Vous êtes sur la page 1sur 11

680

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 13, NO. 2, JUNE 2012

Intelligent Traction Control Model for Speed Sensor


Vehicles in Computer-Based Transit System
Kourosh Noori, Member, IEEE, and Kouroush Jenab, Senior Member, IEEE

AbstractIn this paper, a real-time intelligent traction control


model for speed sensor vehicles in computer-based transit systems
is proposed. Using the Bayesian decision theory, the model analyzes speed sensor data to learn and classify the train traction
conditions (i.e., spin/slip, normal, and slide) that are required
for studying vehicle motion patterns. The patterns are applied
on the sensor input in real-time format to classify train traction
and reduce the error/risk of classification that may cause service
interruptions and incidents. The model can enable us to manage
a number of state natures (i.e., spin/slip, normal, and slide),
features (i.e., delta speed and train speed), and prior knowledge
traction conditions. This model engine can be implemented in any
programming language in onboard or embedded computers. As
a result, the impact of noisy sensors (inaccurate data) and its
delays in such a hard real-time control system is mitigated. This
conceptual model is applied to a case study with promising results
for target and simulation systems.
Index TermsBayesian decision theory, intelligent systems, rail
transit systems, traction control systems, transportation systems.

I. I NTRODUCTION

Fig. 1.

Block diagram of a train traction interface.

accelerates (propulsion), and slide happens when the train


decelerates (brake).
The slip/slide factor (S) is formulated by (1), where train
speed (VTrain ) and wheel speed (VWheel ) are used. If VTrain =
VWheel , then the train is in normal condition; otherwise, the
train is either in slip or in slide condition

N rail transit systems, there are three traction conditions,


i.e., Slip (Spin), Normal, and Slide. The Slip (Spin) is the
condition while the wheel speed is greater than the train speed
in acceleration mode, which may be observed in the stop-togo pattern. The normal traction condition means that the wheel
speed is within the tolerance (likelihood) of train speed, which
is the most desirable condition. The slip traction condition
introduces wear-out of the wheels, which makes unreliable
drive condition for a train with the worn wheels. The slide
traction condition is the condition under which the wheel speed
is less than the train speed, which may be observed in a stopgo
pattern.
The slide traction condition results in uneven wheel wearout causing wheel vibration, unreliable suspension, and driving
condition. The slip (spin) and slide traction conditions have
an impact on the vehicle positioning and safety. The slip
and slide are the same phenomenon with opposite mathematical sign, which means that slip happens when the train

Manuscript received November 7, 2011; accepted November 8, 2011. Date


of publication December 2, 2011; date of current version May 30, 2012. The
Associate Editor for this paper was A. Eskandrian.
K. Noori is with the Department of Mechanical and Industrial Engineering, Ryerson University, Toronto, ON M5B 2K3, Canada (e-mail:
krafizad@ryerson.ca).
K. Jenab is with the Society of Reliability EngineeringOttawa Chapter,
Ottawa, ON K2M 7T9, Canada (e-mail: jenab@ieee.org).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TITS.2011.2176121

S=

(VTrain VWheel )
.
VTrain

(1)

In general, the sign of S is giving the traction condition;


as such, if VTrain > VWheel (S > 0), then the train is in slide
condition. In addition, if VTrain < VWheel , then the train is in
slip condition. There are two special cases. If VTrain > 0 and
VWheel = 0, then the train is in the absolute slide condition, and,
respectively, if VTrain = 0 and VWheel > 0, then the train is in
the absolute slip (spin) condition.
The train traction control system is a module to sense,
control, and monitor the trains motion [1][5]. This includes
propulsion, brake, and traction (see Fig. 1). The traction control
system is considered to be fail-safe due to the safety criticality aspect of the rail transit system with passenger onboard
[2], [3], [6].
There are two traction control methods that exist in the rail
industry, i.e., mechanical traction control and computer-based
traction control. Both methods are vehicle oriented and utilize
propulsion (electrical motor and engine), braking, and other
electromechanical components. The computer-based method
is more recent in the industry and can be integrated into a
larger system. In a computer-based transit system, the traction controller monitors the traction motion against the wheel
traction. This system includes few components such as sensors
(speed sensor, accelerometer, and tachometer), controllers (microcontroller and microprocessor), and actuators (propulsion,
transmission lines, and engine). Fig. 2 shows a tooth wheel

1524-9050/$26.00 2011 IEEE

NOORI AND JENAB: INTELLIGENT TRACTION CONTROL FOR SPEED SENSOR VEHICLE IN TRANSIT SYSTEM

681

Fig. 4. Relationship between pulsewidth and wheel speed.


Fig. 2.

Tooth wheel speed sensor.

The relationship between wheel speed and pulsewidth is


shown by Fig. 4, which means that the wider the pulsewidth,
the lower the wheel speed. Fig. 4(a)(d) shows some example
of pulsewidth patterns with respect to wheel speed. In other
words, the slimmer the pulsewidth, the higher the speed. Pulse
count directly affects the travel distance, which means that the
higher the pulse count, the higher the travel distance. These
characteristics are used to configure a sensor-based traction
control system.
II. R ELATED W ORKS AND P ROBLEM S TATEMENT

Fig. 3. (a) Pulse rise/fall using the sensors potential (in volts). (b) Pulse
rise/fall using the sensors current (in milliamperes).

speed sensor, which can be used in any vehicle in the transit


system [7].
The speed sensors generate motion signals, including necessary information about the train speed, train travel distance,
travel direction, train acceleration, wheel speed, and wheel
acceleration. The controllers process the sensors information
to determine the wheel traction conditions (i.e., slip/spin, normal, and slide) and send a command signal to the actuator for
limiting wheel differential.
Each speed sensor generates electrical pulses that represent
wheel rotation resulting from train movement. The pulsewidth,
pulse count, and pulse phase shift are the three features that
are generated by a sensor. In fact, pulsewidth represents speed,
pulse count represents distance, and pulse phase shift represents
direction. The pulsewidth is the width of the pulse in time unit,
and pulse count is the number of pulses in time unit, as shown in
Fig. 3(a) and (b), using different parameters such as the sensors
current (in milliamperes) or potential (in volts).

