Journal of Physics: Conference Series

OPEN ACCESS You may also like

- Ride performance of a high speed rail
Modern techniques for condition monitoring of vehicle using controlled semi active
suspension system
railway vehicle dynamics Sunil Kumar Sharma and Anil Kumar

- Fault detection and quantitative

assessment of wheel diameter difference
To cite this article: R W Ngigi et al 2012 J. Phys.: Conf. Ser. 364 012016 based on ensemble adaptive tunable Q-
factor wavelet transform and mixed kernel
principal component analysis
Shunqi Sui, Kaiyun Wang and Shiqian
Chen
View the article online for updates and enhancements. - Robust condition monitoring and fault
diagnosis of rolling element bearings using
improved EEMD and statistical features
Fan Jiang, Zhencai Zhu, Wei Li et al.

This content was downloaded from IP address 112.133.246.162 on 02/11/2023 at 04:45

25th International Congress on Condition Monitoring and Diagnostic Engineering IOP Publishing
Journal of Physics: Conference Series 364 (2012) 012016 doi:10.1088/1742-6596/364/1/012016

Modern techniques for condition monitoring of railway

vehicle dynamics

R W Ngigi, C Pislaru, A Ball and F Gu

The Centre for Diagnostic Engineering, University of Huddersfield, Queensgate,
Huddersfield, HD1 3DH, UK

Email: roy.ngigi@hud.ac.uk

Abstract. A modern railway system relies on sophisticated monitoring systems for

maintenance and renewal activities. Some of the existing conditions monitoring techniques
perform fault detection using advanced filtering, system identification and signal analysis
methods. These theoretical approaches do not require complex mathematical models of the
system and can overcome potential difficulties associated with nonlinearities and parameter
variations in the system. Practical applications of condition monitoring tools use sensors which
are mounted either on the track or rolling stock. For instance, monitoring wheelset dynamics
could be done through the use of track-mounted sensors, while vehicle-based sensors are
preferred for monitoring the train infrastructure. This paper attempts to collate and critically
appraise the modern techniques used for condition monitoring of railway vehicle dynamics by
analysing the advantages and shortcomings of these methods.

1. Introduction
The 21st century has brought more pressures to the railway industry stakeholders to deliver more.
Currently, there has been an increase in demand from rail commuters for affordable and high quality
services. Since rail plays a crucial role in stimulating economic growth, legislators are demanding for
a more sustainable industry. These challenges serve as an opportunity for the industry to invest more
on the right technology. For example, a 24 x 7 railway requires the minimisation of disruptions caused
by activities such as inspection, remedial, remove and reactive maintenance, and track renewal. So, it
is necessary to conduct more effective inspection and maintenance in less time by optimising and
automating these activities where possible. To avoid unplanned outage so as to meet the growing
demands on cost efficiency, reliability and safety for railway vehicles, the argument for implementing
intelligent condition monitoring systems is highly desirable.
The dynamics of a railway vehicle represents a balance between forces acting at the wheel-rail
interaction, suspension forces and inertia forces. The excessive response of the rail vehicle to track
irregularities can result in poor guidance and ride quality which may increase wear on the wheel and
rail, and can lead to derailment [1]. The modern intelligent railway vehicle relies on sophisticated
monitoring systems to foresee its overall dynamic behaviour during normal operation and identify
imminent critical conditions. Several concepts and tested hypotheses have been developed to monitor
these dynamics that are unique to railway vehicles.
A number of techniques have been utilised to perform fault detection in railway vehicles. They
include, advanced filtering, system identification and signal analysis methods. These techniques are

Published under licence by IOP Publishing Ltd 1

25th International Congress on Condition Monitoring and Diagnostic Engineering IOP Publishing
Journal of Physics: Conference Series 364 (2012) 012016 doi:10.1088/1742-6596/364/1/012016

used to detect and identify faults that deteriorate with time [2]. Although, the railway vehicle is a
dynamically-complex multi-bodied system that is highly nonlinear, these approaches require less
complicated mathematical models of the system and can overcome potential difficulties associated
with nonlinearities and parameter variations.
The practical application of condition monitoring of the train dynamics are done either through the
employment of track-based sensors or vehicle-based sensors. Mostly, the track bed-based sensors are
used to monitor the condition of wheelset, whereas, the rolling stock-based sensors are concerned with
the monitoring of the rolling stock infrastructure. Modern rolling stock is fitted with high-capacity
communication buses and multiple sensors which require advance processing units for data collection
and management. For instance, an on-board data processing unit should have decision-making
capabilities, hence, should be able to decide how much data to store depending on the severity of the
fault and priority of notification.
This paper aims to give an overview of the existing condition monitoring techniques applied to
monitor railway vehicle dynamics. Section two presents the model-based techniques (Kalman filters;
extended Kalman filter; sequential Monte Carlo method (Rao-Blackwellised particle filter)) used to
estimate the dynamics of the rail vehicle systems. The section also describes the signal - based
techniques (band-pass filter; spectral analysis, wavelet analysis; Fast Fourier Transform) which are
used to detect the wheel faults. Section three describes some practical applications of condition
monitoring systems employing vehicle based and track based sensors. Also the emerging technologies
that may be available in the future for effective wheelset condition monitoring are analysed.

