mei2008

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

295

A mechatronic approach for effective


wheel slip control in railway traction
T X Mei1∗ , J H Yu2 , and D A Wilson2
1
School of Computing, Science, and Engineering, Salford University, Manchester, UK
2
The University of Leeds, Leeds, UK

The manuscript was received on 5 August 2008 and was accepted after revision for publication on 24 October 2008.
DOI: 10.1243/09544097JRRT249

Abstract: This article proposes a radically new approach for the detection of wheel slip/slide
and re-adhesion control of AC traction motors in railway applications, which provides an impor-
tant alternative and advantageous technique in traction/braking control systems to maximize
the use of adhesion in poor contact conditions. The proposed concept explores the variations in
the wheelset dynamic properties caused by the condition changes at the wheel–rail contact, and
detects and controls the slip conditions from the dynamic behaviour of a wheelset, indirectly. All
the dynamic motions of rail wheelset, which are closely related to wheel–rail contact mechanics,
are included in the study; the influence of contact conditions on the relevant modes is assessed,
which is essential in the development of a mechatronic solution based on dynamic interactions.
This is different from the more general approaches in many other studies on traction controls
where the inclusion of only wheelset rotation and longitudinal motions are considered as suffi-
cient. The development of the slip detection and re-adhesion control schemes is presented and
comprehensive simulation results are produced to support the proposed idea.

Keywords: railway adhesion, slip control, wheelset dynamics

1 INTRODUCTION the wheel rotational acceleration below a predefined


threshold [1–4].
Railway traction is a complex mechatronic system, However, there are a number of practical constraints
since the traction motors and the associated con- in the detection approaches. The absolute speed of
trols have to work in harmony with the complex the train is normally obtained from a trailer bogie
mechanical loads at the wheel–rail interface. A wheel (or axle), but the provision of reliable train speed is
slip/slide occurs when the applied tractive effort a problem when all the axles are affected, e.g. in brak-
exceeds the level of maximum adhesion available at ing. The requirement of robust/reliable sensors for
the wheel–rail interface, e.g. in poor weather con- the harsh working environment limits the accuracy
ditions or with contaminated tracks. Apart from the of position/speed measurement. Typically, the resolu-
potential impact on normal operations of a rail net- tion of position encoders used in traction applications
work, the wheel slip/slide causes undesirable wear to is <100 pulses/rev, which is particularly problem-
both wheel/track surfaces and increases the cost of atic for low-speed measurement. The output of the
maintenance. The most conventional wheel slip pro- encoders can be affected by the vertical dynamics of
tection (or re-adhesion control) schemes detect a slip the wheelset/bogie, because the sensors are normally
by measuring and controlling the slip ratio (relative mounted to count the number of teeth of the traction
speed between the two contact surfaces at the wheel– gear and the output noises may be produced due to
rail interface), and in more extreme cases control the vertical vibrations of the gearbox/gears.
Furthermore, the wheel–rail contact characteristics
are subject to large variations and it is difficult to
∗ Corresponding author: School of Computing, Science and Engi- ensure an optimal use of available adhesion because
neering, Salford University, Manchester, M5 4WT, UK. email: there is no fixed relationship between the slip ratio
t.x.mei@theiet.org and the maximum adhesion. There have been studies

JRRT249 © IMechE 2009 Proc. IMechE Vol. 223 Part F: J. Rail and Rapid Transit
Downloaded from pif.sagepub.com at PENNSYLVANIA STATE UNIV on May 25, 2015
296 T X Mei, J H Yu, and D A Wilson

