Vehicle System Dynamics

International Journal of Vehicle Mechanics and Mobility

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

A Three–Dimensional Airspring Model with Friction

and Orifice Damping

Mats Berg

To cite this article: Mats Berg (1999) A Three–Dimensional Airspring Model with
Friction and Orifice Damping, Vehicle System Dynamics, 33:sup1, 528-539, DOI:
10.1080/00423114.1999.12063109

To link to this article: https://doi.org/10.1080/00423114.1999.12063109

Published online: 31 May 2019.

Submit your article to this journal

Article views: 4

Citing articles: 5 View citing articles

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

https://www.tandfonline.com/action/journalInformation?journalCode=nvsd20
Vehicle System Dynamics Supplement 33 (1999), pp. 528-539 © Swets & Zeitlinger

A Three-Dimensional Airspring Model

with Friction and Orifice Damping

MATS BERG

SUMMARY
This paper proposes a non-linear airspring model for general three~irnensional motions. The non-linearity
stems from friction behaviour of rubber pans and from the air flow in the pipe between airbag and surge
reservoir, in particular if the surge pipe is equipped with orifice damping. The model is based on superposition
of elastic, friction and viscous forces. For a given preload the model has 6 and 7 parameters for horizontal
and vertical motions, respectively. Comparsions between model and measurement results show good
agreement for a wide range of cases. The model represents a reasonable compomise between accuracy and
computational effort and should be a suitable tool in rail vehicle dynamics analysis.

I. INTRODUCTION

The airspring, see an example in Figure 1, is an important suspension component in

many rail and road vehicles. It provides vertical and horizontal suspension, low and
level-controlled height, and increased stiffnesses with increased payload. Some
airsprings also provide orifice damping. Trends that will make further demands on
airsprings are wider ranges of payloads (preloads), higher vehicle speeds, higher cant
deficiencies (higher speeds in curves), and improved ride comfort and noise levels.
An improved understanding of the dynamic behaviour of airsprings is therefore
needed. This promotes airspring product development and modelling of airsprings,
which in turn improve vehicle dynamics performance respectively the potential of
vehicle dynamics simulations. Examples of airspring models are given in e.g. [1]-[4).
Here new features, related below, are introduced based on the author's work [5]-[7].

I. Carl>ody interface
2. Bogie interface
3. Air bag
8 4. Additional spring
5. Surge pipe
6. Orifice
7. Surge reservoir
8. Levelling valve

·····~7
·····~
9.
10.
Air container
Compressor

Fig. I. Airspring system.

 THREE-DIMENSIONAL AIRSPRING MODEL 529

2. PROPOSED MODEL

The present paper proposes a three-dimensional airspring model to be used in vehicle

dynamics simulations. With its "medium" complexity it is believed to fill a gap
between very simple linear models, e.g. [1), and highly complex thermodynamic
models, e.g. [4]. The airspring model is partly based on the author's work on models
for conventional rubber springs [5][6]. Like in that work elastic, friction and viscous
effects contribute to the spring forces. Here the airspring preload is also a vital
parameter.
In contrast to [ 1] and [3] the present model is three-dimensional and not only
vertical (axial). Among [1]-[4], friction is only modelled in [2] and only in the lateral
direction. A superposition of elastic, friction and viscous effects for both horizontal
and vertical motions is only used in the present paper. For the surge pipe air flow the
velocity exponent is also introduced as a parameter.
The present airspring model should be representative up to at least 20 Hz of
excitation frequency.

2.1. Basic theory

The proposed airspring model has 12 basic degrees of freedom: three translations and
three rotations at the top (subscript t) and bottom (b) surface respectively. These
translations and rotations, along with their corresponding forces and moments, are
shown in Figure 2 below.

z
Preload P
h

Fig. 2. Basic quantities of the airspring model. Nominal configuration.

