ARTICLE IN PRESS

Control Engineering Practice 12 (2004) 13811391

Active stability control strategies for a high speed bogie

J.T. Pearsona,*, R.M. Goodalla, T.X. Meib, G. Himmelsteinc
a

Department of Electronic and Electrical Engineering, Loughborough University, Ashby Road, Loughborough, Leicestershire LE11 3TU, UK
b
School of Electronic and Electrical Engineering, University of Leeds, Leeds LS2 9JT, UK
c
Bombardier Transportation, Siegen, Germany
Received 31 March 2003; accepted 10 July 2003

Abstract
This paper presents a comparison of control algorithms for actively stabilised wheelsets on high speed railway vehicles. Both
intuitively derived classical strategies and modern optimal strategies are considered. Computer simulations are used to assess and
compare the performance of the strategies in terms of their ability to provide stability across a range of operating conditions, for a
range of parameter uncertainty and also in terms of their actuator requirements. Actuator models are also developed, not only to
properly assess the impact of actuator dynamics on the active stability system performance, but also to further quantify and rene
the actuator requirements needed to implement the system practically. The concept has been implemented practically on a full size
vehicle and some preliminary experimental results are included. Although the paper concentrates upon stability control, some
observations are included regarding the integration of steering controllers.
r 2003 Elsevier Ltd. All rights reserved.
Keywords: Active control; Railways; Kalman lter; Controllers; Feedback stabilisation

1. Introduction
Recently there has been increasing interest in the use
of active control technology within the primary suspensions of railway vehicles. The two fundamental issues at
the primary suspension level are steering and stability;
although this paper focuses upon active stability control,
some observations are included regarding the integration of steering control strategies. Different congurations are possible, for example directly steered wheel
pairs, actively steered independently rotating wheels,
and actively steered conventional axles (Aknin, Ayasse,
& Devallez, 1991; Anon Deutsche Bahn, 1997;
Gretzschel & Bose, 1999). Hedrick (1981) overviews
many of the theoretical active suspension studies
performed in 1970s.
A conventional wheelset for the railway vehicle is
composed of two coned (or proled) wheels rigidly xed
to a common axle. This arrangement has the advantages
*Corresponding author. Tel.: +44-150-922-7015; fax: +44-150-9227014.
E-mail addresses: [email protected], [email protected]
(J.T.
Pearson),
[email protected]
(R.M.
Goodall),
[email protected] (T.X. Mei).
0967-0661/$ - see front matter r 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/S0967-0661(03)00152-7

of natural curving, but when unconstrained it also

exhibits a sustained oscillation referred to as kinematic
oscillation of the wheelset or wheelset hunting
(Wickens, 1991). On conventional rail vehicles, the
wheelset is stabilised by using springs connected to the
bogie of the vehicle, however adding such stiffness is
known to adversely affect the natural curving of the
wheelset. The degree of proling of the wheels, known
as conicity, is also a critical factor because a high
value gives good curving performance but leads to a
higher level and frequency of instability, and it will be
seen that ensuring stability for a range of conicity values
is a key aspect of this study. For many years much effort
has been made to try and solve the difcult design tradeoff between the stability and curving performance of
railway vehicles. Vehicle parameters are carefully
designed such that the critical speed at which the vehicle
becomes unstable is higher than the maximum vehicle
design speed (Dukkipati & Guntur, 1984).
If passive solutions are used then a compromise has to
be made between steering and stability, however it has
been shown (Mei & Goodall, 2000) that the use of active
control concepts can solve this design conict. Active
control schemes can be developed to steer the wheelset
around curves (Mei, Perez, & Goodall, 2000), a number

ARTICLE IN PRESS
1382

J.T. Pearson et al. / Control Engineering Practice 12 (2004) 13811391

of steering strategies are possible for example to follow a

pure rolling line around the curve. This paper compares
a number of control algorithms for actively stabilised
wheelsets on a high speed railway vehicle, using
conventional solid-axle wheelsets. Although the paper
concentrates upon stability control, some observations
are included regarding the integration of steering
controllers. The next section describes the mathematical
model derived for the design of the control strategies,
which are described in Section 3. Consideration is given
to the effects of actuator dynamics on the performance
of the controllers, and this is outlined in Section 4.
Section 5 deals with practical sensors in the formulation
of the control algorithms. Wheelset mounted sensors are
undesirable for a number of reasons identied later and
so the approach adopted in this study was to develop a
model based estimator to provide the controllers with
the required feedback variables. Preliminary full-scale
experimental results are included, although the main
emphasis of the paper is on the formulation of practical
control strategies.