on the use of the so-called disturbance observers


to detect the longitudinal creep forces and hence to
derive the rate of change on the slip curve [5, 6] with
the aim of optimizing the control of slip ratio. How-
ever, there seems to be little report on the practicality
and effectiveness of the approach.
This study investigates an indirect approach for
the wheel slip detection, which explores the desta-
bilizing effect of the poor contact conditions on the
torsional mode of a conventional solid axle wheelset,
and proposes a novel technique to detect the wheel
slip by identifying the variations in wheelset torsional
vibrations. There is no need for the measurement of
slip ratio and/or estimation of creep forces at the Fig. 1 Plane-view diagram of wheelset configuration
wheel–rail interface. The utilization of the maximum
adhesion available is inherent for the control scheme.
The novel concept was first discussed in [7, 8], where surfaces, which are small relative velocities (in longi-
its feasibility was discussed and its control possibilities tudinal and in lateral directions) resulted from elastic
were examined. This article provides a comprehensive deformation of the steel at the point of contact.
study of the new technique, including the develop- The total creepages in the longitudinal direction of
ment of the initial idea, the analysis of the theoretical the two wheels (termed as left and right wheels in the
basis, and the development of a practical slip detec- study) of a wheelset are given in equations (1) and
tion and re-adhesion control scheme with the latest (2), which are the combinations of the wheelset move-
findings from the research. ments (relative to the track) in the rotational, yaw, and
The article is organized as follows. The configuration lateral directions [9]. The contribution from the lateral
of the traction system used in the study is presented movement is caused by the profiled wheel rims at the
and the mathematical models are provided in section contact with the track surface [9]
2. Section 3 describes the general principle and the
ideas behind the proposed slip detection approach. ωl r0 − v Lg ψ̇w ω l γ ( yw − yt ) L g
λxl = + − − (1)
Section 4 presents the development of a simplified v v v R
Kalman filter by using the available position encoders ωr r0 − v Lg ψ̇w ω r γ ( yw − y t ) L g
to overcome the difficulty of measuring the torsional λxr = + − − (2)
v v v R
vibrations from the wheelset axle directly. The over-
all re-adhesion control scheme is also introduced. The The total creepages in the lateral direction are the
performance assessments are given in section 5 and same for the left and right wheel as given in equation
the key conclusions are drawn in section 6. (3), which are the combinations of the wheelsets
movements in the lateral and yaw directions [9]. The
overall creepage of each wheel can be expressed in
2 SYSTEM DESCRIPTION AND MATHEMATICAL equation (4) [10]
MODELS
ẏw
λy = − ψw (3)
v
A standard powered wheelset assembly, as shown in 
Fig. 1, is used in the study, which consists of a conven- λ = λ2y + λ2x (4)
tional solid axle wheelset and a three-phase induction
motor connected to one side of the wheel via a gearbox The overall creep force at the contact point is a
for the provision of traction (for the wheelset and 1/4 of non-linear function of the creep and limited by the
a typical vehicle). Connections between the wheelset adhesion available as confirmed by the measure-
and the bogie/vehicle in the longitudinal direction are ments from many experimental studies [10]. Gen-
assumed to be solid since the stiffness is normally very erally, a creepage–creep force coefficient (µ = creep
high and the associated dynamics is not of significant force/normal load) curve shown in Fig. 2 can be parti-
relevance to this study. tioned into three sections – the low creep or the linear
The models for the mechanical part of the systems section (of the initial slope); the large creep or the non-
are complex. One of the key issues is the modelling linear section (near and before the point of maximum
of the wheel–rail contact forces, which provides the adhesion); and the slip or unstable region (beyond the
essential link between the output of the traction peak point). The dynamic behaviour of a wheelset is
motor and the mechanical load. The wheel–rail con- closely affected by the contact forces and the operating
tact mechanics involves the contact forces caused by regions as it will be further illustrated later in the study.
the so-called creepages between the wheel and rail The overall creep forces at the wheel may be computed

Proc. IMechE Vol. 223 Part F: J. Rail and Rapid Transit JRRT249 © IMechE 2009
Downloaded from pif.sagepub.com at PENNSYLVANIA STATE UNIV on May 25, 2015
Mechatronic approach for effective wheel slip control 297

exist for the left (x = 0) and right (x = L) wheels as


expressed in equations (9) and (10)

