Real Time Vampire ® - Derailment Risk Indicator and
Automatic Track Geometry Accuracy Assessment
S. Patko, K. Hope, B. Gilbert, AEA Technology Rail, Derby, UK
Abstract
Dynamic vehicle simulation using Real Time Vampire® (RTV) is installed on two, highly sophisticated
track recording vehicles in the UK and in France. Once in operation, the operators of these track
recording systems will benefit from the proven capabilities of the high quality vehicle simulation in real
time. RTV provides validation of the recorded track geometry through simulation of the response of the
track recording vehicle itself and comparing that to the behaviour recorded by the on-board transducers
which could not be achieved by other means. A second use of the application is the simulation of various
service vehicles over the measured track to determine their response to the track geometry and hence
detect areas of high derailment risk. This risk assessment is of value as certain track features can lead to
derailment without producing an alarm using the traditional track geometry assessment methods. The
results of the simulation can be visualized in high quality 3D using Vampire® Animator Pro. The
assessment of risk is carried out by the programmed functionality of the Derailment Risk Indicator TM
software. Off line trials using track data from North-American and South-American railway companies
have shown the insight into actual vehicle behaviour over the measured track geometry that this new
technology can provide.
Introduction
Derailment of rail vehicles is a serious and expensive issue for the rail industry. The current approach of
controlling derailment risk by maintaining the track to fixed limits fails to prevent all derailments. It is
generally accepted that the limits are conservative for most vehicle/track combinations, leading to a
higher cost of track maintenance. Despite this derailments have occurred where track geometry
parameters have been inside these limits
On many railway networks, derailments are routinely reconstructed after the event using vehicle dynamic
simulation tools, e.g. Vampire®, to establish root causes. Running such simulations on board track
recording vehicles means that potential derailments can be identified (and therefore prevented) before
they occur. However, with many simulation tools significant simplifications to vehicle models and analysis
algorithms are needed for such analysis to be carried out in real time during track recording. RTV has
been developed to overcome these problems. It has the well-validated and fully non-linear Vampire®
analysis engine at its heart, ensuring that vehicle models are readily available, highly accurate, and have
as many or few degrees of freedom as required. RTV has been used for real time operation in the UK on
Network Rail’s New Measurement Train (NMT) and also in post-processing mode in their Engineering
Support Centre. It is also being installed for real-time operation on the new SNCF high-speed recording
vehicle (MGV). The following chapters describe the two different system designs that were developed
according to the different requirements of Network Rail and SNCF and show the software components
that enable the track maintainers to see easy to understand visualizations of the results.
The system layout of the NMT
Network Rail’s New Measurement Train is one of the world’s most sophisticated railway infrastructure
recording vehicles. In the NMT the Production Vehicle is responsible for recording track geometry and
producing reports for track maintenance activities. This vehicle has three independent track recording
systems installed and supported by AEA Technology Rail’s TrackSysTM track geometry processor. A full
set of track geometry signals are provided by Laserail3000TM, an optical-inertial system manufactured by
Imagemap (USA). Track geometry signals with the exception of gauge and alignment are measured by
TrackLine, a mechanical-inertial system from AEA Technology Rail (UK). The third system installed is the
1/10/
UGMS-Hybrid solution which is an unattended track measurement system, a combination of
ImagemapTM’s UGMS and AEA Technology Rail’s TrackLine.
For reporting within the formal track safety regime, Laserail3000TM, TrackLine and TrackSysTM are used.
TM
TM
The geometry is predominantly taken from Laserail3000 and TrackLine, while TrackSys creates the
reports with positioning information and validates the measured geometry. TrackLine also acts as the
back-up system. As TrackSysTM can also undertake confidence validation of the channels measured by
both system, the scope of RTV has been focused on derailment risk prioritization and rather than
validation.
RTV can receive the track geometry signals from the mathematical processing component of TrackSysTM.
The most important aspect of the signal processing before the vehicle dynamics simulations can be
performed is to produce complimentary alignment and curvature channels from the Laserail3000TM
alignment and curvature outputs. This is achieved using symmetric finite impulse response (FIR) filters.
