A Mechatronic Approach For Anti-Slip Control in Railway Traction
A Mechatronic Approach For Anti-Slip Control in Railway Traction
A Mechatronic Approach For Anti-Slip Control in Railway Traction
Abstract: This paper presents a novel mechatronic approach for the detection of wheel slip/slide and anti-
slip control in railway traction systems, to enable an optimal use of adhesion in poor contact conditions.
The proposed technique explores the variations in wheelset dynamic properties caused by condition
changes at the wheel-rail contact and detects slip conditions from the torsional resonant vibrations of the
wheelset axle indirectly. The modeling of a typical traction system, consisting of an induction traction
motor (with associated power inverter and field-orientated control) connected to a wheelset via a gearbox,
is introduced. The development of the slip detection and control scheme is presented, and the effectiveness
of the proposed technique is demonstrated using computer simulations.
A standard configuration as illustrated in Fig 1 is used in the However, the mechanical models of the systems are much
study, which involves a conventional solid-axle wheelset more complex. One of the key issues is the modelling of the
connected to an AC traction motor through a traction gearbox. wheel-rail contact forces, as it provides the essential link
A typical three-phase induction motor is used for the between the output of the traction motor and the mechanical
provision of traction for the wheelset and ¼ of a typical load. The wheel-rail contact mechanics involves the contact
vehicle. Connections between the wheelset and the forces caused by so-called creepages between the wheel and
bogie/vehicle in the longitudinal direction are assumed to be rail surfaces which are small relative velocities resulted from
solid, as the stiffness is normally very high and the associated elastic deformation of the steel at the point of contact. The
dynamics is not of significant relevance to this study. overall creep force at the contact point is a non-linear
function of the creepage and limited by the available
adhesion. Measurements from many experimental studies
ks have indicated that the creep – creep force relations follow a
general trend as indicated in Fig. 3 [Polach, 2003]. Typically,
a creep – creep force curve can be partitioned into three
different sections – the low creep or the linear section (of the
left motor right initial slope); the large creep or the non-linear section (before
wheel wheel the peak point); and the slip or unstable region (beyond the
peak point).
Fig. 1. Wheelset configuration
The spring ks in Fig. 1 represents the equivalent stiffness for
the first torsional mode of the wheelset. The frequency is
typically about 60Hz, but can vary between perhaps 40Hz for
a soft axle and to around 80Hz for a more rigid one.
An overall diagram of the mathematical models is shown in
Fig 2. The models for the induction motor, the power
electronics and the associated vector control scheme are
fairly standard, some of which are available from
SIMULINK SimPower toolbox [Yu et al 2006].
8276
17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008
The dynamics of the wheelset axle in the rotational direction carried out based on the linearization of the creep - creep
should strictly be modelled using a distributed parameter force curves at individual operating points. Fig. 4 shows the
model, but it can be readily demonstrated that the only two migration of eigenvalues with the increase of the creepage,
dominant modes are necessary to be included in the study – where the two wheels are assumed to have the same creepage,
the common rotation of the wheelset (rolling forward) and the wheelset speed is 30m/s and the resonant frequency of the
the first torsional mode between the two wheels. A simplified torsional mode is set to 60Hz. In the linear (low creep) region
mechanical model is given in equations 1-4, representing the of the creep curve, the positive damping keeps all the modes
rotation of the right wheel, rotation of the left wheel and concerned clearly stable (as indicated by ‘*’). However, once
longitudinal motion of the wheelset (and the vehicle). the creep is increased beyond the peak value (as indicated by
‘o’), there are potentially two unstable modes. One is the
(J w + n 2 J mg )ω R = n ⋅ Te − Ttor − rR ⋅ FR _ crp (γ R , t )
• torsional mode of the axle which is indicated by the pair of
(1) the unstable conjugate poles. The other is the common
rotation of the two wheels indicated by the unstable pole on
• the real axis. Similar observations may be made with axles at
J w ω L = Ttor − rL ⋅ FL _ crp (γ L , t ) (2) different torsional frequency and at different speed/contact
conditions [Mei et al, 2006].