2. Condition monitoring techniques

The significance of employing advanced information technology for condition monitoring purposes in
industries is highly appreciated. The condition monitoring technique has considerably evolved over
the years since it began as a measurement-oriented strategy. More emphasis now has been placed on
computer-based stratagem. More reliance on computer systems is as a result of their efficiency in
sending, storing and analysing large amount of data. Measuring instruments are using standard
computer components and operating systems in order to be cost effective. These changes offer new
possibilities for utilising condition monitoring of various system parameters and also, the integration
of several disciplines in the field of condition monitoring and diagnosis which existed independent of
one another [3].
In order for the railway industry to successfully implement condition-based maintenance, a good
condition monitoring tool which can predict or detect incipient faults in real time is required. It has
been accepted that, when a fault is imminent, there is certainly parametric deviation within the system.
In such instances, parameter and state estimation techniques are more likely preferred for information
extraction. Quite often direct measurements of parameters, especially for the rail vehicle dynamic
system are not readily available due to several limitations, for example, high cost of implementation,
or lack of adequate technology. Generally, condition monitoring for railway vehicle systems aids in
reducing unscheduled downtime by allowing appropriate maintenance to be scheduled.
A proper condition monitoring device to be implemented on a railway vehicle system requires
added features like safety and reliability. Hence, the choice of condition monitoring technique is very
important and should be selected based on how it can handle sever nonlinear system, robustness,
sensitive to disturbances and computation performance. This section presents various techniques used
to estimate unmeasurable parameters for the purpose of condition monitoring of the railway vehicle.

2.1. Model-based techniques

The model-based methodologies are preferred when there is no direct measurement of parameters but
there is access to the relationship between the input and output signals [2]. The model-based
diagnostic techniques have been used to identify faults in dynamic systems through the evaluation of
residuals. One particular interesting technique among the model-based techniques is the observer-
based fault detection filter design (see Figure 1). The observer-based methods are effective in

2
25th International Congress on Condition Monitoring and Diagnostic Engineering IOP Publishing
Journal of Physics: Conference Series 364 (2012) 012016 doi:10.1088/1742-6596/364/1/012016

detecting sensor, actuator, and system component failure but present difficulties when applied to
systems which are subjected to unknown disturbances and model uncertainties [4]. A possible cause of
these difficulties could be in the selection of the initial values and state vector when considering partial
linearization. Even so, there is no guarantee that the estimated parameters will converge [5].

Figure 1. Observer-based fault detection method [5].

The commonly studied method for estimation of dynamic systems that uses an observer for fault
detection is the Kalman filter (KF) for linear systems and the extended Kalman filter (EKF) for
nonlinear stochastic systems [6]. Nowadays, the Kalman filter is one of the most popular methods
used for state and parameter estimation. It utilises measurements linearly related to the state and error
covariance matrices to generate a gain referred to as Kalman Gain. This gain is applied to the prior
state estimate, thus, creating a posterior estimate. The estimation process continues in a predictor-
corrector manner while maintaining a statistically minimal state error covariance matrix. The
interaction between the wheel and the rail profiles influences the dynamic behaviour of a railway
vehicle. This dynamic interaction is nonlinear and is due to the complication arising from the contact
patch, geometry and creep. Typically these dynamics are usually analysed using the creep coefficients
and the conicity. Charles et al [7] used the Kalman filter approach to estimate the nonlinear geometries
as a nonlinear conicity function. Their estimator tracked the parameters well but there were some
degree of uncertainties especially for lower conicity values. Thus, the authors pointed out that
incorporating multiple Kalman filter models was best suited for the investigation. Tsunashima and
Mori [8] demonstrated the possibility of detecting railway vehicle suspension failure using the multi-
model approach. Their method incorporated a set of mathematical models in the initial step (model
design to represent different failure modes) before designing model-based filters based on each model.
They examined the validity of the approach by investigating the secondary lateral damper and spring
failure in railway vehicles using Kalman filter as a mode-matching filter. Despite their simulation
result showing that the mode probability in the interacting multiple model (IMM) is effective for fault
detection, the number of model history increases exponentially with time which can bring
implementation problems [5]. Other notable applications that use Kalman filter approach for
parameter estimation includes creep force [9-10], creep coefficients [11] and suspensions [12].
Another method is the sequential Monte Carlo (also known as particle filter) which in principle
takes the Bayesian approach to estimate the states and parameters, whereby one attempts to accurately
represent the probability distribution function of the parameters of interest. Though it is relatively new,
it has taken many forms over the years and this has made it popular particularly for solving estimation
problems for nonlinear systems. Li et al [13] implemented the Rao-Blackwellised particle filter to
estimate the secondary lateral and anti-yaw damping coefficients of the railway vehicle dynamic

