Vehicle System Dynamics

International Journal of Vehicle Mechanics and Mobility

ISSN: 0042-3114 (Print) 1744-5159 (Online) Journal homepage: http://www.tandfonline.com/loi/nvsd20

Modelling of suspension components in a rail

vehicle dynamics context

Stefano Bruni , Jordi Vinolas , Mats Berg , Oldrich Polach & Sebastian Stichel

To cite this article: Stefano Bruni , Jordi Vinolas , Mats Berg , Oldrich Polach & Sebastian Stichel
(2011) Modelling of suspension components in a rail vehicle dynamics context, Vehicle System
Dynamics, 49:7, 1021-1072, DOI: 10.1080/00423114.2011.586430

To link to this article: https://doi.org/10.1080/00423114.2011.586430

Published online: 24 Jun 2011.

Submit your article to this journal

Article views: 2435

Citing articles: 67 View citing articles

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=nvsd20
Vehicle System Dynamics
Vol. 49, No. 7, July 2011, 1021–1072

Modelling of suspension components in a rail vehicle

dynamics context
Stefano Brunia *, Jordi Vinolasb , Mats Bergc , Oldrich Polachd and Sebastian Stichelc
a Dipartimentodi Meccanica, Politecnico di Milano, Via La Masa 1, Milano 20156, Italy; b Applied
Mechanics, CEIT and Tecnun (University of Navarra), P. Manuel Lardizábal, 15, San Sebastian
20.018, Spain; c Department of Aeronautical and Vehicle Engineering, KTH (Royal Institute of
Technology), Teknikringen 8, Aeronautical and Vehicle Engineering/Rail Vehicles, Stockholm SE-1004,
Sweden; d Bombardier Transportation, Zürcherstrasse 39, Winterthur 8401, Switzerland

(Received 23 February 2011; final version received 27 April 2011 )

Suspension components play key roles in the running behaviour of rail vehicles, and therefore, math-
ematical models of suspension components are essential ingredients of railway vehicle multi-body
models. The aims of this paper are to review existing models for railway vehicle suspension com-
ponents and their use for railway vehicle dynamics multi-body simulations, to describe how model
parameters can be defined and to discuss the required level of detail of component models in view of
the accuracy expected from the overall simulation model. This paper also addresses track models in use
for railway vehicle dynamics simulations, recognising their relevance as an indispensable component
of the system simulation model. Finally, this paper reviews methods presently in use for the checking
and validation of the simulation model.

Keywords: suspension components; rail vehicle modelling; model validation; vehicle acceptance;
running safety; ride quality

1. Introduction

For modern railway vehicles, multi-body vehicle dynamics simulation has become a major
design instrument, allowing the assessment and optimisation of vehicle performance from the
early stage of the design process, before a prototype is built. Typical applications of multi-body
simulation include the verification of running stability and safety, the evaluation of ride quality,
the analysis of wear and other damage effects at the wheel–rail interface, the verification
of dynamic gauging and the numerical estimation of design loads to be used for durability
analysis [1]. Furthermore, multi-body simulation is increasingly being used to complement
and partially replace physical testing to demonstrate running safety and acceptability in view
of the admission of a new or modified rail vehicle into service [2], a topic sometimes referred
to as ‘virtual homologation’ [3].

*Corresponding author. Email: [email protected]

ISSN 0042-3114 print/ISSN 1744-5159 online

© 2011 Taylor & Francis
DOI: 10.1080/00423114.2011.586430
http://www.informaworld.com
1022 S. Bruni et al.

Quite obviously, the above-mentioned applications call for a high degree of reliability and
trustworthiness of the models used, since any significant deviation of the models from the
real vehicle actual behaviour would imply the need for troubleshooting at a late stage of the
design process or even during the acceptance process, especially considering that some of the
issues addressed are safety critical. Therefore, one important challenge regarding modelling
and simulation of railway vehicle dynamics is to define accurate and reliable procedures for
building multi-body vehicle models. A second, strictly related problem is concerned with the
validation of multi-body models, considering that, in a ‘virtual homologation’ perspective,
validation shall not rely on comparison with line tests performed on the same vehicle but
could instead make use of results from line tests performed on similar vehicles and/or of
results of tests performed on single suspension components.
The accuracy of rail vehicle multi-body models is mainly affected by the model of wheel–
rail contact and by the models of vehicle suspension components. As far as the model of
wheel–rail contact is concerned, a very large research effort has been spent in the past to
define modelling approaches ensuring a satisfactory level of detail and, at the same time,
requiring an affordable computational effort, and recent surveys are available on this topic,
such as the one done by Chollet and Piotrowski [4], describing the wheel–rail contact models
most widely used in the state of the art of rail vehicle dynamics simulation. On the other
hand, rail vehicle suspension component models have also been the subject of extensive
research work, especially in recent years, but few updated reviews of the work done are
available, with the two most noteworthy being the survey paper written in 1995 by Eickhoff,
Evans and Minnis [5] and the chapter on simulation in the Handbook of Railway Vehicle
Dynamics [6].
The aims of this paper, which is partly based on the work being undertaken in WP5 of the
European research project DynoTrain, are to provide a comprehensive and updated description
of railway vehicle suspension component models in use for railway vehicle dynamics multi-
body simulations, to describe how model parameter data can be defined and to discuss the
required level of detail of component models in view of the accuracy required for the overall
model of the complete vehicle or train, also considering the specific scope of the multi-body
simulation being performed.
Mathematical models of railway vehicle suspension components may range from relatively
simple linear ones for coil springs to more sophisticated load-sensitive models of friction
elements and nonlinear and frequency-dependent models of rubber springs and bushes and
may also include multi-physics models for example, in the case of air springs and active
suspension components. Even in the case of apparently ‘simple’ suspension components, there
are issues that need to be carefully considered in view of the accuracy of the overall vehicle
model, such as secondary effects related to gravitational loads, uncertainty in the parameters,
geometrical imperfections and deviations from the nominal behaviour.
This paper also addresses track models in use for railway vehicle dynamics simulations:
although it is not intended here to cover track dynamics and train–track interaction models
which are specific to the study of track vibration and track damage problems (such as those
surveyed in [7]), it is recognised that the modelling of track dynamics bears a relevant influence
on the simulation of vehicle dynamics, deserving a presentation of viable approaches and a
discussion of their respective advantages and disadvantages.
Another notable issue which will be discussed in this paper is the effect of model uncertain-
ties on simulation results and how the complete vehicle model shall be checked and validated
before it is used to simulate vehicle dynamics: this subject is particularly relevant to the issue
of ‘virtual homologation’, a task often requiring that model accuracy is assessed based on a
limited amount of line test measurements, possibly referring to a slightly different vehicle or
to different service scenarios than the ones being considered for acceptance.
 Vehicle System Dynamics 1023

Before concluding this paper, a critical review of the challenges and future trends in the
modelling of railway vehicle suspension components is also provided.

2. Rail vehicle models and suspension component models

As indicated above, the physical/practical rail vehicle appearance needs to be represented in

a mathematical/theoretical model, imitating the mechanical behaviour of the vehicle running
along the track. The level of detail in the vehicle modelling and the accuracy of the input data
depend on the purpose of the analysis at hand and how accurate the output quantities need
to be. Based on the review [8], Figure 1 presents the main types of analyses and associated
calculation methods.
From a safety perspective, the key issues in the figure are derailment, track shift forces and
running stability. These issues are emphasised in vehicle acceptance standards such as [9,10].
In Figure 1, it is also obvious that the calculation methods required are dominated by dynamics
simulation in the time domain. A recent review of multi-body simulation in railway vehicle
design is given in [11], including the software packages available.
In the dynamics simulation, the choice of vehicle model is also dependent on the frequency
range of interest; the higher the frequency, the more detailed the model. Traditionally, the
main range of interest is 0–20 Hz. This is certainly sufficient for issues such as car body sway,
gauging, wear and usually also for ride comfort. For wheelset accelerations, wheel–rail forces,
and wheel and rail fatigue issues, this is less obvious.
For the dynamics simulations, the vehicle–track system model is essentially represented
by a number of first-order differential equations including nonlinearities, mainly originating
from the wheel–rail interfaces and many of the suspension components. The unknowns in
these equations are mainly translations and rotations, and the corresponding velocities, of the

Figure 1. Common analysis types and calculation methods applied in rail vehicle engineering. After Polach [8].
1024 S. Bruni et al.

car body, bogie frames and wheelsets of the vehicle. These bodies essentially also host the
inertial properties of the vehicle, including masses of suspension components and equipment
‘lumped’ to these bodies. On the other hand, relative motions between vehicle bodies are
mainly made possible through the flexibility of the suspension components.
As pointed out in the introduction, the modelling of the suspension components is decisive
for the accuracy of the vehicle model and the simulation results. This is due to the fact that
the suspension components play key roles for the vehicle dynamic behaviour, mainly in terms
of vibration isolation and limiting wheel–rail forces. Suspension spring components also
carry static forces due to gravitational effects and quasi-static ones in vehicle curving where
centrifugal effects are also present. Some suspension components are also engaged in vehicle
traction and braking.
The most common suspension components are coil springs, friction-based components such
as leaf springs, rubber springs, air springs and hydraulic dampers, but components such as
traction links, anti-roll bars and bump stops should also be mentioned in this context. All
these components are passive components meaning that for given component properties and
possible pre-load, the forces (and moments) that they give at their interfaces to the vehicle
bodies in question only depend on the displacement and velocities at these interfaces. This
stands in contrast to active suspension components, including semi-active ones, for which
other factors also affect the forces. Modelling of active suspensions is also briefly described
in this review paper.

3. Models and data of suspension components

3.1. Principles of component modelling

The art of suspension component modelling involves several steps and requires engineering
experience and judgement to be successful. Too simple suspension models and/or inaccurate
model parameter data may produce misleading results from vehicle–track simulations in terms
of wheel–rail forces, ride stability, ride comfort, etc.
The mechanical behaviour of suspension components originates from the fields of solid
mechanics, fluid mechanics and tribology. The solid materials dealt with are usually steel or
rubber, whereas the fluids are almost exclusively oil and pressurised air. The tribology here
is often associated with steel-to-steel or rubber-to-steel sliding motion involving friction and
wear issues. In many cases, the suspension components consist of combinations of these parts,
for instance, steel and rubber forming a layered spring, air and rubber for an air spring, and
oil and rubber bushings for a hydraulic damper.
A suspension component may be modelled in detail by describing its geometry and the
material and/or fluid properties as well as possible pre-load and friction interfaces. However,
in many cases, such models would be too complicated and time consuming to be used in the
vehicle–track simulations. Instead, spring elements, dashpots and friction elements are used
and combined in representing suspension components in such simulations. The spring elements
and dashpots often also have linear characteristics. In this way, the key properties of suspension
components are often stiffness, viscous damping, pre-load and friction break-out force. Still,
the more advanced suspension models might be useful in pre-processors, defining stiffnesses,
etc. of the simpler models from the physical appearance of the suspension component in
question.
In some cases, the suspension model parameters might also be obtained from measurements
on the component in question or at least on a similar component. In this process, it is often
convenient to use harmonic displacement excitations and evaluate the corresponding force
 Vehicle System Dynamics 1025

responses. Generally, the response will depend on the excitation frequency, the displacement
amplitude and a possible pre-load. Non-elliptic force–displacement graphs imply that the com-
ponent has a nonlinear behaviour. The component stiffness, often defined as the force amplitude
divided by the displacement amplitude, might be illustrated as a function of excitation fre-
quency, displacement amplitude and/or pre-load. The component damping is sometimes
expressed by the force–displacement phase angle, but for nonlinear component behaviour,
this does not fully represent the hysteresis. The viscous damping of hydraulic dampers may
be seen as the impedance, that is, the force amplitude divided by the velocity amplitude. In
Europe, rail vehicle suspension measurements should be carried out according to EN stan-
dards. In, for instance, spring standards, ‘stiffness’ is often split into a static one and a dynamic
one. Then, great care has to be taken regarding the definition of ‘dynamic stiffness’. As an
example, a single (vertical) dynamic stiffness value is not sufficient in modelling most air
spring systems (Section 3.5).
For air springs and most coil springs, significant horizontal effects take place, meaning
that one-dimensional vertical models are not sufficient. Thus, horizontal forces and bending
moments are introduced at the interfaces with the bodies connected, and the destabilising
effect of the compressive pre-load needs to be considered. For some metal-to-metal interfaces,
two-dimensional friction sliding is possible. In UIC link suspensions, found in some freight
wagons, complicated rolling–sliding phenomena also need to be represented in the modelling.
Spring–dashpot–friction suspension models sometimes resemble the component physical
appearance, but they can also be a pure mathematical representation where such a direct
coupling cannot be identified. For instance, a rubber model may include several parallel sets
of dashpots in series with springs to represent the rubber dynamics over a wide frequency
range. Often, the suspension models include internal degrees of freedom (DoF) (or first-order
differential equations), either to imitate, say, the emergency spring of an air spring system or
to allow for models such as the rubber one just mentioned.
In what follows, the modelling of different suspension components is reviewed. The review
starts with metal (steel) components such as coil springs and friction-based components such
as leaf springs. Then, rubber springs and bushes are covered, including internal friction-like
effects, followed by air springs. Thereafter, the fluid of air is replaced by oil to review models
of hydraulic dampers. Briefly, also modelling of semi-active and active suspension compo-
nents is surveyed. Last but not least, a number of components, strictly not called suspension
components, are reviewed. Examples of such components are traction links, anti-roll bars and
bump stops.

3.2. Coil springs

Modelling coil springs is more difficult than initially expected depending on the deformations
that they are subjected to. In many cases, one has to take into account more than just the axial
stiffness, considering the coupling between different directions and the reaction moments due
to the non-axial displacement of the vertical force. Additionally, the compressive pre-load
provokes destabilising effects in the transverse plane, and consequently, some cases require
a particular analysis of buckling limit (critical axial load or critical axial deflection) which
depends on how the coil spring is able to deform and how its ends are fixed.
The simplest way to model a coil spring is by a single, linear stiffness (which corresponds
to the axial direction). Three-dimensional models include three perpendicular springs. Non-
linearities associated with bump stops or vertical nonlinearity, which commonly appears in
the primary suspension due to inner coil springs with vertical clearance, have to be included.
However, these simple models frequently need an upgrade that includes transverse effects to
1026 S. Bruni et al.

Figure 2. Analytical (Lee–Thompson) and numerical (Abaqus) solutions for the axial dynamic stiffness of a typical
rail vehicle coil spring.

represent adequately real coil spring suspensions. Transverse and bending stiffnesses play an
important role (e.g. coil springs located in the secondary suspension of a railway vehicle) as
transverse shear deformation may compromise vehicle dynamics.
The contribution of coil springs to damping can be neglected as load/unloading curves coin-
cide showing nearly no energy dissipation/hysteresis. Additionally, the dynamic behaviour
of coil springs used in railway applications has nearly no dependency on frequency in the
range of 0–20 Hz. For higher frequencies, some analytical formulae might be used to find
the dynamic response of the spring [12,13], although finite element models can also be used.
Figure 2 compares analytical and numerical solutions for a typical rail vehicle spring and
shows that for frequencies lower than 20 Hz, the coil spring can be modelled using its static
properties.

3.2.1. Stiffness calculations

The European standard EN 13906-1:2008 [14] provides different methods for the calculation
and design of cylindrical helical springs made from round wire and bar. It includes transverse,
buckling and impact loading, stress correction factors over the cross section of the wire,
material property values for the calculation of the spring and a review of the formulae needed
in the design process. Axial stiffness, spring rate R, is estimated using a simple expression that
depends on geometric data of the coil spring. An equation is also proposed for the transverse
spring rate, RQ , although the standard notes that the ‘transverse spring rate is only constant
for short transverse spring deflections, sQ , for a given length L under compression’.
Based on Timoshenko’s work [15–19], the concept for the Haringx model [16] is the division
of the spring into small elements consisting of ordinary, linear springs; it assumes a small helix
angle. By integration, the relationship between axial and shear stiffnesses is obtained.
For the coil spring shown in Figure 3 to be in equilibrium when compressed between two
non-parallel plates, the equations for axial and lateral reactions at the seats are as follows:

Gd 4 (L0 − L)
F= (1)
8nD3
 Vehicle System Dynamics 1027

M1
FQ
1

L
N

FQ
2
M2

Figure 3. Spring subjected to axial, shear forces and moments at both ends.

