Note: Descriptions are shown in the official language in which they were submitted.
CA 02573435 2006-12-01
WO 2005/118366 PCT/GB2005/001600
Processing of Railway Track Data
This invention relates to an apparatus and a method
for processing data, in particular data obtained by
monitoring a railway track, such data for example being
used for assessing the quality of the track.
Track recording vehicles are known, which are used
in surveying a railway track to provide data representing
the undulations of the rails in the vertical and
horizontal planes, and their curvature. Software
packages are also available, for example a software
product under the trade mark VAMPIRE (from AEA Technology
plc), for predicting how a particular vehicle will
respond when travelling at a particular speed along a
track; such software packages, which may be referred to
as vehicle dynamics simulations, require input data
providing an undistorted representation of the track.
The raw data obtained by the sensors on a track recording
vehicle provide information about train movement, and can
be processed to determine track data, in particular being
filtered to distinguish between short wavelength data and
long wavelength data. This filtration process may
introduce phase differences. Data from such a track
recording vehicle can be subjected to a subsequent
filtration process, referred to as "back filtering", to
obtain accurate data about the track. However, this
process requires all the data about an entire section of
track (which might be say 200 km long), and this entire
data stream is then processed in reverse; clearly this
can't be done in real-time.
According to the present invention there is provided
a method of obtaining data on the quality of a railway
track, the method comprising:
CA 02573435 2006-12-01
WO 2005/118366 PCT/GB2005/001600
- 2 -
a) receiving from a track recording vehicle data
concerning variations of a parameter, the data comprising
samples, obtained in either the spatial or the temporal
domain, which have been subjected to a filtration process
having an associated transfer function (H);
b) selecting a transfer function HT which inverts at least
the phase differences of the transfer function H of the
filter;
c) temporarily storing a multiplicity (N) of
sequentially-received samples in a memory, each said
sample being stored with an indication of the
corresponding position or time;
d) generating an output data sample by calculating the
integral of the product of the stored data samples with
an impulse function (F), wherein the impulse function is
deduced from the selected transfer function HT according
to the equation:
HT Ow)= f F(t~i'dt
if time (t) is the appropriate variable, or, if expressed
in terms of distance (s):
HT(Jo))= fF(s)e' 'sds
e) storing the next successive sample of data in the
memory and deleting the oldest sample stored in the
memory, and repeating the step of generating an output
data sample; and
f) repeatedly performing the preceding step.
CA 02573435 2006-12-01
WO 2005/118366 PCT/GB2005/001600
- 3 -
Preferably the multiplicity (N) is an odd number;
and preferably the impulse function is centred on the
middle sample of those stored, that is ((N+1)/2)th sample
if N is odd. It should be appreciated that the impulse
function need not be a symmetrical function; it is
'centred' in the sense that it is a function not of
absolute time (or distance) but of the time (or distance)
relative to that of a specific stored sample.
The method described above enables a series of
output data samples to be generated substantially in
real-time, the only delay being that taken for the
receipt of ((N+1)/2) samples. By appropriately selecting
the impulse function, F, the effect of the filtration
process on phase, or indeed on both amplitude and phase
of the data, can be eliminated.
This method may be performed within a track
recording vehicle. For example it can enable amplitude
and phase distortions of track geometry signals to be
removed, so that the corrected signals can be used as
input for a vehicle dynamics simulation. Another
application is that, once amplitude and phase distortions
of track geometry signals have been removed, the signals
correctly represent the shape of track features such as
dipped rail joints, and so can be used to guide track
maintenance. The method of the invention can also remove
distortions due to anti-aliasing filters.
The present invention also provides an apparatus for
performing this method.
For example, the method of the invention may be used
to provide input data to a vehicle dynamics simulator
carried in a track recording vehicle, so that the
simulator can deduce the risk of derailment of a
CA 02573435 2006-12-01
WO 2005/118366 PCT/GB2005/001600
- - 4 -
particular type of vehicle in substantially real-time.
The vehicle dynamics simulator could give a warning
signal if the corresponding simulated vehicle would be
derailed. Hence the track survey vehicle can,
substantially in real-time, provide warnings of track
sections that would give high derailment risk for a
particular type of vehicle at a particular speed.
Warnings might also be given if the simulated
vehicle would subject passengers to unacceptable jolts,
or if the simulated vehicle would subject the portion of
track to unacceptable track forces, and such information
could also be reported as soon as the vehicle has passed
over that section of the track. This enables track
maintenance to be targeted at those sections of track
most in need of improvement.