3
25th International Congress on Condition Monitoring and Diagnostic Engineering IOP Publishing
Journal of Physics: Conference Series 364 (2012) 012016 doi:10.1088/1742-6596/364/1/012016

model then compared the findings with that of EKF. The idea behind Rao-Blackwellisation is to
reduce the computational requirement and increase the efficiency of the particle filter by reducing the
size of the augmented state space through marginalizing out some of the variables. Their simulations
indicated that there was some degree of uncertainty in the wheelset conicity estimates. This could have
been because of the track roughness and the wheel profile, thus, they concluded that, an explicit non-
linear estimate would suffice
System identification technique entails fitting parametric values to a set of measured states or
regressors to minimise the square of the error of the estimated output to the real output. Due to
unavailability of prior knowledge, ‘black box ‘model is used, but if there is some knowledge of the
system then ‘grey box’ model is employed for parameter identification of the unknown state or
regressor. Charles et al [14], proposed a least-square approach to estimate the conicity function shape.
A piecewise cubic function approach was used to estimate the nonlinear function. The parameters for
the cubic functions were estimated using the least squared approach from measured data collected
from the system. Even though the method estimated the nonlinear wheel geometry function, one of the
assumptions they made was that the track input noise was white, which is normally coloured.
Predicting models developed by Shafiullah et al [15] using regression algorithms were used to
investigate the vertical acceleration behaviour of railway wagons that are attached to a moving
locomotive using modern machine learning techniques. Different types of models were built using a
uniform platform to evaluate their performance. The set of attributes that were used to evaluate the
estimation algorithm are; correlation coefficient, root mean square error, mean absolute error, root
relative squared error, relative absolute error and computational complexity. Both front and rear body
vertical acceleration conditions were predicted using ten common regression algorithms. Although, the
accuracy of the models varied based on several factors, the linear regressor algorithm performed better
overall than any other algorithm.

2.2. Signal-based techniques

In some instances, the output signal is the only signal available; therefore, the signal-based
methodologies are relevant in such instances. The measured signals are analysed in response to some
form of disturbance. The extraction of fault-relevant signal characteristics can in many cases be
restricted to the amplitudes or amplitude densities within a certain bandwidth of the signal. Parametric
signal models can be used to estimate the parameters by analysing the changes in the frequencies and
their amplitudes. The signal processing methods used are spectral analysis, wavelet analysis and band-
pass filters [5].
These methods are useful for extracting the frequency range related to faults. However, it is
difficult to detect faults as the characteristics can vary with conditions and the acceleration state of the
running vehicle. For instance, Oba et al [16] developed a condition monitoring algorithms for fault
detection in Shinkansen bogies. The two algorithms proposed are based on statistical analysis of
vibration acceleration during certain periods. One algorithm analysed the peak vibration distribution of
operation with non-faulty parts and with faulty parts. The vibration states of the front and rear bogies
were compare using the other algorithm for one faulty bogie. One of the assumptions made was that
the faults in the rotating parts on the bogie are related to their rotational frequencies. The faults were
successfully detected by applying an appropriate band-pass filter and evaluating the difference in the
shape of the distribution for the vibration states.
A diagnostic tool is presented by Belloti et al [17] which use wavelet transform to detect wheel-flat
defect of a test train. The rail was instrumented with four accelerometers and an inductive axle-counter
block for assessing the train speed and health status of the wheel. The technique was found to have a
high efficient in detecting damaged wheels as well as measuring the train speed. The estimation
method considered that the impact force from the wheel-flat and the wheel-flat length to be linear for
which they are highly nonlinear.
Zhang et al [18] set an online test to monitor the vibration characteristics of a train wheelset. They
analysed the impacts of vibrations on the axle box bearings using characteristic spectral analysis. The

4
25th International Congress on Condition Monitoring and Diagnostic Engineering IOP Publishing
Journal of Physics: Conference Series 364 (2012) 012016 doi:10.1088/1742-6596/364/1/012016

vibration sensors were mounted on the axle box bearings to extract the vibration signals then sent to a
data acquisition circuit through the signal condition and demodulation circuit. Another signal detected
by a photo-electric speed sensor was sent to a measurement and sampling unit for analysis. The
measured signals were then transformed from time to angle domain discrete signals and then to
characteristic domain signals. From this characteristic signal, the fault in the axle box bearing was
identified. Their studies showed that, the diagnostic system performed as expected and as far much
better than the conventional spectral analysis method.
In the paper by Mehrpouya and Ahmadian[19], a superelement technique is proposed to identify
the forces that are exerted on the wheelset. A finite element (FE) model of a railway freight vehicle
was adopted for the analysis. The model developed comprised of two-axle bogies and a freight wagon.
In their studies, the model update was performed in two stages. The first stage, the bogie model was
update was updated using data measured from the actual bogies, and the second stage, the whole
vehicle model was updated from data obtained from the actual vehicle system. Due to the increased
computational cost of their technique, a superelement analysis method was found to be appropriate in
order to reduce the model to an acceptable level for computational purposes. The resultant model was
applied for force identification scheme conducted on the frequency domain. Their results indicated
that the procedure gave acceptable estimates of the low frequency range forces. They concluded by
indicating that the technique investigate is suitable for detecting forces during normal operation of the
rail vehicle and also viable in predicting parts of the track where the forces exceed the permissible
range.