F(ψ1 + ψ2 )
FQ = (2)
(4G/E + 2)(L0 − L)/(qD2 tan (qL/2)) − 2
    
EGqLd 4 FQ 4G L0 − L
M1 = ψ1 + 1− +2
16(2G + E)(nD)tan(qL/2) F E qD2 sin qL (3)
M2 = M1 + FQ L,

where D is the mean coil diameter; n is the number of active coils; d is the nominal diameter
of the wire; E is the Young modulus; G is the shear modulus; L0 is the unloaded spring length;
ψ1 and ψ2 are the angles of the ends of the spring or seat angles when the spring is seated and
q is the so-called buckling factor. Buckling refers to the loss of stability of a component and
is usually independent of material strength. In this case, it can be calculated as

 2    2
1 2G 2G E L0 L0
q= + −1 + −1 . (4)
D E E G L L

Krettek and Sobczak [17] worked on the same problem. They estimated different correction
factors following extensive tests which relate the axial and shear deformations for a given
loading factor, as they found differences between experiments and previously proposed for-
mulae. These correction factors, ai , depend on the relative axial deformation of the spring and
are foreseen for slenderness ratios between 1.5 and 3:

• a1 = 1.9619ξ + 0.6740 Correction factor for lateral stiffness

• a2 = 1.3822ξ + 0.6513 Correction factor for mixed term (lateral/bending coupling)
• a3 = 0.8945ξ + 0.6690 Correction factor for bending stiffness

s L0 − L
ξ= = . (5)
L0 L0
1028 S. Bruni et al.

They also provided the following formulae for lateral and bending stiffness (Ky∗ and Kψ∗ ):

F
Ky∗= a1 Ky = a1 · lateral stiffness
2/c(1 + F/RQ ) · tan(cL/2) − L
  (6a)
∗
Rψ F cL
Kψ = a3 Kψ = a3 . Ky . 1+  − bending stiffness.
F RQ tan(cL)

The contribution of bending angle to lateral force and the contribution of lateral displacement
to bending moment can be included using a ‘mixed term’, with it being
  
∗
cRψ cL
Kyψ = a2 Kyψ = a2 · Ky . tan , (6b)
F 2
where c is a non-dimensional factor which relates the axial load with the bending and shear
properties of the spring through the equation:

F F
c =  1+ 
2
(7)
Rψ RQ

and
Ed 4 L
Rψ =
32nD(ν + 2)
Gd 4 L
RQ = (ν + 1) (8)
4nD3
Gd 4 L
R = RL =
8nD3
Figure 4 compares the lateral stiffness for a particular coil spring using the different approaches.
This coil spring has a slenderness ratio of 1.63; in the case of higher values of λ, the results
of the Timoshenko formulae would give less accurate results. The influence of the correction
factors proposed by Krettek–Sobczak is also appreciated, which justifies why in applications
where the transverse stability of the spring is an important operational factor, the calculated
values should be verified by practical tests as suggested by EN 13906-1:2008.

3.2.2. Modelling

There are two main options to model spring components, in general, in a multi-body vehicle
model [20]. The first one is the so-called point-to-point force element (PtP), which exerts
only an axial force along the line of action. The second one is the compact force element
(Cmp), which enables axial and shear forces and reaction moments. In both kinds of elements,
linear and nonlinear characteristics can be modelled, if necessary pre-compression can also
be accounted for.
All the force elements have in common being mass-less. The mass and inertia of the compo-
nent are very small compared with those of other vehicle bodies. The options are to neglect the
masses or to share them among the bodies connected to retain the mass of the actual vehicle.
PtP elements act along the connecting line of their coupling markers (Mi , Mj ) with all their
outputs (forces/torques) applied in this direction. An example is shown in Figure 5. At t = 0,
Mi and Mj are the coupled markers defining the line (Mi − Mj ) in which the forces are acting.
 Vehicle System Dynamics 1029

Figure 4. Lateral stiffness (transverse spring rate) of a coil spring vs. axial load, F (coil spring data L0 = 0.32 m;
d = 0.036 m; D = 0.196 m and n = 4.5).

Mi
Mi’

Bi Bi

Mj

Bj

Figure 5. Point-to-point force element diagram.

At t = t1 , Bi moves towards the final position defining the final position of marker Mi , which
is Mi ; Bj does not change its position. Consequently, at t1 , a different line of action is defined
(Mi − Mj ), that is, the direction of the acting force has changed, as well as its magnitude,
because of the new distance between these markers (|Mi − Mj |).
Compact spring elements allow the user to take into consideration the three main directions:
X, Y and Z. In addition, the stiffness curve of each direction can be different. In this case,
1030 S. Bruni et al.

-F 1 -F
·(r x F )
2

1
(r x F ) ·(r x F )
F F 2

Figure 6. Compact spring element connecting two bodies. Left, reaction moment only at one end. Right, reaction
moment distributed between both ends.

the reaction moments are not neglected; therefore, moments are generated from the offset, as
indicated in Figure 6. The reaction moments can be distributed between the connected bodies.
This element is recommended when bending and reaction moments are important.
Frequently, the coil spring has rubber seats at both ends in order to reduce vibration trans-
mission and improve its seating. Not only does the geometric configuration of the spring and
seats affect the force line, but it has also been observed experimentally that the spring seat
material has an effect [18].

3.3. Friction-based suspension components

In many suspension components in railway vehicles, friction damping is used, especially in

freight wagons or deflated air springs. In some components, such as rubber and (inflated)
air springs, friction damping is only a side effect. These components are described in other
sections of this paper. This section will concentrate on components where damping in the
system is generated by two – in many cases metallic – surfaces that slide on each other. One
advantage of such friction components is that they are relatively cheap and almost maintenance
free. Another major advantage is that the amount of damping in many arrangements is more or
less proportional to the axle load. This is an important property, especially in freight wagons
where a loaded wagon can have up to five times higher axle load than an empty one. A
disadvantage is that the efficiency of friction dampers depends on parameters such as the
friction coefficient in the contacting surfaces and the flexibility in the damper arrangement.
The friction coefficient can vary quite significantly during operation of a vehicle, which can
change the running behaviour of a freight wagon. The uncertainty regarding the parameters
also makes it a challenge to simulate the running behaviour of vehicles, which, to a great
extent, rely on friction damping [21].
 Vehicle System Dynamics 1031

Fx Fx

x x

Figure 7. Force–displacement curve of Coulomb friction model (right) and Coulomb model with spring in series
like those shown in Figure 8 (left).

N
F m F

ks

Figure 8. Friction element with spring in series.

3.3.1. Modelling of friction

In most cases, friction in rail vehicle suspensions is modelled as dry Coulomb friction. In this
model, the friction force is proportional to the normal load. Another possibility is to model the
friction considering two friction coefficients, static and kinetic. The static friction coefficient
during sticking is higher than the kinetic one during sliding. The disadvantage of the Coulomb
model is that it is non-smooth, multi-valued and non-differentiable, which causes numerical
problems in simulations, see the force-displacement curve in Figure 7 (right). Therefore, most
authors presenting work on simulation of dry friction dampers apply regularisation to avoid
the difficulties mentioned above, e.g. [22,23]. One possibility to avoid the problem of a multi-
valued function is using a linear spring in series with a friction slider (cf. Figure 8). The
resulting force-displacement characteristic is shown in Figure 7 (left). Since most structures
containing friction have a finite flexibility, such models could also be regarded as more realistic.
Another possibility for regularisation can be found in the model proposed by Bosso et al.
[24]. The friction force is calculated according to the following expression:

v·X
F= , (9)
1 − (v · X/μ · N)2

where F is the friction force and v is the relative velocity. The X parameter represents the angle
between the velocity axis and the friction curve around the origin, μ is the kinetic friction
coefficient and N is the force normal to the friction surfaces. The force–velocity curve of the
element is shown in Figure 9. A disadvantage of this model is the missing stick state. When
the velocity is very less, the body interface slips too. For dynamic analysis, the model is a
good representation for simulation of friction, but even for very slow force variations, there is
always a slip between the bodies, which does not reflect the real behaviour. The impact of the
regularisation on simulation results depends on the regularisation function used.
Piotrowski [25,26] has developed a non-smooth rheological model that does not resort to
the regularisation. Instead, it employs the notion of the differential succession involving a
contingent derivative of the non-smooth, multi-valued characteristics of Coulomb friction. In
his model, the action of the friction slider in series with spring ks is replaced by the friction
force T (cf. Figure 10). In order to derive the differential equation for the friction force, the
1032 S. Bruni et al.

Friction force

Continuous
function

Relative velocity

If-else function

Figure 9. Continuous friction model.

Figure 10. The slider and the spring replaced by the friction force T [25].

continuity condition for the slider and spring in series is considered. It has the form

Ṫ
+ vs = ς̇ − Ẏ , (10)
ks

where ς̇ is the velocity of the base of the element, Ẏ is the velocity of the top and vs is the
sliding velocity.
Tan and Rogers [27] proposed equivalent viscous damping models to avoid the numerical
problems of Coulomb friction. They claim that this substitution works very well for cases
where sliding motions predominate.
The friction model can be one dimensional or two dimensional. In a two-dimensional model,
sliding in two directions is possible.
A two-dimensional Coulomb friction model can be found, for example, in [28]. The model
is shown in Figure 11. The components of the friction force must satisfy the relation

Fμ = Fxμ 2 + F 2 ≤ μN.
yμ (11)

The sliding velocity which determines the direction of the kinematic force reads

|v| = ẋ 2 + ẏ2 . (12)

Another two-dimensional model for anisotropic friction was derived in [26]. It is based on the
one-dimensional model of Piotrowski described above and is shown in Figure 12.

3.3.2. The influence of dither in friction damping on the running behaviour of freight wagon

Piotrowski [26] also described how dither (i.e. high-frequency vibrations superimposed on
low-frequency vibrations), generated, for example, in the wheel–rail contact, influences the
 Vehicle System Dynamics 1033

Figure 11. A two-dimensional friction oscillator [28].

Figure 12. Two-dimensional friction model for anisotropic friction proposed by Piotrowski [26].

running behaviour of freight wagons. With the help of experiments, it is shown that a friction
damper that is influenced by dither, either perpendicular or co-linear to the sliding velocity of
the body, more or less acts like a viscous damper. In a simulation example, it is shown that
mid-frequency dither in some operational conditions could remove the instability of a two-axle
freight wagon. Therefore, this effect has to be included to achieve relevant simulation results.

3.3.3. Friction wedges in three-piece bogies

The three-piece bogie is probably the most common type of freight wagon bogie in the world.
Therefore, a large number of authors have developed simulation models of this type of bogie,
see e.g. [29–33]. The most critical components to model in a three-piece bogie are the friction
wedges (cf. Figure 13). The friction damping can either be independent of the load (b) or be
dependent on the load (a). The friction between bolster and wedge is usually modelled in one

Figure 13. Friction damping in three-piece bogie with load-dependent damping (a) and constant damping (b).
1034 S. Bruni et al.

Figure 14. (a) Suspension of the standard UIC Y25 bogie. 1, bogie frame; 2, inner spring; 3, outer spring; 4, friction
piston; 5, sliding surface; 6, Lenoir link; 7, axle box and 8, pivot. (b) A photograph of the suspension [26].

dimension, while the friction between wedge and side frame is modelled in two dimensions,
because both vertical sliding and lateral sliding are possible. A typical value of around μ = 0.2
for the friction between bolster and wedge and a value of ca. μ = 0.3 for that between wedge
and side frame could be found in, for example, [34,35]. The difference in the friction is due to
the inclined surfaces between bolster and wedge usually being greased. These values, however,
will probably vary significantly during operation.

3.3.4. Lenoir link in Y25 bogie

The Lenoir link in theY25 bogie shown in Figure 14 is another example of a friction component
in a freight wagon running gear. Models of the Y25 bogie have been developed for example,
by Eickhoff et al. [5], Evans and Rogers [36], Jendel [37], Keudel [38] and Stichel [39]. The
Lenoir link transfers a normal force onto the friction surfaces, which is needed to obtain the
damping. As shown in Figure 14, the outer coil spring is connected to this link via a spring
holder. When the bogie frame moves downwards, the link will pull the spring holder down and
via the friction piston apply a normal load to the friction surfaces. A higher load on the wagon
will thus result in a larger force in the link. The longitudinal component of the link force is the
initial normal force on the friction surface, that is, when the wheelsets are not yawed relative
to the bogie.
The Lenoir link also affects the stiffness in the primary suspension. Since the link is inclined,
there will be a coupling stiffness in the vertical and longitudinal directions. According to Jendel
[37], the stiffness matrix is written as
  
Fx kxx kxz x kz tan2 α kz tan α x
= = . (13)
Fz kzx kzz z kz tan α kz z

Primary damping in the Y25 bogie is provided by two-dimensional friction damping acting
in the Y –Z plane. There are friction surfaces on both sides (front/rear) of the axle boxes, as
can be seen in Figure 15. The normal force on these surfaces is the longitudinal component
of the link force plus the force from the metallic stop between the friction piston and the
bogie frame if the 4-mm clearance is overcome. When the wheelset begins to yaw, one of the
friction surfaces out from bogie centre, left or right side of the bogie, will lose its contact,
and the damping will be zero. This behaviour can be modelled using a nonlinear stiffness
property, having the value of the longitudinal component of the pre-loaded link force at zero
yaw (x = 0) and a zero force if the contact is lost (x > 0).
 Vehicle System Dynamics 1035

Figure 15. Friction surfaces in primary suspension of Y25 bogie.

Friction damping will also occur in the lateral bump stops. Damping will be provided in
the vertical and longitudinal directions when a 10 mm play has been overcome. The stiffness
in the normal direction will be the result of a ‘metallic’ contact though and thus is very high.

3.3.5. Leaf springs

Leaf springs and link suspensions – either single or double links – are still the most common
suspension components in two-axle freight wagons in Europe. They are also frequently used
in four-axle bogie freight wagons. Leaf springs and links provide both stiffness and damping
in one component. In these components, the suspension characteristics also adapt to the axle
load, that is, they are more or less constant when normalised with the axle box load. A typical
leaf spring and double-link arrangement of a two-axle freight wagon is shown in Figure 16.

3.3.6. Typical model of leaf spring and link suspension

Leaf springs are used as vertical suspensions. The horizontal flexibility is usually provided
by other suspension components, for example, double links (cf. Figure 16). The leaf spring in
these cases is regarded as rigid in both the longitudinal and lateral directions. In Figure 17 (left),
typical force–displacement curves of a single-stage and a two-stage (progressive) leaf spring
for large displacements are shown. For dynamic displacements around a static equilibrium
position, leaf springs are characterised by a relatively high stiffness for small displacements
and a significantly lower stiffness for larger displacement (Figure 17, right). Leaf springs are
described in the ORE reports [40,41].

Figure 16. Suspension arrangement of two-axle freight wagon.

1036 S. Bruni et al.

Figure 17. Typical force–displacement diagram of leaf spring/link suspension. Example of a curve for large
displacements of leaf spring (left). Example of a curve for small displacements around a static equilibrium (right).

Figure 18. Model for leaf spring or link suspension as used, for example, by KTH [39]. See also Figure 19.

Since link suspensions show very similar characteristics, they are often modelled in a way
that is similar to the way leaf springs are modelled, at least for the lateral link behaviour. The
initial higher stiffness k1 is caused by friction, for example, the leaves of a leaf spring stick
together for small displacements and start to slide on each other for larger displacements. In the
same way, the link rolls in the end bearing as long as there is no sliding in the contact area. The
lower stiffness k2 is the value for sliding in the leaf spring or the so-called pendulum stiffness
of a link. The force Fd determines the amount of damping in the hysteresis. It should be taken
into account that the characteristic of leaf springs strongly varies due to run-in, deterioration
or lubrication state. A commonly used method to represent the two different stiffness values
with the hysteresis is to use a linear spring and a friction element in series, in parallel with
another linear spring (cf. Figure 18).
The three parameters in the model described above can be derived from measurements of
the components. Measurement results and more detailed descriptions of link suspensions can
be found in [42–51]. This model, however, is simplified, since the shape of the hysteresis curve
is usually rounded (cf. Figure 17).

3.3.7. Model for longitudinal link stiffness

For lateral displacements of a double link, all four joints are assumed to start to slide at the
same time. Therefore, the model shown in Figure 18 is sufficient. In the longitudinal direction,
however, it is more likely that the joints start to slide at different displacements as shown, for
example, by Piotrowski [25]. He used a set of four sliders and spring elements with different
 Vehicle System Dynamics 1037

Figure 19. Model for leaf spring or link suspension with several friction elements in parallel as used, for example,
by Stiepel and Zeipel [51].

break-out forces in parallel to describe these characteristics. Also, in a model used by Alstom,
several elements in parallel are used, as can be seen in Figure 19.

3.3.8. Models using an exponential expression

To give a better representation of the rounded shape of the hysteresis curves, Fancher developed
a model for truck leaf springs [52,53] using exponential expressions. The equations given below
were suggested by Jönsson et al. [48] and are based on Fancher’s model. The total force over
the suspension component is separated into piecewise elastic and friction forces. The model
is used for both leaf springs and double links:

F = F e + F f = F0 + K e · δ + F f (14)

where F0 is the static pre-load, Ke is the stiffness and δ is the deformation over the coupling.
The friction force is given by

kfA kfA
F f = Ff 1 + · (1 − eαA ·(δ1 −δ) ), δ̇ ≥ 0, αA = . (15a)
αA FfA − Ff1
kfB kfB
Ff = Ff 2 + · (1 − eαB ·(δ−δ2 ) ), δ̇ < 0, αB = . (15b)
αB FfB − Ff2

The difference compared to Fancher’s approach is that the force gradient, ∂F/∂δ, is assumed
to be constant at every point when the direction of loading is changed:
 
∂Ff  ∂Ff 
= kfA , = kfB . (16)
∂δ 1 ∂δ 2

The characteristics described above are shown in Figure 20. If the direction of loading is
changed at point 1, the response from both models, α1 and β1 , forms a closed loop and hence
energy is dissipated. At point 2, the response from Fancher’s model, β2 , does not form a closed
loop.
Another possibility to describe a hysteresis with rounded shape for link suspensions is using
rolling contact theory. This was proposed by Piotrowski [54]. Based on the slip velocity, the
creepage in the contact is calculated. With the help of the heuristic formula of Shen et al.
[55], the creep force and the angle of the pendulum are calculated, which gives a transition
1038 S. Bruni et al.