∂ 2 θ (x, t) ∂ 2 θ (x, t)
ρJ = GJ (8)
∂t 2 ∂x 2
∂θ(0, t) ∂ 2 θ (0, t)
GJ + τ0 (t) = Il (9)
∂x ∂t 2
∂θ(L, t) ∂ 2 θ (L, t)
− GJ + τL (t) = Ir (10)
∂x ∂t 2
The partial differential equation is solved in the
computer simulations in the study of the performance
Fig. 2 Creepage and creep force characteristics evaluations of the proposed anti-slip control tech-
nique, but is not useful for the control development.
A detailed analysis in [13] shows that the common
using creepage–creep force coefficient curves for any rotational and the first torsional modes are dominant
given creepages. in a wheelset, and therefore the torsional dynamics
There are strong interactions between different may be approximated by using a simplified lumped
motions of a wheelset through the creep forces at the parameter model, in which only the first-order tor-
wheel–rail contact acting in both longitudinal and lat- sional mode must be included as indicated in equation
eral directions. It is, therefore, essential in the study (11). The rotational dynamics of the right wheel, which
of wheelset dynamics and the development of effec- is driven directly by a traction motor, and the left
tive slip control solutions, to develop and use a more wheel are therefore governed by equations (12) and
comprehensive model that include all motions of the (13), respectively
wheelset related to the contact forces. The total creep 
force at the two wheels in the longitudinal direction Ts = ks (ωr − ωl )dt + C(ωr − ωl ) (11)
provides the essential tractive effort for 1/4 of a vehi-
cle, and therefore the equation of the forward motion Ir (dωr /dt) = Tm − Ts − Ftr_x · r0 (12)
can be expressed in equation (5). The yaw movement Il (dωl /dt) = Ts − Ftl_x · r0 (13)
of the wheelset is affected by the difference in the lon-
gitudinal creep force between the two wheels and the The modern traction systems are mostly equipped
equation of motion is governed by equation (6), and with induction motors controlled by the sophisticated
the lateral dynamics are determined by the total creep techniques such as field orientated vector control.
force of the two wheels in the lateral direction as shown The dynamics of the traction motor and its associ-
in equation (7) ated control is an integral part of the traction system
and are included in the study to ensure that there will
Mv v̇ = Ftr_x + Ftl_x (5) be no potential interactions with the anti-slip con-
Iws ψ̈w = Ftr_x · Lg − Ftl_x · Lg − kw · ψw (6) trol. An indirect vector control scheme is designed
in the study and used to provide a fast and accurate
mws ÿw = Ftr_ y + Ftl_ y (7) torque control of the motor for the provision of trac-
tive effort. The vector control techniques have been
The torsional dynamics of the wheelset are often well-reported in the past [14] and the schematic dia-
neglected in the studies for traction control, as the axle gram of the system modelled in the study is shown
is relatively rigid and the natural frequency is typically in Fig. 3, where the model of a voltage source pulse
in the range of ∼60 Hz or above. However, the torsional width modulation inverter (switched at a frequency
vibrations in the slip conditions have caused serious of 1 kHz) is included to provide the voltages to the
problems in practice [11] and the detection of the tor- traction motor.
sional vibrations forms the basis of this study. In this
section, a rigorous analysis of the wheelset axle is pro-
vided by considering the shaft elasticity and treating 3 WHEELSET DYNAMICS AND SLIP DETECTION
the axle as a distributed parameter system. A sim-
plified model consisting of the dominant modes of The contact forces at the wheel–rail interface are
the wheelset is then deduced for the development of essential in the provision of the guidance control
the practical slip detection and re-adhesion scheme and the delivery of traction for railway vehicles, but
proposed in the study can produce undesirable dynamic effect in more
The free torsional vibration of the wheelset is gov- extremely contact conditions.When the adhesion level
erned by the partial differential equation as shown in is high, contact forces at the wheel–rail interface are
equation (8) [12], where the two boundary conditions approximately linear functions to the creepages, i.e.

JRRT249 © IMechE 2009 Proc. IMechE Vol. 223 Part F: J. Rail and Rapid Transit
Downloaded from pif.sagepub.com at PENNSYLVANIA STATE UNIV on May 25, 2015
298 T X Mei, J H Yu, and D A Wilson