3. Examples of practical applications of condition monitoring systems employing vehicle-based

and track-based sensors
Today, most of the commercially available products for condition monitoring of railway vehicles are
predominantly focused on the bogie system; this is because some of its critical components change
their parameters rapidly when in operation and can pose safety related issues. The key concept here is
the ability of the existing technology to monitor and identify these parameters in real time for
condition monitoring and predictive maintenance purposes. Different sensor configurations are
currently being implemented in the industry for monitoring railway vehicle parameters; but they
mainly fall as either on-board (vehicle-based) or track-based systems (see Figure 2a and 2b).

Bogie-mounted lateral Track

accelerometer and Bogie-mounted vertical
pitch rate gyro accelerometer, yaw rate
gyro Sensor

Axlebox-
mounted Displacement Bogie-mounted vertical
vertical transducer and lateral
accelerometer accelerometers; roll rate
Axlebox-mounted gyro
vertical and lateral Trackside system
accelerometers

Figure 2a. Bogie and wheelset sensor position [2]. Figure 2b. Trackside sensor configuration.

Condition monitoring technology within the railway industry has proliferated in recent years; this is
due to the continuous improvement of electronic-based systems. This has created a unique situation

5
25th International Congress on Condition Monitoring and Diagnostic Engineering IOP Publishing
Journal of Physics: Conference Series 364 (2012) 012016 doi:10.1088/1742-6596/364/1/012016

for implementing proactive condition monitoring technology in the railway industry. This approach
will create the possibilities of identifying failing systems while the asset is in operation before they
create catastrophic damage. Economically, most of these proactive products are wayside condition
monitoring systems and very few sensors are few sensors are directly mounted on the vehicles.
Another reason is that the cost of monitoring the condition of the bogie will be expensive than the
faults they are handling. There are enormous amount of trains in operations and to equip them with
several detectors is a challenge in regards to cost and maintaining the overall detector technology [20].

3.1. Vehicle-based sensors

Sensors mounted on a railway vehicle can be used to identify track irregularities, bogie dynamic
performance and absolute train speed. The sensor networks on the bogies are instrumental in
identifying track irregularities [2]:
Pitch rate gyro can be used to obtain the mean vertical alignment of the track at longer
wavelength.
Axlebox accelerometers can be used to measure the vertical track irregularities at shorter
wavelength.
Bogie roll rate gyro are used to approximate the track cross level for longer wavelength.
The combination of the roll and yaw rate gyro together with the lateral sensing accelerometer can
be used to estimate the absolute roll of the track. The use of these sensors allows the twist from the
design transitions to be included in the absolute twist estimate [2]. The speed of the vehicle is very
important while acquiring the data because it is necessary for performing conversion between time and
displacement along the track.
In spite of the fact that the bogie mounted sensors are adequate for monitoring track irregularities,
they can also be used to identify deviation in the rail vehicle performance. Mei and Li [21] used the
inertial sensors which were mounted on the bogie to monitor the vehicle dynamic response to track
excitation. The advantage of this method is that, it eliminates the inaccuracy inherent with position
encoders especially when the wheel is in slip/slide mode. The rail vehicle pitch and bounce
accelerations were successfully estimated using two separate filters. This approach was used for filter
design simplification by decoupling the interactions in the system. The drawback with this scheme is
that at high speeds the time shift (delay) between two signals is small, thereby introducing large errors
in the measurement.
In another application Monje et al [22], developed an intelligent sensor that can measure the rolling
contact fatigue at the wheel rail interaction. The technique comprised of an optical sensor (emitter and
detector photodiodes) attached to the bogie (i.e. positioned facing the wheels) and a radio transmitter
for sending radio frequency to a configuration circuit. In the analysis, the detectors were able to detect
the sliding effect at the wheel rail contact point. The test deduced that the wheels do not roll evenly at
all time. The major bottlenecks of using the optical sensors for such applications is that it can only be
used for short term application and that things like dirt and dust clogging on the detector may induce
measurement disparities.
Matsumoto et al [23] adopted a different approach to estimate the forces at the wheel-rail interface.
The proposed method relied on non-contact gap sensors to detect wheel distortion rather than strain
gauges or load cells. Unlike conventional sensors which are usually attached to the wheel rim, the gap
sensors are mounted on the non-rolling part of the bogie (sensor base is attached on the bearing box)
but close to wheel rim. Since the movement of the wheelset cannot be neglected, two gap sensors were
employed to compensate for the movement. Although, the measuring unit could be improved by
inclusion of more gap sensors to compensate for contact point difference and the structure of the wheel
surface, their results showed that the method can be used to extract satisfactory data.
Bleakley and Senini [24] presented an online tool for analysing the acceleration signals acquired
from accelerometers mounted on the body frame of a wagon. The dual-axis sensors were configured to
measure the lateral and vertical accelerations. The signals were converted from time to frequency
domain using fast Fourier transform (FFT) in order to establish their spectral composition. Though, the

6
25th International Congress on Condition Monitoring and Diagnostic Engineering IOP Publishing
Journal of Physics: Conference Series 364 (2012) 012016 doi:10.1088/1742-6596/364/1/012016

detection method developed provided a set of coefficients according to the frequencies interest, the
results they obtained were similar to the short term RMS value via its frequency domain. From their
results, there was some significant improvement on the signal to noise ratio (from 5/4 to 4/2) thereby
making it suitable for online detection. Even though their approach performed as expected, they
pointed out that a further improvement on the weighting and summation vector could enhance the
detection capabilities significantly.