Figure 20. Force–displacement characteristics for the friction force model proposed in [48].

between the state of pure rolling and the state of pure sliding. Piotrowski concludes, however,
that ‘neglecting the creep in the joints leads to marginal discrepancies in the parameters, as
compared to the parameters resulting from the model accounting for creep’.

3.3.9. Buffers

Side buffers are design components that – if considered – are usually modelled with friction
elements. Side buffers, in general, stabilise a train. Therefore, to achieve a realistic running
behaviour, it is important that several vehicles be modelled with buffers in between. Especially,
the so-called low-frequency instability, a kind of lateral snaking mode, usually at a frequency
of 1–2 Hz, can be influenced significantly by taking into account the side buffers between
vehicles. In derailment studies, in the case of push operation of a passenger train or in the case
of braking of a long freight train, detailed modelling of the buffers is also important. While
dealing with freight wagons, however, one has to take into account that the authorities usually
require a wagon to fulfil the running behaviour requirements without help from other wagons.
This means that the draw coupling between wagons is not tightened so that the buffers are not
in contact or are in contact without pre-load.
A buffer is usually an energy-absorbing component at least if the buffer force in the longi-
tudinal direction exceeds a certain level. Most designs rely on friction damping. The simplest
model for the longitudinal dynamics of side buffers is a nonlinear spring in parallel with a
dashpot. Also, models with Coulomb friction as those shown in Figures 8 and 18 could be
used depending on the buffer design.
The friction that arises in the lateral and vertical directions between two pre-loaded buffers
is also significant. The two-dimensional friction could be modelled as shown in Figure 11.
The pre-load of the buffer is given by the longitudinal springs in the buffers. The quasi-static
load is produced by the draw coupling at the centre of the vehicle ends. In reality, the buffer
heads are not planar, but at least the buffer on one vehicle has a curved shape. Especially
when studying the risk of derailment, it is necessary to take the real shape of the buffer into
consideration to get realistic results.
Belforte et al. [56] and Cheli and Melzi [57,58] developed a very detailed model of the lon-
gitudinal dynamics of side buffers taking into account also misaligned loading of the buffer
in the case of lateral displacement. The model, shown in Figure 21, is based on laboratory
measurements on a side buffer under quasi-static and dynamic loading conditions. The param-
eters are derived by least-square approximation with time series of measured buffer forces on
a freight train. The experiments reveal that an increase in the excitation frequency increases
 Vehicle System Dynamics 1039

Figure 21. Rheological model of side buffer (from [58]).

the stiffness and lowers the dissipation force. Internal friction in the buffer increases when
misaligned loads or tangential contact forces are applied.
The model consists of a linear spring kp to reproduce the presence of a pull rod that generates
a pre-compression of the elastomer rings. A nonlinear spring k allows modelling of the quasi-
static response to the elastomer. A first nonlinear Maxwell element [59] is introduced to
reproduce the change in stiffness and damping properties of the elastomer associated with
dynamic loads. A second Maxwell element is used to fit the changes in stiffness and damping
properties of the elastomer at higher frequencies. Finally, a friction element allows the effect
of internal friction to be included, which is emphasised by the application of misaligned load
or tangential contact forces.

3.4. Rubber springs and bushes

Rubber springs found in rail vehicles serve as part of the primary and secondary suspension
systems [60]. An example of a rubber spring is the layered rubber-metal spring (chevron) used
in the primary suspension (Figure 22). Sometimes, these components can provide flexibility
in all three directions, whereas they only give flexibility in the horizontal plane as shown in
Figure 22 (right). Rubber in the secondary suspension mainly shows up as the bellow of air
springs, hosting the pressurised air but also affecting the lateral air spring characteristics. Air
springs are discussed in Section 3.5. Rubber is also used as bushes for hydraulic dampers,
traction rods, etc.

Figure 22. Example of layered rubber-metal components: single component (left) and part of primary suspension
(right).
1040 S. Bruni et al.

Figure 23. One-dimensional model often used to represent rubber (springs). Parallel stiffness k, viscous damping
c and series stiffness ks .

In contrast to the traditional coil springs, rubber springs provide damping as well as increased
stiffness with increased excitation frequency and decreased amplitude. Rubber spring mod-
elling in the present context mainly focuses on the frequency range of 0–20 Hz and on
displacement (deformation) amplitudes typically found in primary and secondary suspensions.

3.4.1. One-dimensional models

In many cases, the very simple model of a spring and dashpot in parallel is used. This might
be acceptable if only a very limited frequency range is studied, but the increase in stiffness
with frequency is significant and often not representative for rubber. A common rubber spring
model is instead the one shown in Figure 23 [61,62]. Here, a series spring has been added to
the dashpot (Maxwell element), which means that the model stiffness will stay in the range
of k to k + ks and monotonically increase with increasing excitation frequency. This better
matches the results of rubber spring measurements. The characteristic frequency ks /c will
determine where the main transition between the two stiffness levels will take place.
To improve the frequency dependency, additional Maxwell elements may be added in paral-
lel to the model components shown in Figure 23 [63,64]. But such, and more advanced, linear
models, in principle, need additional measurements to be justified and more input parameters.
The models described above imply that for harmonic excitation, the hysteresis effect will
tend to zero when the excitation frequency tends to zero. But such behaviour is generally not
supported by measurements. Instead, a certain hysteresis effect remains even for very low
frequencies. This can be related to internal rubber friction associated with the introduction of
carbon black in the rubber manufacturing. Some rubber springs also allow for friction sliding
between rubber and metal parts. In such cases, the hysteresis can be very significant, also
for low frequencies. The internal and external rubber friction will also increase the rubber
spring stiffness, in particular, for low-amplitude motions. The trend of higher rubber stiffness
for higher frequencies is also due to the lower displacement amplitudes associated with high-
frequency motions.
To extend the models described above to account for the phenomena described, friction needs
to be represented. In this way, the models will become nonlinear. One way of introducing the
friction is to add another parallel element to the model shown in Figure 23. This approach is
shown in Figure 24, where the elastic and viscous forces may, for instance, originate from,
respectively, the k and c and ks components shown in Figure 23.
The friction part of the model shown in Figure 24 may be represented by a Coulomb friction
element with a series spring. As for the viscous part, several such friction components may
be added in parallel to imitate a successive stick/slip, see, for instance, Figure 25 referring to
[65]. Models used for leaf springs may also come into question (Section 3.3). Other examples
are given in [66–71]. In [66–70], two-parameter models have been used to represent ‘smooth
 Vehicle System Dynamics 1041

Figure 24. One-dimensional model principle to represent rubber (springs). The total force is the sum of the elastic,
friction and viscous force contributions.

Figure 25. Example of one-dimensional model for elastomeric components: generalised Zener model from [65].

friction’. In [66,67], a logarithmic function was used, whereas a fractional expression was
used in the work of Berg [68–70]. In [71], the friction was included in a nonlinear viscous
part by a velocity-dependent friction model. Also, a nonlinear elastic part was used through a
three-parameter polynomial expression. Also, see models for lateral behaviour of air springs
(Section 3.5).
In principle, the parameters of rubber spring models need to be determined by compo-
nent measurements. Measurements on rubber specimens can also support this process. In
the component measurements, the dynamic behaviour at different excitation frequencies and
displacement amplitudes need to be evaluated, often also different pre-loads.
Traditionally, rubber (spring) damping is expressed as a loss angle, or the tangent of this
angle, describing the phase angle by which displacement lags force at harmonic excitation.
For strong frictional rubber behaviour, cf. above, alternative damping definitions may be
considered, since at low-frequency, large-amplitude excitation, the frictional hysteresis can be
significant but still the phase angle can be very small.

3.4.2. Multi-dimensional models

For two- or three-dimensional rubber spring models, one-dimensional ones are often superim-
posed. This may also be the case for nonlinear one-dimensional models, although this is not
theoretically correct. Aspects such as reference points/levels and possible resisting moments
must be considered as described for coil springs in Section 3.2.
1042 S. Bruni et al.

3.5. Air springs

Air springs are often used in the secondary suspension of passenger railway vehicles, and their
modelling has important implications for the accuracy of quasi-static and dynamic multi-body
simulations. The overall behaviour of this suspension element can be described in terms of
vertical and horizontal behaviour, generally with a weak interaction between the two, although
the vertical pre-load has an important influence on the lateral behaviour of the suspension.
In the vertical direction, air spring suspensions show behaviour highly dependent on the pre-
load and on the amplitude and frequency of dynamic displacements. Hence, specific models
have been defined, which are reviewed in this section. In the horizontal plane, air springs
represent a particular case of shear springs, and the relationship between shear and rotational
deformations and the shear forces and moments reacted by the spring is often not negligible,
requiring a three-dimensional modelling approach. When deflated, air springs sit down on
a rubber emergency spring and can be modelled using the models for rubber elements and
models for friction elements in two dimensions, discussed in Sections 3.4 and 3.3, respectively.
Models of the vertical air spring behaviour can be classified into ‘equivalent mechanical
models’ and ‘thermodynamic models’. Equivalent mechanical models are based on the use of
lumped parameter springs, dashpots and masses. These allow a relatively simple mathematical
description of the suspension, but they generally do not account for the levelling system
behaviour and do not provide an estimate of air consumption. Furthermore, these models may
not be well suited to consider non-conventional suspension configurations (e.g. cross piping
of the bellows) or active/semi-active suspension control.
Thermodynamic models instead aim at representing the actual mechanical and thermody-
namic processes occurring in the air spring suspension, and hence, all parameters in such
models have a clear physical meaning. Despite this, tuning may be needed to define the val-
ues of model parameters describing the concentrated and distributed losses in the pneumatic
circuit.
More details on equivalent mechanical and thermodynamic models are provided in the next
two subsections, while it is not the intention of this paper to cover in detail all existing air
spring models, it focuses on some representative ones. In a subsequent subsection, models of
the air spring suspension in the horizontal plane are reviewed. The final subsection deals with
the definition of model data.

3.5.1. Equivalent mechanical models

The simplest model of the air spring suspension in the vertical direction consists of a spring with
a viscous dashpot in parallel. This model, however, only reproduces the quasi-static stiffness
of the suspension, and it is difficult to define a correct value for the damper parameter, because
the actual dissipative effects in the suspension are far from linear.
A different model, appropriate in a wider frequency range and known as a ‘Nishimura
model’ [72], is shown in Figure 26(a). It consists of a spring K1 representing the bellows, in
series with the parallel combination of a spring K2 representing the compressibility of air in
the surge reservoir and a viscous damper C accounting for dissipations in the surge pipe and,
in suspension systems with air damping to replace hydraulic dampers, also the dissipation
from the orifice between the bellow and the reservoir. Optionally, a spring K3 may be added in
parallel to represent the additional stiffness effect due to the change in the effective area with the
suspension height [73]. A similar model is the ‘linear air-spring element FE83’, implemented
in Simpack [73], see Figure 26(b): this model also includes an additional damper in parallel
to spring K3 and a spring-damper element K4 and C4 in series to the rest of the suspension,
representing the emergency spring.
 Vehicle System Dynamics 1043

Figure 26. (a) The Oda–Nishimura model (from [73]). (b) The Simpack FE83 model [73].

Figure 27. (a) The Vampire model in vertical direction (from [1]). (b) The Berg model in vertical direction (from
[75]).

The simplest Nishimura model with K3 = 0 is sufficient to define a frequency-dependent

behaviour of the air spring: when the frequency of deformation is low, the force in the damper
C is negligible, and the model behaves as the series of springs K1 and K2 , but the damper
force increases with frequency until the deformations of the damper and of spring K2 become
negligible and the complete model approaches the behaviour of the single spring K1 .
Hence, the Nishimura model allows the reproduction of the transition between the ‘low-
frequency’ and ‘high-frequency’ stiffness of the pneumatic suspension, but it does not include
internal suspension resonances which may occur on account of inertial effects in the air mass
contained in the surge pipe. As pointed out in [5], this is a relatively small mass, but it is
accelerated to high velocities when air is exchanged between the bellows and the reservoir,
and therefore, the equivalent inertia can be important. Two models allowing consideration of
this effect are the ‘Vampire model’ [5,74] and the ‘Berg model’ [75].
The Vampire vertical model is shown in Figure 27(a): it consists of a Nishimura model to
which a lumped mass M and a series stack stiffness K4 are added. The K4 spring represents the
emergency spring and may be taken to be infinite. The lumped mass M represents the inertia
of the air mass in the surge pipe. It is worth remarking that in the Vampire model, the damping
term can follow a square law, which is considered to be more accurate than linear damping.
Berg [75] defined a three-dimensional model, of which, only the vertical part is described here
(Figure 27(b)). This is similar to the Vampire vertical model, but it also includes a friction
force element and a velocity exponent parameter (β).
1044 S. Bruni et al.

5
x 10
14
K
Nishimura
12
VAMPIRE

10
Dynamic Stiffness (N/n)

0
0 5 10 15 20 25 30 35 40
Freq (Hz)

Figure 28. Comparison of the vertical dynamic stiffness for an air spring system, Nishimura and Vampire models
(K1 = 7.6E5 N/m, K2 = 5.4E5 N/m, C = 1.29E5 Ns/m, M = 5.4E-3 kg, κ = 209) from [84].

When compared with the Nishimura model, both the Vampire and Berg models are capable
of accounting for an internal resonance of the air spring suspension, leading to a maximum
value of the dynamic stiffnesses in an intermediate frequency range, with the quasi-static and
high-frequency stiffness approaching the same values as those for the Nishimura model, see
Figure 28 for the Vampire model and Figure 29 for the Berg model.

3.5.2. Thermodynamic models

Thermodynamic models of an air spring suspension [76–81] include the following main
elements (Figure 30):
• a model of the bellows and surge reservoir;
• a model of the surge pipe connecting the bellows with the reservoir.
Depending on the scope of the air spring suspension model, a model of the levelling system
can also be introduced, as done in [80,81].
The bellows and reservoir are modelled, respectively, as variable size and constant size
air volumes, whose thermodynamic states are varying on account of the boundary conditions
applied and of fluid exchange between the volumes. The volume of the bellows is expressed
as a function of the air spring height: Quaglia and Sorli [76], Nieto et al. [78] and Docquier
[81] reported examples of measured volume vs. height functions for different types of bellows,
and they all observed that the actual nonlinear relationship between these quantities is well
approximated by a linearised expression.
The vertical force F generated by the air spring is expressed as

F = prel Ae , (17)

where prel is the relative pressure of the air in the bellows and Ae is a geometric parameter
called ‘effective area’. This can be defined either based on the geometry of the bellows or
from measurements. Examples of the relationship between the effective area and the air spring
height are also reported in [76,78,81], where it was observed that the effective area can be
 Vehicle System Dynamics 1045

Figure 29. Dynamic stiffness and damping for different air spring models (from [77]). Model 1: oscillating air
mass. Model 2: incompressible differential. Model 3: incompressible algebraic. Model 4: ISO 6358. Model 5: Berg’s
equivalent mechanical model.

well approximated using either a constant value or a linear expression as a function of the air
spring height.
The thermodynamic state of the air volumes in the bellows and surge reservoir can be
described using a formulation based on mass and energy balances. Instead of the energy
conservation equation, one can consider a polytropic law of the type:
 k  k
V V0
p = p0 , (18)
M M0
where V is the volume of the bellows or tank, M is the air mass in the bellows or tank and
p is the (absolute) air pressure in the bellows or tank, while p0 , V0 and M0 are the given
reference conditions and k is the polytropic exponent whose value depends on assumptions
made on the energy balance of the system, k = 1 describing an isothermal transformation
and k = γ (the specific heat ratio) representing the case of an adiabatic transformation. It
shall be pointed out that the suspension behaviour is significantly affected by the exponent
k used in the polytropic law (25): this issue is specifically addressed in [81,82], where a
sensitivity analysis on the effect of heat transfer was performed, including the two extreme
cases of zero and infinite heat exchange capacity, corresponding, respectively, to the adiabatic
and isothermal transformations. The case of a metro railway car is considered, showing that
different assumptions on the heat transfer coefficient lead to significant differences in wheel
unloading while the vehicle negotiates a track twist and also heavily affect the estimate of
air consumption associated with passenger loading/unloading; note, however, that both the
conditions considered are happening on a time scale in the range of 102 s, much larger than the

Figure 30. Thermodynamic model of an air spring suspension (from [77]).