For static equilibrium, with the vehicle on tangent and plane track, all displacements
are set to zero and the only non-zero forces are caused by the static compressive
= =
preload P giving Fz 1 P and F zb -P. Thus it is assumed that the nominal axial
direction (Z) of the airspring is vertical. This almost always applies to rail vehicles.
The nominal height of the airspring is denoted by h. For an airspring with levelling
valve this height is independent of the preload.
530 M.BERG

It is now assumed that inertial effects of the airspring are negligible. In most vehicle
dynamics applications this in not a major restriction since the frequency range of
interest is typically 0-20 Hz. (Note, however, the special mass effect introduced later).
The following six equilibrium conditions should therefore hold during general
airspring motions
Fx 1 + Fxb = 0 (la)

Fy 1 + Fyb =0 (lb)

Fzt + Fzb = 0 (lc)

Mt/1, + Mt/lb + Fy1 • (h - z, + zb) + Fzt · (y, - Yb) =0 (ld)

Mx, + Mxb- Fx1 · (h - z, + zb) - Fv · (x 1 - xb) =0 (le)

MI/J, + MI/Jb = 0 (lf)

where it assumed that the X-, Y- and Z-axes are principal axes of the airspring. The
moment equations (ld) and (le) refer to the bottom of the spring. Note that a deformed
configuration is considered in these equations since the translations of the airspring
ends are present. The relative vertical displacement -z, + zb (change of airspring
height) is often less than, say, 0.05h and is therefore neglected in the following.
However, the influence of the relative displacements x 1 - xb and y 1 - Yb• in
combination with the large compressive force Fzt• is significant and gives important
moment contributions.
In the airspring model the forces and moments of Equation ( 1) should now be
related to airspring motions and the preload P. It is sufficient, and appropriate, to
formulate these relations for the forces and moments of one of the airspring ends, and
then use Equation (1) to determine the forces and moments of the other end. Here the
airspring top forces and moments are selected for the formulation.
It is here assumed that the top forces and moments can be expressed as
~=~+~+~ ~

Fyr = Feyt + FlY,+ Fvyt (2b)

Fzt = Fezt + Fftt + Fvzt (2c)

Mt/1, = Me;r (2d)

Mx, = Mex, (2e)

~=~ ~
with elastic (subscript e), friction (j) and viscous (v) contributions. This force
superpostion has been suggested in the rubber spring work of the author [5][6]. Here
the preload effect is included implicitly in the elastic contribution, see Section 2.2.
The three contributions are derived in Sections 2.2-2.4. In these sections the
6+ 7=13 model parameters, for any given preload, of Figure 3 will be introduced and
explained.
 THREE-DIMENSIONAL AIRSPRING MODEL 531

(a) (b)

2M

Fig. 3. Principle airspring model and model parameters. (a) Horizontal motion, six parameters; (b) Vertical
motion, seven parameters. See Sections 2.2-2.4 for further explanations.

2.2. Elastic behaviour

For the simple case of linear elastic behaviour the forces and moments of the airspring
top are obtained as, see [7],
Ftxt = Fzr · Xb + Kex · (x, - xb + hXb) + Kexx · (X, - Xb) (3a)

Feyt = - Fzt · <Pb + Key · (y,- Yb- h</Jb) + Key.; · (</Jr- </Jb) (3b)
Fezt = P + Kez · (z, - zb) (3c)

Me,, = Key.; · (y, - Yb - h</Jb) + Ke, · (</J 1 - </Jb) (3d)

Mex 1 = Kexx · (x, - xb + hXb) + Kex · (X 1 - Xb) (3e)

Me"'' = Ke"' · (1/Jt- 1/Jb) (3f)

where the preload effect also is included explicitly by the first term in Equations
(3a)-(3c). The (small) airspring inclinations Xb and ¢b give with the preload a
destabilizing effect. In Equations (3a) and (3b) the total vertical force Fz1 is used, for
completeness, instead of only P or Feu·
In Equation (3) eight different elastic stiffnesses are introduced. If for instance
x 1 ¢0 and the other displacements are zero, i.e. a longitudinal shear motion, the
stiffness Kex relates x 1 to Fext by the simple equation Fext = Kex· x 1• The elastic
stiffnesses Key• Kez• Ke;• Kez and Ke., can be explained analogously.
The remaining two stiffnesses, Kexx and Key.;• couple the longitudinal force/
translation to the pitch rotation/moment, respectively the lateral force/translation to
the roll rotation/moment. For instance, the case above with x 1 .e0 and the other
displacements equal to zero, also requires a moment Mez, = Kax · x 1•
Now assuming axisymmetric airsprings with respect to the Z-axis, which is the
most common situation, the number of different elastic stiffnesses can be reduced
=
from eight to jive since Key= Ka. Ke; Kez and Key;= -Kexx (note the minus sign).
In addition, the five stiffnesses can often be reduced to only three since the
=
moments related to Ke; Kez and Ke., can often be neglected as compared to the roll
and yaw moments transferred by an airspring pair (left and right) through vertical and
longitudinal airspring forces respectively.
532 M.BERG