3.2. Track-based sensors

Incorporating sensors on the track for condition monitoring purposes is to ensure the smooth running
of rail vehicle and no sudden disruption on the railway line. However, these systems are not that
reliable and in most cases the inspection of railway vehicles takes place in the depot before it leaves
for operation. Such inspections are time-consuming and prone to human error. These techniques have
been around for many years, but an increase in damaged wheelset due to higher speeds, heavier loads
and modified operating conditions has led the rail stakeholders to re-evaluate these inspections
strategies.
Brickle et al [25] produced a report on the current and emerging state-of-art automated systems that
employ track-based sensors for wheelset condition monitoring (WCM). The research commissioned
by the Rail Safety and Standard Board (RSSB) UK identified the following functional categories of
automated WCM inspection systems:
Wheel profile monitoring systems – these systems extracts data from the actual wheel profile
which will be used to compare with measurement from a new wheel profile so as to make
key analysis. Most of the available wheel profile measuring systems employs non-contact
technique to monitor the wear on the wheel as the train passes. A laser line or high-
intensity strobe light illuminates the wheels and the images are captured using high-speed
digital cameras. The extraction of the wheel parameters is done using specialized computer
software. The problem with such systems is the ability to identify cracks in the plate, rim,
and flange and tread region. Some of the current systems that are commercially available
for wheel profile measurement include DeltaRail’s Treadview, ImageMap’s WheelSpec,
Beena Vision’s WheelView and LynxRail’s ATEx. The treadView developed by DeltaRail
(UK) comprises of a series of lasers and cameras installed on the track. As the train passes
by at low-speed (less than 10mph) the images of the wheel are captured then sent to a
computer for image analysis. The wheel parameters (flange height and width, tread hollow
and rim thickness, for example) are calculated and then stored so as to build each wheel
wear history [26].
Wheel impact load detectors (WILD) – This type of measuring system detect the presence of
a defective wheel by measuring the magnitude of the load (amount of force the wheel
exerts to the rail) and comparing it to the specified threshold. WILD depend on optical
sensor, accelerometers, load cells or strain gauges to measure and detect wheel defects.
Out-of-round, shelling or flat spots characteristics on the wheels induce excessive impact
on the rail and can contribute to wear and tear of track and vehicle. The rail deflection
caused by the vertical forces exerted by the wheels is measured and analysed to determine
the wheel tread irregularities. The available systems in market are GE Transportation’s
MATTILD, DeltaRail’s Wheelchex, Teknis’ WCM and Salient System’s WILD. Also
Gotcha-QuoVadis by Baas R&D is another WILD product that uses optical sensors to
measure wheel defects. The force applied by the wheels as the rail vehicle passes is
measured through the deflection of the rails. The data collected from the sensors is
analysed using Gotcha software to identify the overall quality of the wheel.
Bogie performance detectors – These are wayside systems implemented to identify the level
of performance for railway vehicles. Different systems employ different measurement
strategies, for instance, the hunting truck detector (HTD) by Salient systems measures the
lateral force exerted on the rail in order to identify vehicles that shows excessive hunting

7
25th International Congress on Condition Monitoring and Diagnostic Engineering IOP Publishing
Journal of Physics: Conference Series 364 (2012) 012016 doi:10.1088/1742-6596/364/1/012016