2. High speed bogie mathematical model

The vehicle consists of vehicle body and two bogie
frames, each bogie has two solid axle wheelsets. The
wheelsets are connected to the bogie frames via springs
and dampers in the longitudinal and lateral directions.
Since for the purposes of stability and steering the plan
view dynamics of the vehicle are of prime interest the
vertical and roll modes, and consequently the vertical
suspension, have not been modelled. Fig. 1 shows the
plan-view half-vehicle model used for this study, and the
parameters for the model are given in Appendix 1 and
are representative of a modern high-speed railway
vehicle. The vehicle model contains seven degrees-offreedom, i.e. lateral and yaw modes for each wheelset
and for the bogie frame, and a lateral mode for the
vehicle body (dened by Eqs. (1)(7)). The model is
therefore 14th order overall, and is a highly coupled

complex MIMO system. It can be represented in state

space form by Eq. (8) (Li & Goodall, 1998; Mei &
Goodall, 1999a).
An actuator is connected between each wheelset and
the bogie frame, in the yaw direction for the implementation of active control. This actuator can be used for
both steering and stability commands:



2f22
mw y. w1
Cs y w1 Ks yw1  2f22 yw1  Cs y g
Vs

2

Vs

 Ks yg  Cs Lv yg  Ks Lv yg mw
 gyc1 ;
1
R1
2f22 L2g
2f11 lLg
yw1
y w1
Vs
r0
2f11 L2g 2f11 lLg


yt1 Tw1 ;
R1
r0



2f22
mw y. w2
Cs y w2 Ks yw2  2f22 yw2  Cs y g
Vs

2

Vs

 Ks yg Cs Lv yg Ks Lv yg mw
 gyc2 ;
R2
Iw y. w1

2f22 L2g
2f11 lLg
yw2
y w2
Vs
r0
2f11 L2g 2f11 lLg


yt2 Tw2 ;
R2
r0

Iw y. w2

mg y. g 2Cs Csc y g 2Ks Ksc yg  Cs y w1

 Ks yw1  Cs y w2  Ks yw2  Csc y v  Ksc yv






1
1
yc1 yc2

mg Vs2

 mg g
;
2R1 2R2
2
2

Ig y. g 2L2v Cs y g 2L2v Cs yg  Lv Cs y w1 Lv Cs y w2
 Lv Ks yw1 Lv Ks yw2 Tw1 Tw2 ;
mv y. v Csc y v Ksc yv  Csc y g  Ksc yg






1
1
yc1 yc2

mv Vs2

 mv g
;
2R1 2R2
2
2
x A  x B  u G  w;

7
8

where
x y w1 yw1 y w1 yw1 y w2 yw2 y w2 yw2 y g yg y g yg y v yv T ;
w 1=R1 yc1 yt1 1=R2 yc2 yt2 T ;
u Tw1 Tw2 T :

Fig. 1. Plan view bogie model.

The input vector u represents the two control torques

(one actuator per wheelset), and the vector w is used to
dene the inputs from the railway track (curvature, cant
and lateral displacement). Track curvature and cant are
deterministic inputs associated with curves, whereas
the lateral track displacement is a random input, which

ARTICLE IN PRESS
J.T. Pearson et al. / Control Engineering Practice 12 (2004) 13811391
Table 1
Plan-view vehicle model eigenvalues
Eigenvalues

1383

3.1. Classical control strategy

Speed 200 km=h
Frequency
(Hz)

Percentage
damping

Body lateral

0.92
0.92

20.15
20.15

Kinematic 1

5.99
5.99

0.80
0.80

Kinematic 2

13.68
13.68

87.06
87.06

Bogie yaw and lateral

14.71
14.71
26.69
26.69

26.75
26.75
99.98
99.98

High frequency wheelset modes

65.83
65.83
72.84
72.84

31.67
31.67
22.20
22.20

represents the deviation of the track from its intended

alignment.
The main dynamic modes for the wheelset and bogie
lie in the range from 2 to 10 Hz; and there is also a body
mode at a little under 1 Hz (see Table 1).