Fig. 3 Induction motor and indirect vector control

in the low creep and linear region, and the creep coef- the torsional torque waveform of a 40 Hz wheelset and
ficients are normally of large values. The large creep 80 Hz wheelset, respectively under the same contact
coefficients (in the order of MNs) provide a high level condition, where the frequency of the dynamic oscilla-
of damping to all dynamic motions of a wheelset, with tions agrees with that of the dominant torsional mode
the exception of the kinematic instability, which are of the wheelset axle.
in practice stabilized separately in the design of pri- The problem of the torsional vibrations may be tack-
mary suspensions [15]. In low adhesion conditions, it led at the design stage of rail vehicles by including
is much more likely that a wheelset would operate in sufficient damping to the system. However, the link
the non-linear or even the unstable region of the slip between the level of damping (or stability) for the
curve, where the damping effect is significantly lower
or becomes negative.
A stability analysis based on a simplified wheelset
model and linearized creep–creep force curves at indi-
vidual operating points is given in [7], which demon-
strates a strong link between the wheelset dynamic
behaviours and different contact conditions. Most
noticeably, it has been observed that there are poten-
tially two unstable modes in the wheelset when the
creep is increased beyond the point of maximum
adhesion. One is an unbounded increase in speed for
the common rotation of the two wheels indicated by
an unstable pole on the real axis. The other is the
torsional oscillations of the axle, which is indicated
by a pair of unstable conjugate poles [7]. The former
is obvious as the majority of the tractive effort from Fig. 4 Torsional torque (60 Hz axle)
the motor will be applied to accelerate the wheels
when the wheel slip occurs. The latter is considered
in the railway industry as a highly undesirable prob-
lem because they may lead to component failure or rail
corrugation [11, 16].
The presence of the axle torsional vibrations
observed in the practical applications may be con-
firmed in computer simulations by using the math-
ematical models presented in the previous section.
Figure 4 shows the torsional torque waveform of a
60 Hz wheelset (i.e. the frequency of the dominant tor-
sional mode is 60 Hz), where the wheel–rail contact
condition encounters a sudden drop of adhesion at
the time t = 4 s. The result reveals a typical torsional
oscillation at the torsional natural mode when a slip
occurs at the wheel–rail interface. Figures 5 and 6 give Fig. 5 Torsional torque (40 Hz axle)

Proc. IMechE Vol. 223 Part F: J. Rail and Rapid Transit JRRT249 © IMechE 2009
Downloaded from pif.sagepub.com at PENNSYLVANIA STATE UNIV on May 25, 2015
Mechatronic approach for effective wheel slip control 299

a simplification is first made by considering only the


motions that have a direct and the most dominant
effect on the torsional vibration, which are the rota-
tional motions of the two wheels – equations (12) and
(14). The torsional torque of the axle is included as
one of the states in the formulation, because it is the
focus of the estimation. The linearization needed in
this study is for the contact forces at the two wheels,
which have the non-linear creep–creep force charac-
teristics (as illustrated in Fig. 2) and it is achieved by
deriving a small signal model linearized at a selected
operating point of the non-linear curves based on the
Fig. 6 Torsional torque (80 Hz axle) principle of first-order approximation. The resulted
linear small signal model is shown in equations (14)
and (15), where the output measurement (i.e. input
torsional mode of a wheelset and the wheel–rail con- to the Kalman filter) is the rotational speed of the
tact conditions provides an excellent indication for the right wheel, which is where the traction motor and
wheel slip, and therefore it is proposed in this study to the gearbox is assumed to be mounted in the study
be explored for the development of an anti-slip control ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
scheme. As long as the wheel slip/slide can be effec- ω −k /I 0 −1/Ir ωr
d ⎣ r⎦ ⎣ 1 r
tively controlled, no lasting oscillations and associated ωl = 0 −k2 /Il 1/Il ⎦ · ⎣ ωl ⎦
damages to the contact surfaces are expected in the dt Ts ks −ks 0 Ts
wheelset. ⎡ ⎤
1/Ir
+ ⎣ 0 ⎦ Tm + w (14)
4 SLIP DETECTION AND RE-ADHESION 0
CONTROL ⎡ ⎤
  ωr
y = 1 0 0 · ⎣ ωl ⎦ + v (15)
4.1 Estimation of torsional vibrations with Ts
practical sensing
where E(wwT ) = Qkf , E(vvT ) = Rkf , k1 , and k2 are lin-
Although the torsional dynamics provides an excel- earization coefficients that reflect the slope at any
lent possibility for the wheel slip detection, the direct particular operation points (i.e. the rate of change)
implementation of the idea will be problematic in on the creepage–creep force curve at the right and left
practice as it requires a measurement of the tor- wheels, respectively.
sional torque. The strain gauges (or other torque The values of k1 and k2 are dependent upon the oper-
sensors) would have to be mounted on the (rotat- ating point of the creep–creep force curve where the
ing) axle, which would present very difficult challenges linearization is carried out, and strictly the small signal
in terms of the sensor reliability and transmission of model is only valid at the particular point. To provide a
the measured signals because of the harsh working more precise estimation of the torsional torque across
environment. Therefore, it is essential that the prac- the entire operating conditions, either an extended
ticality of sensing is properly tackled. In this study, a Kalman filter or the use of a bank of filters linearized at
Kalman filter is developed to provide an estimation of different points may be more desirable. However, this
the torsional oscillations from the speed of the trac- would lead to a much more complex structure and
tion motor in the system. The speed measurement may even require additional sensor measurements.
in railway traction is always available and is normally In this study, a compromise is made between the
achieved through the use of inductive sensors to detect model complexity and the estimation accuracy. A sin-
the time shifts between the teeth of the traction trans- gle Kalman filter is designed based on k1 and k2 values
mission gears; or to count the number of teeth, the deduced from a point in the saturation region of the
output of which is then used to derive the rotational creep–creep force curve. The resulted Kalman filter
speed of the wheel or that of the traction motor as the in the form of equation (16) is simple in structure
diameter of the wheel and the gear ratio are known (third-order) and therefore easy for practical imple-
parameters. mentation, where L is the Kalman gain which can be
The full mathematical model given in section 2 precalculated
is unsuitable for the Kalman filter design since it
would lead to unnecessary complexity for the estima- ẋˆ = Ax̂ + Bu + L(y − Cx̂) (16)
tor. There are also severe non-linearities especially the
creep–creep force characteristics that directly affect Due to the compromise in the design, only a partial
the behaviour of the axle torsional mode. In this study, estimation of the torsional vibration (variation) can