motion through the evaluation of the hunting index (railway safety technology). Some
other systems like the TBOGI by wayside inspection devices (WID) uses laser technology
to assess the bogie wheelset angle of attack and its respective position on the track. The
data collected from the TBOGI system is analysed to detect faults in bogie like
misalignment, skewed or warped. Even though such systems are effective in their
applications but they fall short in monitoring defective springs. Other available systems
include GE Transportation system’s MATTILD, LynxRail’s ATEx and Salient systems’
truck performance detector (TPD) [26].
Tread condition detectors – Most systems in this category applies the non-destructive
ultrasonic sensor technology, to detect the presence of any discontinuity caused by surface
breakings/cracks on the tread surface of the wheel. The ultrasonic waves transmitted by the
detector transducers propagates through the wheel, and in case there is a crack, then the
signal will be attenuated which will be sensed by the transducers. Lasers have been
successfully implemented in ultrasonic wave generations which are detected by non-
contact ultrasonic transducers. For example the Module 2000 DSR by Talgo utilizes the
nondestructive ultrasonic detection technique to measure the wheel tread surface. The
system (DSR) is adequate in determining surface breaking or cracks on the wheel as the
train moves at speed of 6mph or less. Another detector that uses ultrasonic detection is the
Argus by Hegenscheidt MFD.
Hot axle bearing – They are used to detect anomalous hot wheel bearings. Thermal sensors
extract heat signature from the bearings to establish any indication of failure. The hot axle
bearing detection technology has developed over the years since it started in the early
1960s. The systems then depend on thermal sensitive resistors to detect infrared radiation.
On the contrary, current systems utilize infrared image processing techniques to obtain
more accurate measurement while the train is operating at speeds of about 310mph
maximum. The Pegasus by ITSS is a good example of a hot bearing detector that uses
multi-element sensors configuration to provide thermal data of bearings, wheel discs and
brakes. The pegasus can measure bearing temperatures between 0 to 150C when the train is
travelling at 310mph. Additional systems that are available include, Harbin VEIC’s HTK
499 hot bearing detector system, GE Transportation system’s micro hot bearing detector
and Schenck’s MULTIRAIL hot box detector system.
Hot and cold wheel detectors – The technology used for hot and cold wheel detection is
quite similar to that of hot bearing detectors because both techniques rely on infrared
images for wheel temperature analysis as the train is moving at high speed. When the rail
vehicle’s brakes are stuck or fully released during normal operation this is an indication of
hot wheel temperature whereas for cold wheel temperature, the brakes have malfunction
(failed to apply). The systems that are current used in the railway industry are mostly
provided as an add-on module to the hot bearing detector, however, the GE Transportation
systems’ micro hot wheel detector can be supplied as an add-on or standalone module for
brake inspection.
Acoustic bearing defect detectors – Microphone array detector are used to record the
characteristic of sound made by bearings as the train passes. Noise and excessive vibration
of the bearings will be produced as they start to fail. This type of detectors are more mature
than the hot axle bearing by reason of, that they are highly sensitive and can predict failing
bearings in advance. According to [], the recent acoustic detectors should be able to detect
at least 35 % of hot bearing failures. Examples of acoustic bearing defect detectors are
TTCI’s Trackside Acoustic Detection System (TADS®) and VIPAC’s RailBAM®.
Automatic vehicle identification (AVI) systems – The substantial use of AVI system is
significant as it helps in transmission of data from rail vehicles with defective components
to the maintenance crew at the depot for corrective action. For instance, rail vehicles are
equipped with radio frequency identification (RFID) tags which are preprogrammed for

8
25th International Congress on Condition Monitoring and Diagnostic Engineering IOP Publishing
Journal of Physics: Conference Series 364 (2012) 012016 doi:10.1088/1742-6596/364/1/012016

specific vehicles. As the train passes an area with vehicle identification reader, radio
frequency energy is sent to a certain area of the track to analyse the tread and wear patterns
of moving trains. When the train has passed, the data is sent back via GPRS to centralized
database for optimization and analysis. TransCore and TagMaster are some of the leading
manufacturers of RFID tags and readers used in the railway industry for AVI systems.
Brake pad inspection systems – This type of inspection system captures digital images of
brake pads and then use the current machine vision technology to ascertain the wear rates,
uneven wear and also detect missing brake pad. PadView by DeltaRail uses cameras and
strobe light mounted beneath the track to assess the condition of the brake pad. Other
systems that use machine vision technology include; Brake block measurement system by
MRX Technologies and FactIS by Lynxrail and TTCI.
The identified technologies by Brickle et al [25] identified several emerging technologies that
utilise track-based sensors for effective railway vehicle condition monitoring.
Train fault detection system (TFDS) – TFDS is a system that integrates high speed digital
image acquisition, real-time image data processing and pattern recognition technology to
assess various safety critical components of railway vehicles. The machine vision
technology is used to capture images of the bogie system to detect key parts that are
essential for safe operation. The components that are inspected include springs, brake shoe
and pin, bearings adapter and end cap, brake beam, and coupling components. High speed
video cameras collects the images of a moving train, analyse and process them using
computer-aided technology to detect deficiency, fractures, and other faults on the train. The
apparent application of TFDS has transformed from manual detection and maintenance to
computer-based detection and manual maintenance.
Laser-based ultrasonic cracked axle detection – This state-of-art technology for crack axle
detection utilises a laser in conjunction with standard ultrasonic transducer to detect flaws
on the axle. The ultrasonic waves generated by the high-energy pulsed laser introduces
high frequency sound waves to the axle, the feature of the received signal by air-coupled
transducer is sent to a signal processing unit for analysis to determine cracks across the
axle circumference. In order to achieve a comprehensive measurement data, multiple
inspections are done on the axle.
Automated ultrasonic-based cracked wheel detection - This is a trackside oriented system for
monitoring railway wheels using ultrasonic sensor approach. The wayside cracked wheel
detection system by Dapco utilises the ultrasonic technology to detect cracked wheels as
the train travels at speeds of up to 5mph. The data collected from the four test stations is
analysed using pattern recognition technique so as to provide real-time evaluation of the
flaw size, type, and location. The system is designed to identify cracked/shattered wheels
which are above 0.5 inches. In consequence of the fact that the detector needs to access the
wheel tread for maximum data extraction, a flange bearing rails is supplied on the track and
a couplant (to facilitate ultrasonic wave transmission) that is pumped through the coupling
of the ultrasonic sensor and wheel. This type of system is robust and can be applied to
different environmental conditions.
Displacement sensor-based bogie hunting detection – Excessive hunting experienced by a
train during high-speed operation can cause derailment, thus, a wayside system proposed
by LynxRail that utilises an array of inductive displacement sensors is used to detect the
hunting phenomenon. The displacement sensors used are capable of providing quality
information about various wheelset parameters relative to the track. The LynxRail system
is capable of detecting hunting for railway vehicle traveling at lower speeds.