A control-oriented analysis of an unconstrained single

wheelset reveals that it is possible to stabilise a solidaxle wheelset with a relatively simple feedback control
(Goodall & Li, 1999). Figs. 2 and 3 show a linearised
and simplied model of a single solid-axle wheelset,
where high frequency modes related with the mass and
moment of inertia are neglected. f11 and f22 are the creep
coefcients. yw and c represents wheelset lateral
displacement and yaw angle. L; r0 and l are the gauge,
radius and conicity of the wheelset. The two diagrams
given in the gures are for the same wheelset model,
except that they are drawn to illustrate the use of
different input and output variables.
It is clear that the local passive damping (either yaw
damping or lateral damping) does not assist with
stability and it merely affects the value of frequency
at which the instability takes place. However two types
of active damping will give stability for a solid-axle
wheelset. One approach is to apply a yaw torque Tw
proportional to the lateral velocity of the wheelset to
the wheelset as shown in dashed line in Fig. 2 (hence
termed active yaw damping). The other active damping
approach is given in dashed line in Fig. 3 and the
stabilisation is achieved by applying a lateral force Fy
proportional to the yaw velocity of the wheelset (hence
termed active lateral damping).
The classical control strategy used in this study is one
that stabilises the solid axles by applying a yaw torque to

3. Basic control strategy design and performance

Fy

Active stabilisation of a solid axle wheelset can be

realised by applying either a yaw torque or a lateral
force between the bogie frame and the wheelset (axle)
(Mei & Goodall, 2000). The approach used in this study
is to apply a yaw torque to the wheelset, an arrangement
that enables the stability control actuators also to be
used for steering purposes. In addition previous work
(Mei & Goodall, 2000) has shown that ride quality is
degraded when a lateral force is used. Two basic control
strategies have been studiedan intuitively derived
classical control strategy and an optimal control
strategy (Mei & Goodall, 1999b).
The primary design objective for the stability controllers is to provide at least 20% damping across all
modes at a vehicle speed of 65 m=s and with a conicity
of 0.3. Since railway vehicles are subject to parameter
variations, such as conicity deviating from its nominal
value during operation a secondary, but nonetheless
essential, design goal is to provide at least 5% damping
across all modes at a higher conicity of 0.5. It is
important that the control strategies designed are robust
against such uncertainties. Practical issues relating to
sensors and actuators are considered in later sections.

Tw

2 f22

2f11L s

yw

2f22s

sCs
2f11L /r0
Fig. 2. Linearised and simplied wheelset model (1).

Tw
Fy

v
2f22s

yw

2f11L /r0

v
2f11L2

sCs
2 f22
Fig. 3. Linearised and simplied wheelset model (2).

ARTICLE IN PRESS
J.T. Pearson et al. / Control Engineering Practice 12 (2004) 13811391

1384

.s+1

Lateral Velocity
Wheelset 1

Kyd

Control Torque
Wheelset 1

.
.s+1

Kyd

.s+1
.
.s+1

Lateral Velocity
Wheelset 2

Control Torque
Wheelset 2

Fig. 4. Classical control strategy structure.

each axle proportional to its lateral velocity (Eq. (9)):

Tw1 Kyd y w1 ;
Tw2 Kyd y w2 :

However, the best performance that can be achieved by

applying this basic strategy is a damping level of 11.9%
at 0.3 conicity, and it also fails to provide a stable
control solution at the higher conicity case. Adding
a phase lead term into the feedback loop, as shown in
Fig. 4 and in Eq. (10), improves the situation, and this
controller provides 20.9% damping at 0.3 conicity and
also 17.1% damping at the higher conicity case of 0.5.



ts 1
Tw1
Kyd y w1 ;
ats 1



ts 1
10
Tw2
Kyd y w2 :
ats 1
Kyd is the overall proportional gain, and t and a o1
are tuned to give an adequate phase lead, and Fig. 5
shows the variation with gain of the minimum damping
level, with and without the phase lead term included in
the feedback path.