These outputs then are forwarded to the RTV processor box, where the simulation takes place in real
time. The calculated derailment risk indicator figures are sent back to TrackSysTM, where they are
inserted into the track geometry stream and sent to the reporting subsystem. Fig. 1 details this process:
Laserail3000 track geometry
TrackLine transducer signals
Positioning information
TrackSys processing
Data stream
generation from
the input sources
Mathematical
processing
Report
generation
Outputs
Real Time Vampire processing
Interface for real
time TrackSys
data stream
Vampire
processing
Interface for
returning data
to TrackSys
Fig. 1: The data stream between TrackSysTM and the RTV slave computers
In this setup the RTV processing is an extention of the TrackSysTM mathematical processing that is
redirected to a separate computer. This separation will not be necessary in the future as increasing
computing power will enable the full incorporation of RTV into the TrackSysTM processor. However, it
does allow RTV to operate with any system capable of providing the necessary track geometry data.
Whilst this method places the RTV outputs and the main track geometry in the same reports, it has a
drawback in that TrackSysTM has to generate the data channels for RTV using symmetric FIR filters.
These filters cause considerable spatial delay in the data stream; additionally TrackSysTM has to wait for
the RTV processor to return the results. These two delays may amount to several hundred metres so the
real time charts and reports on a track recording system using this configuration will be available for
viewing later than without RTV processing.
The system layout of the MGV
The high-speed measurement train (MGV) of SNCF uses just one track recording system, ImagemapTM’s
Laserail4000TM with TrackSysTM to generate the reports. The other important difference in requirements is
that the system has to produce defined outputs within 250m whilst travelling at 320kph. Processing
delays within the acquisition system at this speed means that is no processing time available for the FIR
filtration within TrackSysTM for RTV. An alternative processing architecture has therefore been deployed.
TrackSysTM is still the main hub of all information. However, after the track geometry signal has been
associated with the positioning information, it is processed in two different ways. One processing branch
TM
uses TrackSys functionality to produce standard track geometry reports as soon as possible for the
2/10/
availability of the real time charts. The other processing branch provides data to the RTV processor boxes
for simulation. The related mathematical pre-processing takes place in the separate computers followed
by the RTV simulation. The report summarizing the RTV results and any subsequent analysis can then be
saved with the output from the standard track geometry branch. Fig. 2 shows the layout of this process.
Laserail4000 track geometry
TrackLine transducer signals
Positioning information
TrackSys processing
Data stream
generation from
the input sources
Mathematical
processing
Report
generation
Outputs
Report
generation
Outputs
Real Time Vampire processing
Interface for real
time TrackSys
data stream
Vampire
processing
Fig. 2: The alternative solution used on the MGV
Track geometry validation using RTV
An important feature of the Vampire® technology is that it enables the usage of virtual transducers. Using
virtual transducers it is possible not only to build a model of the vehicle but also to simulate the transducer
signals of the track recording system. All types of transducers can be simulated - accelerometers,
gyroscopes and displacement sensors. Until recently virtual transducers in Vampire® were used for virtual
prototyping of track recording vehicles. The simulated outputs could be used to design and test the digital
signal processing algorithms used for track geometry reconstruction. In RTV, however, the virtual
transducer signals are used to validate the recorded track geometry.
The validated accuracy of the Vampire® technology makes it possible to simulate the transducer signals
of the track recording vehicle with great accuracy provided that the recorded track geometry is accurate.
Using RTV the comparison of the raw transducer signals of the track recording system and the simulated
transducer signals can be undertaken on board the track recording vehicle during the recording session
itself. Inaccuracies of the recording are detected immediately and the system operator can be warned
about the deteriorated condition of the recorded track geometry.
Calibration error can, for example, lead to offset in the track cant signal. If this offset is small it may not
even be detected, as cant signals are seldom used for alarm generation and the printed scale is usually
smaller than would allow detection of the offset. However, when comparing the average of the real and
simulated lateral accelerometers over a short piece of track, the offset can be automatically detected and
reported to the operator.
Another recurring example of corrupted track geometry records is where there are missing samples.