The invention will now be further and more
particularly described, by way of example only, and with
reference to the accompanying drawings which represents
as a block diagram apparatus incorporating the present
invention.
In this example, an apparatus incorporating the
present invention is installed in a track recording
vehicle 10, that is to say a rail vehicle incorporating
transducers monitoring displacements and accelerations of
the bogie and/or the body as the vehicle 10 moves along
the track 11. For example it might incorporate an
accelerometer monitoring vertical accelerations of the
bogie, and a displacement transducef monitoring vertical
displacement of the axle relative to the bogie; data from
such transducers would enable undulations in the vertical
plane of each rail of the track to be monitored.
Similarly accelerometers measuring horizontal
accelerations, along with a displacement transducer to
CA 02573435 2006-12-01
WO 2005/118366 PCT/GB2005/001600
- 5 -
monitor the wheel relative to the bogie, enable
undulations of the track in the horizontal plane to be
monitored. Track recording vehicles normally incorporate
several different transducers, data from the transducers
being sampled every 1/8 m and digitized, and the output
data may involve calculations that combine data from
several such transducers. In any event the data is
subjected to signal processing (represented
diagrammatically by box 12) that includes filtration so
as to generate track data, which would typically be
displayed to an operator, for example using a graphical
interface, and stored for subsequent processing. The
data may also be stored in conjunction with data from
other sensors, for example positional data from a GPS
sensor.
As regards the lateral plane, the data typically
would represent alignment (a measure of the offset of the
rails from the required smooth curve, measured in mm),
and curvature (indicating the reciprocal of the radius of
the curve followed by the track, measured in km-i).
Typically the cutoff wavelength is set at 70 m,
horizontal displacements of shorter wavelength than this
being treated as alignment, and horizontal displacements
of longer wavelength being treated as curvature. As
regards the vertical plane, the data typically would
represent "top" (a measure of the displacement of the
rails from the required smooth curve, measured in mm),
and gradient (indicating the slope of the track, in
mm/mrn). The cutoff wavelength in this case is typically
also set to 70 m.
In the apparatus shown, the track data streams from
the processor 12 representing alignment, curvature, and
top (and possibly also gradient), and possibly other data
streams such as positional information, are transmitted
CA 02573435 2006-12-01
WO 2005/118366 PCT/GB2005/001600
- 6 -
to a data post-processing server 14, and thence to a
reporting server 16, and so to various display interfaces
18 and to a data store 20.
Data streams representing alignment, curvature, and
top (and possibly also gradient) are also supplied by the
post-processing server 14 to several different vehicle
dynamics modules 22 (three such modules are represented).
Each such module 22 consists of a microprocessor arranged
to model the dynamics of a particular vehicle travelling
along the track 11 at a particular speed. The output of
these vehicle dynamics modules 22 is fed back to the data
post-processing server 14, and is supplied to the
reporting server 16 along with the corresponding track
data (processed as described below).
The data post-processing server 14 is programmed to
subject the track data streams from the processor 12 to
the filtration process of the invention.
As mentioned above, the processor 12 is used to
separate high frequency (short wavelength) components
from low frequency (long wavelength) components.
Analogue filters or digital infinite impulse response
(IIR) filters can perform these tasks efficiently, but
they introduce distortion. Methods are known to
eliminate this phase distortion, either avoiding it by
using finite impulse response (FIR) filters instead of
IIR filters, or by back filtering the already distorted
data with an identical IIR filter to restore the original
phase content. Hbwever, there are cases where a signal
has already been distorted by an analogue or IIR filter,
and an undistorted signal is required. This is taken to
be the case here.
CA 02573435 2006-12-01
WO 2005/118366 PCT/GB2005/001600
- 7 -
The server 14 performs signal shaping of the
incoming data, and forwards it to the rest of the system
for storage and/or further processing. The signal
processing method can deal with both spatially and
temporally sampled data streams. It can also perform
'cross-domain' operations, as well, that is to say to
perform temporally defined operations in spatially
sampled (taken at equal distances) data, and vice versa.
The server 14 consists of:
- Digital input and output interfaces
- A buffer memory to store N samples of the data stream,
including the measured value and a time or distance
stamp, indicating the time or distance the measurement
was taken. The type of the stamp data depends on the
actual operation: if temporal operation is needed, then
time stamp, if spatial operation is needed the distance
stamp has to be attached to each measured value. The
actual sampling method (equal time or equal distances)
does not affect the operation of the filter. For example,
usually the measurements are taken at equal distances, so
if the vehicle speed is increasing, then the differences
between the consecutive time stamps will decrease, but
the system operation will not change.