The overall robustness of the nal controller is

assessed by varying the vehicle velocity from 20 to
85 m=s; and varying the conicity from 0.1 to 0.7. The
controller exceeded the two design objectives specied
previously, and also provided a minimum of 3%
damping over this entire robustness range (see Fig. 6).
Clearly, this is a very desirable controller solution,
yielding a controller that provides stability over a wide
operating range, and it does not require scheduling.
These results assume the wheelset lateral velocity can be
measured directly. (Section 5 considers the issue of
practical measurements and sensors, and the effect these
have on the controller formulation and performance.)
3.2. Optimal control strategy
An optimal control formulation for the active stability
controller was a natural choice since it inherently
provides a stable controller, and it produces a single
solution to several design goals. This latter capability is
particularly useful when integrating steering objectives
into the controller design. It is assumed at this initial
stage that the complete state vector is available for
feedback. (A state estimator is designed in Section 5 to
estimate the state variables that are impractical to
measure.) A general diagram of the control structure is
shown in Fig. 7.
When designing the optimal controller, it is not
necessary to minimise all the states. For stability control
it has been found that minimisation of the lateral
displacements or the yaw rates of the wheelsets results in

Optimisation - Classical Yaw Damping Strategy

25
20.8% with Phase Advance Filters

Minimum Damping Level (%)

20
15
11.9% GainTerm Only
10
5
0
-5
-10
-15
-20
105

106

Control Gain
Fig. 5. Classical yaw damping strategy tuning.

ARTICLE IN PRESS
J.T. Pearson et al. / Control Engineering Practice 12 (2004) 13811391

1385

Fig. 6. Classical yaw damping strategy robustness.

obtained. Selection of appropriate weighting factors

Q; R is very important to ensure a good design of the
optimal controller:
x A  x B  u G  w;

11

y C  x D:u;
Z
J yT  Q  y uT  R  u dt;

12

where
"
Q

1:7e  5
0

#
0
;
1:7e  5

13
"

#
0:01
0
;
0
0:01

y yw1 yw2 T :
The control input vector consisting of the two actuator
torque signals is dened by Eq. (14):
Fig. 7. Optimal control strategy structure.

a controller that provides the desired degree of stability.

For steering control it may be necessary to include other
state variables such as the angle of attack (i.e. the yaw
angle relative to the track) of the two wheelsets in the
cost function as well.
Eqs. (11) and (12) dene a state space representation
of the vehicle model suitable for optimal control design,
where x is a vector of state variables (dened in Section
2), u is the control input vector, w is a vector of track
inputs, and y is a vector of outputs or measurements.
The cost function is given in Eq. (13), the rst term
reects the expected performance of the controller and
the second term reects the control effort requirement.
By tuning the weighting matrices a control gain matrix is

u Kopt x:

14

Fig. 8 shows the variation in the minimum damping

level with control input weighting, both for the design
case of 0.3 conicity and for the high conicity robustness
case of 0.5.
The highest damping levels that the optimal controller
provided was 21.0% damping at 0.3 conicity and 18.3%
damping at the higher conicity case of 0.5.
The overall robustness of the controller is assessed as
before (velocity 2085 m=s; conicity 0.10.7). A controller weighting was selected that gave a minimum of
21.0% damping at 0.3 conicity, and 18.3% at 0.5
conicity. This proved to be very robust, providing a
minimum of 11.3% damping over this entire robustness
range (see Fig. 9). Note however that this controller
requires the direct feedback of all 14 states.

ARTICLE IN PRESS
J.T. Pearson et al. / Control Engineering Practice 12 (2004) 13811391

1386

Optimisation - Optimal Control Strategy

22
20
21.0% 0.3 Conicity (Design Point)

Minimum Damping Level (%)

18
16
14
12
10
8
18.3% 0.5 Conicity
6
4
2
0
-5

10

10

10

Control Input Weighting (R)

Fig. 8. Optimal control strategy tuning.

Fig. 9. Optimal control strategy robustness.

4. Actuator modelling and actuator requirements

An important aspect in the design of any integrated
active stability and steering control scheme are the
specication of realisable actuator requirements, and the
assessment of real actuator dynamics on controller
performance.