Several causes can lead to this condition; for example data processing errors in the software can result in
lost samples in the geometry stream. Similarly, drop-outs which are difficult to detect can result from the
very brief disturbance of the cameras used to measure the displacement of the rails. Where there is a
very short loss of visible rail profile, the recording system usually substitutes the missing values using
interpolation; this is a tolerable approach as the only other option would be to drop the outputs to zero
values. Often these lost or damaged samples are undetected for years because a few such samples in
the geometry data cannot be seen in overview charts and they rarely lead to alarm reporting. However
even a single lost sample will lead to an unexpectedly rapid change in the recorded track geometry that
will produce a stronger acceleration response in the virtual vehicle than that observable in the real
3/10/
system. Additionally, if the interpolated samples in the case of camera drop-outs are inaccurate, the
simulated train behaviour will be significantly different from that of the real train. While these hidden faults
in the geometry data will normally go unnoticed, sudden high peaks in the virtual signals are easily
detected and reported. Fig. 3 shows an extreme example of corrupted track geometry.
Fig. 3: Close-up of errors in recorded track geometry
Similar features when only a few samples are affected cannot be detected visually
Fig. 4: Single missing sample in the track geometry resulted in violent simulated response
Visualization of RTV results with Animator ProTM
During the course of the measurement session the easiest and most spectacular way to see the result of
the Vampire® simulation is to use Vampire® Animator ProTM to show the 3D model of the simulated
vehicle travelling on the measured track. The animator displays the track and, optionally, 3D models of
track-side features found in the route information provided. The operator can freely choose his point of
view; for example, he can zoom close to one of the wheels and observe lost contacts or can have an
overview of the entire car or rake. The corruption of track geometry data is easiest seen in the animator
4/10/
screen because the violent response of the vehicle model is impossible to miss, so the operator can take
the necessary actions to improve the state of the recording system immediately. Figs. 5 and 6 show an
overview and a close-up view of the simulated vehicles shown by Animator ProTM.
Fig. 5: Overview of a hopper
Fig.: 6: Close-up of the contact area
5/10/
The Derailment Risk Indicator: the tool for the track maintainer
The complex analysis in RTV output data is summarized in reports that contain the location and the value
of the dynamic features that indicate derailment or possible derailment during the simulation. This data
TM
file is visualized using the Derailment Risk Indicator (DRI) software developed for the RTV system. The
DRI provides easy access to the calculated data and aids decision-making in track maintenance by
offering a number of ways to display and sort the information. The chart view gives visualization of the
number of detected risk locations over the measured track. This enables the maintainer to have a quick
overview of the results and choose the locations highlighted by the selected derailment risk parameters.
More detailed analysis is possible in the table view, where the individual detections can be seen. In
addition to selecting and sorting the data the maintainer is interested in, clicking the “Details” button
shows track geometry in the area surrounding of the derailment risk parameter alarm.
Figs. 7 and 8 show the two main views:
Fig. 7: DRI chart view
6/10/
Fig. 8: DRI table view
Analysis of derailment risk on North-American track sections
In 2004 a North-American railways (NA) commissioned an off-line pilot study into the applicability of RTV
technology. For the study NA provided two track sections. On both tracks three different vehicle models
were tested:
- “Good” tank car, i.e. the vehicle in design condition
- “Bad” tank car, that had a twisted bogie setup
- 286000 lb covered hopper
During the simulations the vehicles were run at three different steady speeds - 60mph, 55mph and
50mph, giving a spectrum of usual travelling speeds.
Three different derailment risk indicators were used to assess the vehicle behaviour:
- Wheel unloading (DQ/Q)
- Lateral/Vertical wheel load (Y/Q)
- Wheel lift
The assessment was done through counting the number of alarms reported using the three derailment
risk indicators in 1/8th of a mile sections.
This study had produced several results that were anticipated from previous experience, and also a
number of surprising ones.
The best overall performance was produced by the covered hopper having negligible number of sites with
risk of derailment. The good tank car usually ran better on the track than the bad one. These results were
according to experience. However, on a few locations, the good tank car produced more DRI alarms than
the bad one.