1046 S. Bruni et al.

time scales typically associated with vehicle dynamics. Vehicle dynamics on a shorter time
scale were dealt in [76,77,83] under the assumption of heat exchange being negligible in a fast
manoeuvre and hence using the adiabatic exponent; in [83], this assumption was corroborated
by a good fit with full-scale measurements. On the other hand, in [78], it was reported that
measurements performed on an air spring suspension in working condition supported the
assumption of an isothermal transformation, but no information was given on the time scale
of the experiments performed.
The second main component of air spring thermodynamic models is the model of the
surge pipe connecting the bellows with the reservoir. This can be defined at different levels
of complexity: the simplest model, used, for example, in [76], consists of a simple fluidic
resistance, defined according to ISO 6358 and accounting for concentrated and distributed
losses, and may include the effect of an orifice introduced in the duct to increase the damping of
the suspension. A more detailed model, used, for example, in [77,81,83], is the ‘incompressible
differential’ model, which accounts for the inertial effects associated with the oscillation of the
fluid in the pipe, assuming a one-dimensional incompressible flow; in this case, a first-order
differential equation is written for the pipe, and one additional state is introduced for each
pipe. A similar approach is implemented in the Simpack FE82 element [79], a thermodynamic
model in which the mass of the air pipe is considered as a rigid mass oscillating between the
bellows and the reservoir, subject to dissipative forces representing losses in the pipe.
Docquier et al. [77] and Docquier [81] also proposed an ‘incompressible algebraic’ pipe
model, which is derived from the incompressible differential one neglecting the inertial effects
related to the air contained in the pipe. Like the case of the model based on ISO 6358 formulae,
this model results in an algebraic equation relating the mass flow rate in the pipe to the pressure
drop between the reservoir and the bellows. The different pipe models were compared in
[77,81], see Figure 29, showing that the incompressible differential model (or similar ones)
is able to account for an internal resonance of the air spring in the same way as the Vampire
and Berg equivalent mechanical models.
Docquier [81] in his PhD thesis also proposed the use of different kinds of compressible flow
models to describe fluid motion in the piping and showed that by considering fluid flexibility, a
second internal resonance of the suspension can be described, which, depending on the length
of the pipes, may fall below 20 Hz.

3.5.3. Air spring models in the horizontal plane

The behaviour of the air spring in the horizontal plane is determined by the structural stiffness
of the bellows and by the effect of the vertical load, which combined with shear and rotational
deformation of the bellows give rise to shear forces and moments on the end mountings.
Friction effects at the rubber–metal interface at the bellow upper perimeter may also contribute.
In the Vampire model (Figure 31(a)), the horizontal air spring behaviour is modelled using an
elastic force element having either a linear or a nonlinear characteristic, in parallel with a model
of rubber hysteresis and a visco-elastic element consisting of a damper with series stiffness
[73,74]. Additionally, a balancing moment is introduced to satisfy the static equilibrium of the
air spring, given the pre-load, shear force and deformation of the spring. The user is allowed
to define a non-uniform share of the balancing moment on the upper and lower ends of the air
spring.
The Berg [75] model in the horizontal plane consists of three components: a set of elastic
forces, a frictional contribution and a viscous one. The elastic component consists of two
shear forces, a roll moment and a pitch moment applied at the air spring upper end, defined as
linear functions of the shear deformation in longitudinal and lateral directions and of the roll
 Vehicle System Dynamics 1047

Figure 31. (a) The Vampire model in lateral direction (from [73]). (b) The Berg model in lateral direction (from
[75]).

and pitch rotations of the end mountings. The forces and moments at the lower end mounting
are obtained from the static equilibrium of the air spring. The friction and viscous forces are
instead defined only in the lateral and longitudinal directions. The complete Berg model for
the lateral direction, shown in Figure 31(b), is hence similar to the Vampire model in the lateral
direction, with the difference being the linearity of the elastic term and the different function
for the ‘smooth’ friction.
Despite the fact that the elastic component of the Berg model accounts for the coupling of
shear deformation and rotations of the end mountings and for the effect of pre-load, in [75],
only the direct shear stiffness term was identified, and comparisons with measurements were
limited to the shear force generated under pure shear deformation.
Facchinetti et al. [83] proposed a quasi-static, linear elastic model for the air spring suspen-
sion in the horizontal plane, similar to the elastic component of the Berg model but limited to
lateral and roll movements. All stiffness parameters were identified from full-scale measure-
ments performed by applying combinations of shear and roll deformation. In the same paper,
vehicle dynamics simulations were performed using the proposed model and a simplified one
neglecting the direct roll stiffness term and the shear–roll coupling term, and the results were
compared in terms of wheel–rail contact forces, showing that the stiffness terms under exami-
nation have a non-negligible effect on the load transfer effects when the vehicle is negotiating
a curve or is subject to the effect of crosswinds.

3.5.4. Definition of air spring model parameters

Presthus [73] collected formulae to define the parameters of various equivalent mechanical
models, including the Nishimura, Simpack FE83 and Berg models, based on the physical
properties of the air spring suspension (effective area, air volumes, concentrated and distributed
loss coefficients), and similar expressions are available for the Vampire model [74]. Alonso
et al. [84] reported a comparison between the measured and simulated dynamic stiffnesses of
an air spring suspension, showing a good agreement between the two sets of data for different
cases of concentrated losses in the surge pipe.
For thermodynamic models, the definition of the input parameters is more straightforward,
since these are directly represented by the physical parameters of the system. However, the
accuracy of both equivalent mechanical and thermodynamic air spring models may be affected
by uncertainties in some physical parameters: in particular, the variation of the effective area
with the air spring height can hardly be defined other than by a direct measurement and the range
of validity of semi-empirical formulae defining the loss coefficients based on the geometry of
1048 S. Bruni et al.

the surge pipe needs to be carefully considered, especially in the case of complicated geometry
of the pipes and orifices.
Based on the examination of the state of the art, the direct measurement remains at present
the most frequently used way to define some critical air spring geometric parameters, namely
the relationships between the effective area and the volume of the bellows and the height of the
air spring, although attempts have been made, for example, by Qing and Shi [85], to compute
some of these parameters based on the analysis of the air spring geometry.

3.6. Hydraulic dampers

The damping mechanism in hydraulic dampers comes from oil passing through restricted
orifices. It is the most common suspension component for many types of vehicles. Different
technical and manufacturing solutions exist depending on the application (bikes, automobiles,
trucks, buses, rail vehicles, etc.) which defines displacements, piston speed, maximum force
allowed, temperature, sealing requirements, impacts, etc. The fluid inside the damper is gen-
erally oil as gas dampers are not used in the railway sector, although they are common in other
types of vehicles. Hydraulic dampers, also called shock absorbers, behave in a nonlinear and
time-variant manner, and consequently, modelling their dynamic behaviour is not straight-
forward. Damper designers need to introduce nonlinear characteristics in their components
in order to satisfy the conflicting requirements of comfort, stability and reduced train–track
interaction forces, among other criteria. Force–velocity diagrams are the straightforward way
to represent damper dynamics. However, on the track or road, the behaviour of the vehi-
cle depends not only on some absolute values of these diagrams but also on the shape of the
curves, which are influenced by the excitation frequency and amplitude of the imposed motion
and by temperature. This leads to the use of force–displacement diagrams and force history
(force–time) plots if good characterisation and understanding of the damper are the goals.
Hydraulic dampers are used in vertical and lateral secondary suspensions of rail vehicles
and often also in primary vertical suspensions. Yaw dampers need a special approach and
accurate characterisation. They are critical for the vehicle running dynamics as they assure
bogie stability with respect to hunting. Therefore, suitable models are required to replicate
the actual operating conditions of the component which, during vehicle motion, undergoes
displacements with different amplitudes: small ones 1–3 mm, as far as hunting motion is
concerned, higher ones if curve negotiation is considered, and at various frequencies, the
frequency varies from relatively low values during curve negotiation to high values (e.g.
4–8 Hz) when hunting motion occurs.
An introduction to hydraulic damper modelling can be found in [5], and also in Duym
et al. [86] and Carrera et al. [87], related to automotive applications. Most detailed damper
models are those that relate their parameters directly to the physical properties of the different
damper parts. These mechanistic models help to improve the understanding of the damper’s
behaviour. An example of a detailed physical model of a shock absorber is that of Lang [88].
It is based on the processes of the oil/gas flow through the various internal chambers and
describes the behaviour of the damper in a broad range of operating conditions. The results
of the model with 87 parameters show a very good agreement with experiments; however, it
is rather uneconomical in terms of computational effort and modelling complexity.
Other simpler physical models are possible. Alonso and Giménez [89] recently defined a
detailed physical model of a hydraulic damper of a rail vehicle, based on differential equations
reproducing the dynamics of oil fluid. The results were compared with measurements on a
real damper, showing a satisfactory agreement. Furthermore, a simplified model suitable for
being introduced in a multi-body model of a complete rail vehicle was developed in the same
 Vehicle System Dynamics 1049

Figure 32. The ‘dashpot with series stiffness’ model of hydraulic dampers (top). The flexibility and damping of the
rubber bushings can be included (bottom). If Cs = 0, then 1/K = 2/Ks + 1/Ka and C = Ca .

paper, consisting of one first-order differential equation, having two different mathematical
expressions to separately represent the compression and extension phases. The simplified
model was validated by comparing it to the more detailed one, showing very good agreement.
Van Kasteel et al. [90] defined a damper model based on the physics of the component, a
‘white box’ model, which allows for insight into the damper physics and includes all non-
linearities (e.g. nonlinear effect associated with leakage) and dynamics within a frequency
range up to 30 Hz. In the paper, the issue of modelling the damper component was developed,
providing information on the valve and pressure modelling.
The alternative to ‘physical models’ is the use of elementary models constructed with
combinations of spring and ideal viscous damping elements whose values are fitted from
experiments or from component specifications.
The model consisting of a dashpot with series stiffness is the most widely used (Figure 32).
The dashpot can have a linear/piecewise linear/fully nonlinear characteristic curve, relating
the velocity to the force generated, whereas the series stiffness is mostly assumed to be linear,
with the notable exception of [91], as discussed below. The ‘dashpot with series stiffness’
model is also one of the modelling options for hydraulic dampers available in multi-body
simulation software packages. Additionally, the model should take into account the effect of
blow-off valves by limiting the maximum force of the damper model.
In the ‘dashpot with series stiffness’ model, the stiffness is used to represent the effect of
internal damper flexibility, the deformation of the end bushings and other possible effects (e.g
the flexibility of the damper supports). The combination of series stiffness and dashpot acts
effectively as a pure damper at low frequencies and a pure stiffness at high frequencies. The
cut-off frequency fc depends on the values of damping C and series stiffness K, f = K/(2π C).
Low series stiffness reduces the energy transmission at high frequencies, so the use of softer
damper bushes can help to reduce vibration levels and improve running comfort. However, it
is important to ensure that the cut-off frequency of the damper is sufficiently higher than the
frequency intended to be damped.
When flexibility effects are neglected, the model reduces to a single dashpot with either a
linear or a nonlinear behaviour. However, the use of a correctly tuned series stiffness parameter
can be important to represent the damper’s behaviour at high frequency of deformation, with
relevant implications on both the transmission of high-frequency vibrations through the damper
and the effectiveness of yaw dampers to control the hunting oscillation of the bogie. Reasons
for modelling flexibility effects in yaw dampers are detailed by Wrang [92] and Bruni et al.
[93], where the force generated by the damper is shown to be the sum of an elastic term and
a dissipative one, and the importance of the two force components is discussed regarding the
frequency and the amplitude of the excitation.
Furthermore, Conde Mellado et al. [91] compared the results of the ‘dashpot with series
stiffness’ model with the results of experiments performed on a yaw damper designed for a
high-speed vehicle and found some important deviations from the actual behaviour of the real
1050 S. Bruni et al.

Figure 33. The ‘pinlink element’ model of hydraulic dampers.

component. The paper shows that the yaw damper model could be improved substantially by
introducing two modifications:

• replacement of the linear series stiffness with a bilinear, asymmetric stiffness element to
reproduce a decrease in stiffness close to 35% in compression and 25% in extension when
exposed to small displacements.
• use of an asymmetric force–velocity characteristic to reproduce the different behaviour of
the component in compression and extension.

In the same paper, the influence of the yaw damper model on the numerical estimation of the
vehicle critical speed was examined, showing a variation of the critical speed from 480 km/h
when the original yaw damper model was used to 400 km/h using the modified yaw damper
model.
The sensitivity of the critical speed calculations to the yaw damper model was also addressed
by Alfi et al. [94]. In this paper, three different yaw damper models were compared: a dashpot
without series stiffness, a dashpot with series stiffness and a dashpot with stiffness in paral-
lel. All the three models are linear, and their parameters are tuned based on laboratory tests
performed on a yaw damper for high-speed vehicles. By means of running dynamics calcula-
tions performed using a nonlinear wheel–rail contact model, critical speed values in the range
between 300 and 355 km/h are obtained depending on the model used for the yaw damper:
this shows a sensitivity to the yaw damper model in the range of 18%, which is comparable
to the corresponding results reported in [91].
A different model of hydraulic dampers including yaw dampers is the ‘pinlink’ element
available in Vampire [74], shown in Figure 33. With respect to the ‘dashpot with series stiffness’
model described previously, this element additionally includes a nonlinear stiffness and a
friction element in parallel with the dashpot. The nonlinear stiffness K and friction element
F can be disabled, obtaining a model similar to the ‘dashpot with series stiffness’ one (called
‘damper element’ in Vampire), with the notable difference that the line of action of the force
generated by the damper element remains fixed irrespective of the relative movements of its
ends in Vampire, whereas in the pinlink element, the line of action changes orientation to keep
the direction parallel to the relative position of the two ends of the element. When the friction
element is activated, a finite value of the series stiffness Ks shall be prescribed, and a series
damping Cs is calculated automatically, to avoid the occurrence of numerical oscillations
produced by stick–slip effects.
In summary, many options are available for hydraulic damper models. The simplest option
is just a linear or nonlinear viscous damper, which can be upgraded using a linear or nonlinear
viscous damping element, to which a friction damping element could also be added. For yaw
dampers, in most cases, the choice is a model based on a nonlinear viscous element with series
 Vehicle System Dynamics 1051

stiffness. For other dampers, the preferred model varies as the selection of a particular damper
model is related not only to the type and properties of the damper itself but also to the kind of
suspension the damper is part of. In general, due to the combined effect of rubber end bushes
and oil/structural flexibility, the modelling of the internal damper flexibility is recognised as
a fundamental effect which cannot be disregarded.

3.7. Active and semi-active suspensions

The use of semi-active and active devices is being increasingly considered in passenger rail
vehicle suspensions, mostly at the secondary suspension level, as a means to improve ride
comfort or to provide the same level of comfort for increased service speed, and a number of
railway vehicles in commercial service are nowadays being equipped with active or semi-active
secondary suspensions. On a more research-related level, active primary suspensions have also
been proposed, particularly to remove the design conflict between curving and stability which
is typical for the passive primary suspension in a railway vehicle [95].
Appropriate models for active and semi-active suspensions are hence being developed to
assess the system performance, verify the implications of active control on system stability,
estimate energy consumption, define actuator requirements and analyse the behaviour of the
system in fault conditions.
Active and semi-active suspensions can be categorised based on the function that they
perform in the vehicle, for example, primary vs. secondary or vertical vs. lateral actuation, or
otherwise based on the principle of actuation, with pneumatic, hydraulic, electro-mechanical
and magneto-rheological being the most widely used. The choice in this section is to categorise
the models based on their principle of actuation, as this profoundly affects the modelling.

3.7.1. Pneumatic semi-active and active suspensions

Active and semi-active pneumatic suspensions are used in the secondary suspension stage of
passenger railway vehicles and can act either in the vertical direction (active/semi-active air
springs) or in the lateral direction (lateral secondary suspension).
Mathematical models of active and semi-active vertical suspensions are derived from both
the ‘equivalent mechanic’ and ‘thermodynamic’ air spring suspension models presented in
Section 3.5.
Tang [96] proposed an equivalent mechanical model for a semi-active air spring with a
controlled variable size orifice in the surge pipe: the model consists of a modified Nishimura
model in which the effect of the variable size orifice is represented by changing the damping
rate C of the viscous damper. The same modelling approach was followed by Sugahara et
al. [97], where a model for the vertical vibration of a complete rail vehicle equipped with
semi-active air springs and adaptive primary dampers was also defined.
A thermodynamic model of an active air spring suspension was defined by Alfi et al. [98] as
a modification of the model shown in Figure 30, by modelling one additional constant air vol-
ume representing an additional supply reservoir and the servo-valve connecting the additional
reservoir with the surge reservoir. The model of the active suspension was compared with
experiments performed on a full-scale prototype, showing a good match of numerical simula-
tions with the measurements. The model of the active suspension was then used in combination
with a multi-body model of the entire vehicle to investigate the use of the active suspension to
improve ride comfort in curves and to increase the vehicle’s resistance to overturning in the
presence of crosswind.
1052 S. Bruni et al.