Not surprisingly the actuator requirements are dominated by the stability controller, since it is applying
control torques to stabilise the wheelset kinematic modes
which are relatively high frequency inputs when compared to the frequencies required for steering control
torques. However the stroke of the actuator will be
determined by the maximum steering angle required.

ARTICLE IN PRESS
J.T. Pearson et al. / Control Engineering Practice 12 (2004) 13811391

Typical good quality main line track data was used

for the simulation studies to provide r.m.s. and peak
values for actuator torque, velocity and power requirements. The results (averaged for both actuators) are
given in Table 1.
Based upon the results in Table 2 a motor/gearbox
combination was selected from a range of actuators
supplied by Moog. From the parameters and specication for this motor/gearbox combination, a mathematical model was developed and included into the
simulation. The actuator model consists of two sections:
the rst models the electrical dynamics of a dc motor
and the second models the mechanical aspects of the
motor inertia, shaft resilience, gearbox ratio and gearbox inertia (Fig. 10). The parameters for the actuator
model are contained in the appendix.
The results showed that there was negligible degradation in performance, and that the difference in power

1387

Table 2
Actuator requirements
Classical

Optimal

Torque
kN m

r.m.s.
Peak

0.38
1.24

0.33
2.42

Velocity
(m/s)

r.m.s.
Peak

0.31
1.14

0.42
1.87

Power
(kW)

r.m.s.
Peak

0.08
1.25

0.08
1.49

required for the controllers with no actuator dynamics

and those with real actuator dynamics was also
negligible.
The small difference between the graphs is essentially
the power that is required to accelerate the actuator and
gearbox inertias. Fig. 11 shows the power spectrum of

Fig. 10. Electromechanical actuator model.

Power Spectrum Actuator Velocity Wheelset 1

101

with actuator dynamics

no actuator dynamics
100

(rad/s)2/Hz

10-1

10-2

-3

10

-4

10

-5

10

10

12

14

Frequency (Hz)
Fig. 11. Power spectrum of actuator velocity.

16

18

20

ARTICLE IN PRESS
1388

J.T. Pearson et al. / Control Engineering Practice 12 (2004) 13811391

the actuator velocity for the case where the actuator has
no dynamics (i.e. is perfect) and the case where real
actuator dynamics have been modelled.

5. Practical sensors and estimator design

In Section 3 control strategies for the active stability
controller were developed assuming perfect measurements of the required feedback variables. For the
classical control approaches the lateral velocity of the
wheelsets is required, an obvious was to obtain this is to
mount accelerometers onto the axles or axles boxes and
integrate the measured acceleration to provide the
required velocity information. However, in practice
inertial sensors mounted on the wheelsets are extremely
undesirable, primarily due to the severe vibration

environment and the consequent expense required to

provide sensors that have sufcient accuracy and
reliability in this environment.
For the optimal control approaches, full state feedback (14 variables) is required, and a majority of the
state variables are not readily measurable. In practice it
is extremely difcult and expensive to measure parameters such as wheel-rail deection and wheelset angle
of attack. Therefore some form of state estimation is
necessary in order to enable the implementation of the
active stability controllers to be both practical and
effective. The approach adopted in this research study
was to develop a model-based estimator (Kalman Filter)
to provide the required feedback variables using only
bogie-mounted inertial sensors. The complete classical
controller structure is shown in Figs. 12 and 13 for the
optimal controller; the sensors feed measurements to the

Fig. 12. Controller structure (classical control).

Fig. 13. Controller structure (optimal control).

ARTICLE IN PRESS
J.T. Pearson et al. / Control Engineering Practice 12 (2004) 13811391

Kalman Filter which provides state estimates to the

controller; the controller calculates torque demands
which are sent to the actuators and the actuators apply
torques to the wheelsets.
Initially the following sensors were used:
(i) Bogie mounted lateral accelerometer and
(ii) Bogie mounted yaw gyro.
The Kalman Filter was formulated to estimate 14
bogie and wheelset states and 2 lateral track velocity
states:
xk1 y w1 yw1 y w1 yw1 y w2 yw2 y w2 yw2 y g yg y g yg y v yv y t1 y t2 T :