9
25th International Congress on Condition Monitoring and Diagnostic Engineering IOP Publishing
Journal of Physics: Conference Series 364 (2012) 012016 doi:10.1088/1742-6596/364/1/012016

Wei et al [27] presented a real-time wheel defect monitoring system that used Fibre Bragg Grating
(FBG) sensors. The sensors were mounted on the track to measure the wheel-rail interaction in relation
to the strain imposed on the track by this interface. The strain signal of the wheels collected from the
sensors was analysed using a condition index (CI) system. The problem of using the CI is that, if one
of the wheels was in bad condition, the system will not be able make a distinction of which wheel is
faulty. One way round it, is by using low-pass filter to filter out the low-frequency of the strain signal.
Nevertheless, the condition monitoring system using FBG sensors was able to identify wheel defects
in real time and that the FBG sensors were immune to the electromagnetic interference. Nenov et al
[28] proposed an improvement of a measuring system to detect unusual wheel loading for railway
vehicles in motion. The measurement device incorporates two specialised force sensors which were
attached to both sides of the track. Figure 3 shows the set-up of the force sensor on one side of the
track. The sensors comprised of two strain gauges which measures the tangential forces at the wheel-
rail contact. Due to the power lines interference with the measured data, a notch filter was used to
cancel out the 50Hz harmonics.

Figure 3. Wheel loading measuring device [28].

The data collected as the railway vehicle passes through the sensors was analysed and the errors in
loads due to the non-perfect track were determined. The improvements made to the wheel load
measuring system are:
Adopting an appropriate notch filter to cancel the power-line interference on the signal.
Errors in the load due to the non-perfect track section measured.

4. Conclusions
The modern intelligent railway relies on sophisticated monitoring systems to allow informed decision
making on asset management actions especially in maintenance and renewals activities. An overview
of the existing techniques used for condition monitoring of railway vehicle dynamics have been
presented. Some of the existing condition monitoring techniques presented performs the fault
detection in railway based problems (critical wheel/rail contact area, for example) by using advanced
filtering, system identification and signal analysis methods. Other techniques are focussing on the fault
detection and condition monitoring of vehicle suspensions by analyzing the dynamic interactions
between different vehicle modes caused by component failures in the system and leading to simple but
effective solutions. These approaches do not require complex mathematical models of the system and
can overcome potential difficulties associated with nonlinearities and parameter variations in the
system. This paper will study the feasibility of implementing these theoretical approaches in practical
condition monitoring systems.
The practical monitoring of wheel defects for trains could be done through track mounted sensors and
the measured data processed by an advanced calculation programme before being combined with the
identification tag of a locomotive or a coach. This technique is employed by existing condition

10
25th International Congress on Condition Monitoring and Diagnostic Engineering IOP Publishing
Journal of Physics: Conference Series 364 (2012) 012016 doi:10.1088/1742-6596/364/1/012016

monitoring systems and determines precisely which part of the train is faulty/damaged, and to what
extent. Some of these wayside monitoring devices have encouraged the adoption of condition-based
maintenance thereby saving the industry valuable time and costs. Sensors can also be mounted on the
rolling stock in order to monitor the condition of the railway vehicle infrastructure. So modern rolling
stock is fitted with high-capacity communication buses and multiple sensors and will result in the
potential for advanced processing of collected data. This approach requires intelligent image
acquisition and analysis systems capable of processing large amounts of data and various ongoing
research projects are tackling this task. This paper was an attempt to collate and critically appraise the
techniques used for condition monitoring of railway vehicle dynamics.
The challenge is to find the right measurement technologies, since reliable and valid measurements
are a necessity for an effective condition monitoring approach. There is the question of finding
relevant and correct parameters that can be measured to provide the most relevant measuring data,
because the measured data must then be transformed into relevant and understandable information that
can then be used as decision support in the maintenance management process. Conclusively, these are
some of the corner stones that are needed to be able to arrive at a condition-based maintenance
strategy.