Thus the linear elastic airspring behaviour can be represented by the three elastic
stiffnesses Kex• Kexx and Kez.• cf. Figure 3. The elastic contributions to the airspring
forces and moments are most pronounced for motions of large amplitudes at low
frequency. The vertical elastic stiffness K ez. depends on the so called air bag stiffness
and reservoir stiffness and, optionally, also on the so called change of area stiffness
and the vertical stiffness of an additional spring in series. Note that all three stiffnesses
depend upon the preload, see Section 2.5.

2.3. Friction behaviour

When studying harmonic excitation of any of the airspring degrees of freedom, while
the other are constrained to zero, Equation (3) suggests that any resulting force/
moment describes a straight line in a corresponding force/moment versus
displacement graph. However, not even for low excitation frequency, say f :s; 0.1 Hz
giving negligible viscous effects, this straight line can be identified. This especially
applies to horizontal motion, for which the air bag rubber and a possible skirt
contribute. The additional rubber spring also contributes. One possible effect is
non-linear elasticity, another is friction hysteresis.
The friction model part of the present airspring model gives increased stiffnesses
for small displacement amplitudes and a frequency independent (rate-independent)
hysteresis. This also means that the airspring model becomes non-linear. A
one-dimensional friction model developed by the author [5][6], and mainly intended
to describe the friction behaviour of ordinary rubber springs, forms the basis for the
current airspring friction modelling.

2.3.1. Vertical motion

Starting with vertical (one-dimensional) motion, the developed friction model [5](6]
can be adopted directly. In contrast to a simple Coulomb element with a series spring,
the present friction model can represent the smoother friction behaviour seen in rubber
spring measurements. As input data the maximum friction force is given along with
a friction displacement parameter that describes the smoothness.
For a monotonical increase in relative vertical displacement z = z, - zb of the
airspring, the vertical friction force F ft.t in Equation (2c) follows the equation
F: = _z_F: (4)
ft.t Z2 +z ft.,max

where Fft..max and .z 2 are the two input parameters implied above, cf. Figure 3b. For
= =
z z2 half the maximum friction force is reached, i.e. Fft.t Fft..max /2. It can be shown
that the initial tangent stiffness due to friction equals Fft. .max fz 2• The tangent stiffness
then decreases and tends to zero for z ~ z 2. However, when z is about to decrease, the
tangent stiffness again equals Fft..max f z 2 and then approaches zero as the friction force
tends to -Fft..max· A somewhat different equation than (4) applies here. The
one-dimensional friction behaviour is examplified in Figure 5b for lateral motion. For
more details, see [5][6].
 THREE-DIMENSIONAL AIRSPRING MODEL 533

2.3.2. Horizontal motion

As mentioned above, the frictional airspring behaviour is usually most pronounced for
horizontal motion. In principle the horizontal friction forces Ffxt and Ffyr• cf.
Equations (2a-b), can be determined by the one-dimensional friction model in
Section 2.3.1 with one model in longitudinal direction and one in lateral direction.
However, denoting the maximum friction force in these directions by Ffx,max, the
maximum friction force for any radial motion differing from the longitudinal and
lateral motions would exceed the value Ffx,max. This is clearly unphysical for an
axisymmetric airspring, which by definition has no "preferred" radial direction.
As a consequence, the non-linear friction behaviour has to be represented by a true
two-dimensional (horizontal) friction model. Still, for a one-dimensional motion
along any straight line in the horizontal plane such a model should predict the same
friction force history as the one-dimensional model would do. The one-dimensional
friction model has also been verified by measurements.
The goal of formulating a two-dimensional friction model is to relate the horizontal
friction forces Ffxt and FIY1 to general, horizontal airspring motion. This horizontal
motion is here represented by the relative longitudinal motion x = x1 - xb + hXb
and the relative lateral motion y = y1 - Yb - h¢b· However, the friction forces Ffxt
and Ffyr do not only depend on the instantaneous displacements x and y but also on
the histories of these forces and displacements.
In Figure 4 the basis for the formulation of the present new two-dimensional
friction model is illustrated. This incremental formulation raises the question: Given
the latest reference state (x,, y,, Ffxr.r, Ffyr,r> and the second latest reference state
(Xrr.Yrr• Ffxt,rr• Ffyt,rr), what will the forces (Ffxt' Ffy1 ) be for the next given
displacements (x,y) ?
Longitudinal
displacement
FtrY, = ?
X