JRRT249 © IMechE 2009 Proc. IMechE Vol. 223 Part F: J. Rail and Rapid Transit
Downloaded from pif.sagepub.com at PENNSYLVANIA STATE UNIV on May 25, 2015
300 T X Mei, J H Yu, and D A Wilson

Fig. 7 Torsional torque variation (60 Hz axle)


Fig. 9 Estimated torsional torque variation with posi-
tion sensing error (60 Hz axle)

do not necessarily appear at the frequency that coin-


cides with that of the torsional mode, and hence the
effect on the estimation of the torque to detect the
wheel slip at a specific frequency is expected to be lim-
ited as illustrated next and in section 5, even though
the performance of the Kalman filter appears to be
compromised.

Fig. 8 Estimated torsional torque variation (60 Hz axle)


4.2 Kalman filter based slip detection
The estimated output of the Kalman filters may be
further processed to extract the magnitude of the tor-
be achieved. However, as it is revealed in the simu- sional vibration signal at the frequency corresponding
lations, the Kalman filter can provide the estimation to the axle torsional mode by using conventional fil-
sufficient for the slip detection and appears to be rea- tering methods. A band pass filter is used and tuned to
sonably robust in a variety of different slip conditions. coincide with that of the torsional mode of a particu-
Note that the Kalman filter estimates the variations in lar wheelset that is normally known. The output of the
the torsional vibration rather than the absolute vibra- filter is then processed in a rectifier to detect the mag-
tions because of the small signal model. An initial nitude before a low pass filter that removes any noises
performance assessment can be made using simula- and other disturbances in the system.
tion examples given in Figs 7 and 8, which give the Figures 10 and 11 show the obtained magnitude of
original torsional torque variation from the simulation the estimated vibration variations – one with the ideal
where the vehicle is accelerated from an initial speed measurement and the other with practical sensing as
of 10 km/h and the estimated result from the Kalman given illustrated in Figs 8 and 9, respectively. Both
filter, respectively. Although a close match is not results indicate a rapid increase in magnitude in the
achieved and there are clearly differences, the Kalman slip condition after the time t = 4 s, whereas the mag-
filter manages to capture the essential component of nitude in the normal condition (i.e. non-slip) is very
the torsional oscillations at the right frequency. small as can be seen before t = 4 s. Therefore, a thresh-
The estimations may be further compromised by old detector may be used to enable a fast detection of
the accuracy and quality of the speed measurement the slip conditions.
that is available from the practical sensors used in
the railway vehicles. The type of sensors that are reli- 4.3 Overall re-adhesion control scheme
able and robust enough for the railway applications
typically provide a measurement with the resolution Once a wheel slip is detected, the control actions
no more than 100 pulses/rev, which is much less pre- can be applied to reduce the tractive effort until the
cise than the optical encoders used in conventional torsional oscillations disappear that would be at a
motor drives and can substantially affect the estima- level just below the maximum adhesion available in
tion results of the Kalman filter. Figure 9 shows the the stable region of the creep–creep force curves.
estimation result of the Kalman filter (with the sensor The reduction of the tractive effort must be appropri-
resolution of 100 pulses/rev), which is much noisier ate such that (a) the problem of wheel slip is overcome
than that shown in Fig. 8. On the other hand, the noises and (b) the use of adhesion is optimized to deliver the