Currently, most of existing traction control systems are vehicle oriented and are the application of vehicle mechanics and/or
vehicle dynamics, in which existing rail traction controllers
use either pure mechanical components or electromechanical
devices to detect and maintain the traction conditions in the
component level [8][12].
In some approaches, the traction was considered as an application of electrical propulsion, along with pulsewidth modulation in [13] and [14], control, and torque mitigation [15]. These
components can be controlled through the logic circuit, track
circuit, and power lines [16][20]. In addition, the electrical
and power applications of the traction can be sensor based and
are controlled by an embedded computer. These systems are
considered hard real time and work in very low application
cycles (e.g., 10 ms); there is no time for traction compensation
due to the short application cycle and noisy nature of the
environment [21].
Embedded traction control processors are mainly independent, with no intrusive interaction with the central computer
due to the fast action module and slow communication backbone. Power traction control is necessary for railway signaling;
however, according to the pertaining literature, there is still no
connection through the integrated system under automatic train
protection (ATP) and automatic train operation (ATO) [6].
Modern traction control has become very popular over the
past few years due to the ability of new methods to address uncertainty, learning ability, and adaptability [22][26].
Since slip/slide is an uncertain phenomenon- and time-varying
process, intelligent logic is a good choice for slip/slide
control. An artificial intelligent method was introduced in [27]
for processing, simulation of the traction control model, and

682

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 13, NO. 2, JUNE 2012

traction conditions are conducted into slip, normal, and slide,


as long as the condition persists.
There exist several drawbacks in the traditional sensor-based
traction system mainly due to the existence of noise in the
sensors, lack of vehicle characteristics (motion patterns), and
high cost of processing time (i.e., persistency and delay), These
drawbacks are identified here.

Fig. 5. Sensor-based train traction control.

analysis of how the artificial intelligence is taken to account for


the rail industry. In addition, a method was introduced in [21]
to address the fault tolerance of sensors in speed sensor and
traction control systems. Although the aforementioned models
are traction related, they are considered as component-driven
solutions [23] rather than system-driven models that cannot be
controlled by ATO/ATP.
A fuzzy approach is introduced for train control and is very
popular due to the component-driven approach of the industry
[28][31]. Frylmark and Johnson [32] and Khatun et al. [33]
[35], along with many more research studies on intelligent
methods such as [36][41], have introduced several methods
of slip control such as intelligent adhesion-observer-based controllers, fuzzy logic slip controllers, and hybrid slip control
methods. Although the proposed methods resulted in improving
traditional models, they cannot be part of an integrated system
such as ATP and ATO.
The traction system can be configured using sensors [42]; the
following are the key features of sensor-based traction control
(see Fig. 5):
Sensor signal: Sensors generate electrical pulses and pass
them to the input/output card.
Signal filtration: The input signal is filtered (low-pass
filter) to remove noises.
Data conversation: The filtered signal is converted, and the
outcome is data.
Wheel data: All converted data are the wheel data (e.g.,
wheel speed).
Train data: Wheel data are converted to train data (e.g.,
train speed).
Plausibility: All measured data have to be cross checked
and plausible.
Traction control: Upon plausible data, the traction condition is controlled.
To trust the data, plausibility check or cross check is performed among the sensors. As a result of this process, the

There is real-time data dependence rather than vehicle


behavior dependence.
Sensors records could be noisy, and data availability/
integrity could be low.
Sensor persistency limits the processing of the speed of
traction detection.
There is sensor noise detection delay in processing and
higher tolerance.
Higher tolerance in the traction detection introduces lower
precision.
Multiple sensors are needed for better reliability and
precision.
Traction model does not follow any pattern and is very
complex.
Train characteristics do not affect model effectiveness and
efficiency.
Traditional systems have high implementation cost due to
a high number of field tests for fine tuning the system.
There is no decision management applied to existing traction controls.
The main purpose of the intelligent model is to address
the aforementioned drawbacks of the traditional sensor-based
traction control system. In this paper, the focus is on speed
sensor inputs to recognize the motion patterns and to classify
the traction conditions in the intelligent format. To recognize
the traction conditions (i.e., slip/spin, normal, and slide), the
traction patterns need to be learned through training data samples. The patterns can be formulated and classified under the
Bayesian decision theory [43], [44].
Due to limited motion dynamism, having enough field data
and effective vehicle characteristics, a probabilistic approach
can be applied for the benefit of precision, safety, and reliability
to alleviate service interruptions during revenue service in rail
transit systems. As a result of literature review and domain
knowledge of the industry, the Bayesian decision theory is
chosen because of precision of Bayesian classification among
other intelligent models [43], manageable number of state of
natures (i.e., three traction conditions), and manageable number
of features (i.e., two features: delta speed and train speed).
The Bayesian decision model works for the system with available prior knowledge, in which the state of natures and features are engineered and precisely extracted from the training
samples [43].
Prior knowledge is one of the most important parts of the
Bayesian decision theory that should be calculated for each
state of nature in the training samples prior to holding any
design and implementation of the system. The prior knowledge can be derived from the field-simulated (filtered) logs,
along with knowing the vehicle/track characteristics and behaviors. Some of these behaviors are guideway layout, motion
jerk, max/min train acceleration, max/min train speed, weather