- Memory to store the parameters of the calculations.
- Arithmetic processing capability.
It will be appreciated that the details such as the
data transfer protocols, memory type etc. must be
adjusted to the system in which the server 14 is used.
In certain cases it may be a separate instrument
connected to the data bus of the measurement system, in
other cases it may be fully integrated into the
measurement system.
The operation of the server 14 is as follows:
CA 02573435 2006-12-01
WO 2005/118366 PCT/GB2005/001600
- 8 -
1. The samples of the incoming data are stored in an N-
element first-in-first-out (FIFO) buffer, which is
initialized with zeros as measured values. Each new
sample enters the first slot of the buffer, moving the
previous measurements one slot forward. The data that had
been in the Nth slot is deleted, since it is replaced by
the one coming from the (N-1) th slot. N is preferably odd.
2. After the new data sample is inserted into the
buffer, the following calculation is performed:
TZ
Y"To )= fF(t - To )1X(t)dt
T, Eq.1
where:
Y(To) is the output data, time stamped as taken at To.
To is the actual time stamp of the ((N+1)/2)th data in
the buffer. In a certain sense, the calculation above is
centred on To, and the output data stream is always
delayed by (N+1)/2 samples.
T1 is the time stamp of the oldest (Nth) data in the
buffer.
T2 is the time stamp of the latest (15t) data in the
buffer. It is also true, that TI<To<T2.
X(t) is the data stream stored in the buffer.
F(t) is the finite impulse function, derived from the
desired restoration. F(t) is integratable between any
possible t values.
It will be appreciated that Equation 1, which is
expressed above as an integral (implying continuous
functions), must in practice be performed as a
summation, by a suitable discrete calculation method.
Since each sample is processed separately, and has an
associated time stamp, if the time intervals or spatial
distances between successive samples vary, or there are
randomly missing samples, overall operation is not
affected. This is a significant advantage.
CA 02573435 2006-12-01
WO 2005/118366 PCT/GB2005/001600
- 9 -
Eq. 1 is shown in the temporal domain. The formula
is still valid in the spatial domain, where the time
values have to be replaced with distance values:
sZ
Y(So)= fF(s-So)X(s~s
S, Eq. lb
3. The calculated output is forwarded for further
processing.
The operation clearly depends on correctly
determining the impulse function, F(t) or F(s). The
impulse function is defined from the desired system
behaviour, described by a transfer function. Transfer
functions are complex equations that describe the system
behaviour as a function of the cyclic frequency, w. If
H(jc)) is the transfer function of a filter, then:
IH(jo))l is the ratio of the output to the input
amplitude,
~(H (jc,o) ) is angle of the phase delay,
where j is the square root of -1.
The selected transfer function HT is one that
reverses at least the phase change, and may also be
selected so as to return the amplitude to its original
value. The relationship between the transfer function and
F(t) is:
HT0co)= fF(t)e' ''dt
-W E q . 2
The equation above has to be solved for F(t). Analytical
and numerical solutions are both suitable to get a
functional F(t), and some examples are discussed below.
CA 02573435 2006-12-01
WO 2005/118366 PCT/GB2005/001600
- 10 -
The final step is to define the size of the buffer
memory. First we calculate T1 and Th, such that the
following approximation will be true:
T
HT 0ro)= f F(t)e'wtdt,,:e f F(t~*dt
T, E q . 3
Once T, and Th are found, the size of the buffer (N) can
be calculated as follows:
1. The temporal window (time period) over which
integration is performed is Tw=Th-Tl.
2. The number of samples in this time period will
change as the vehicle changes speed, but if the
maximum speed of the vehicle is known, then the
number of samples will not exceed
N = Tw=(Top speed)=(Samples per metre)
3. If the vehicle is going slower than the top speed,
some of the stored samples will fall outside this
specified time period. However, Eq. 3 shows that we
can take F(t)=0 for such samples.
This derivation assumes operation in the temporal
domain. If spatial domain operation is needed, F(s) can
be generated by replacing the temporal terms with spatial
terms, as in Eq. lb.
Example 1: Restoring the original phase content of an
anti-aliased signal
This describes an operation in the temporal domain..
As mentioned earlier, a track recording vehicle 10
will include various transducers which measure aspects of
the vehicle movement, such as an accelerometer, gyroscope
etc. Typically the signal from such a transducer, which
CA 02573435 2006-12-01
WO 2005/118366 PCT/GB2005/001600
- 11 -
is an analogue signal, is first fed into an anti-aliasing
filter, in order to avoid interference of high frequency
signals with the digital sampling rate, called aliasing.