As far as pneumatic active lateral suspensions are concerned, a thermodynamic model was
proposed by Sorli et al. [99], which was integrated into a multi-physics simulator of the lateral
dynamics of a Pendolino vehicle; numerical results from the model were compared with line
measurements, showing good agreement.
Conde et al. [100] proposed a model of an active lateral suspension interconnected to the
vertical pneumatic secondary suspension, and the model was used to compare the performances
of the proposed active suspension to a conventional passive one.

3.7.2. Active and semi-active hydraulic suspensions

Active and semi-active hydraulic devices are being proposed in primary and secondary rail
vehicle suspensions. Hydraulic devices allow a larger pass-band than pneumatic actuators
and can be used to improve the performance of the active or semi-active suspension, also
allowing the use of these devices in the primary suspension [97]. However, hydraulic sus-
pensions imply a higher transmissibility at high frequency and, therefore, their use needs to
be carefully considered in view of high-frequency vibration transmitted to the car body. Fur-
thermore, hydraulic actuation requires that a dedicated hydraulic circuit be introduced in the
vehicle, whereas pneumatic actuators may use the same circuit feeding brakes, doors and other
pneumatic devices in the train.
Sugahara et al. [97] proposed a model for a semi-active vertical damper for use in the
primary suspension of Shinkansen trains. The model consists of an ideal orifice in parallel to
a controlled blow-off valve. The ideal orifice produces a quadratic relationship between the
speed of deformation and the force generated by the damper, which is saturated by the blow-off
valve to a linear characteristic. The value of the damping force at which the blow-off valve
intervenes is controlled by an input signal representing the current fed into the valve’s solenoid.
The time lag between the current signal and the opening of the blow-off valve is reproduced
by a first-order state equation, whose time constant is calibrated based on experiments.
Cheli et al. [101] used a third-order linear model of hydraulic actuators, in combination
with a four-bar linkage model of the tilting bolster to model body tilt actuation in a Pendolino
vehicle. Results of the model were compared with those obtained from the experiments per-
formed on a test rig which allows the reproduction of curve negotiation on a single railway
vehicle by imposing independent roll rotations to each wheelset and with line tests. A similar
model of a servo-hydraulic actuator for use in the active secondary suspension of a railway
vehicle was proposed by Foo and Goodall [102].

3.7.3. Electro-mechanical active suspensions

The use of electro-mechanical actuators has been proposed for rail vehicle active suspensions,
on account of their good dynamic properties and high bandwidth.
Modelling an electro-mechanical actuator requires a multi-physics approach, by which the
dynamics of both the electrical motor and the mechanical gear are accounted for. Examples
of DC motor models applied in active suspensions for rail vehicles can be found in Pearson
et al. [103] and Pacchioni et al. [104].
A model of an AC synchronous motor and mechanical actuator controlling the yaw motion
of the bogie was defined by Resta and Bruni [105], and numerical results were compared with
those obtained from the laboratory tests. The same model was applied to define the expected
performance of the actuator as a means to increase vehicle stability and curving performance
and the results obtained were compared with line test measurements, partly performed with
the actively controlled vehicle [106].
 Vehicle System Dynamics 1053

3.7.4. Magneto-rheological dampers

Magneto-rheological dampers are semi-active devices exploiting a physical property of con-

trollable fluids capable of changing their viscosity properties when exposed to a magnetic
field. A number of applications have been proposed for magneto-rheological dampers in the
automotive field as semi-active suspension components, and a limited number of references
can be traced where the same technology has been proposed for rail vehicle suspensions.
In particular, Lau and Liao [107] and Wang and Liao [108] proposed a magneto-rheological
semi-active lateral secondary suspension in a railway vehicle, introduced two slightly different
equivalent mechanical models of the suspension and compared the numerically predicted
force–displacement hysteresis loop with laboratory measurements. The model of the magneto-
rheological suspension was then implemented in Simulink and integrated into a Vampire model
of the complete vehicle to assess the semi-active suspension performance.

3.8. Linkages and other vehicle components

The vehicle may possess other components to be represented in the model, for example,
traction connections between car body and bogies, anti-roll devices, bump stops, metallic
stops, links, car body tilting mechanism, wheelset guides, steering linkages, drive system,
brake system, articulation joints or coupling devices between car bodies, inter-car dampers,
buffers and couplers between vehicles. The modelling of these components is very rarely a
topic of publications, but the review papers [1,5] mention some of them. Simulation tools
usually do not contain special elements for modelling of these components.
Besides the functional stiffness of the component such as the longitudinal stiffness of a trac-
tion link or the roll stiffness of an anti-roll bar, several components inevitably provide stiffness
in other directions too. The term ‘parasitic stiffness’ is used to characterise the stiffening of
the suspension which was not intended during the design. The typical sources of such stiff-
ening are the traction link, anti-roll device, brake piping, traction motor cables and hydraulic
dampers (due to parasitic friction in series with the end stiffness). It is important to include
the parasitic stiffness in the vehicle model unless it is proven that this effect is negligible.
Basically, there are two ways to model complex components:

• simplified representation by equivalent stiffness and damping parameters derived from the
parameters of the bushes and other connecting parts and
• detailed modelling with rigid bodies representing each single body (link, rod, etc.) and
coupling elements representing each physical component (rubber parts, slide bushes, etc.).

Both ways will be explained in the following two subsections using examples of a traction
link and an anti-roll bar.

3.8.1. Traction connection between car body and bogie

The function of the traction connection is to transfer the longitudinal traction and braking
forces between the bogie frame and car body. Many different solutions exist, for example,

• traction rod,
• centre pivot and Watts (lemniscate) linkage,
• centre pivot with play and rubber bump stop and
• pre-stressed layered rubber-to-metal components.
1054 S. Bruni et al.

k 1 k 2
kx1 kx2

kz1 kz2

Figure 34. Horizontal traction link and its stiffness parameters (dashpots representing the damping are not shown).

We have explained the different ways to model such a component using an example of a
traction link consisting of a horizontal steel rod with spherical rubber joints on both ends
(Figure 34).
Considering infinite stiffness of the steel rod, the equivalent longitudinal stiffness kx of this
component yields
1 kx1 kx2
kx = = , (19)
1/kx1 + 1/kx2 kx1 + kx2
with kx1 , kx2 – stiffness of spherical joints in the longitudinal direction.
This is the intended design stiffness, while the stiffness of this component in the vertical
and lateral directions should be as low as possible. The presented traction link design results
in parasitic stiffness in the vertical direction:

1 kz1 kz2 (kφ1 + kφ2 )

kz = = , (20)
1/(1/(1/kz1 + 1/kz2 )) + 1
(1/b2 )(kφ1 +kφ2 )
(kφ1 + kφ2 )(kz1 + kz2 ) + b2 kz1 kz2

with kz1 , kz2 – stiffness of spherical joints in the vertical direction and kφ1 , kφ2 – torsional
stiffness of spherical joints around the horizontal lateral axis.
The parasitic stiffness in the lateral direction can be calculated similarly by replacing the
stiffness parameters by the values in the horizontal plane.
A model of this traction link design can consist of
• longitudinal spring and damper to model the equivalent stiffness and damping between the
bogie and car body, while the parasitic stiffness is considered in other suspension elements,
• a rigid body, representing the rod, and two bushes, each representing the stiffness of the
actual rubber bush.
Another design solution for the transfer of forces between the bogie and the car body is a
combination of traction connection with the car body support to a flat or a spherical centre
bowl as typically used in freight wagons. The paper by Fergusson et al. [109] presents the
modelling of the rotational friction at the centre plate of a freight wagon used in simulations
related to the optimisation of self-steering bogies regarding curving.

3.8.2. Anti-roll bars

To achieve soft ride characteristics in the vertical direction and, at the same time, operation
of the vehicle within the limits of the operating gauge, passenger vehicles are often equipped
with anti-roll bars. They provide an additional rotational stiffness between bogie and car body
around a longitudinal axis. A complete anti-roll assembly comprises a torsion bar, levers,
links and support bearings (Figure 35). The anti-roll bar may be a single-piece bent bar or a
three-piece subassembly, where the separate levers are shrunk onto a straight torsion bar. The
spherical joints and support bearings can be either of rubber or a sliding type.
 Vehicle System Dynamics 1055

Longitudinal axis

Vertical links

Bent torsion bar

Spherical joints

Support bearings Lever

Figure 35. Design of anti-roll device.

Although the anti-roll bars primarily provide rotational stiffness, they will, in practice, also
generate parasitic stiffness in the lateral and vertical directions that should be considered and
included in the vehicle model.
Similarly to the previous example of the traction link, the model of an anti-roll bar can
be realised either by the equivalent rotational stiffness of the complete anti-roll bar while the
parasitic stiffness is considered in other suspension elements or by a detailed model, consisting
of rigid bodies and connecting elements, respectively. Both these ways have advantages and
disadvantages. The first method leads to a less complex model and is well suited for numerical
integration. However, it requires the calculation of equivalent parameters representing the
complete component, which can be a source of errors. Furthermore, the forces acting on
bodies from the anti-roll bar are not represented at the correct positions, so that the model may
be insufficient for simulations considering a flexible car body or for calculation of loads on
the bogie frame and car body. The second method leads to a rather complicated model with
bodies of small masses, connected by stiff elements, which is disadvantageous for numerical
simulations. The parameters of coupling components, however, can be directly derived from
data sheet or measurements, and the forces act on the bodies in the correct locations.
In the questionnaire carried out in the project DynoTrain within 15 partners (rolling stock
suppliers, operators, universities, research institutes and a safety body), 70% responded to
be using the roll stiffness only to model the anti-roll bar. This confirms that the simplified
modelling of the anti-roll bars can be considered as the state of the art. This simplified model,
however, may be insufficient for certain anti-roll bar designs. For example, the anti-roll device
with inclined vertical links (Figure 36) leads to a kinematic effect similar to a pendulum
suspension, which would not be considered by using only a torsional spring between car body
and bogie.

3.8.3. Bump stops and metallic stops

Bump stops and metallic stops are modelled as a single-sided or double-sided nonlinear stiff-
ness with a zero force inside the clearance. It is important to consider the change of the play
of vertical stops and bump stops when varying the loading conditions. Even for the nominal
conditions, these plays can change if there is any transient movement during the calculation
1056 S. Bruni et al.

Figure 36. Kinematics of an anti-roll bar with vertical (left) and inclined links (right).

of nominal forces after the model assembly. In such a case, the vehicle static equilibrium has
to be checked and the plays in stops have to be approved before using the model.
Stops or bump stops with horizontal play dependent on curve radius are sometimes used
to allow a widening of the car body by reducing the maximum lateral displacement between
bogie and car body in very small-radius curves (typically smaller than 250 m). The coulisse
used in this component reduces the play in the lateral direction depending on the yaw rotation
of the bogie frame against the car body in curves. This specific design of stops either can be
modelled by a two-dimensional elastic contact element (if available) or requires a specific set
of parameters for the simulation of each single curve radius, respectively.

3.8.4. Wheelset guidance

The function of wheelset guidance can be integrated in the primary suspension using rub-
ber springs or it can be performed by a separate component, respectively. Possible wheelset
guidance design solutions are
• sliding wheelset guide with play,
• pivot sliding guide,
• guidance by links (rods): longitudinal link, triangular link and lemniscate link,
• swing arm link and
• guidance by layered rubber-to-metal springs.
The model of the wheelset guide can sometimes be included in the model of primary suspension
even if this functionality is not integrated in the real vehicle. This simplified modelling can
be used if the effect of the guide does not influence the direction of forces and moments. This
is valid if the guide acts horizontally in the wheelset axis as, for example, sliding guide or
lemniscate link. For other wheelset guide designs, this simplification can lead to significant
differences in the results, see the analysis in [5,110] for a swing arm.

3.8.5. Steering linkages

Steering linkages are used for a cross-coupling of wheelsets or a coupling of the wheelset
yaw angle with the yaw angle between the car body and bogie or between two car bodies,
respectively. These or other linkage mechanisms can be modelled either simply by introducing
the links between DoF or in more detail by rigid bodies coupled through elastic components.
Though the simplified modelling is better suited for numerical integration, it neglects the
parameters which influence the wheelset yawing due to the longitudinal elasticity. Polach
 Vehicle System Dynamics 1057

analysed the equivalent axle guidance stiffness of a self-steering wheelset cross-coupling

mechanism consisting of a lateral torsional shaft fixed through rubber bearings on the bogie
frame in [111]. The paper investigated the sensitivity of this self-steering mechanism to the
influence of operating conditions such as curve radius, rail inclination, rail wear, tractive force
and wheel flange lubrication.

3.8.6. Drive system

The drive system can be unsuspended or suspended (partly or fully) by primary suspension or
by both primary and secondary suspensions. The drive components can be fixed
• in the car body (inside or below the floor of the coach),
• on the bogie frame and
• around the wheelset axle or independent wheel.
While the last option is very rare, the drive system suspended in the bogie frame represents
the design most widely used in electric locomotives as well as in other traction units.
The drive system model can be simplified for the running dynamics assessment, if the
traction motor is rigidly fixed to the bogie frame, car body or other parts of the vehicle.
The mass of drive components must, however, be included in the mass and other inertia
parameters.
A traction motor and/or other parts of the drive system suspended in the bogie frame
may have a significant influence on the stability assessment and the dynamics of the
bogie, see [112]. They should be modelled by separate rigid bodies and relevant connecting
elements.
Simulation of running dynamic performance is usually carried out without considering the
tractive effort. For heavy traction vehicles, there can be an important impact of the tractive
effort on track loading, wear and damage of wheels and rails. The effect of traction effort on
the wheelset radial steering was investigated by Polach [113]. For this kind of assessment,
the traction torque can be modelled as a moment on the wheelset. Investigations of interac-
tion between the running dynamics, traction chain dynamics and traction control, however,
require not only a detailed modelling of the torsional drive system but also an extension of
the creep force model with the negative slope at large creepages as described and illustrated
in [114].

3.8.7. Brake system

Similarly as for tractive effort, simulations of running dynamics are usually carried out without
considering any braking forces. However, the impact of braking can be important for long trains
due to the delay of brake onset and for the assessment of the derailment risk and track loading
forces when braking in curves.
Because of a rather small mass of brake equipment, the mass of brake components can be
added to the relevant body masses and moments of inertia and the centres of gravity (CoG)
replaced by the actual values considering the mass of the brake equipment. Heavier brake
components, however, may require to be modelled as a separate rigid body. This can be the
case, for example, for a magnetic brake suspended in the bogie frame or a low hanging magnetic
brake fixed to the axle boxes as used in trams.
Braking usually does not influence the running performance. In specific cases, the brakes
can have relevant influence on running performance and should be considered when modelling
the vehicle as illustrated in the recently published investigation by Kovtun et al. [115]. They
1058 S. Bruni et al.

investigated the influence of braking on the running performance of a freight wagon due to
the interaction of the conventional brake rig with the wheelsets and frame of a three-piece
bogie. The results obtained show that the brake blocks are excited to virtually undamped
longitudinal oscillations by the up and down motions of the bogie bolster which worsen the
wagon dynamics.

3.8.8. Coupling between vehicles and between car bodies of articulated vehicles

Vehicles in a train are connected by coupler and buffers (Section 3.3). Car bodies of a traction
unit are usually coupled by a short coupler instead of by a standard coupler and buffers.
Such compositions as well as articulated vehicles require a model consisting of the whole
composition or at least its sequence with more than one car body. The modelling of the
inter-car joints can be carried out either using a constraint reducing one or more DoF or
using a bush element considering all six stiffness parameters. Besides the articulation joint
connecting the neighbouring car bodies, the vehicle can also possess bump stops and/or stops,
inter-car dampers and possibly also linkages between the car bodies. The modelling options of
these components are the same as those described in the previous sections about the suspension
components.

4. Track models for use in rail vehicle dynamics simulation

Track modelling is taken to include the definition of

• Track design geometry.

• Rail profile.
• Track irregularities.
• Track dynamics.

The track design geometry definition includes the track layout (curve radius, curve transition,
track cants and track gauge), which can be either design or measured data.
The rail profile can be obtained from the standard EN 13674-1 [116] or other standards and
include the nominal rail inclination. It might be important to consider worn rail profiles too.
An additional effect which modifies the wheel–rail contact conditions is the effect of rail roll
due to lateral forces [117].
If available, measured data should preferably be used for track irregularities. It is important
to use irregularity data which contain wavelengths covering all investigated frequencies of the
vehicle under study for the required speed range. The track irregularity data should contain
sufficiently long wavelengths to excite the lowest car body modes at the maximum speed
and also sufficiently short wavelengths to excite the highest frequency at low speeds. This
requirement is often difficult to fulfil, especially for long wavelengths due to the limitations
present in the measuring vehicles.
When classifying the irregularity data, the difference between stochastic and periodic irreg-
ularities should be mentioned first and then the different irregularity types in each group,
for example, sleepers or rail joints as a periodic irregularity. Track irregularities could also
be classified according to their wavelength [118]. The stochastic track irregularities can also
be randomly generated based on the specified power spectra density (PSD). PSDs based on
measured irregularity data ‘Low Level’ and ‘High Level’ according to ORE B176 [119] are
often used in specifications for simulations in continental Europe.
 Vehicle System Dynamics 1059

Other interesting disturbances are

• voided sleepers,
• irregular sleeper pitches,
• varying rail support stiffness (e.g. failure of rail fastening) and
• track transitions before and after bridges and tunnels.