NOORI AND JENAB: INTELLIGENT TRACTION CONTROL FOR SPEED SENSOR VEHICLE IN TRANSIT SYSTEM

683

The Bayesian decision rule for two-category classes (c = 2)


and any given feature (x) is given by
Decide on 1 :
if P (1 | x) > P (2 | x); otherwise, decide on 2 . (5)
The decision rule can be extended from the posterior probability to classification error; thus, for a given feature (x), the
probability of error given by
P (error | x) = min {P (1 | x), P (2 | x)} .

Fig. 6.

(6)

In addition, risk and loss function is another term to calculate the cost of decision making and expected loss (risk).
Let {1 , 2 , . . . , c } be the set of classes (state of natures),
and let {1 , 2 , . . . , a } be the set of possible actions to be
taken. Then, (i |j ) is the loss function of action (1 ) when
the state of nature is (j ). As a result, Bayesian risk can be
formulated by
c

R (i | X) =
(i | j ) P (j | X) i.
(7)

Bayesian intelligent traction control steps.

j=1

condition, and other mechanical aspects of the motion, such as


tilt, banking, braking, and weighting.
Fig. 6 shows the process steps in an intelligent traction
control system that is proposed in this study. However, data
conversion and noise filtration are not part of the model, and the
data used for modeling have already been filtered. The details
of the data conversion are described in Section IV-A.

Now, the decision can be formulated for minimum conditional risk that is given by
Decide on j :
if R (j | X) = min {R (1 | X), R(2 | X), . . . , R (a | X)}
(8)
where i = 1, 2 . . . a, and j = 1, 2 . . . c.

III. M ODEL F ORMULATION

IV. T RACTION C ONDITIONS C LASSIFICATION A LGORITHM

The Bayesian decision theory is a method to solve pattern


recognition problems when those problems are posed in a
particular way, along with prior knowledge. This pattern is
extracted by examining certain features of the objects and
classifying them by using known patterns. This is exactly the
pattern classification problem described in [43]. It makes a
probabilistic decision on the area with higher likelihood that
is expressed by

To apply the decision formulas, the necessary features, state


of natures (classes), probabilistic distribution, parameters, and
assumptions are defined. All configuration and assumptions are
discussed as follows:
Configuration: Two independent speed sensors are mounted
on two independent train axels, and one accelerometer is
mounted on the train. In addition, the real-time application
cycle is set, for example, to 100 ms. The application cycle can
be set to any value, and it does not affect the result resolution;
however, the application cycle is important for the hardware and
software architecture and configuration. Note that the engineers
have to take into account the vital criticality of the rail transit
application in terms of time resolution and redundancy.
1) Features: The delta speed (x1 ) and train speed (x2 ) are
the two features in the feature vector (X). The delta speed
is the wheel speed difference over the cycle time, and
the train speed is the snapshot of the train speed for each
cycle. These two features are independent; however, the
train speed can be the average of two speed sensors for
some application.
State of nature (classes): There are three states
of nature (classes), i.e., slip (1 ), normal (2 ), and
slide (3 ).
2) Probabilistic distribution: Gaussian probabilistic distribution (4) is selected due to goodness-of-fit analysis by
BestFit application over the training data set and many
more filed logs. The distribution can be different from

posterior = likelihood prior


evidence.

(2)

In technical terms, the formula states that the posterior


probability of a class or state of nature (j ), given feature (x),
can be formulated by the prior probability of the class (j )
and the class-conditional probability of feature (x), given the
class (j ). In general, for a multidimensional feature input multivariate vector X = [x1 , x2 , . . . , xd ], the posterior probability
formula could be written in
P (j | X) =

p(X | j )P (j )
.
p(X)

(3)

The class conditional probability density of X (multivariate


feature vector) is Gaussian; the general form of the multivariate
Gaussian density is given by


1
1
t 1
p(X) =
(X
)
exp

(X
)
. (4)
2
(2)d/2 ||1/2

684

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 13, NO. 2, JUNE 2012

project to project due to different train characteristics


and different guideway topologies and weather conditions. The mean and variance should be calculated for
each traction condition (1 , 2 , and 3 ) in the training
data set.
3) Decision rules: The Bayesian decision rules are applied
on the inputs to decide the traction conditions that have
higher posterior probability with minimum error and risk.
In addition, the decision boundary is the area in which
the posterior probability of one condition is greater than
the other conditions, along with less error and risk. Since
there are three states of nature (1 , 2 , and 3 ) for traction conditions (i.e., spin, normal, and slide), the decision
rules are formulated by
Decide on 1 :
if P (1 | X) > P (2 | X) AND P (1 | X) > P (3 | X)
(9)
Decide on 2 :
if P (2 | X) > P (1 | X) AND P (2 | X) > P (3 | X)
(10)
Decide on 3 :
if P (3 | X) > P (1 | X) AND P (3 | X) > P (2 | X).
(11)
Special cases like equal posterior probability among the
classes will be discussed in Section IV-C (Classification Step).
The aforementioned decision rules (9)(11) can be pseudointerpreted, respectively, as given here.
If P (Slip | X) > P (Normal | X) ANDP (Slip | X) >
P (Slide | X), Decide on Slip (1 ).
If P (Normal | X) > P (Slip | X) ANDP (Normal | X) >
P (Slide | X), Decide on Normal (2 ).
If P (Slide | X) > P (Normal | X) ANDP (Slide | X) >
P (Slip | X), Decide on Slide (3 ).
Therefore, the preceding decision rules should be applied in
each speed sensor data and then cross compared for plausibility
purpose. To reach a decision, the prior probability and class
conditional probability for the available state of nature (i.e.,
slip, normal, and slide) and features (i.e., delta speed and train
speed) are required. The patterns need to be extracted from the
training set through the intelligent model and can be applied on
real-time input data. The following steps should be taken for the
core model to calculate the model elements such as training data
set, extract pattern, pattern classification, and safety/plausibility
steps.
A. Training Data Set Preparation
The training data set has five columns of data, which
include speed sensor 1 (wheel speed in meters per second), speed sensor 2 (wheel speed in meters per second),
train speed (in meters per second), train acceleration (in meters per second per second), and traction states conditions
(slip = 1, normal = 2, and slide = 3). The training data set has
been partially simulated to include all aspects of the operations
that cannot be seen in the field log due to either infrequent or
lack of valid scenarios. This part needs robust engineering to