x,

0~--~----~------------+-----~
Lateral
0 Yrr Yr y displacement
fig. 4. Basic principle for determination of horizontal friction force vector Ffryt = ( fjn, fh.,) based on the
two latest reference states. Length I and angle 0.

Note that the force vectors are generally not parallel to the motion directions.
However, the basic idea of the present two-dimensional friction model is that the force
direction should accomodate to the present motion direction. Thus if a certain motion
direction prevails, the force direction should converge towards that direction.
534 M.BERG

Moreover, the force magnitude should converge towards the maximum friction force.
These convergences are described by a friction displacement denoted x 2• Together
with the maximum friction force in the horizontal plane, Ffx,max• the displacement x 2
define the two input parameters for the present two-<limensional friction model,
which is thus analogous to the one-dimensional model parameters. Cf. Figure 3.
=
To determine the force vector Ffxyr (Ffxr• Ffy1) for any displacements (x,y) the
procedure is as follows:
a) Determine the length land the angle (J (change of motion direction).
b) Among the four angles -90, 0, 90 and 180°, choose the two closest to the current 0.
c) Determine the force vectors for the two chosen angles and the current l.
d) Determine the current force vector Ffxyt for the angle (J by linear interpolation.
The question is now how the force vectors for the angles -90, 0, 90 and 180° are
determined. As Figure 4 is drawn, 0 < (} :;; 90° and the angles 0 and 90° are thus
closest. The principle determination of the force vectors for these two angles is now
considered, cf. part c) above.
For the angle 0° the motion trend defined by the two reference states rr and r
prevails. As implied above the force component perpendicular to the prevailing
motion direction should converge towards zero whereas the the force component
parallel to the motion direction should tend to Ffx,max. For the perpendicular force
component only the state r is used, but for the parallel force component both states r
and rr are utilized. (At the first step of the displacement excitation only state r is
available; then the one-dimensonal friction model is used to calculate the force).
For the angle 90° the motion direction defined by the two reference states rr and
r is interrupted by a 90° tum to the right. Only the reference state r is used here to
determine the force components perpendicular and parallel to latest motion direction.
For more details on the horizontal friction force model, see [7]. Here the
two-dimensional friction behaviour is examplifed in Figure Sa for five idealized
displacement excitation~. All excitations start with a 20 mm displacement along the
Y-axis. Then an additional displacement of 20 mm takes place in five different
directions. The friction force response is shown by force vectors along the different
displacement paths.
For case (1) the displacement continues along the Y-axis and the friction force
tends to the maximum for9e, as the one-dimensional model in [5][6] would do. For
case (3) the displacement makes a 90° tum. First the force direction is far from the
motion direction, but after 20 mm the force vector is almost parallel to the new motion
direction and the magnitude almost that of (1).
In case (2) the tum is 45°. The force vector after one step is the average of the
pertinent force vectors for (1) and (3). As in (3), the force vector in (2) then tends to
the motion direction and increases its magnitude. Case (5) is a 180° tum with a force
vector parallel to the motion direction. But according to Figure 5b it takes about 7.5
mm. from y = 20 to 12.5 mm, until the force vector is directed in the new motion
direction. As for (1) the results of (5) are identical to those of the one-dimensional
model. Case (4) makes a 135° tum and, like in (2), the first step gives an average force
vector of the corresponding forces of (3) and (5).
 THREE-DIMENSIONAL AIRSPRING MODEL 535

Note that the angle (} introduced above equals zero for all displacements steps in
Figure Sa except for the first step from (x,y) =(0,20) in the directions of cases (2}-(5).
But of course the displacement excitation cases here are highly idealized. For a rail
vehicle in operation the horizontal motion of an airspring is more or less random. This
implies that the parameter e rarely equals zero, but that there is a change of motion
direction at virtually all time.
Force F. 1 (kN)
r-,~~~---'-~..:-.-~~-~~-. 0.6