Track dynamics is a very wide field, and in the last 150 years, a great effort has been made to
characterise the dynamic behaviour of the track. The level of complexity of the track model
varies depending on the phenomena to be studied. It is not intended to cover here track
dynamics and train–track interaction models which are specific to the study of track vibration
problems [118,120,121]. This section focuses on answering two basic questions: how is track
flexibility included in MBS, and when is it important to consider track flexibility in a vehicle
dynamic study.

4.1. How is the track flexibility included in MBS?

The flexibility of the track might be included in MBS software in different ways. Usually,
a track model with moving equivalent mass is used in multi-body simulations. This moving
equivalent mass represents the sleeper and a part of the track oscillating together with the
wheelset. This simple model of moving elastic track contains stiffness and damping (linear
or nonlinear) in the vertical and lateral directions. Default values are often used, for example,
the ones suggested for ballasted track in [119,122] (Table 1).
This approach is available in several MBS codes. Figure 37 shows the Discrete Track
Model option in SIMPACK. In the default model, each wheelset rests on a rigid body. This
rigid body represents the equivalent mass and inertia of the track section under each wheelset
(for ballasted track, the sum of sleeper mass and ballast mass). Roll, vertical and lateral
displacements of the track rigid body are allowed. The track is connected to the ground by
spring-damper elements acting in the vertical and lateral directions. This track model could
be easily modified in order to represent additional elastic layers of the track and also include
springs and dampers connecting masses under each wheel or wheelset. A critical issue is
then to find the correct values of these parameters to represent adequately the actual track
behaviour.
This simple track model is adequate for simulating vehicle–track interaction and vehicle
dynamics for frequencies up to about 20 Hz, but it is not suited to study other effects such as
forces in individual track components, effects of wheel flats [123], local variations in track
support and broken rail [124].
A more detailed modelling option for the track consists of adding to the vehicle model,
one by one, each rail and possibly sleepers as flexible bodies. The flexibility of these bodies
is given by a previous modal analysis. This might be useful for investigations considering

Table 1. Measured values of a ballasted track for a simple model of

moving elastic track containing stiffness and damping, published in
[119].

Unit Value

Lateral stiffness (per rail) MN/m 20

Vertical stiffness (per rail) MN/m 75
Relative lateral damping (per rail) – 0.3
Relative vertical damping (per rail) – (0.1) · · · 0.3
1060 S. Bruni et al.

Figure 37. SIMPACK discrete track model (default configuration).

higher frequencies. However, as the frequency increases, with no doubt, it is wise to simplify
the vehicle model (just the unsprung mass) as the dynamic behaviour is decoupled by the
soft secondary suspension. Consequently, as frequency increases, approaches other than MBS
models are more efficient.
The question then is obviously which track model has to be used when studying a particular
problem. Simple models are preferred, as they avoid numerical complications and limit the
number of parameters which, in fact, are either not known or difficult to obtain. A moving
equivalent mass track model with parameters obtained usually from the literature is a widely
used option. Due to lack of track dynamic properties knowledge, a simple rigid track model
is also used for analyses in the frequency range up to 20 Hz.
Németh and Schleinzer [125] compared typical track models used in MBS tools and con-
cluded that there is no difference between wheel displacements or wheel–rail forces among the
chosen track models up to 40 Hz. Alfi et al. [126] confirmed that the spectrum of wheel–rail
contact forces, up to 20 Hz, is not affected by the introduction of track flexibility. However,
they reported changes in the vehicle stability calculations with the three different track mod-
els used (a rigid track, a 3 DoF moving track model and a FE model of the track): use of a
rigid track gives an overestimation up to more than 10% of the vehicle critical speed and an
overestimation of 20% in the hunting frequency.
Table 2 recalls three typical deformation shapes and their occurrence frequencies (F1, F2
and F3) of a ballasted track on sleepers. This is the result of calculating frequency response
functions or measuring frequency response functions of the track. Clearly, if the phenomena to
be studied are below 20 Hz, it is unlikely that the flexibility of the track will affect the results.
Additionally, as F2 is much higher than F1, it is more important to include the flexibility of the
ballast/sleeper system than the stiffness of the rail pad itself which will affect the phenomena
occurring at higher frequencies.
An interesting comparison of moving track models, and correlation with experiments, was
reported by Chaar and Berg [127] (Figure 38). The model A does not consider the track
flexibility under the sleepers in contrast to models B and C. Model B does not consider the
vertical flexibility at the rail. Up to 80 Hz, the magnitude of the vertical receptance from model
A agrees rather well, and models B and C agree with the measurements up to 120 Hz. Above
 Vehicle System Dynamics 1061

Table 2. Typical deformation shapes and their occurrence frequencies (F1, F2 and F3) of a ballasted track on
sleepers.

F1: around 100 Hz. Rail and sleepers on phase vibration over ballast

F2: around 300–500 Hz. The rail is vibrating over rail pads

F3: pinned–pinned frequency, 700–1100 Hz. The rail is vibrating

with a wavelength which value is twice the sleeper spacing

Figure 38. Moving track models A, B and C (from [127]).

120 Hz, model C agrees with the measurements far better than models A and B. The calculated
lateral receptances agree very well with measurements up to nearly 200 Hz.

5. Assembly, checking and validation of vehicle models

5.1. Introduction

An adequate representation of a real vehicle–track system through a multi-body model requires

an experienced specialist, a proven computer tool able to represent the phenomena under
1062 S. Bruni et al.

investigation and an application of appropriate modelling technique. The work has to be

carried out according to quality requirements to avoid mistakes and errors. Nevertheless,
every model needs to be checked and tested before being used for the assessment of vehicle
performance. The plausibility check can help to identify errors or mistakes at an early project
stage; however, it does not provide satisfactory approval of a vehicle model. The validation
of the final model by comparison with measurements should demonstrate that the model is a
sufficient representation of the investigated system for the intended range of simulations and
that it is ‘fit for purpose’.

5.2. Model assembly and checking

During the assembly of a vehicle model, the substructures, which can be used with different
models, are compiled into a single model and the model parameters are adapted to the specific
case under consideration (e.g. empty or loaded vehicle).
The testing and checking of the assembled model should be carried out with all model
versions, including different vehicle loading variants. The change of vehicle payload does not
only lead to another car body mass and moment of inertia. The suspension displacements due
to loading may lead to variation of stiffness and damping parameters in coupling elements.
The suspension displacements due to loading lead to changed positions of CoG and of the
end points of various components. While many of the parameter changes can possibly be
neglected, if they are very small and do not change the direction of the suspension component
action, some of them are very important, for example, the consideration of changed plays in
the bump stops or pre-loads in the vertical suspensions. Other effects such as stiffening of
rubber elements or air springs can be either included in the modelling or represented by new
parameter values, which means another model.
The calculation of static equilibrium, that is, the computation of nominal forces, either can
be repeated with a new parameter set or the model can be ‘set down’ by numerical simulation
to a new position of bodies and CoGs. The second variant allows the changes of all position
parameters to be omitted; however, it is not correct, if the stiffening of components such as
rubber springs and changes of play in the bump stops are not implemented in the component
models used.
Model checking can be carried out by
• plausibility check and
• comparison with simulation results of similar vehicles.
The comparisons with results of similar vehicles can help to identify and eliminate errors and
mistakes. However, they could support incorrect results or raise doubts about correct results
if the differences between the compared similar models are underestimated.
A set of a few simulations representing all possible applications (eigenvalue calculation, runs
on an ideal straight track, straight track with irregularities, ideal curved track, curved track with
irregularities, etc.) can be used to test the vehicle model. The plausibility of simulation results
can be assessed by checks used to prove the signals measured during the tests on a real vehicle.

5.3. Model validation

The experience with validation of railway vehicle multi-body models differs significantly
between countries, institutions and dynamics specialists. The publications on validation of
railway vehicle models usually deal with a particular simulation tool, vehicle type and appli-
cation case without assessing and generalising the validation process. The survey done by
 Vehicle System Dynamics 1063

Cooperrider and Law [128] in 1978 deals with testing for validation and theoretical rail vehi-
cle dynamic analyses in the advent of modern computer simulation techniques. Meanwhile,
computer simulations are now widely being used in the design of railway rolling stock and in
research studies, but the progress of validation methodologies is rather limited. From recent
publications on this topic, [2,129] can be mentioned. The review paper by Evans and Berg
[1] dealing with simulation of rail vehicle dynamics provides some input regarding the model
validation too. In the present paper, the validation procedure is related mainly to the applica-
tion of simulations for the acceptance of running characteristics of railway vehicles according
to EN 14363 [10].
Significant experience with validation of railway vehicle models suited for the assessment
of running characteristics has been gained in the past in the UK, where there is a long history of
simulation as part of the acceptance process. The model validation is described in Appendix
G of the draft Railway Group Standard [130]. Although this draft has never been issued,
the principles described in this document are used by dynamics specialists in the UK. The
validation of a vehicle dynamics simulation model is mentioned in the draft of the standard
prEN 15827 [131], which is related to bogies and running gear. The recently issued revision of
UIC 518 [9] specifies conditions for application of simulations for vehicle acceptance instead
of tests. The revision of EN 14363, currently in preparation, should introduce the possibility
of using simulations of railway vehicle dynamics as a part of the vehicle acceptance process,
whereby the starting point is the document UIC 518 [9].
The validation of vehicle models is commonly carried out by comparisons with measure-
ments. The measurement error, scatter and the deviation of repeated measurements, however,
lead to many questions. The best agreement with a particular measurement does not mean that
the model correctly represents the statistical behaviour of the investigated system. Generally,
there are two validation approaches; the validation can be realised either by checking the
correctness of the physical relationships in each model component or a signal-based verifi-
cation can be used by comparing the input and output data (i.e. using a ‘black box’ model),
respectively. The verification of suspension and coupling component modelling is a suitable
part of the validation process; however, the validation of component modelling does not mean
that the complete model is validated. Contrarily, the use of the black box model can lead to
the risk of a good agreement for the complete model by chance. This kind of validation may
be insufficient for railway vehicles because of a limited number of measurements and a large
number of unknown or uncertain parameters.
The best practice of model validation in railway applications is a combination of physical
and signal-based verification. Each particular model is validated by a synthesis of stationary
tests, low-speed tests and on-track measurements. The comparisons between simulations and
measurements should be carried out for various conditions (static, quasi-static, dynamic) and
for several signals (wheel–rail forces, accelerations, displacements, etc.). If simulations are
to be used for a vehicle in different conditions (e.g. tare, laden and inflated and deflated air
suspension), separate models will need to be validated for each condition.
In the UIC 518 revision from 2009 [9] as well as in the revision of EN 14363, now under
preparation, the validation of the model relies upon the static, slow-speed and on-track tests
as specified in EN 14363 for testing of running characteristics.
The static and slow-speed tests are intended to be used to validate different aspects of the
vehicle model, namely

• wheel loads and load distribution,

• behaviour on twisted track,
• yaw moment during bogie rotation and
• sway or roll coefficient.
1064 S. Bruni et al.

Simulations of the static and slow-speed tests should be carried out and the output values
should be compared with the test results. For example, for the measurements of wheel loads
and load distribution, the following values should be compared:

• load on each individual wheel,

• load on each axle (sum of two wheels),
• load on each bogie (sum of wheels) and
• load on each side of the vehicle (sum of wheels on that side).

For the main part of the proposed assessment comparisons, however, there is no limit required
for a successful validation. Consequently, the validation may mature to a very subjective
assessment. Even the few criteria specified in [9] do not necessarily ensure a well-validated
model. For example, the limit prescribed for the comparison of individual wheel loads related
to the appropriated test results is 15%. The acceptance of such large deviations between the
simulation and measurement was introduced because of the wheel load variation between suc-
cessive measurements of the same vehicle, particularly, for vehicles with friction damping like
freight wagons. Fulfilment of this modest criterion, however, cannot guarantee a satisfactory
representation of the actual real vehicle.
Another stationary test suited for model validation but not mentioned in EN 14363 and
UIC 518 could be the modal analysis. A simple identification of the eigenfrequencies and
eigenmodes of the vehicle car body is carried out by analysing the vehicle reaction on a
single excitation in a so-called wedge test, rolling with a walking speed over 15–25-mm-high
wedges. This test can provide a useful validation input for modes with rather low damping. It
is well suited for conventional vehicles but more difficult to apply to articulated vehicles. An
advanced modal analysis is only rarely used as it requires a special test device.
Other kinds of tests could be useful too, for example, a test on a special track section with
extreme track irregularities to allow an extrapolation of the vehicle assessment. Such tests
would, however, need an adaptation of existing test centres or building of new test centres.
The questionnaire carried out in the frame of the project DynoTrain identified the following
stationary tests as the most useful for model validation:

• Measurement of static wheel loads.

• Wheel unloading test on a test rig or twisted track.
• Measurement of guiding force in a flat curve with a radius of 150 m as specified in the test
of safety against derailment according to Method 2 in EN 14363 [10].

The on-track tests represent another relevant part of model validation exercises. The dynamic
behaviour can differ from the static and slow-speed behaviour due to the effect of dynamic
stiffening and due to the contact geometry in the coupling between wheel and rail. Simulation
of on-track test and comparison with measured values, therefore, represent an indispensable
part of model validation.
The revision of UIC 518 from 2009 [9] as well as the revision of EN 14363 (in prepara-
tion) proposes the validation of vehicle models using the on-track tests specified for testing of
running characteristics in the same documents. The approval of running characteristics of rail-
way vehicles according to EN 14363 [10] yields the maximum estimated values of measured
quantities in each of the four test zones (straight, curves with large radius, curves with small
radius and curves with very small radius). These maximum estimated values are based on the
statistical assessment (either one- or two-dimensional) of the measurement results in the spec-
ified number of sections from the relevant zone, considering normal (Gaussian) distribution
of measured quantities. The maximum estimated values together with the results of statistical
analysis as provided in the reports for vehicle acceptance according to EN 14363 are, however,
 Vehicle System Dynamics 1065

not well suited for validation. For comparison between simulation and measurement, the data
from single sections are required together with a detailed knowledge of boundary conditions
such as track irregularity, track gauge, rail profiles and inclination and weather conditions
(friction coefficient between wheel and rail).
The results of statistical analysis can be used for model validation too. The two-dimensional
analysis as a function of the lateral acceleration at track level as used today is less suited for
validation, because the effect of lateral acceleration on the measurement results is rather
weak and ambiguous. The dependency of measured quantities on the curve radius is more
significant, at least for the curves with very small radius, and should preferably be used for
the two-dimensional statistical analysis intended for model validation.
The validation of the simulation model by comparison of simulations with the on-track test
results as described in UIC 518 [9] and in the latest draft of the revision of EN 14363 includes
the following parameters:

(1) assessment quantities according to UIC 518 or EN 14363, respectively (section values,
mean, standard deviation and estimated maximum as appropriate);
(2) power spectral densities (PSDs) and key frequencies of the following measurement
quantities:
• vehicle body lateral and vertical accelerations at each end,
• vehicle body bounce and pitch accelerations (derived from in- and out-of-phase values
of body end vertical accelerations),
• calculated vehicle body lateral and yaw accelerations (derived from in- and out-of-phase
values of body end lateral accelerations),
• bogie lateral and yaw accelerations,
• bogie vertical and pitch accelerations and
• sum of guiding forces (key frequencies);
(3) distribution plot of values for lateral and vertical wheel–rail forces as function of curve
radius or cant deficiency, respectively;
(4) sample time histories over straight and curved track sections for all the measurement
quantities.

Unfortunately, there is no specification of allowable differences between simulation and

test results for a well-validated model, so that the comparison of time histories results in a
subjective assessment.
The experience with model validation using the acceptance tests according to EN 14363
is rather limited. Two trials were initiated by UIC during the preparation of the revision of
UIC 518. The experience with one of them is published in [2]. As the measured rail profiles
were not available, the validation started with the theoretical, design rail profile. However,
no satisfactory agreement could be achieved. An application of a worn rail profile measured
previously at another location than at the investigated test line improved the agreement between
the simulations and measurements and confirmed the importance of the boundary condition
knowledge for a successful validation.
The actual track irregularities and rail profile data along the line during the on-track tests,
which represent important input data, are very often missing. The track irregularity data are
nowadays being recorded regularly; however, the data are often either not available outside
of the companies handling the infrastructure maintenance or measured using a chord system
which is not well suited for simulations. Even more problematic is the knowledge of rail
profile shapes. A continuous measurement of rail profile shapes has recently been introduced
in some countries; however, the measurements are only used for calculation of equivalent
conicity or of the radial steering index according to UIC 518 [9], which both represent
1066 S. Bruni et al.

insufficient information regarding the validation of multi-body railway vehicle models. The
availability and accessibility of measured track irregularities and measured rail profiles are
crucial for successful model validation and for the use of simulations partly replacing physical
testing.