Although this formulation provided adequate estimates

of wheelset lateral velocity for the classical stability
controller it suffered from a lack of robustness.
Consequently a body mounted lateral accelerometer
was included for the classical controller. For the optimal
control strategies it was also necessary to include the
primary suspension longitudinal displacement. This
fourth measurement was necessary in order for the
Kalman Filter to provide adequate estimates of all
the states, as opposed to just the lateral velocity of the
wheelsets required for the classical control strategy.
The Kalman Filter designed using this formulation
produces accurate results on straight track, however
preliminary analyses have shown that large errors will
result when the vehicle travels a curved track. This is
because the track curvature and cant angles are deterministic track features which are not estimated by or known
a priori by the Kalman Filter. A solution is to reformulate

1389

the state equation of the Kalman Filter to include states

for the track curvature and cant at each axle:

T
1
1
xk1 y w1 yw1 y w1 yw1 y w2 yw2 y w2 yw2 y g yg y g yg y v yv y t1 y t2
yc1
yc2 :
R1
R2

Fig. 14 shows the wheel-rail lateral displacement for the

leading wheelset along with its estimation error. An
analysis of the complete classical controller, including
actuator dynamics and state estimator, showed that the
minimum damping levels provided by the classical
controller had fallen slightly to 17.9% at 0.3 conicity
(the design point) and 5.1% at 0.5 conicity. This drop in
robustness is attributed to the estimator in the feedback
loop which has a model scheduled on velocity only. The
optimal controllers performance (also including estimator and actuator dynamics) dropped to 14.9% at 0.3
conicity, and 11.1% at 0.5 conicity.
6. Conclusions
This paper has studied optimal control and classical
control strategies for an active stabilisation control
system of a high-speed railway vehicle. The aim of the
system is to provide a bogie which is stable at high
speeds without the need for heavy secondary yaw
dampers (these also form an undesirable path for the
transmission of noise).
Individually steerable axles also offer the ability to
reduce wear and noise generation through an optimal
(not necessarily minimal) angle of attack, and the
opportunity to balance lateral rail forces, both of which
are important benets.

Wheel-Rail Lateral Displacement

3

Displacement (mm)

-1

-2

-3
0

Time (secs)
Fig. 14. Wheel-rail lateral displacement and estimation error.

10

ARTICLE IN PRESS
J.T. Pearson et al. / Control Engineering Practice 12 (2004) 13811391

1390

Active versus Passive Vehicle Stabilisation

Lateral Wheelset Displacement
6
4

Passive

Displacement (mm)

2
0
-2
-4
Active
-6
-8
-10
-12
7

10

11

12

13

14

15

16

17

Time (seconds)
Fig. 15. Experimental roller rig testslateral wheel-rail deection.

An important issue relates to the integration of the

stability and steering controllers. These are being
developed separately because the former acts at typically
2 Hz and higher, whereas the latter will be relatively
slow-acting with a bandwidth signicantly less than
1 Hz: This area is the subject of on-going investigations,
but is not anticipated to present major difculties.
The requirement for practical sensors is a restrictive
one and it affects both control strategies. The requirement for full state feedback for the optimal controller
means that more sensors are needed in this case in order
to provide sufcient levels of accuracy in the state
estimates, particularly with respect to robustness.
Real actuator dynamics have been shown not to be a
problemthe effects on controller performance were
negligible.
The active yaw stabilisation (classical control algorithm see Section 3) concept has been implemented on a
full size vehicle. The actuators were supplied by Moog
and were selected based upon the analysis described in
Section 4. The system has been proven experimentally
following a comprehensive testing phase on a roller rig.
The vehicle was successfully stabilised at speeds over
300 km=h: The analysis of the results is on-going, but
Fig. 15 shows some preliminary results from the testing
phase. The gure shows two time histories for the lateral
displacement of the leading wheelset in response to
standard stability test input (a 1-cosine displacement of
the rails). The passive vehicle is marginally stable
whereas the active controller shows good levels of
damping.