Anti-aliasing filters are low frequency pass analogue
filters, eliminating the undesired frequency content.
The data processor 12 would then produce digital output
signals by sampling the analogue signal at equal
distances along the track. Anti-aliasing is essential,
but it introduces a non-linear phase delay of the
incoming signal. This phase delay will distort the shape
of the signal. Until now back-filtering was only the way
to restore the original phase content. However, back-
filtering changes the amplitudes in the transition band
and cannot be used if the results are needed in real
time.
The transfer function H of the analogue anti-
aliasing filter can be given by the amplitude and phase
responses as a function of the cyclic frequency:
A(c.o) = IH(Jco) I , and ~(cu) =~(H(j(o) ), Eq.4
where (o is in radians per second. These two functions can
be analytically derived, or measured. We must select or
create a transfer function HT which leaves the amplitude
intact, but reverses the phase delay. Hence the amplitude
and phase responses of the selected transfer function HT
should be as follows:
. AT((o) = 1, ~T(cO) Eqs. 5
This is satisfied by the transfer function:
HT(j(O) = cos (~T((O) )+jsin(~T(CO) ) = cos (-~(co) )+jsin(-~((o) )
Eq.6
CA 02573435 2006-12-01
WO 2005/118366 PCT/GB2005/001600
- 12 -
Knowing the target transfer function, F(t) and N can
be calculated as described above. Once these have been
calculated, the server 14 can restore the original phase
content of the incoming signal.
Example 2: Restoring broadband curvature signal from
asymmetric versine input
This describes an operation in the spatial domain.
If we model the railway track as a planar curve, it
may be described by its curvature. Curvature for any
planar curve is defined as:
C(S)= dzPW
z
, ds Eq. 7
where p is the vector pointing to a location on the
track, s is the path taken on the track. Usually, the
track curvature is split into long and short wavelength
parts: the long wavelength part describes the track
design, all the bends and straight sections needed to
lead the train from A to B, while the short wavelength
part describes the local deviations from the design,
affecting the ride quality along the track.
It is difficult to measure curvature directly, so
different indirect methods are used. One of them is
asymmetric versine; the asymmetric versine, v, is
measured by considering a fixed length chord between two
points on the rail. The chord is divided by a point Y
into two unequal parts, L1 and L2, and v is the distance
of the rail from the point Y measured along a line
perpendicular to the chord. Asymmetric versine is easy
to measure both manually and automatically. It gives a
CA 02573435 2006-12-01
WO 2005/118366 PCT/GB2005/001600
- 13 -
broadband description of the lateral track geometry,
recording both short and long wavelengths components in
the same output. Unfortunately, to determine curvature
from asymmetric versine a complicated transfer function
is required, which also introduces phase distortion.
Previously-known methods were unable to give a proper
reconstruction of curvature from versine in real time.
The server 14 can be configured to reproduce
broadband curvature from digital asymmetric versine input
in real time.
The transfer function from curvature to versine is:
Hcv 0(0) _ - 1 a 1- L2 e~L L, ejL,o)
m Li +Lz L1+L2 E s
q=
where co is in radians per metre, and Ll and L2 are in
metres.
The inverse transfer function:
r 1
HvclJ(0)
Hcv Eq. 9
is the required transfer function (i.e. the selected
transfer function HT), and hence F(s) and N can be
calculated as described above. Once these are
calculated, the server 14 can restore the original
curvature.
It will be appreciated that a track recording
vehicle 10 might include several such vehicle dynamics
modules 22 operating in parallel, for example twelve
rather than the three modules 22 shown here. Operation
of this one vehicle 10 is therefore equivalent to running
CA 02573435 2006-12-01
WO 2005/118366 PCT/GB2005/001600
- 14 -
a fleet of a dozen different vehicles that may use this
particular route, each at their own speed, and each of
the virtual vehicles is effectively instrumented for
assessing the risk of derailment, and also other
parameters such as passenger comfort, track forces,
vehicle kinematic movements etc.. This information is
obtained in real-time, and is reported as part of the
data provided to the display interfaces 18 as soon as the
track recording vehicle 10 has passed over a portion of
the track 11. The information is embedded in the same
stream of data as the information on track geometry.
Hence it can be readily interfaced to track management
software.
Although the method has been described as being
performed within a track recording vehicle 10, and so
giving information in real-time, it will also be
appreciated that data previously obtained using a track
recording vehicle 10 may be supplied later to such a
phase and amplitude correction microprocessor (equivalent
to the post processing server 14), and hence if desired
to a plurality of vehicle dynamics modules 22.