Proc. IMechE Vol. 223 Part F: J. Rail and Rapid Transit JRRT249 © IMechE 2009
Downloaded from pif.sagepub.com at PENNSYLVANIA STATE UNIV on May 25, 2015
Mechatronic approach for effective wheel slip control 301

the magnitude Tmag exceeds a pretuned threshold and


an enabling command will be generated. A classical
proportional integral (PI) controller is used to deter-
mine the deduction of the tractive demand, until the
adhesion is restored and Tmag falls below the threshold.
The input to the control is the detected magnitude of
the torsional vibrations so that the level of wheel slip
(hence that of the oscillations) will also affect the rate
of deduction in the tractive effort.

Fig. 10 The magnitude of the estimated torsional torque


at the torsional frequency 5 PERFORMANCE EVALUATIONS

For the performance assessments, a typical profile of


the tractive effort demand is used in the simulation
as shown in Fig. 13. After an initial time period of t0 =
1.5 s, the torque demand is increased linearly with time
to limit the rate of acceleration (or jerk) for the reason
of passenger comfort. At the time ta = 2 s, the motor
output reaches the rated torque, which is also limited
in the design for passenger comfort and the demand is
then kept at the constant level. Once the rated power
of the motor is reached (at tb ), the torque demand is
reduced to maintain a constant power output when
the vehicle speed is increased.
The performance of the proposed control scheme
Fig. 11 The magnitude of the estimated torsional torque
is evaluated by using a variety of different operation
with position sensing error
conditions and condition changes, as presented below.

Condition 1: An initial speed of 10 km/h, the tor-


sional frequency of the axle 60 Hz; the maximum
adhesion is reduced suddenly at the time t = 4 s from
a dry condition (µ = 0.4) to a low adhesion at µ =
0.05. Figure 14 compares the wheel speed with that
of the vehicle (converted to the rational equivalent).
At t = 4 s the contact condition is changed causing a
severe wheel slip, where the rotational speed of the
wheels becomes substantially greater than the equiv-
alent of the vehicle speed. The wheel slip in turn causes
the torsional vibrations in the axle, the magnitude of
which at the torsional frequency is detected from the
estimations by using the Kalman filter with a posi-
Fig. 12 Re-adhesion control scheme
tion sensor as illustrated in Fig. 15. The magnitude
of the vibration is then detected in a small fraction

most traction possible. Figure 12 shows a block dia-


gram of the proposed re-adhesion control scheme. The
blocks outside the dashed frame represent the main
components in a conventional traction control sys-
tem without the re-adhesion control. The proposed
scheme for the wheel-slip detection and re-adhesion
is shown in the dashed frame. The measurement of
the motor speed is fed to the Kalman filter, which is
tuned to estimate torsional torque variations Ts_est .
The estimated output is then fed to the filtering unit
to extract the magnitude of the signal at the natural
frequency of the wheelset axle Tmag . In the threshold
detection unit, a slip is deemed to have occurred if Fig. 13 Tractive effort demand

JRRT249 © IMechE 2009 Proc. IMechE Vol. 223 Part F: J. Rail and Rapid Transit
Downloaded from pif.sagepub.com at PENNSYLVANIA STATE UNIV on May 25, 2015
302 T X Mei, J H Yu, and D A Wilson

Fig. 14 Re-adhesion performance with position sensing


error (60 Hz axle) Fig. 17 Re-adhesion performance with position sensing
error (80 Hz axle)