Fig. 7.

Bayesian pattern extraction.

come up with extensive field log analysis with million hours of


operation.
The first two columns are directly converted from speed
sensor data such as pulsewidth and pulse count. The sensor
pulse data transformation can be formulated from the simple
mechanical law in vehicle dynamics. Similar to [7], the train
wheel speed (V ) is calculated based on the pulsewidth (pw ),
number of pulses per wheel rotation (ppr ), and wheel diameter
(wl ) as follows:
V =

( wl )
.
(pw ppr )

(12)

Similarly, the traveled wheel distance (D) can be calculated


based on the same parameters as stated in (12), including speed
sensor pulse count (pc ) as follows:
pc ( wl )
D=
.
(13)
ppr
The delta speed is (x1 ), in which x1 is the delta speed
between the current application cycle and the previous application cycle. The train speed (x2 ) is normally calculated
as a result of the mathematical average of two wheel speeds
from two speed sensors. The tractions (1 = slip, 2 = normal,
and 3 = slide) are given based on the vehicle characteristics
and applicable mechanical dynamism. The train acceleration
(fourth column) is given and used for cross check and plausibility purposes.
The last column is the traction condition, which is given
by simulation based on the real mechanical dynamism of the
wheels and train with respect to the guideway topology. The
traction condition is defined by the comparison of wheel speeds,
train speed, and train acceleration over the application cycle.
B. Pattern Extraction Step
This step is to extract the motion and traction patterns from
the training data set. This step has to be run one and only one
time. There are four substeps here to calculate prior probability,
mean, variance, and class conditional probability. Fig. 7 shows
the flowchart of the pattern extraction steps.

NOORI AND JENAB: INTELLIGENT TRACTION CONTROL FOR SPEED SENSOR VEHICLE IN TRANSIT SYSTEM

Fig. 8.

685

Bayesian traction classification flow.

Prior estimated probabilities of the states of the nature are


defined by
P (i ) =

(14)

Total

where j , j = 1, 2, 3 is the number of j observations in the


training set. The mean of the state of natures is given by
j


(j ) =

Oi
(15)

Total

where Oi , i = 1, 2, 3 is the number of observations, and


Total is the total number of observations in the training set.
The variance of the state of natures is expressed by
j


v(j ) =

[Oi (j )]2
Total

(16)

The class conditional probabilities of the delta speed and


train speed for the given state of natures are given by


1

1
1
t
(X (j )) .
P (X | j ) =
 1 exp (X(j ))
2
2| | 2
(17)
C. Classification Step
In this step, the extracted patterns are applied on the input
data to classify the traction conditions. The processing step
should be applied on the real-time data input stream to classify
the inputs into the known traction conditions. Fig. 8 shows the
flowcharts of the processing steps for input classification. The
posterior probabilities for each state of natures (tractions) for
given delta speed and train speed are formulated by
P (j | X) =

P (X | j )P (j )
.
P (X)

(18)

1) Prior probabilities of conditions are calculated by


(14)(16). The class conditional probabilities are calculated by (17). Upon receiving a real-time input, (18) is

686

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 13, NO. 2, JUNE 2012

applied for all possible conditions; however, the classification is based on the greatest value of all posterior
probabilities. On the other hand, the greatest posterior
probability is the classified traction for that input data.
Repeat this process until the end of the input stream, or
stop the application in case of failure or implausible conditions. In case of the equal posterior probability or other
issues such as noise and lack of data, which might happen
in some application cycles, the error and risk of classification need to be calculated, and the classification would
be based on the lowest error and/or lowest risk of action.
2) Error classification is used when the posterior probability
does not give the certain classification (either the posterior probability normal condition is equal to slip condition
or the posterior probability normal condition is equal to
slide condition). The error of classification for each state
of nature for the given delta speed and train is calculated
by (6). Overall, the model decides on the state of nature
with minimum error of decision, which is formulated by
P (error | X) = min {P (1 |X), P (2 |X), P (3 |X)} . (19)
There is no possibility that either all posterior probabilities are equal or posterior probability slip is equal to
slide. In such case, the data input is discarded and waiting
for the next cycle input (see Fig. 8).
3) Risk is used when posterior probability does not give the
precise decision (the same applied conditions that kick
in the error calculation). There is a risk associated with
each decision, and loss factors can be assigned to each
decision action on the state of natures (i.e., slip, normal,
and slide). These loss factors are used to calculate risk by
using (7) and (8). The classification is made based on the
lowest risk of action on the state of natures for the given
delta speed and train speed. The risks of action in (8) are
given by
R(i | X) = [(i | 1 )P (1 | X)]+[(i | 2 )P (2 | X)]+
(20)
where i = 1, 2, 3, . . . , a. In addition, the decision rules in
(7) are formulated by
Decide on 1 :
if R(1 | X) = min [R(1 | X), R(2 | X), R(3 | X), . . .]
(21)
Decide on 2 :
if R(2 | X) = min [R(1 | X), R(2 | X), R(3 | X), . . .]
(22)
Decide on 3 :
if R(3 | X) = min [R(1 | X), R(2 | X), R(3 | X), . . .]
(23)
respectively.
D. Safety and Plausibility Step
This step is to make sure that the traction classification is
safe and plausible among the sensors. Basically, the detected
traction for speed sensors 1 and 2 should be the same or within
the defined tolerance (see Fig. 9). In addition, the acceleration