7/10/
th
The 1/8 mile sections were classified by NA into six classes from 5 (best) to 0 (worst). One would
normally anticipate strict correlation between the traditional classification and the risk of derailment. The
investigation, however, did not support this view. For example, there were no DRI alarms reported by any
vehicles between mileposts 304 and 305 on the first track in spite of the fact that one section was
classified as bad as class 1. On the other hand, several alarms were found between mileposts 308 and
309 of the same track, while this whole mile is classified of the best quality.
The quick method to mitigate derailment risk where track faults are found is to lower the permitted vehicle
speed. This study has shown that it may not be a good solution in some cases. In general, running at
50mph resulted in much lower number of derailment risk alarms. However, , reduction from 60mph to 55
resulted in more alarms for a few sections – this was possibly because of a resonant performance
feature.
The overall conclusion of the study is that due to the complexity of the vehicle behaviour over a given
piece of track the traditional quality figures may not indicate the true risk of derailment. Vehicle dynamics
simulation, however, can confirm the quality of the track and draw attention to areas where the traditional
methods failed to accurately predict the risk.
Fig. 9: The chart showing the unexpected behaviour of the good tanker model over
the second track provided
Off-line RTV trial commissioned by a South-American railway company (SA)
This analysis was commissioned by SA as an off-line trial of RTV using actual SA track data to assess the
derailment risk of hopper cars used on SA routes at typical operating speeds.
RTV worked well with SA track geometry data. The track was divided into 100m long sections, each
having a track quality classification from 5 (best) to 0 (worst).
There were three vehicle models used for the simulation. All three were developed from the same hopper
design.
- “Good” hopper car, i.e. the vehicle in design condition unloaded (tare)
- “Good” hopper car, fully loaded (laden)
8/10/
- “Worn” hopper car. unloaded (tare)
The tare vehicle models were run at 40 and 50kph; the laden model was tested at 30 and 40kph.
The examined derailment risk parameters were the same as in the study for NA, i.e.:
- Wheel unloading (DQ/Q)
- Lateral/Vertical wheel load (Y/Q)
- Wheel lift
The main conclusion of the report was that the standards based entirely on fixed limits applied to track
geometry parameters are not always a reliable indicator of safety, because there was a great variation of
track classes on top ten worst locations from a derailment risk perspective. The simulations have shown
that there was a considerable difference between the vehicle response in all aspects of the variables. The
state of the cars, their loading and the running speed all made observable changes in the derailment risk
indicators. Eight out of the worst ten locations, in terms of derailment risk, on the route were all found to
pose a risk only to tare vehicles (both design case and worn) running at 50kph. A reduction in operating
speed to 40kph for both vehicle types was demonstrated to avoid the high risk of derailment at these
locations. This shows how using RTV makes it possible to determine an operating regime which avoids
high derailment risk.
tare,50
laden,40
tare,40
laden,30
worn,50
worn,40
Low speed
Track Class
6
20
5
16
4
t
n
u
o
c
I
R
D
s
s
a
l
c
12
3
8
k
c
ra
t
A
R
F
2
4
1
0
350.0
355.0
360.0
365.0
370.0
375.0
380.0
385.0
390.0
395.0
0
400.0
Distance (km)
Fig. 10: Chart showing that the sites with derailment risk are often classified as good quality track
Conclusions
The recent increase in the power of vehicle dynamics simulation has enabled the development and
application of RTV to produce reliable analysis of vehicle behaviour on board track recording vehicles. In
off-line studies, the system has been shown to be able to reveal errors in the track geometry signal that
were previously difficult to detect. The simulation can also classify the tracks according to the actual risk
of derailment of the modelled vehicles that do not always correlate with the traditional track quality
TM
indicators. The results are displayed by two different tools, Animator Pro and DRI. These tools are easy
to use when scheduling the track maintenance activities. The process of derailment risk indication, as
9/10/
®
TM
delivered by AEA Technology Rail’s Real Time Vampire - Derailment Risk Indicator , can therefore
contribute considerably to railway maintenance and safety regimes.
10/10/