0.4

0.2

~.2

r., 0.6 kN
~.4
I (a) (b)
~-~~~~~~~~~~~~~.6~-~-~-~--~-~-~
20 25 30 35 40 -5 0 5 10 15 20 25
Displacement y (mm) Displacement y (mm)
Fig. 5. (a) Horizontal friction force vs. horizontal displacement. Force response ofthe present friction model
for five displacement paths. Step length I= 5 mm. Model data: Fp;.max = 0.6 kN and x 2 = 5 mm. (b)
Friction force vs. displacement for case (5), i.e. for 20 mmy-displacement and back again (I= I mm).

2.4. Viscous behaviour

So far only preload and displacement dependent forces have been represented in the
airspring model, resulting in frequency independent response. But measurements,
especially vertical ones, show a strong frequency dependence. Thus a viscous model
part needs to be introduced allowing for rate-dependent hysteresis (damping) and
increasing stiffnesses with increasing excitation frequency.

2.4.1. Horizontal motion

In the horizontal viscous model the viscous forces Fvxt and Fvy1 , cf. Equations (2a-b ),
should be related to airspring velocities of general horizontal airspring motion. This
horizontal motion is here represented by relative longitudinal and lateral velocities
obtained by time differentiation of the displacements x andy, respectively, introduced
in Section 2.3.2.
The one--dimensional viscous model used in [5][6] is adopted here, one in
longitudinal direction and one in lateral direction. The one--dimensional model
consists of a linear viscous damper with a linear series spring. Due to the assumed
airspring axisymmetry the same parameters can be used for both motion directions.
Here they are denoted Cx and Kv:n cf. Figure 3a. For one-dimensional, harmonic
excitation such a model gives a maximum energy loss per cycle at the excitation
frequency w = Kvx!Cx. One "internal" first order differential equation is thus
introduced in both directions. In [2] the same viscous model has been used, but only
in lateral direction.
536 M.BERG

2.4.2. Vertical motion

Here the force Fvzr in Equation (2c) is to be expressed as a function of vertical airspring
motion. A reasonable start would be to adopt the one--dimesional viscous model
introduced above. However, from vertical measurements it can be shown that the
viscous behaviour is not linear due to interaction between air bag and reservoir
volumes. Usually, see e.g. [2] and [3]. the viscous force is instead assumed to be
proportional to the velocity squared (turbulent air flow in the surge pipe). The present
measurements indicate, cf. Chapter 3, that a velocity exponent somewhat Jess than 2
is more accurate. Therefore an exponent parameter {3, which may take any value
between I (linear) and 2 (quadratic), is introuced. To emphasize the relation to {3. the
viscous damping is denoted cljJ. with unit of the form force. (timeflength)P.
In addition, an airspring system without orifice damping (low damping) shows a
significant mass effect for excitations of small displament amplitudes. It must be a
mass effect since a stiffness lower than the static one is obtained in a certain frequency
span. The mass effect originates from the inertia of the air mass in the pipe. This air
mass is very small, usally Jess than 0.1 kg, but due to the high air acceleration the mass
effect cannot in general be neglected. In the present airspring model the mass effect
is represented by a mass M which gives a mass effect proportional to the relative
acceleration, cf. Figure 3b where 2M is subjected to half the relative acceleration.
The vertical viscous force, including the mass effect, can thus be expressed as
.,{3 . ..
I
Fvzt = Kvz(Z-Zv) = cz{J Zv sign(zv) + Mzv (5)
where z=z1- zb as before and where Zv is the displacement over the damper and mass.
A dot denotes time differentiation. The two corresponding first order differential
equations can be written as (q is the velocity over the damper and mass)
Zv =q (6a)
• CzfJ .B . Kvz
q = -Miqrsign(q) + M(Z-Zv) (6b)