Fig. 9.

Bayesian traction plausibility.

from speed sensors should be the same as the acceleration from


the accelerometer or within the tolerance of it (see Fig. 9). In
case of failure, the data are discarded, and the process will start
over for the next set of input data, as shown in Fig. 8. This part
of the flowchart is considered as the safety reaction for vital
critical application. There is a counter to count the number of
implausible failures, and if the number of failures exceeds the
tolerance, it will shut down the application to comply with the
safety critical reaction and fail-safe model.
V. S IMULATION AND P ERFORMANCE A NALYSIS
As projected in the Model Formulation section, the model
is implemented in four phases, i.e., training data preparation,
pattern extraction step, classification step, and safety and plausibility step. In all phases, the authors have used C/C++ to code
the program, along with using MS Excel, Matlab, and BestFit
to analyze the input and output data; drawing the charts; and
simulating the data. Although the authors used certain software
tools, the model can be implemented by any programming
language (i.e., C/C++, Ada, etc.) that is suitable for a real-time
hardware platform (i.e., embedded Intel 386, 486, Pentium,
AMD, TI TMS, and Motorola). This flexibility is because of
the independence of the model from the tools, software, and
hardware.
The application has three modules: 1) extracting patterns
from the training set; 2) using the extracted patterns to classify
the input data; and 3) validating the results by comparing the

NOORI AND JENAB: INTELLIGENT TRACTION CONTROL FOR SPEED SENSOR VEHICLE IN TRANSIT SYSTEM

687

Fig. 11. PP plot for sample data integrity.


Fig. 10. Training sample speed distribution.

classified traction with the traditional tractions. Patter recognition is done during the engineering phase, and the technical
team needs to know all the aspects of the project from vehicle
characteristics to guideway topology. Pattern classification and
plausibility phase are done in real time and can be loaded in
the target system, depending on the application and configuration. The succeeding sections are to discuss the effectiveness,
performance, and results of the implemented model.
A. Training Data Discussion
A training data set is simulated and compiled from the field
data. Field data need many filtration and deletion due to the
nature of field data. For instance, in the real field data, a train
can sit in the line for hours, which will cost multimegabyte data
with no use; in addition, some of the logs may have no slip/slide
condition and thus are not good candidates for training. The
training data set needs to have all traction conditions and all
vehicle behavioral characteristics; this can be achieved during
the system-engineering phase of the project and would alleviate
the later cost of field tests. In this paper, the VisSim application
is used for simulation of the stop-and-go motion patterns,
which include all possible traction conditions during a course of
daily operation. The training sample includes engineered train
characteristics such as maximum acceleration/deceleration and
braking curve to comply with the actual condition in the field.
Acceleration and deceleration are the key players in the traction because, generally, the slip (spin) happens in acceleration
during the traction and the slide happens in the deceleration
during the braking. Weather conditions (snow and rain) contribute in the traction condition, as such; the wheel adhesion to
the rail is very dependent on the dryness and wetness of the
road. However, in the training samples, the weather condition
is ignored.
For the input analysis, the wheel speed in the training set is
analyzed by BestFit software to find out the best probabilistic
distribution that matches the curve. The distribution chart in
Fig. 10 for wheel speed data samples shows the goodness-of-

Fig. 12. QQ plot for sample data integrity.

fit of Gaussian distribution using the chi-square test, along with


data integrity shown in Figs. 11 and 12 using the PP plot and
QQ plot methods. In Fig. 12, there is no straight line because
data had been collected from different operational days. The
speed unit in this example is in meters per second, and the
acceleration unit is in meters per second per second.
In contrast, 5% of the P-value from each side of the region in
Fig. 10 shows that the probability of error in those areas is 5%,
which is altogether 10%. Although the P-value is 10% based on
the chi-square test, there is not much data (speed) distributed
in those erroneous regions. For example, there is no negative
speed, and there is no speed more than 24.5 m/s due to the
physical limitation and speed profile that is applied to train and
guideway.
A higher number of data records in the training samples
produces a higher resolution. Increasing the training samples
is the performance criterion that needs to be considered for any
intelligent project using the probabilistic approach such as the
Bayesian decision theory. In fact, a good convergence rate is a
result of increases in the number of training samples (learning

688

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 13, NO. 2, JUNE 2012

rate for finite separable elements [43]), which will decrease the
variance and estimation as it gets closer to density. However,
for the sake of practicality, in this paper, 2000 samples are
introduced for training purposes that includes all applicable
traction conditions and vehicle behavior based on the field
data and simulated/engineered data. This number of record
(2000) is sufficiently enough to prove the practicality and
precision of the proposed model. Although we used this number
of training samples, for real application, a higher number of
training records is suggested. For other projects, the number of
records may go up to millions, depending on the application
and required conditions.
B. Input Data Discussion
Fig. 13.