References
[1] Wickens A H Fundamentals of Rail Vehicle Dynamics: Guidance and Stability Taylor & Francis,
2003
[2] Ward C P Weston P F Stewart E J C Li H Goodall R M Roberts C Mei T X Charles G and Dixon
R Condition monitoring opportunities using vehicle-based sensors Proceedings of the Institution
of Mechanical Engineers Part F: Journal of Rail and Rapid Transit vol 225 no 2 pp 202–218
Mar 2011
[3] Rao B K N Handbook of condition monitoring Elsevier 1996
[4] Chen J and Patton R Robust model-based fault diagnosis for dynamic systems Kluwer Academic
Publishers 1999
[5] Bruni S Goodall R Mei T X and Tsunashima H Control and monitoring for railway vehicle
dynamics Vehicle System Dynamics vol 45 no 7–8 pp 743–779 2007
[6] Kalman R A new approach to linear filtering and prediction problems Transactions of the ASME –
Journal of Basic Engineering no 82 (Series D) pp 35–45 1960
[7] Charles G Goodall R and Dixon R Wheel-rail profile estimation in Railway Condition Monitoring
2006 The Institution of Engineering and Technology International Conference on 2006 pp 32 –
37
[8] Tsunashima H and Mori H Condition monitoring of railway vehicle suspension using adaptive
multiple model approach in Control Automation and Systems (ICCAS) 2010 International
Conferensce on 2010 pp 584 –589
[9] Christopher W Goodall R M Dixon R Contact force estimation in the railway vehicle wheel-rail
interface pp 4398–4403 2011
[10] Xia F Cole C and Wolfs P Grey box-based inverse wagon model to predict wheel–rail contact
forces from measured wagon body responses Vehicle System Dynamics vol 46 no 1 pp 469–479
2008
[11] Hussain I and Mei T X Multi kalman filtering approach for estimation of wheel-rail contact
conditions in In the proceeding of the UKACC International Conference on Control 2010
Coventry UK
[12] Fan H Wei X Jia L and Qin Y Fault detection of railway vehicle suspension systems in Computer
Science and Education (ICCSE) 2010 5th International Conference on 2010 pp 1264 –1269
[13] Li P Goodall R Weston P Seng L C Goodman C and Roberts C Estimation of railway vehicle
suspension parameters for condition monitoring Control Engineering Practice vol 15 no 1 pp
43–55 Jan 2007

11
25th International Congress on Condition Monitoring and Diagnostic Engineering IOP Publishing
Journal of Physics: Conference Series 364 (2012) 012016 doi:10.1088/1742-6596/364/1/012016

[14] Charles G Goodall R and Dixon R A least mean squared approach to wheel-rail profile estimation
in 4th IET International Conference on Railway Condition Monitoring 2008 vol 1
[15] Shafiullah G M Ali A B M S Thompson A and Wolfs P J Predicting vertical acceleration of
railway wagons using regression algorithms Intelligent Transportation Systems IEEE
Transactions on vol 11 no 2 pp 290 –299 Jun 2010
[16] Oba T Yamada K Okada N and Tanifuji K Condition monitoring for shinkansen bogies based on
vibration analysis Journal of Mechanical Systems for Transportation and Logistics vol 2 no 2
pp 133–144 2009
[17] Belotti V Crenna F Michelini R C and Rossi G B Wheel-flat diagnostic tool via wavelet transform
Mechanical Systems and Signal Processing vol 20 no 8 pp 1953–1966 Nov 2006
[18] Zhang J Zhou S and Huang C A new method for the fault diagnosis of the train wheelset based on
characteristic spectrum analysis in Control Conference (CCC) 2010 29th Chinese 2010 pp
3988–3992
[19] Mehrpouya M and Ahmadian H Estimation of applied forces on railway vehicle wheelsets from
measured vehicle responses International Journal of Vehicle Structures & Systems vol 1 no 4
Nov 2009
[20] Lagnebäck R Evaluation of wayside condition monitoring technologies for condition-based
maintenance of railway vehicles Licentiate Thesis Luleå University of Technology Department
of Civil Mining and Environmental Engineering Division of Operation and Maintenance
Engineering Luleå Sweden 2007
[21] Mei T X and Li H Measurement of vehicle ground speed using bogie-based inertial sensors
Proceedings of the Institution of Mechanical Engineers Part F: Journal of Rail and Rapid
Transit vol 222 no 2 pp 107–116 Mar 2008
[22] Monje P Aranguren G Martinez B and Casado L Using bogie-mounted sensors to measure wheel
rolling and sliding in railway tracks Proceedings of the Institution of Mechanical Engineers
Part F: Journal of Rail and Rapid Transit Dec 2011
[23]Matsumoto A Sato Y Ohno H Tomeoka M Matsumoto K Kurihara J Ogino T Tanimoto M
Kishimoto Y Sato Y and Nakai T A new measuring method of wheel–rail contact forces and
related considerations Wear vol 265 no 9–10 pp 1518–1525 Oct 2008
[24] Bleakely S and Senini S Autonomous time frequency analysis of wagon body accelerations in
Proceedings of the Fifth Asia Pacific Industrial Engineering and Management Systems
Conference 2004 Gold Coast Australia 2004 pp 34.6.1–34.6.12
[25] Brickle B Morgan R Smith E Brosseau J and Pinney C Identification of existing and new
technologies for wheelset condition monitoring TTCI Ltd UK RSSB Report for Task T607 Jul
2008
[26] Moynihan T W and English G W Railway safety technologies Canada Research and traffic group
Jul 2007
[27] Wei C Xin Q Chung W H Liu S Tam H and Ho S L Real-time train wheel condition monitoring
by fiber bragg grating sensors International Journal of Distributed Sensor Networks vol 2012
pp 1–7 2012
[28] Nenov N G Dirmitrov E N Mihov G S Ruzhekov T G and Piskulev P M A study on sensors for
measuring load of railway vehicle wheels in motion in Electronics Technology 2008 ISSE ’08
31st International Spring Seminar on 2008 pp 550 –555

12