2.5. Varying preload

Strictly speaking, the 13 model parameter values only apply to the current preload.
This is because most of them change with changing preload. In vehicle simulations
with various preloads (payloads) at least two sets of parameters for two different
preloads are needed, preferrably for no payload and maximum payload. For other
preloads the 13 input data can be obtained by linear interpolation. For three sets of
parameters, parabolic interpolation applies.
2.6. Computer implementation, linearization and other simplifications
For computer implementation of the model and verification against measurements
the software MATLAB [8] is used. For full-scale rail vehicle dynamics simulations,
using the present model, the software package GENSYS [9] is utilized.
For linear vehicle dynamics analysis the horizontal/vertical friction and vertical
viscous damping, if f3 ~ I, need to be linearized. If desired certain effects may be
omitted, e.g. by setting the maximum friction forces, the series spring stiffnesses or
the mass to zero. But linear elastic behaviour should always be represented.
 THREE-DIMENSIONAL AIRSPRING MODEL 537

3. COMPARISONS OF SIMULATED AND MEASURED RESULTS

Systematic airspring measurements in laboratory have been perfonned to verify the
present airspring model. The evaluation was carried out for different preloads and
horizontal and vertical motions. The (diaphragm) airspring was mainly excited
harmonically with different displacement amplitudes and frequencies.
The air bag and reservoir volumes of 35 and 100 litres, respectively, were
connected by a 2.2 m long surge pipe with inner diameter of 40 mm. Locally the
diameter could be reduced through orifice damping. A conical additional spring was
used in series, like in Figure 1. The airspring height was 234 mm.

3.1. Horizontal motion

A comparison of radial force response is shown in Figure 6 for a very low frequency
and two displacement amplitudes. The friction hysteresis is obvious here and the
agreement for x 0 = 12 mm is good. For x 0 = 45 mm, however, a degressive elastic
behaviour shows up in the measurements. This may call for an extension of the present
model to non-:-linear elastic behaviour during significant horizontal motion.

Force F:x1 (kN) Force Fxr (kN)

3 I Xo = 12 mml 10 l.xo=45mml
2
5

0 0

-1
-5
-2
(a) -10 (b)
-3
-10 -5 0 5 10 --40 -20 0 20 40
Displacement x (mm) Displacement x (mm)
Fig. 6. Force-displacement loops for radial harmonic excitations. P = 80 kN,f= 0.1 Hz. Calculation (solid
=
line) and measurement (dashed line). (a) x 0 12 nun, (b) x 0 = 45 mm. Model data: Ku = 0.220
kN/mm, Ff.r.III4X = 1.0 kN and x2 = 7 mm. K•xx• Cx and Kvx are not relevant here.

3.2. Vertical motion

A comparison of results for vertical stiffness is shown in Figure 7, both without and
=
with orifice damping. The stiffness is defined as Sz (Fz 1 - P)o /z 0 , or dynamic force
amplitude divided by displacement amplitude.
The inertial effect is pronounced for low amplitude excitation and no orifice
damping, giving a ratio between highest and lowest stiffness of as much as 10.
Non-linearities due to friction and viscous damping are obvious in Figure 7a. The
increase in amplitude of Figure 7a results in a behaviour similar to that of orifice
damping in Figure 7b. Stiffness increase for higher frequencies is mitigated through
the rather soft additional spring. The model results are in good agreement with the
measured ones.
538 M.BERG

Stiffness Sz (kN/mm) Stiffness Sz (kN/mm)

1.5 1.5
I
d=40mml ll<J =2 mm I

1.0 1.0

0.5 0.5

(a) (b)
0 0~~--~~--~~~~~~~
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
Frequency I (Hz) Frequency I (Hz)
Fig. 7. Vertical stiffness vs. frequency. P = 80 kN. Calculation (solid line) and measurement (dashed line).
(a) Without orifice damping: amplitudes 2 and 12 mm, (b) Amplitude 2 mm: without/with orifice
damping. Model data: K.. =0.250kN/mm, F/z..IMx =0.5 kN, Z2 = !Omm, Kvz=0.420kN/mm, C:P
=0.000065/0.000400 kN(slmm)P (no orifice I orifice diameter d =15 mm), p=1.8 and M =215 kg.
A corresponding comparison for vertical damping is shown in Figure 8. Here the
damping is defined as Dz = Ez I (( Fzt - P)0 • Z0 ), or energy dissipation per cycle
divided by the product of dynamic force amplitude and displacement amplitude. This
general damping measure can for linear systems be related to the force-displacement
=
phase angle 0 by the relation Dz :rc sin 6, cf. [5][6]. The case with Zo =2 mm and
=
d 40 mm gives, somewhat surprisingly, the highest damping values (above 3).
However, increase in displacement amplitude and/or orifice damping give higher
damping in the lower frequency range. As above the model results are in good
agreement with the measured ones.
Finally, in Figure 9 corresponding examples of force-displacement graphs are
shown. In Figure 9a the extreme stiffness ratio related above is indicated. Also here
the simulaton-measurement agreement is good.
Damping Dz (-) Damping Dz (-)
3.5 3.5
3.0
d=40mm I 3.0 I =2mm ltc I
2.5 2.5
2.0 2.0
1.5 1.5