The input data set is collected from a field test that includes
2000 records. These records are sorted into four columns of
data, i.e., speed sensor 1 signal (wheel speed 1), speed sensor
2 signal (wheel speed 2), train speed (witness data), and actual
traction condition (witness data). Each record of data represents
data for an application cycle. As such, the number of inputs
does not affect the classification result. This number may go
up to millions, depending on the hours of operations and
application conditions.
Since the models are implemented and tested in the laboratory, the input data injected into the model (record by record)
and the classification results (record by record) are collected accordingly. The implemented model runs through all the records
to classify the traction for each record (each row). The given
train speed and the given traction condition are for witness and
validation purposes only. In fact, the model calculates the wheel
speed and train speed, and cross compares for plausibility, as
discussed in Section IV-C.
The model classifies the data input, regardless of the number
of records; however, for sake of practicality, the authors run the
program with 2000 input records, and it proves the effectiveness
and precision of the model. The model is able to classify the
traction based on the vehicle pattern in case of noise, but
the traditional sensor-based traction control failed to classify
the condition (unknown). In most of the cases, due to the safety
criticality of the rail application, the lack of robustness and
healthiness of the input data may cause application halt.
C. Result Discussion
The output of the model is traction classification, along
with plausibility check among the sensors (speed sensors 1
and 2, and the accelerometer). As a result of the classification
exercise and as a part of the validation, the model output
is compared to the field test traction result to validate the
results, as described in the previous section. For the sake of
comparison, the proposed model and traditional model run on
the same data collected from the field and compare the results.
As a result of this comparison analysis, the intelligent traction
classification is improved by 25% over the traditional results.
Improvement of 25% is conducted from the counting of the
number of misclassifications in the field test log versus the
classification output. Therefore, the intelligent model demonstrates better classification, more precise decision on noise, and

Traction classification comparison between Bayesian and traditional.

possible missing data, along with low cost of error and risk of
misclassification.
In fact, the model works with the data, which has neither
signal nor noise that the traditional traction classifier cannot
process. The traditional sensor-based model is easily misled by
noise or mistaken as input, but the intelligent model ignores
those noisy data and follows the vehicle motion pattern. Fig. 13
shows a snapshot of the 100 cycles for illustration purposes
only and can be extended to the rest of the cycles but cannot be
displayed in the chart due to lack of space. The perforated line
represents traditional classification, and the solid line represents
Bayesian classification. The traction results (Unknown = 0,
Slip = 1, Normal = 2, and Slide = 3) are on the y-axis, and
the records are on the x-axis. The traditional method has
many unknown tractions, along with many misclassifications,
compared with the proposed model. Traditional data come from
the field test log, and it is known that the field test suffers from
missing data due to noise and misclassification.
VI. C ONCLUSION
The proposed model has focused on the speed sensor input
(wheel speed and train speed) to detect the traction condition in
an intelligent format. The intelligent model has been derived
from the Bayesian decision theory, which was implemented
in the form of pattern recognition of traction conditions. The
practice of implementation has shown that the proposed model
can be a suitable candidate for such a system due to fact that
the number of traction conditions is limited to the few motion
conditions, which is very manageable for computation.
Although the model itself is independent from the data, for
the sake of accuracy of the patterns, we need to train the system
with an ample number of training samples. Due to the high
rate of convergence and fast learning rate, the model works
with a limited number (i.e., 2000) of training samples. The
proposed model is relatively simpler than the traditional system
and is more precise in terms of traction classification. In fact,
the training sample should include all aspects of the traction
conditions and characteristics, which need to be engineered
prior to the design of such system.
The computer-based intelligent train traction safety model
based on Bayesian decision theory has less error and less risk
of misclassification, compared with a similar system that is