Fig. 15 Detected magnitude of the axle torsional vibra-


tion (60 Hz axle)

of a second and used to reduce the demand in trac- Fig. 18 Re-adhesion performance with position sensing
tive effort. Then the torque output from the traction error (60 Hz axle)
motor is reduced accordingly after the torque reduc-
tion controller is enabled and the wheel speeds are
‘pulled’ back to the vehicle’s speed in ∼1 s to restore Condition 3: An initial speed of 10 km/h , the tor-
the adhesion. sional frequency of the axle 60 Hz;but the vehicle
is accelerated on the track with the low (initial) at
Condition 2: An initial speed of 10 km/h, the tor- µ = 0.05. Figure 18 shows the speed difference in the
sional frequency of the axle 40 Hz and 80 Hz; the slip condition and Fig. 19 shows the magnitude of the
maximum adhesion is reduced suddenly at the time natural frequency component (60 Hz) from the esti-
t = 4 s from a dry condition (µ = 0.4) to a low adhe- mated torsional torque variation. In this case, the slip
sion at µ = 0.05. For the wheelset with different tor- develops in a slower manner than the condition of
sional frequencies, the proposed slip detection and the sudden change and the re-adhesion is achieved
re-adhesion control scheme works equally effectively. after two attempts. The torque reduction was not large
Figures 16 and 17 present the results indicating that the enough in the first instance, because of the more gen-
re-adhesion is achieved in < 1.5 s and 1 s, respectively. tle occurrence in the wheel slip with a slower increase

Fig. 16 Re-adhesion performance with position sensing Fig. 19 Detected magnitude of the torsional vibration at
error (40 Hz axle) the natural frequency (60 Hz axle)

Proc. IMechE Vol. 223 Part F: J. Rail and Rapid Transit JRRT249 © IMechE 2009
Downloaded from pif.sagepub.com at PENNSYLVANIA STATE UNIV on May 25, 2015
Mechatronic approach for effective wheel slip control 303

REFERENCES

1 Watanabe, T. and Yamanaka, A. Optimisation of read-


hesion control of Shinkansen trains with wheel–rail
adhesion. In Proceedings of the IEEE Power Conversion
Conference (PCC’97), Nagaoka, Japan, 1997.
2 Schwartz, H. J. and Krebe, R. Implementation of an
advanced wheel creep control with searching strategy on
a light rail vehicle. In Proceedings of the European Power
Electronics Conference (EPE’97), Nagaoka, Japan, 1997.
3 Yasuoka, I. and Henmi, T. Improvement of re-adhesion
for commuter trains with vector control traction
Fig. 20 Re-adhesion performance with position sensing
inverter. In Proceedings of the IEEE Power Conversion
error (60 Hz axle)
Conference (PCC’97), Nagaoka, Japan, 1997.
4 Park, D., Kim, M., Hwang, D., Lee, J., and Kim, Y. Hybrid
in the level of torsional vibrations. Since the controller re-adhesion control method for traction system of
keeps monitor and adjusts the traction demand con- high speed railway. In Proceedings of the IEEE Interna-
tinuously, any long lasting wheel slip is prevented tional Conference on Electrical Machines and Systems
(ICEMS’01), Shenyan, China, 2001.
though the tractive effort available is still applied.
5 Kim, W. S., Kim, Y. S., Kang, J. K., and Sul, S. K.
Electro-mechanical re-adhesion control simulator for
Condition 4: An initial speed of 100 km/h, the tor- inverter-driven railway electric vehicle. In Proceedings
sional frequency of the axle 60 Hz; the maximum of the IEEE Industry Applications Conference, 1999, vol.
adhesion is reduced suddenly at the time t = 4 s from 2, pp. 1026–1032.
a dry condition (µ = 0.4) to a low adhesion at µ = 0.05. 6 Kadowaki, S., Ohishi, K., Miyashita, I., and Yasukawa, S.
At the higher speed of 100 km/h under the sudden Re-adhesion control of electric motor coach based on
change of adhesion, the detection of the wheel slip disturbance observer and sensor-less vector control. In
(in ∼0.7 s) appears to be slower because of the less Proceedings of the Power Conversion Conference, 2002,
vol. 3, pp. 1020–1025.
rapid increase in the relative wheel speed, i.e. the
7 Mei, T. X., Yu, J. H., and Wilson, D. A. Wheelset dynamics
wheel slip is less severe. The entire time for slip detec-
and wheel slip detection. In the STECH2006, Chengdu,
tion and re-adhesion is ∼3 s as demonstrated in Fig. 20. China, 2006.
8 Yu, J. H., Mei,T. X., and Wilson, D. A. Re-adhesion control
based on wheelset dynamics in railway traction system.
6 CONCLUSIONS
In the UKACC2006, Glasgow, UK, 2006.
9 Yu, J. H. Re-adhesion control for AC traction system
An effective use of the maximum adhesion avail- in railway application. PhD Thesis, The University of
able without causing any damaging wheel slip is an Leeds, September 2007.
important and challenging requirement in the railway 10 Polach, O. Creep forces in simulations of traction vehicle
traction. A new concept for the slip detection and re- running on adhesion limit. Wear, 2005, 258, 992–1000.
adhesion control has been presented in this article. 11 Lu, G. and Harwood, N. A. Prediction of torsional vibra-
Unlike the conventional anti-slip control techniques, tion on mass transit vehicle. In Computers in railways
the new control strategy does not require the accu- ’96, vol. 2, 1996 (Railway Technology and Environment).
12 Leonard, M. Principles and techniques of vibrations,
rate measurement of train speed or information about
1997 (Prentice-Hall, New Jersey, USA).
the slip ratio, either of which can be problematic to
13 Yu, J. H. Re-adhesion control for AC traction system
obtain in practice. Since the slip detection is based on in railway application. PhD Thesis, The University of
the observation that the wheel slip is closely linked to Leeds, 2007.
the wheelset torsional vibrations, it inherently enables 14 Chen, J. Control of AC motors, 1989 (National Defence
an optimal use of the adhesion regardless of the Industry Press, Beijing, People’s Republic of China).
uncertainties/variations of the contact characteristics. 15 Wichens, A. H. Fundamentals of rail vehicle dynam-
The novelty and potential usefulness of the pro- ics – guidance and stability, 2003 (Swets & Zeitlinger
posed new anti-slip control strategy have been clearly Publishers, Lisse).
demonstrated, which are also supported by the 16 Hardwood, N. A. and Keogh, P. S. Self-excited oscillation
simulation-based performance assessments in a vari- in locomotive transmission systems under loss of adhe-
sion. Institute of Mechanical Engineers, C414/063, 1991.
ety of adverse conditions. Further work is on going
to improve the estimation of the torsional vibrations
and to verify experimentally the proposed technique. APPENDIX
It may also be needed to include and study possible
Notation
interferences of other wheelset dynamic modes on the
torsional vibrations and investigate the potential effect C material damping of the wheelset axle
on the proposed control scheme. (Nm/rad/s)