1.0 1.0

4 6 8 4 6 8

Fig. 8. Vertical damping vs. frequency. P =80 kN. Calculation (solid line) and measurement (dashed line).
(a) Without orifice damping: amplitudes 2 and 12 mm. (b) Amplitude 2 mm: without/with orifice
damping. Model data according to Figure 7.
 THREE-DIMENSIONAL AIRSPRING MODEL 539

Force Fv (k.N) Force Fv (leN)

82 I d=40 mm I 82 I d= 15 mm I /=8Hz

81 81

80

79 79

78 (a) 78 (b)
0 0.5 1.0 1.5 2.0 -2.0 -1.5 -1.0 -0.5 0 0.5 1.0 1.5 2.0
Displacement z (mm) Disolacement z (mm)
Fig. 9. Force-displacement loops for vertical hannonicexcitations. P = 80 kN,/=4 and 8Hz. z0 ... 2 mrn.
Calculation (solid line) and measurement (dashed line). (a) Without orifice damping (d = 40 mm),
(b) With orifice damping (d = 15 mm). Model data according to Figure 7.

4. CONCLUSIONS

The model results are in general in good agreement with the measured ones. The
model parameter values can also be obtained by only a limited number of
measurements. In addition, the computational effort is moderate. It is therefore
concluded that the proposed non-linear three-dimensional airspring model should be
a suitable tool in the field of vehicle dynamics simulation.
If needed the model may be extended to account for non-linear elastic behaviour
during horiozontal motion. Then additional parameters are needed.

ACKNOWLEDGEMENTS
The author acknowledges the financial support of Adtranz Sweden and NUTEK (The Swedish National
Board for Industrial and Technical Development). Many thanks to Mr L Ohlsson and Mr N Nilstarn at Adtranz
sweden, Mr I Persson at DEsolver, and Prof. E Andersson, Royal Institute of Technology, for their support.

REFERENCES
1. Nishimura S, Kikuchi I, KidaJ, ShimaS and Yoshioka H: Fore-Aft SIU1Iiride, The Sumitomo Search, No.
17, pp. 144-150, May 1977.
2. AEA Technology Rail ; Introduction to Railway Vehicle Dynamics, Course, Derby, 2-4 November, 1998.
3. Homeyer A and M~mann C ; Mode me Verfahren zur Auslegung von Luftfedersystemen, ZEV+DET
Glas. Ann. 122, Nr. 9/10, 1998.
4. Grajnert J and Krettek 0; Zur Phlinomenologie und Ersarunodtllbildung von Luftfedem, ZEV+DET
Glas. Ann. I 15, Nr. 7/8, 1991.
s. Berg, M.: A model for rubber springs in the dynamic analysis of rail vehicles, Journal of Rail and Rapid
Transit, Institute of Mechanical Engineers, Vol. 211, Part F, pp. 95-108, 1997.
6. Berg, M.: A Non-Linear Rubber Spring Model for Rail Vehicle Dynamics Analysis, Journal of Vehicle
System Dynamics, Vol. 30, pp. 197-212, 1998.
7. Berg M: An airspring model for dynamic analysis of rail vehicles, TRITA-FKT Report 1999:32, Div. of
Railway Technology, Dep. of Vehicle Engineering, Royal Institute of Technology, Stockholm. 1999.
s. The Math Works, Inc.: MATLAB User's Manual, Natick, Massachussets, USA, 1998.
9. Persson. 1.: GENSYS in railway vehicle modeling, Journal of Vehicle System Dynamics, to appear.