NOORI AND JENAB: INTELLIGENT TRACTION CONTROL FOR SPEED SENSOR VEHICLE IN TRANSIT SYSTEM

configured the same. Overall, the result is very promising with


high precision of traction classification. The model handles the
data affected by noise, along with low cost of execution and
delay. The program is small and very fast in processing the
input. Furthermore, it needs less delay and tolerance. The study
experience gives strong feedback on the flexibility of the model,
i.e., it is independent from hardware and software that can be
ported to any target platform (hardware and software). During
the exercise, the models precision proves the effectiveness of
the model for such a safety critical system.
Traction mitigation and a feedback system are not discussed
in this paper; however, it can be elaborated in the future to make
this intelligent model an adaptive model, in conjunction with
braking and propulsion modules as they are all needed in any
rail transit system.
ACKNOWLEDGMENT
The authors would like to express their appreciation to
anonymous referees for their constructive comments, which
enhanced the quality of this paper.
R EFERENCES
[1] Rail Transit Vehicle Interface Standards Committee of the IEEE Vehicular
Technology Society, IEEE Standard for the Functioning of and Interfaces
Among Propulsion, Friction Brake, and Train-borne Master Control on
Rail Rapid Transit Vehicles, 1999.
[2] Rail Transit Vehicle Interface Standards Committee of the IEEE Vehicular
Technology Society, IEEE Standard for Communications-Based Train
Control (CBTC) Performance and Functional Requirements, 1999.
[3] T. J. McGean, Developing IEEE rail transit vehicle standards, in Proc.
ASME/IEEE Joint Railroad Conf., 1998, pp. 95105.
[4] P. Liljas, Speed and positioning systems. The traditional way, in Proc.
IEEE Colloquium Where Are We Going (And How Fast!) Seminar Exploring Speed Positioning Syst. Transp. Sector, London, U.K., 1997, vol. 395,
pp. 2/12/9.
[5] G. Yelloz, Urban mass transit goes driverless, in Proc. 4th Int. Conf.
Adv. Mater. Process., Rome, Italy, 2000.
[6] C. Braban, Unattended/driverless communications-based train control
for new lines and re-signalling applications, in Proc. 4th Int. Conf. Adv.
Mater. Process., Rome, Italy, 2007.
[7] Leonard and Bauer, Technical Document TI2471: Speed Sensor for Electrically Conducting Toothed-Wheels. Oberhausen, Germany: Leonard,
Bauer, 2005.
[8] R. J. Hill, Electric railway traction. I. Electric traction and DC traction
motor drives, Power Eng. J., vol. 8, no. 1, pp. 4756, Feb. 1994.
[9] R. J. Hill, Electric railway traction. II. Traction drives with three-phase
induction motors, Power Eng. J., vol. 8, no. 3, pp. 143152, Jun. 1994.
[10] R. J. Hill, Electric railway traction. III: Traction power supplies, Power
Eng. J., vol. 8, no. 6, pp. 275286, Dec. 1994.
[11] R. J. Hill, Electric railway traction. IV: Signalling and interlocking,
Power Eng. J., vol. 9, no. 4, pp. 201206, Aug. 1995.
[12] R. J. Hill, Electric railway traction. V: Communications and supervisions, Power Eng. J., vol. 10, no. 2, pp. 8795, Apr. 1996.
[13] T. Kulworawanichpong and C. J. Goodman, Optimal area control of AC
railway systems via PWM traction drives, Proc. Inst. Elect. Eng.Elect.
Power Appl., vol. 152, no. 1, pp. 3340, Jan. 2005.
[14] J. Shen and N. Buttherworth, Analysis and design of three-level PWM
convertor system for railway traction application, Proc. Inst. Elect.
Eng.Elect. Power Appl., vol. 144, no. 5, pp. 357371, Sep. 1997.
[15] A. Steimel, Direct self-control and synchronous pulse techniques for
high-power traction inverters in comparison, IEEE Trans. Ind. Electron.,
vol. 51, no. 4, pp. 810820, Aug. 2004.
[16] A. Mariscotti and P. Pozzobon, Determination of the electrical parameters of railway traction lines: Calculation, measurement, and reference
data, IEEE Trans. Power Del., vol. 19, no. 4, pp. 15381546, Oct. 2004.
[17] A. Mariscotti and P. Pozzobon, Distribution of the traction return current
in AT electric railway systems, IEEE Trans. Power Del., vol. 20, no. 3,
pp. 21192128, Jul. 2005.