6. Challenges and future trends

Based on the state-of-the-art examination proposed above, some challenges and trends for
future research are foreseen.
Modelling of rubber suspension components: the examination of the state of the art shows that
there are not many references dealing with the modelling of rubber for railway suspension
components, although rubber springs and bushes are widely used in rail vehicles and their
behaviour is critical to vehicle dynamics. It is believed that this is a field where more research
is needed in the future, to define models able to correctly capture the hysteretic behaviour
of these materials, considering, for example, frequency- and amplitude-dependent behaviour
at the same time and possibly also the influence of temperature and other parameters. Also,
the modelling of rubber-to-steel contact and the effect of friction and wear is a subject that
deserves further research.
Multi-physics modelling of suspension components: passenger train suspensions increas-
ingly tend to be composed of several interacting subsystems, including mechanical,
pneumatic/hydraulic and electronic ones. Active and semi-active suspensions are an obvi-
ous example of this trend, but the increased popularity of pneumatic suspensions represents
another example of the same tendency. This calls for multi-physics approaches, incorpo-
rating a mathematical description of all the different physical phenomena influencing the
force–deformation characteristics in the suspension.
At the same time, the trend in vehicle modelling is increasingly tending towards the definition
of one single numerical model able to handle different types of analyses such as running
dynamics, comfort and gauging rather than using different ‘specialised’ models for each task.
This means that using simple, equivalent mechanical suspension models is likely to become
less attractive than in the past, since these models might turn out to be not suited to represent
the multifaceted suspension behaviour in a variety of operating conditions.
Hence, increased focus towards multi-physics models of suspension components is expected
in the future, posing new challenges to consider, for example, the effect of high-frequency
dynamics in the control and actuation system and the specificities of the validation process
for an active vehicle model.
Interaction between testing and modelling of components: as discussed in Section 3 of this
paper, many suspension component models rely on the identification of some key model
parameters from specific tests performed on the single component. For instance, air springs
require the definition of the bellows volume and effective area as a function of height, which
is often measured to ensure the required accuracy. Similarly, the parameters of an oil damper
model can often be defined in a more accurate and reliable way through the direct measurement
of the force–deformation curves rather than from a detailed physical model of the damper.
However, the definition of the complete vehicle model shall not entail a large experimental
work for component testing to take the best advantage from the use of simulation instead of
vehicle testing. Hence, a best deal has to be sought between the proven accuracy of the vehicle
model and the effort and time required to set up the model. This also entails the relationship
between the validation of single-component models and the validation of the whole-vehicle
 Vehicle System Dynamics 1067

model, as treated in Section 5 of this paper, and how the accuracy of a vehicle model can be
reliably quantified.
Effect of suspension parameter dispersion: one of the most important perceived advantages
of extending the scope of vehicle dynamics simulations, especially in the framework of the
vehicle homologation process, is the possibility of considering the effect of non-deterministic
dispersion of parameters between nominally identical suspension components, which can
be expected on account of the variability implied by the component-manufacturing process.
Hence, not only the ‘average’ behaviour of a vehicle but also the maximum deviation from
the average shall be quantified to ensure that all vehicles in the fleet will meet performance,
safety and comfort requirements.
This issue can hardly be handled by means of physical testing, since for obvious reasons,
only a very limited number of vehicles will be tested, but it can be solved by the proper use
of numerical simulation. Indeed, by the use of appropriate mathematical and statistical tools,
it is possible to quantify the dispersion of typical vehicle running dynamics indicators such
as safety indexes and comfort indexes based on the assumed scatter of suspension parameters
in the fleet described in statistical terms. To this end, the use of Monte Carlo methods [3],
FORM/SORM methods and the generalised polynomial chaos theory [132,133] has been
proposed, but a well-established procedure is not available yet, and more research work and
alternative approaches to treat the problem are likely to appear in the near future.
Payload modelling: to date, the effect of payload in rail vehicle models has been considered
as mainly affecting the mass properties of the car body. Recent work has shown, however, that
the passengers possess damping properties, which shall be considered in dynamic analysis,
especially when ride comfort is at stake.
For freight vehicles, the modelling of liquid in a tank wagon is also a challenging problem,
which has been studied quite extensively in road vehicles and has also been approached for
railway vehicles, but surely deserves a more comprehensive investigation.

7. Conclusions

This paper has reviewed existing models for various types of railway vehicle suspension
components, showing the key role that they play in the building of the overall multi-body
vehicle model. Considering the increased accuracy of wheel–rail contact models which has
been achieved after decades of model refinements, it is likely that future research shall also
look at improved models of suspension components to allow for further improvements in the
accuracy of rail vehicle models.
Although reliable models are now well established for nearly all suspension components in
a rail vehicle, the availability of modelling approaches having different levels of complexity,
accuracy and ease of use calls for an increased awareness of the implications that choices
made at the suspension component modelling level have on the accuracy and trustworthiness
of the overall rail vehicle model.
Furthermore, no matter how accurate the suspension component model is, it will produce
correct results only when fed by an appropriate set of parameter values. Hence, the development
of new suspension component models shall be considered together with the definition of
appropriate (theoretical and experimental) methods to derive the parameter data for use in the
simulations. For most components considered in this paper, the experimental approach based
on laboratory tests presently represents the preferred option, but in the future, extended use
of refined modelling and simulation techniques can be anticipated to reduce the experimental
effort and provide improved theoretical background to the data-definition process.
1068 S. Bruni et al.

This paper has also recognised the track flexibility model as an essential component of the
overall vehicle model. Various track models are available, with different levels of complexity
from simple lumped parameter ones to a combination of multi-body and finite elements. The
usefulness of each approach depends on the frequency range of interest. Simple track models
work well for frequencies up to, at least, 20 Hz, but to extend the frequency range of the
analysis, a more detailed track model is needed.
Finally, model checking and validation represent key steps of the overall model-building
process. To date, no clear specification of a validation process is available in the standards
in force: although there is probably a wide consensus on the good practices required to build
and verify a railway vehicle model, a formal definition of the model verification process is
still lacking, as well as the quantitative definition of the expected matching error of the model
with respect to line test measurements, so that model validation remains a matter of subjective
assessment. It is expected, however, that the work being performed in the DynoTrain project
and in working groups attending to the revision of railway standards will in the future achieve
a clearer and more objective definition of the model validation process.

Acknowledgements
This paper describes the work partly undertaken in the context of the DynoTrain project, Railway Vehicle Dynamics
and Track Interactions: Total Regulatory Acceptance for the Interoperable Network (www.triotrain.eu). DynoTrain is a
collaborative project – medium-scale focused research project supported by the European 7th Framework Programme,
contract number: 234079.

References

[1] J. Evans and M. Berg, Challenges in simulation of rail vehicle dynamics, Veh. Syst. Dyn. 47 (2009), pp. 1023–
1048.
[2] L.-O. Jönsson, N. Nilstam, and I. Persson, Using simulations for approval of railway vehicles: A comparison
between measured and simulated track forces, Veh. Syst. Dyn. 46 (Suppl.) (2008), pp. 869–881.
[3] L. Mazzola, Possibilities and challenges for the virtual homologation of the railway rolling stock. Ph.D. thesis,
Politecnico di Milano, 2009.
[4] J. Piotrowski and H. Chollet, Wheel–rail contact models for vehicle system dynamics including multi-point
contact, Veh. Syst. Dyn. 43 (2005), pp. 455–483.
[5] B.M. Eickhoff, J.R. Evans, and A.J. Minnis, A review of modelling methods for railway vehicle suspension
components, Veh. Syst. Dyn. 24(6–7) (1995), pp. 469–496.
[6] O. Polach, M. Berg, and S. Iwnicki, 2006, Simulation, in Handbook of Railway Vehicle Dynamics, S. Iwnicki,
ed., Chapter 12, CRC Taylor & Francis Group, London. pp. 359–421.
[7] J.C.O. Nielsen, R. Lundén, A. Johansson, and T. Vernersson, Train–track interaction and mechanisms of
irregular wear on wheel and rail surfaces, Veh. Syst. Dyn. 40(1) (2003), pp. 3–54.
[8] O. Polach, 2008, Simulations of running dynamics in bogie design and development, Eur. Railw. Rev. 14(3),
pp. 76–81.
[9] UIC Code 518: Testing and approval of railway vehicles from the point of view of their dynamic behaviour –
Safety – Track fatigue – Running behaviour, International Union of Railways, September 2009.
[10] EN 14363 Railway Applications – Testing for the Acceptance of Running Characteristics of Railway Vehicles
– Testing of Running Behaviour and Stationary Tests, CEN, Brussels, 2005.
[11] C. Weidemann, State-of-the-Art Railway Vehicle Design with Multi-Body Simulation, J. Mech. Syst. Transp.
Logist. 3(1) (2010), pp. 12–26.
[12] W.H. Wittrick, On elastic wave propagation in helical springs, Int. J. Mech. Sci. 8 (1966), pp. 25–47.
[13] J. Lee and D.J. Thompson, Dynamic stiffness formulation, free vibration and wave motion of helical springs,
J. Sound Vib. 239(2) (2001), pp. 297–320.
[14] EN 13906-1:2008. Cylindrical Helical Springs made from Round Wire and Bars. Calculation and Design.
Part 1: Compression Springs.
[15] S.P. Timoshenko and J.M. Gere, Theory of Elastic Stability, McGraw-Hill, New York, 1961.
[16] J.A. Haringx, On highly compressible helical springs and rubber rods, and their application for vibration –
free mountings, Report 4/1949, Philips Res. Lab., Eindhoven.
[17] O. Krettek and M. Sobczak, Zur Berechnung der Quer- und Biegekennung von Schraubenfedern für
Schienenfahrzeuge, ZEV-GlasersAnnalen 112 No, 9 September, Aachen, 1988, pp. 321–326.
 Vehicle System Dynamics 1069

[18] S. Nishizawa, M. Ikeda, J. Logsdon, H. Enomoto, N. Sato, and T. Hamano, The effect of rubber seats on coil
spring force line, Paper Number: 2002-01-0317, SAE, 2002, doi:10.4271/2002-01-0317.
[19] C.M. Wang, J.N. Reddy, and K.H. Lee, Shear Deformable Beams and Plates: Relationships with Classical
Solutions, Elsevier, Oxford, UK, 2000, ISBN: 978-0-08-043784-2.
[20] C. Weidemann and S. Mulski, Modelling Helical Springs, SIMPACK News 11, 1st issue, July 2007.
[21] S. Iwnicki (ed.), Handbook of Railway Vehicle Dynamics, CRC Press, Boca Raton, FL, 2006, ISBN: 978-0-
8493-3321-7.
[22] F. Xia, Modelling of wedge dampers in the presence of two-dimensional friction, Veh. Syst. Dyn. 35 (2003),
pp. 755–762.
[23] A.B. Kaiser, J.P. Cusumano, and J.F. Gardner, Modelling and dynamics of friction wedge dampers in railroad
trucks, Veh. Syst. Dyn. 38(1) (2002), pp. 55–82.
[24] N. Bosso, A. Gugliotta, and A. Somà, Multibody Simulation of Freight Bogie with Friction Dampers, Proc. of
the 2002 ASME/IEEE Joint Rail Conference, Washington DC, April 23–25, 2002, pp. 47–56.
[25] J. Piotrowski, Model of the UIC link suspension for freight wagons, Arch. Appl. Mech. 73 (2003), pp. 517–532.
[26] J. Piotrowski, Smoothing dry friction damping by dither generated in rolling contact of wheel and rail and its
influence on ride dynamics of freight wagons, Veh. Syst. Dyn. 48(6) (2010), pp. 675–703.
[27] X. Tan and J. Rogers, Equivalent viscous damping models of Coulomb friction in multi-degree-of-freedom
vibration systems, J. Sound Vib. 185(1) (1995), pp. 33–50.
[28] F. Xia, Modelling of a two-dimensional Coulomb friction oscillator, J. Sound Vib. 265 (2003),
pp. 1063–1074.
[29] R.F. Harder, Dynamic Modelling and Simulation of Three-Piece North American Freight Vehicle Suspensions
with Non-linear Frictional Behaviour Using ADAMS/Rail, 5th ADAMS/Rail User’s Conference, Haarlem,
May 10–12, 2000.
[30] Yu. Boronenko, A. Orlova, and E. Rudakova, Influence of construction schemes and parameters, Veh. Syst.
Dyn. 44(1) (2006), pp. 402–414.
[31] A. Berghuvud, Curving performance and nonlinear dynamic behaviour of freight cars with three-piece bogies,
Ph.D. thesis, Luleå tekniska universitet, 2001.
[32] Y.Q. Sun and S. Simson, Nonlinear three-dimensional wagon-track model for the investigation of rail
corrugation initiation on curved track, Veh. Syst. Dyn. 45(2) (2007), pp. 113–132.
[33] L.H. Ren, G. Shen, and Y.S. Hu, A test-rig for measuring three-piece bogie dynamic parameters applied to a
freight car application, Veh. Syst. Dyn. (44)1 (2006), pp. 853–861.
[34] A. Orlova and Yu. Romenb, Refining the wedge friction damper of three-piece freight bogies, Veh. Syst. Dyn.
46(1) (2008), pp. 445–455.
[35] F. Xia and H. True, On the Dynamics of the Three-piece-freight truck, Proceedings of the 2003 IEEOASME
Joint Rail Conference, 22–24 April, 2003, Chicago, IL.
[36] J.R Evans and P.J. Rogers, Validation of dynamic simulations of rail vehicles with friction damped Y25 bogies,
Veh. Syst. Dyn. 28 (1998), pp. 219–233.
[37] T. Jendel, Dynamic analysis of a freight wagon with modified Y25 bogies, Master thesis, TRITA-FKT Report
1997:48, ISSN1103-470X, Stockholm, 1997.
[38] J. Keudel, Modellierung des Güterwagendrehgestells Y25 für das Mehrkörperdynamikprogramm MEDYNA,
Master thesis, TU Berlin, 2002.
[39] S. Stichel, On freight wagon dynamics and track deterioration, Proc. Inst. Mech. Eng. F, J. Rail Rapid Transit
(UK) 213(F4) (1999), pp. 243–254.
[40] ORE, Flexibility of trapezoidal springs, ORE B12 Rp 25, 2nd ed., Utrecht, 1986.
[41] ORE, Parabolic springs for wagons (design, calculation, treatment), ORE B12 Rp 43, Utrecht, 1988.
[42] ORE B56/Rp2, Improvement of the running stability of existing RIV wagons required to run under any loading
conditions at speeds of 80 km/h, ORE, Utrecht, March, 1967.
[43] ORE DT30, Etude de la stabilité transversale d’un véhicule ferroviaire à deux essieux, par M. R. Joly, Technical
Document, ORE, Utrecht, January, 1974.
[44] J.B. Ayasse et al. Computer simulation of freight vehicles with leaf springs: A comparison between different
packages, Tech. Rep., INRETS, INRETS/RE-01-046-FR, 2001.
[45] P.-A. Jönsson, Modelling and laboratory investigations on freight wagon link suspension, with respect
to vehicle-track dynamic interaction, Licentiate thesis, TRITA AVE 2004:48, KTH Railway Technology,
Stockholm, 2004.
[46] P.-A. Jönsson, E. Anderson, and S. Stichel, Experimental and theoretical analysis of freight wagon link
suspension, Proc. Inst. Mech. Eng. F, J. Rail Rapid Transit (UK) 220(F4) (2006), pp. 361–372.
[47] P.-A. Jönsson, E. Anderson, and S. Stichel, Influence of link suspension characteristics variation on two-axle
freight wagon dynamics, Veh. Syst. Dyn. 44(Suppl.) (2006), pp. 415–423.
[48] P.-A. Jönsson, S. Stichel, and I. Persson, New simulation model for freight wagons with UIC link suspension,
Veh. Syst. Dyn. 46(Suppl.) (2008), pp. 695–704.
[49] M. Hoffmann, Dynamics of European two-axle freight wagons, PhD thesis, DTU Lyngby. IMM-PhD:ISBN
87-643-0128-1, 2006.
[50] H. Lange, Dynamic analysis of a freight car with standard UIC single-axle running gear, TRITA – FKT Rep.
1996:34, KTH Railway Technology, Stockholm, 1996.
[51] M. Stiepel and S. Zeipel, Freight wagon running gears with leaf spring and ring suspension, Paper presented
at the SIMPACK user group meeting, 2004.
1070 S. Bruni et al.

