer International Publishing AG 2017 1

The investigation of load-bearing behavior is essentially the field of vertical

1.2 Significance of Dynamics on the Operation of Rail

Table 1.1 Assessment criteria for short-term dynamics

Table 1.2 Assessment criteria for long-term behavior

1.3 On the History of Research in the Field of Railway

In Germany, Ferdinand Redtenbacher,2 from Karlsruhe, was the first to try to

the Polytechnikum in Karlsruhe, where he studied engineering science. In 1840, he successfully

Fig. 1.2 Hunting motion of single wheelset according to Klingel [9]

Fig. 1.3 Nonlinear unstable Y

processes occurring during tangential contact, which were described by Boedecker

1.3.2 A New Start After 1945: Japan and France

Fig. 1.4 Track displacement

(Office de Recherche et d’Essais de l’Union International des Chemins de Fer, the

waves.” ([28], p. 82).

1.3.3 Research and Development Aimed at Overcoming

Support Programs in the Federal Republic of Germany.

1.4 Railway Industry in Europe

12 German Research Foundation.

Konzentrationsprozess der Systemhäuser in der Schienenfahrzeugindustrie

AEG, D MAN / MBB, D LEV, D Von Roll, CH

Quellen: Siemens; Fortschreibung: DIW.

1.5 Overview of This Book

The subsequent Chaps. 2–15 can be classified into six groups:

• Chapter 14 deals with curving behavior investigations, including wear processes

1.5.2 Vertical and Lateral Vibrations

1.5.3 Curving Behavior

1.5.4 Frequency- and Time-Domain Calculations

Although the ever increasing performance of computers is leading to the implemen-

1. B. Bergander, Private communication. Guest lecture TU (Springer, Berlin, 1993)

2.1 Prior Considerations and Coordinate Systems

© Springer International Publishing AG 2017 17

Fig. 2.1 Coordinate systems; nomenclature of motions

which we consider rigid bodies for simplicity. We differentiate between a rail-bound

Rolling in combination with lateral motion is called sway or body rolling.

2.2 Vehicle Modeling

2.2.1 Bogie Frame Design, Primary Suspension,

lateral bump stop longitudinal bump stop

bolster beam bolster beam

vertical bolster beam

dampers installed with an inclination result in effective damping in the vertical as

curve dependent lateral play

pendulum with links

lateral bolster guidance

Fig. 2.3 Details of bolster pendulum arrangement of MD-bogie, from [2]

Fig. 2.4 Characteristics of 40

suspended side-bearers centre-bowl

centre-bowl Lenoir - link