Acknowledgements
The authors wish to acknowledge the support of
Bombardier Transportation.
Appendix A
The symbols and parameters are shown in Table 3 of
Appendix A.
Table 3
Symbols and parameters
yw1 ; yw2 ; yg ; yv
yw1 ; yw2 ; yg ; yv
Vs
mw ; Iw
lg ; lv
r0 ; l
mg ; Ig
Ks ; Cs
mv
Ksc ; Csc
f1 ; f2
R1 ; R2

Lateral displacement of leading, trailing wheelset,

bogie frame and vehicle body
Yaw displacement of leading, trailing wheelset,
bogie frame and vehicle body
Vehicle travel speed 83:3 m=s or 300 km=h
Wheelset mass 1376 kg and Yaw inertia
766 kg m2 ; respectively
Half gauge of wheelset 0:75 m and half
spacing of axles 1:225 m
Wheel radius 0:445 m and conicity (0.3),
respectively
Bogie frame mass 3447 kg and yaw inertia
3200 kg m2 ; respectively
Lateral stiffness 511 kN=m and damping
per wheelset 37 kN s=m; respectively
Vehicle mass 34; 460 kg
Secondary lateral stiffness 471 kN=m and
damping per wheelset 12 kN s=m; respectively
Longitudinal and lateral creepage coefcients
(10 and 10 MN)
Radius of the curved track at the leading and
trailing wheelsets 3500 m

ARTICLE IN PRESS
J.T. Pearson et al. / Control Engineering Practice 12 (2004) 13811391
Table 3 (continued)
yc1 ; yc2
yt1 ; yt2
Tw1 ; Tw2
g
Im
Ra
La
Kt
Kv
Cm
Ig1
n
Kg
C-g

Cant angle of the curved track at the leading and

trailing wheelsets 6
Track lateral displacement (irregularities)
Controlled torque for leading and
trailing wheelsets, respectively
Gravity 9:8 m=s2
Motor moment of inertia 0:00115 kg m2
Motor armature resistance 0:112 O
Motor armature inductance 9:04e-4 H
Motor torque constant 0:537 N m=A
Motor back emf constant (0.435)
Motor-gearbox shaft Damping 0:0084 N m=rads
Gearbox moment of inertia (motor end)
3:864e-4 kg m2
Gear ratio (1/87)
Gearbox drive stiffness 1:131102e6 N m=rad
Gearbox drive damping (7540.7) N m s=rad

References
Aknin, P., Ayasse, J. B., & Devallez, A. (1991). Active steering of
railway wheelsets. 12th IAVSD conference, Lyon, France.
Anon. (1997). Dosaged torque. BahnTech (report of Deutsche Bahn
Ag), 3/97 (pp. 1011).

1391

Dukkipati, R. V., & Guntur, R. R. (1984). Design of a longitudinal

railway vehicle suspension system to ensure the stability of a
wheelset. International Journal of Vehicle Design, 5(5), 451466.
Goodall, R., & Li, H. (1999). Solid axle and independently rotating
wheelsetsa control engineering assessment. Journal of Vehicle
System Dynamics, 33, 5767.
Gretzschel, M., & Bose, L. (1999). A mechatronic approach for active
inuence on railway vehicle running behaviour. 16th IAVSD
symposium dynamics of vehicles on roads and tracks, Pretoria,
August.
Hedrick, J. K. (1981). Railway vehicle active suspensions. Vehicle
System Dynamics, 10(45), 267283.
Li, H., & Goodall, R. M. (1998). Modelling and analysis of a railway
wheelset for active control. UK Control 98, Swansea, September.
Mei, T. X., & Goodall, R. M. (1999a). Kalman Filter for the state
estimation of a 2-axle railway vehicle. Fifth ECC99 conference,
Karlsruhe, Germany, AugustSeptember.
Mei, T. X., & Goodall, R. M. (1999b). Optimal control strategies for
active steering of railway vehicles. Proceedings of IFAC world
congress 99, Beijing, China, Vol. F (pp. 251256).
Mei, T. X., & Goodall, R. M. (2000). Wheelset control strategies for a
2-axle railway vehicle. Journal of Vehicle System Dynamics, 33,
653664.
Mei, T. X., Perez, J., & Goodall, R. M. (2000). Design of optimal
controls for perfect curving of solid axle wheelset. UKACC control
2000, UK.
Wickens, A. H. (1991). Dynamics of actively guided vehicles. Vehicle
System Dynamics, 20(34), 219242.