689

[18] C. J. Goodman, Train performance and simulation, fourth vacation


school on electric traction systems, IEE Power Div. J., vol. 2, no. A,
pp. 137, Apr. 1997.
[19] R. W. White, AC supply systems and protection, fourth vacation school
on electric traction systems, IEE Power Div., vol. 1, no. D, pp. 122,
1997.
[20] G. W. Chang, H. Lin, and S. Chen, Modeling characteristics of harmonic
currents generated by high-speed railway traction drive converters, IEEE
Trans. Power Del., vol. 19, no. 2, pp. 766773, Apr. 2004.
[21] S. M. Bennett, R. J. Patton, and S. Daley, Sensor fault-tolerant control of
a rail traction drive, Control Eng. Pract. J., vol. 7, no. 2, pp. 217225,
Feb. 1999.
[22] H. P. Jong and Y. K. Chan, Wheel slip control in traction control
system for vehicle stability, Veh. Syst. Dyn., vol. 31, no. 4, pp. 263278,
1999.
[23] T. X. Mei, J. H. Yu, and D. A. Wilson, A mechatronic approach for
effective wheel slip control in railway traction, Proc. Inst. Mech. Eng.
F, J. Rail Rapid Transit, vol. 223, no. 3, pp. 295304, May 2009.
[24] H. Lee and T. Masayoshi, Adaptive vehicle traction force control for
intelligent vehicle highway systems, IEEE Trans. Ind. Electron., vol. 50,
no. 1, pp. 3747, Feb. 2003.
[25] H. J. Ryoo, S. J. Kim, G. H. Rim, Y. J. Kim, and M. S. Kim, Novel
antislip/slide control algorithm for Korean high-speed train, in Proc. 29th
Annu. Conf. IEEE, 2003, vol. 3, pp. 25702574.
[26] D. Bonta, R. Festila, and V. Tulbure, The problem of speed measurements in the slip-slide control for electric railway traction, in Proc. IEEE
Int. Conf. Autom., Qual. Testing, Robot., 2006, vol. 1, pp. 321324.
[27] T. T. Chan, AI applications and solution techniques for AC railway
system control and simulation, M.S. thesis, Hong Kong Polytechnic,
Hong Kong, 1990.
[28] C. Lee, Fuzzy logic in control systems: Fuzzy logic controllerPart I,
IEEE Trans. Syst., Man, Cybern., vol. 20, no. 2, pp. 404418,
Mar./Apr. 1990.
[29] L. X. Wang and J. M. Mendel, Fuzzy basis functions, universal approximation, and orthogonal least square learning, IEEE Trans. Neural Netw.,
vol. 3, no. 5, pp. 807814, Sep. 1992.
[30] L. X. Wang, Stable adaptive fuzzy control of nonlinear systems, IEEE
Trans. Fuzzy Syst., vol. 1, no. 2, pp. 146155, May 1993.
[31] V. A. Constantin, Fuzzy logic in automotive engineering, in Proc.
Embedded Syst. Conf., 1996.
[32] D. Frylmark and S. Johnson, Automatic slip control for railway
vehicles, M.S. thesis, Dept. Elect. Eng., Linkping Univ., Linkping,
Sweden, 2003.
[33] P. Khatun, C. M. Bingham, P. H. Mellor, and N. Schofield, Application
of fuzzy control algorithms for electric vehicle antilock braking/traction
control systems, IEEE Trans. Veh. Technol., vol. 52, no. 5, pp. 1356
1364, Sep. 2003.
[34] P. Khatun, C. M. Bingham, N. Schofield, and P. H. Mellor, Antilock braking/traction control for a high-performance all-electric racing vehicle, in
Proc. 41st Int. Conf. PCIM Power Convers. Intell. Motion, Nuremburg,
Germany, 1999, pp. 251273.
[35] P. Khatun, C. M. Bingham, N. Schofield, and P. H. Mellor, Comparison
of control methods for electric vehicle antilock braking/traction control
systems, presented at the SAE World Congr., Detroit, MI, 2001, Paper
2001-01-0596.
[36] A. D. Cheok and S. Shogo, A fuzzy logic based antiskid control system
for railway applications, in Proc. 2nd Int. Conf. Know.-Based Intell.
Electron. Syst., Adelaide, SA, 1998, vol. 1, pp. 195201.
[37] A. D. Cheok and S. Shiomi, Combined heuristic knowledge and limited
measurement based fuzzy logic antiskid control for railway application,
IEEE Trans. Syst., Man, Cybern. C, Appl. Rev., vol. 30, no. 4, pp. 557568,
Nov. 2000.
[38] K. Hiroaki, S. Hideo, S. Shin-Ichiro, and H. Yoichi, Optimal slip ratio
estimator for traction control system of electric vehicle based on fuzzy
inference, Trans. Inst. Elect. Eng. Jpn., vol. 120-D, no. 4, pp. 581586,
2001.
[39] M. Garcia-Riviera, R. Sanz, and J. A. Perez-Rodriguez, An antislipping
fuzzy logic controller for a railway traction system, in Proc. 6th IEEE
Int. Conf. Fuzzy Syst., 1997, vol. 1, pp. 119124.
[40] M. Amodeo, A. Ferrara, R. Terzaghi, and C. Vecchio, Wheel slip control
via second-order, sliding-mode generation, IEEE Trans. Intell. Transp.
Syst., vol. 11, no. 1, pp. 122131, Mar. 2010.
[41] M. Yamashita and T. Soeda, A novel slip control method considering
axle-weight transfer for electric locomotive, in Proc. IEEE Trans. VPPC,
2010, pp. 16.
[42] S. Kristensen, Sensor planning with Bayesian decision theory, Robot.
Auton. Syst., vol. 19, no. 3/4, pp. 273286, Mar. 1997.

690

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 13, NO. 2, JUNE 2012

[43] R. O. Duda, P. E. Hart, and D. G. Stork, Pattern Classification, 2nd ed.


New York: Wiley, 2001.
[44] C. M. Bishop, Pattern Recognition and Machine Learning, 1st ed.
Singapore: Springer, 2007.

Kourosh Noori (M10) received the B.A.Sc. degree


in applied mathematics in computer science from
the University of Kerman, Kerman, Iran, in 1994
and the M.A.Sc. degree in mechanical engineering in 2009 from Ryerson University, Toronto, ON,
Canada, where he worked on the safe and reliable
system design of intelligent transportation systems in
rail transit.
He is currently with the Department of Mechanical
and Industrial Engineering, Ryerson University. He
has 18 years of experience in diverse engineering
capacities encompassing infrastructure and turn-key projects, along with supporting numerous local and international projects. He has focused on signaling
and automatic train control systems, including automatic train protection, automatic train operation, and automatic train supervision. His research interests
include train-borne equipment, train positioning systems (i.e., tags, loops,
radios, Global Positioning System, and axle counters), wayside and station
controllers, control center systems, platform door controllers, communications
backbones, network management systems, supervisory control and data acquisition systems, closed-circuit television, passenger announcement systems,
fare-collection equipment, power supplies, intrusion detection, security system,
and event recorders.

Kouroush Jenab (SM10) received the B.Sc. degree


from Isfahan University of Technology, Isfahan, Iran,
in 1989, the M.Sc. degree from Tehran Polytechnic,
Tehran, Iran, in 1992, and the Ph.D. degree from the
University of Ottawa, Ottawa, ON, Canada, 2005.
He was a Senior Engineer/Manager in the automotive and high-tech industries for 18 years. He joined
the National Research Council Canada as a Research
Officer, where he participated in several international
research projects. In 2006, he joined the Department
of Mechanical and Industrial Engineering, Ryerson
University, Toronto, ON, as an Assistant Professor. He is currently the Education Chair of the Ottawa Chapter of the Society of Reliability Engineering, and
the Vice President of the Journal and Conference Affairs of the International
Association of Journals and Conferences. He has published several papers in
international journals based on his experiences in industry.
Dr. Jenab has been the Editor-in-Chief of IJE and on the Editorial Board
of the International Journal of Industrial Electronics and Control and steering
committees of several conferences.

Vous aimerez peut-être aussi