⎛ m ⎞•
⎜ m w + b ⎟ V = FR _ crp (γ R , t ) + FL _ crp (γ L , t ) (3) 400
⎝ 4 ⎠
Ttor = k s ⋅ ∆θ s + cs (ω R − ω L )
200
(4)
Imaginary axis
where, Te is the driving torque transmitted from the motor 0
through a gearbox, Ttor is the torsional torque, FR-crp is the
creep force at the right wheel and FL-crp is that at the left -200
wheel, Jw and Jmg are moment of inertias of wheel and
motor/gearbox respectively, mw and mb are wheelset and
vehicle masses respectively, n is the gearbox ratio, ks is the -400
-400 -200 0 200
torsional stiffness and cs represents the material damping of Real axis
the axle, ωR and ωL are the angular velocity of the right and
the left wheel respectively, ∆θs is the relative rotation Fig. 4. Eigenvalue migrations with contact conditions
between the two wheels, γR and γL are the creepages at the (Vs=30m/s; f=60Hz; adhesion coeff=0.1)
right and left wheels.
Because the existence of sustained external disturbances to
The interactions with the lateral and yaw motions of the the railway vehicles, in particular those related to the track
wheelset are considered in the full performance assessment, inputs, an unstable or even lightly damped wheelset would
although the models are not given in the paper as they are not lead to potentially damaging oscillations at a frequency that
concerned with the control design. is very high for a mechanical system. In the railway industry,
the vibrations have been largely considered as a ‘problem’
that must be overcome, because they may lead to component
3. WHEEL-RAIL CONTACT AND SYSTEM DYNAMICS failure or rail corrugation [Hardwood et al, 1991 and Lu et al,
The contact forces are essential in the provision of the 1996].
guidance control and the delivery of traction for railway
vehicles, but can produce undesirable dynamic effects under However, the link between the level of damping (or stability)
more extreme contact conditions. When the adhesion level is for the torsional mode and the wheel-rail contact conditions
high, contact forces at the wheel-rail interface are provides an excellent opportunity for control engineers to
approximately a linear function of the creepages (i.e. in the explore in the pursuit of alternative slip detection and anti-
low creep and linear region) and the creep coefficients are slip control methods. This paper will build upon the general
normally large. The large creep coefficients (in the order of principle of the novel slip detection that have been introduced
MNs) provide a high level of damping to all dynamic before [Mei et al 2006 and Yu et al 2006], and present more
motions of a wheelset, with the exception of the kinematic technical detail on the control design and performance
mode (related to the yaw and lateral modes of the wheelset) assessment.
which are in practice stabilised as a part of the design for
primary suspensions. In low adhesion conditions, it is much 4. CONTROL SYSTEM DESIGN AND ASSESSMENT
more likely that a wheelset would operate in the non-linear or
even the unstable region of the slip curve, where the damping
effect is significantly lower (or becomes negative). 4.1 Slip detection with axle based sensor
A stability analysis to study how the wheelset dynamic If a reliable sensor can be provided to measure directly the
behaviours are affected by different contact conditions is torsional torque in the axle, the vibration signals at the
resonant frequency would clearly provide an easy solution to
8277
17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008
for detecting wheel slip. This may be achieved by the use of which is normally provided in rail traction systems, is
some fairly standard data processing techniques as shown in explored in the development of a Kalman filter that estimates
Fig. 5. The band pass filter can be designed to provide a the torsional vibrations.
narrow pass band for the frequency of the torsional mode to
avoid possible interferences from other dynamic modes in the To reduce the complexity of the Kalman filter, a linearised
system. The low pass filter is used to remove the effect of the model is derived to represent the key features of the wheelset
change of traction effort from the traction motor. The dynamics as given in equations 5 and 6.