JRRT249 © IMechE 2009 Proc. IMechE Vol. 223 Part F: J. Rail and Rapid Transit
Downloaded from pif.sagepub.com at PENNSYLVANIA STATE UNIV on May 25, 2015
304 T X Mei, J H Yu, and D A Wilson

Ftl_x longitudinal creep force at the left v measurement noise vector


wheel (N) v vehicle speed (m/s)
Ftr_x longitudinal creep force at the right w process noise vector
wheel (N) x state variable vector
Ftl_ y lateral creep force at the left wheel (N) yt track lateral displacement (m)
Ftr_ y lateral creep force at the right wheel yw wheelset lateral displacement (m)
(N)
Il inertia of the left wheel (kg m2 ) γ wheel conicity
Ir inertia of the right wheel (kg m2 ) θ rotation angle of wheel (rad)
Iws wheelset yaw inertia (kg m2 ) λ total creep at a contact point
k1 , k2 linearization coefficients (Nm/rad/s) λxl longitudinal creep at the left wheel
ks torsional stiffness of wheelset axle λxr longitudinal creep at the right wheel
(Nm/rad) λy lateral creep
kw yaw stiffness of wheelset (Nm/rad) µ adhesion coefficient
Lg half gauge (m) µo , µm , µs adhesion coefficients in the linear
mws wheelset mass (kg) region, maximum adhesion and slip
Mv equivalent vehicle mass driven region respectively
by the traction system on the wheelset τ0 , τl the external torque applied to the right
(kg) and left wheels of a wheelset (Nm)
r0 nominal wheel radius (m) ψw wheelset yaw angle (rad)
R curve radius of track input (m) ωl , ωl rotational speed and speed difference
Tm tractive effort demand (Nm) of the left wheel (rad/s)
Ts , Ts torsional torque and torque difference ωr , ωr rotational speed and speed difference
(Nm) of the right wheel (rad/s)

Proc. IMechE Vol. 223 Part F: J. Rail and Rapid Transit JRRT249 © IMechE 2009
Downloaded from pif.sagepub.com at PENNSYLVANIA STATE UNIV on May 25, 2015

You might also like