[52] D. Cebon, Simulation of the response of leaf springs to broad band random excitation, Veh. Syst. Dyn. 15(1986),
pp. 375–392.
[53] P.S. Fancher, R.D. Ervin, C.C. MacAdam, and C.B. Winkler, Measurements and representation of the
mechanical properties of truck leaf springs, SAE-P 800905, 1980.
[54] J. Piotrowski, On application of the rolling contact theory for modelling of the UIC link suspension for freight
wagons, Internal Report, Warzaw University of Technology, Institute of Vehicles, Zeszyty Naukowe Institutu
Pojazdow, 3, 2003.
[55] Z.Y. Shen, J.A. Hedrick, and J.A. Elkins, A comparison of an alternative creep force model for rail vehicle
dynamic analysis, Veh. Syst. Dyn. 12 (1983), pp. 70–82.
[56] P. Belforte, F. Cheli, G. Diana, and S. Melzi, Numerical and experimental approach for the evaluation of severe
longitudinal dynamics of heavy freight trains, Veh. Syst. Dyn. 46(Suppl. 1) (2008), pp. 937–955.
[57] F. Cheli and S. Melzi, Rheological model of a side buffer for freight trains, 7th EUROMECH Solid Mechanics
Conference, Lisbon, Portugal, 2009.
[58] F. Cheli and S. Melzi, Experimental characterization and modelling of a side buffer for freight trains, Proc.
Inst. Mech. Eng. F, J. Rail Rapid Transit (UK) 337(6) (2010), pp. 535–546.
[59] H.N. Onah and L. Garibaldi, Viscoelastic Material Damping Technology, Becchis Osiride, Torino, 1996.
[60] W.J. Mortel and D. Weston, The development of rubber suspension components on rail vehicles, Paper
C478/10/033, Seminar on Railway Division of IMechE, UK, 24–26 May, 1994.
[61] P.K. Freakley and A.R. Payne, Theory and Practice of Engineering with Rubber, Applied Science Publishers
Ltd, London, 1978.
[62] J.D. Ferry, Viscoelastic Properties of Polymers, 3rd ed., Wiley, New York, 1980.
[63] J. Nicolin and T. Dellmann, Über die modellhafte Nachbildung der dynamischen Eigenschaften einer
Gummifeder, ZEV-Glas. Ann. 109 (1985), Nr. 4.
[64] P.E. Austrell, Mechanical Properties of Carbon Black Filled Rubber Vulcanizates, Proc. of Rubbercon ’95,
Göteborg, Sweden, 10–12 May, 1995.
[65] S. Bruni and A. Collina, Modelling the viscoelastic behaviour of elastomeric components: An application to
the simulation of train-track interaction, Veh. Syst. Dyn. 34 (2000), pp. 283–301.
[66] H. Peeken and S. Lambertz, Nichtlineares Materialmodell zur Beschreibung des Spannungs-Dehnungs
verhaltens gefüllter Kautschuk-Vulkanisate, Konstruktion 46, 9–15, Springer-Verlag, 1994.
[67] P. Pfeffer and K. Hofer, Einfaches nichtlineares Modell für Elastomer- und Hydrolager zur Optimierung der
Gesamtfahrzeug-Simulation, ATZ 104(5) (2002), pp. 442–451.
[68] M. Berg, A rubber spring model for dynamic analysis of rail vehicles, TRITA – FKT Report 1995:53, KTH
Railway Technology, Stockholm, 1995.
[69] M. Berg, A model for rubber springs in the dynamic analysis of rail vehicles, Proc. Inst. Mech. Eng. F 211
(1997), pp. 95–108.
[70] M. Berg, A non-linear rubber spring model for rail vehicle dynamics analysis, Veh. Syst. Dyn. 30 (1998),
pp. 197–212.
[71] P. Allen, A. Hameed, and H. Goyder, Measurement and modelling of carbon black filled natural rubber
components, Int. J. Veh. Syst. Model. Test. 2(4) (2007), pp. 315–344.
[72] N. Oda and S. Nishimura, Vibration of air suspension bogies and their design, Bull. JSME 13(55) (1970),
pp. 43–50.
[73] M. Presthus, Derivation of air spring model parameters for train simulation, Master’s Thesis, Luleå University
of Technology, DeltaRail Group, Derby, UK, 2002.
[74] Vampire, User’s Manual Version 5.03, 2009.
[75] M. Berg, A three-dimensional airspring model with friction and orifice damping, Veh. Syst. Dyn. 33(Suppl.)
(1999), pp. 528–539.
[76] G. Quaglia and M. Sorli, Air suspension dimensionless analysis and design procedure, Veh. Syst. Dyn. 35(6)
(2001), pp. 443–475.
[77] N. Docquier, P. Fisette, and H. Jeanmart, Multiphysic modelling of railway vehicles equipped with pneumatic
suspensions, Veh. Syst. Dyn. 45(6) (2007), pp. 505–524.
[78] A.J. Nieto, A.L. Morales, A. Gonzalez, J.M. Chicharro, and P. Pintado, An analytical model of pneumatic
suspensions based on an experimental characterization, J. Sound Vib. 313 (2008), pp. 290–307.
[79] C. Weidemann, Air-springs in Simpack, Simpack News, 2003, pp. 10–11.Available at http://www.simpack.com/
fileadmin/simpack/doc/newsletter/2003/sn-3-03-air-springs.pdf.
[80] N. Doquier, P. Fisette, and H. Jeanmart, Model based evaluation of railway pneumatic suspension, Veh. Syst.
Dyn. 46(Supp.) (2008), pp. 481–493.
[81] N. Doquier, Multiphysics modelling of multibody systems – application to railway pneumatic suspensions,
Ph.D. thesis, Université Catholique de Louvain, November, 2010.
[82] N. Docquier, P. Fisette, and H. Jeanmart, Influence of heat transfer on railway pneumatic suspension dynamics,
21st International Symposium on Dynamics of Vehicles on Roads and Tracks (IAVSD’09), Stockholm, Sweden,
17–21 August, 2009.
[83] A. Facchinetti, L. Mazzola, S. Alfi, and S. Bruni, Mathematical modelling of railway vehicles with secondary
airspring suspension: Safety and ride comfort, Veh. Syst. Dyn. 48(1) (2010), pp. 429–449.
[84] A. Alonso, J.G. Giménez, J. Nieto, and J. Vinolas, Air suspension characterization. Increasing air suspension
effectiveness by means of a variable area orifice, Veh. Syst. Dyn. 48(1) (2010), pp. 271–286.
[85] O. Qing and S. Yin, The non-linear mechanical properties of an airspring, Mech. Syst. Signal Process. 17(3)
(2003), pp. 705–711.
 Vehicle System Dynamics 1071

[86] S. Duym, R. Stiens, and K. Reybrouck, Evaluation of shock absorber models, Veh. Syst. Dyn. 27(2) (1997),
pp. 109–127.
[87] X. Carrera, J. Vinolas, J. Savall, and J. Biera, A parametric damper model validated on a track, Int. J. Heavy
Veh. Syst. 13(3) (2006), pp. 145–162.
[88] H.H. Lang, A Study of the Characteristics of Automotive Hydraulic Dampers at High Stroking Frequencies,
Mechanical Engineering, The University of Michigan, Ann Arbor, 1977, p. 288.
[89] A. Alonso and J. Giménez, Damper Modelling and its Implementation in Railway Simulation Program,
Proceedings of the Euromech 500 Colloquium, Ed. Springer, 2010, pp. 123–135.
[90] R. Van Kasteel, W. Cheng-Guo, Q. Lixin, L. Jin-Zhao, and Y. Guo-Hong, A new shock absorber model for use
in vehicle dynamics studies, Veh. Syst. Dyn. 43 (2005), pp. 613–631.
[91] A. Conde Mellado, E. Gomez, and J. Vinolas, Advances on railway yaw damper characterisation exposed to
small displacements, Int. J. Heavy Veh. Syst. 13(4) (2006), pp. 263–280.
[92] M. Wrang, Instability phenomena of a passenger coach, caused by internal damper flexibility, Veh. Syst. Dyn.
33 (Suppl.) (2000), pp. 406–417.
[93] S. Bruni, P. Belforte, G. Mancini, A. Cera, H. Hartwig, and L. Mazzola, Experimental investigation of yaw
damper performances: An improved and harmonised testing methodology developed within ModTrain EU
project, WCRR’08 International Conference, Seoul (KR), 18–22 May, 2008.
[94] S. Alfi, S. Bruni, and L. Mazzola, Impact of Suspension Component Modelling on the Accuracy of Rail Vehicle
Dynamics Simulation, VSDIA, Budapest, Hungary, 2008.
[95] S. Bruni, R. Goodall, T.X. Mei, and H. Tsunashima, Control and monitoring for railway vehicle dynamics,
Veh. Syst. Dyn. 45(7) (2007), pp. 743–779.
[96] J.S. Tang, Passive and semi-active airspring suspension for rail passenger vehicle – theory and practice, Proc.
Inst. Mech. Eng. F, J. Rail Rapid Transit (UK) 210 (1996), pp. 103–117.
[97] Y. Sugahara, A. Kazato, R. Koganei, M. Sampei, and S. Nakaura, Suppression of vertical bending and rigid-
body-mode vibration in railway vehicle carbody by primary and secondary suspension control: Results of
simulations and running tests using Shinkansen vehicle, Proc. Inst. Mech. Eng. F, J. Rail Rapid Transit (UK)
223 (2009), pp. 517–531.
[98] S. Alfi, S. Bruni, G. Diana, A. Facchinetti, and L. Mazzola, Active control of airspring secondary suspension
to improve ride quality and safety against crosswind, Proc. IMechE Part F: J. Rail and Rapid Transit 225(F1)
(2011), pp. 84–98.
[99] M. Sorli, W. Franco, S. Mauro, G. Quaglia, R. Giuzio, and G. Vernillo, Features of the lateral active pneumatic
suspensions in the ETR 470 high speed train, 6th UK Mechatronics Forum International Conference Skovde,
Sweden, 9–11 September, 1998.
[100] A. Conde Mellado, C. Casanueva, J. Vinolas, and J.G. Giménez, A lateral active suspension for conventional
railway bogies, Veh. Syst. Dyn. 47(1) (2008), pp. 1–14.
[101] F. Cheli, G. Diana, and F. Resta, Numerical model of a tilting body railway vehicle compared with rig and on
track tests, Veh. Syst. Dyn. 35(6) (2001), pp. 417–442.
[102] E. Foo and R.M. Goodall, Active suspension control of flexible-bodied railway vehicles using electro-hydraulic
and electro-magnetic actuators, Control Eng. Pract. 8 (2000), pp. 507–518.
[103] J.T. Pearson, R.M. Goodall, T.X. Mei, and G. Himmelstein, Active stability control strategies for a high speed
bogie, Control Eng. Pract. 12 (2004), pp. 1381–1391.
[104] A. Pacchioni, R.M. Goodall, and S. Bruni, Active suspension for a two-axle railway vehicle, Veh. Syst. Dyn.
48(1) (2010), pp. 105–120.
[105] F. Resta and S. Bruni, Active control of railway vehicles to avoid hunting instability, IEEE/ASME Int.
Conference on Advanced Intelligent Mechatronics, Como, Italy, 8–12 July, 2001.
[106] F. Braghin, S. Bruni, and F. Resta, Active yaw damper for the improvement of railway vehicle stability and
curving performances: Simulations and experimental results, Veh. Syst. Dyn. 44(11) (2006), pp. 857–869.
[107] Y.K. Lau and W.H. Liao, Design and analysis of magnetorheological dampers for train suspension, Proc. Inst.
Mech. Eng. F, J. Rail Rapid Transit (UK) 219 (2005), pp. 261–276.
[108] D.H. Wang and W.H. Liao, Semi-active suspension systems for railway vehicles using magnetorheological
dampers. Part I: System integration and modelling, Veh. Syst. Dyn. 47(11) (2009), pp. 1305–1325.
[109] S.N. Fergusson, R.D. Fröhling, and H. Klopper, Minimising wheel wear by optimising the primary suspension
stiffness and centre plate friction of self-steering bogies, Veh. Syst. Dyn. 46(Suppl.) (2008), pp. 457–468.
[110] J.R. Evans, The Modelling of Railway Passenger Vehicles, the Dynamics of Vehicles on Roads and Tracks,
Proceedings of the 12th IAVSD Symposium held in Lyon, France, August 26–30, 1991, Swets&Zeitlinger,
Amsterdam/Lisse, 1992, pp. 144–156.
[111] O. Polach, Curving and stability optimisation of locomotive bogies using interconnected wheelsets, Veh. Syst.
Dyn. 41 (2004), pp. 53–62.
[112] S. Alfi, L. Mazzola, and S. Bruni, Effect of motor connection on the critical speed of high-speed railway
vehicles, Veh. Syst. Dyn., 46(Suppl.) (2008), pp. 201–214.
[113] O. Polach, Influence of locomotive tractive effort on the forces between wheel and rail, Veh. Syst. Dyn.
35(Suppl.) (2001), pp. 7–22.
[114] O. Polach, Creep forces in simulations of traction vehicles running on adhesion limit, Wear 258 (2005),
pp. 992–1000.
[115] H. Kovtun, O. Markova, V. Litvin, W. Kik, and H. Scheffel, Brake rig influence on vehicle dynamics, Proc.
Inst. Mech. Eng. F, J. Rail Rapid Transit (UK) 224 (2010), pp. 558–566.
1072 S. Bruni et al.

[116] EN 13674-1:2003+A1:2007: ‘Railway Applications. Track. Rail: Part 1: Vignole railway rails 46 kg/m and
above’.
[117] Y. Bezin, S.D. Iwnicki, and M. Cavalletti, The effect of dynamic rail roll on the wheel–rail contact conditions,
Veh. Syst. Dyn. 46(Suppl.) (2008), pp. 107–117.
[118] K. Popp, H. Kruse, and I. Kaiser, Vehicle–track dynamics in the mid-frequency range, Veh. Syst. Dyn. 31
(1999), pp. 423–464.
[119] ORE B 176, Bogies with steered or steering wheelsets, Report No. 1: Specifications and preliminary studies,
Vol. 2, Specification for a bogie with improved curving characteristics, ORE, Utrecht, 1989.
[120] K.L. Knothe and S.L. Grassie, Modelling of railway track and vehicle/track interaction at high frequencies,
Veh. Syst. Dyn. 22(3–4) (1993), pp. 209–262.
[121] A. Igeland and J. Oscarsson, Modelling of Railway Track for Computer Simulation of Dynamic Train/track
Interaction, Proceedings XVth IMAC (International Modal Analysis Conference), Tokyo, September, 1997,
pp. 464–470.
[122] SIMPACK, Wheel Rail Element Reference, Release v.8.904. SIMPACK AG, 31st August, 2010.
[123] M. Seco, E. Sanchez, and J. Vinolas, A time domain wheelflat/track FE model, IET International Confer-
ence on Railway Condition Monitoring, 29–30 November 2006, Institution of Engineering and Technology,
Birmingham, UK, 2006, pp. 130–135, ISBN: 0-86341-732-9.
[124] F.J. Gonzalez, B. Suarez, J. Paulin, and I. Fernandez, Safety assessment of underground vehicles passing over
highly resilient straight track in the presence of a broken rail, J. Rail Rapid Transit 222 (2008), pp. 69–84.
[125] I. Németh and G. Schleinzer, Assessment of railway track models from dynamical aspects, 7th International
Conference on Railway Bogie and running Gears, Budapest, 3–7 September 2007, pp. 133–142.
[126] S. Alfi, S. Bruni, and E. Di Gialleonardo, The Influence of Track Modelling Options on the Simulation of Train-
Track Interaction, Proceedings of the Tenth International Conference on Computational Structures Technology,
Civil-Comp Press, Stirlingshire, UK, Paper 12, 2010, doi: 10.4203/ccp.93.12.
[127] N. Chaar and M. Berg, Simulation of vehicle–track interaction with flexible wheelsets, moving track models
and field tests, Veh. Syst. Dyn. 44(Suppl.) (2006), pp. 921–931.
[128] N.K. Cooperrider and E.H. Law, A survey of rail vehicle testing for validation of theoretical dynamic analyses,
J. Dyn. Syst. Meas. Control 100 (1978), pp. 238–251.
[129] F. Cheli, R. Corradi, G. Diana, and A. Facchinetti, Validation of a numerical model for the simulation of
tramcar vehicle dynamics by means of comparison with experimental data, J. Comput. Nonlinear Dyn. 2
(2007), pp. 299–307.
[130] GM/RT2141 Resistance of Railway Vehicles to Derailment and Roll-Over, Issue Three Draft 1m, December,
2004.
[131] prEN 15827 Railway applications – Bogies and running gear. Railway applications – Bogies and running
gear, Draft, September, 2009.
[132] A. Sandu, C. Sandu, and M. Ahmadian, Modeling multibody systems with uncertainties. Part I: Theoretical
and computational aspects, Multibody Syst. Dyn. 15 (2006), pp. 373–395.
[133] C. Sandu, A. Sandu, and M. Ahmadian, Modeling multibody systems with uncertainties. Part II: Numerical
applications, Multibody Syst. Dyn. 15 (2006), pp. 241–262.