threshold will have to be carefully tuned to provide a reliable
detection of wheel slip and in the meantime avoid false ⎡ k1 ks ⎤
detection. ⎢− • 0 −
J R ⎥ ⎡∆ω ⎤ ⎢ ⎥
⎡1⎤
⎡∆ωR ⎤ ⎢ J R ⎥ R J
Sensor
⎢ ∆ω ⎥ = ⎢ 0 −
k2 ks ⎥ ⎢
∆ω ⎥ + ⎢ 0R ⎥ ⋅ ∆T
output ⎢ L⎥ ⎢ JL JL ⎥⎢
L⎥ ⎢ ⎥ e
⎢⎣ ∆θ s ⎥⎦ ⎢ ⎢ ⎥
0 ⎥ ⎣ ∆θ s ⎦ ⎢ 0 ⎥
Bandpass Magnitude Low pass ⎢ ⎥
detection 1 −1
filter filter ⎢ ⎥ ⎣ ⎦ (5)
⎣ ⎦
⎡∆ω R ⎤
Threshold
detection Y = [1 0 0]⋅ ⎢⎢ ∆ω L ⎥⎥
(6)
⎢⎣ ∆θ s ⎥⎦
Wheel slip
indicator
where additional variables JR (=Jw +n2 Jmg ) and JL (=Jw) are
Fig. 5. Wheel slip detection with axle based sensors moment of inertias of right and left wheels respectively, and
Y represents the output measurement; and k1 and k2 are the
Figure 6 shows how the magnitude of the torsional vibration rate of change on the creep-creep force curve corresponding
(the output of the low pass filter) is linked to the wheel slip, to the creepages at the two wheels.
where the maximum adhesion coefficient is changed from 0.4
to 0.05 at the time of 4s. This is clearly a small signal model which, in a strict sense, is
only valid at defined operating point(s) on a particular creep-
creep force curve and therefore a Kalman filter obtained from
the model should ideally only be used to provide estimations
Magnitude (Nm)
8278
17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008
A comparison of the original and estimated torsional motions A hysteresis control for the switching would help to
is given in Figs 7 and 8. Although a close match is not increase/restore the original torque demand after the contact
achieved in this case, the Kalman filter provides an condition has improved.
estimation that captures the essential vibrations which is
sufficient to detect the occurrence of a wheel slip. 4.4 Performances in different conditions
Motor tractive
effort Kalman Veh.
filter
Magnitude Time
Torque of vibration Filtering
Fig. 10. Re-adhesion control – reduced adhesion at t=4s
reduction etc
(initial vehicle speed of 10km/h)
8279
17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008
REFERENCES
Watanabe T, and Yamanaka A et al (1997), Optimisation of
Veh. readhesion control of Shinkansen Trains with wheel-rail
adhesion, IEEE Power Conversion Conference (PCC’97).
Time Schwartz H J, and Krebe R (1997), Implementation of an
advanced wheel creep control with searching strategy on
Fig. 11. Re-adhesion control – low adhesion from t=0s a light rail vehicle, European Power Electronics
(initial vehicle speed of 10km/h) Conference (EPE’97).
Yasuoka I, and Henmi T et al (1997), Improvement of re-
adhesion for commuter trains with vector control traction
inverter, IEEE Power Conversion Conference (PCC’97).
Park D, and Kim M S et al (2001), Hybrid re-adhesion
Speeds (rad/s)
wheel
and wheel slip detection, STECH2006, Chengdu, China.
Yu J H, Mei T X and Wilson D A (2006), Re-adhesion
control based on wheelset dynamics in railway traction
system, UKACC2006, Glasgow, UK
Polach O (2003), Creep forces in simulations of traction
vehicle vehicle running on adhesion limit”, 6th International
conference on contact mechanics and wear of rail/wheel
Time systems (CM’2003)
Hardwood N A, and Keogh P S (1991), Self-excited
oscillation in locomotive transmission systems under loss
Fig. 13. Re-adhesion control – reduced adhesion at t=4s (a of adhesion”, Institute of Mechanical Engineers,
more rigid axle, initial vehicle speed of 10km/h C414/063.
Lu G, and Harwood N A (1996), Prediction of torsional
5. CONCLUSIONS vibration on mass transit vehicle”, Computers in
Railways’96, Volume 2: Railway Technology and
Effective use of the maximum adhesion available without Environment.
causing any damaging wheel slip is an important and Anon (2001), SAW-based transducers for non contact torque,
challenging requirement in railway traction. A new concept Electronicstalk, 21 Feb 2001
for the slip detection and re-adhesion control has been
presented in this paper. Unlike conventional anti-slip control
8280