GB2516663A - Train speed determination - Google Patents

Abstract

The speed of a train 1 is determined from the output from two vibration sensors 3, 4 at spaced-apart locations a known longitudinal distance d apart inside the same car - or different cars - of the train. The vibration sensors measure vibration during movement of the train; the measured vibration patterns A1, A2 are correlated to identify patterns detected by each sensor relating to a common source; the time difference is determined between measurement of the identified patterns at each sensor; and the time difference is related to the distance between the vibration sensors to determine the speed. The sensors may be located over respective bogies. Weighting may be applied to the sensor outputs. The speed may be combined with an independently determined train speed.

Description

Intellectual Property Office Applicacion Nc,. (lB 1313505.5 RTM Dacc:20 Dircinbcr 2013 The following terms are registered trade marks and should he rcad as such wherever they occur in this document: Wi-FL Bluetooth Inlelleclual Property Office is an operaling name of the Pateni Office www.ipo.gov.uk Train Speed Determination This invention relates to a method for determining the speed of a train, train speed determination apparatus and a train.

Background

Speed measurement and positioning systems are key elements in railway operation as parts of the Automatic Train Control (ATC) and Automatic Train Protection (ATP) systems, since braking curves and maximum distance permissible between trains are calculated through train position and speed. Therefore, improvements in the measurement precision and quality will result in safer, faster and more efficient railway networks. Desirable characteristics of position and speed estimation systems include: reliability, high precision, accuracy and availability over different weather conditions, low latency, protection against electromagnetic interferences, high data-rate, low cost and ease of installation and maintenance.

Unfortunately, to date, no single sensor is able to meet all these requirements.

Therefore several disparate technologies of measurement and integration are commonly used, leading to multisensor systems. The most frequent technologies used include balises, tachometers, track circuits, Doppler radar, inertial navigation systems, global navigation satellite systems, and artificial vision. Advantages and drawbacks of each may be balanced by using sensor fusion systems with or without map-matching strategies.

A known-type of speed estimator system is based on the delay in the pass through irregularities of the rail track. In this model two sensors are placed separated by a distance d, in the direction of motion, and convert rail irregularities into two signals xl and x2. Ideally x2 is only a delayed version of xl. If the delay time is t then the speed can be estimated as V=d/t. This estimator system has been proposed for several types of sensors including optical, eddy current and vibration sensors. In particular respect of the use of vibration sensors, the use of vibrations produced by wheel/rail interaction to detect rail irregularities has been studied and a speed estimation method based on the delay in the excitations of two wheel-sets of the same bogie has been proposed -in other words it has been shown to be theoretically practical to use bogie-mounted vibration sensors to determine train speed.

However, the practical installation and maintenance of a system with sensors in the bogies is complex. The sensors are exposed to very hostile mechanical conditions, and carrying signals to a processing unit presents several technical problems.

As prior art may be mentioned:

1] George Achakji,, <(A Review of State-of-the-Art Train Control Systems Technology>>. Transportation Development Centre Safety and Security Transport Canada, Mar-i 998.

[2] P. Liljas, <<Speed and positioning systems. The traditional way>), presented at the lEE Colloquium on Where Are We Going? ( And How Fast!) Seminar Exploring Speed And Positioning Systems For The Transport Sector (1997/395), 1997, págs.

2/1-2/9.

[3] A. J. Beesley, <<Distance/velocity measurement by Doppler [rail traffic control]>>, presented at the lEE Colloquium on Where Are We Going? ( And How Fast!) Seminar Exploring Speed And Positioning Systems For The Transport Sector (1 997/395), 1997, págs. 5/i -512.

[4] Richard Shenton, <<Train Video Positioning>>, presented at the IRSE Aspect, London, 2008.

[5] A. Mirabadi, N. Mort, y F. Schmid, <<Application of sensor fusion to railway systems>>, in IEEE/SICE/RSJ International Conference on Multisensor Fusion and Integration for Intelligent Systems, 1996, 1996, pegs. 185-192.

[6] A. Acharya, S. Sadhu, y T. K. Ghoshal, <<Train localization and parting detection using data fusion>>, Transportation Research Part C: Emerging Technologies, vol. 19, no.1, págs. 75-84, Feb. 2011.

[7] K. Gerlach y C. Rahmig, <<Multi-Hypothesis Based Map-Matching Algorithm for Precise Train Positioning>>, Jul-2009. 12th International Conference on Information Fusion.

[8] Samer S. Saab, <<A Map Matching Approach for Train Positioning Part I: Development and Analysis>>, IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, vol. 49, Mar. 2000.

[9] J. Bohmann, H. Meyr, R. Peters, y G. Spies, ((A signal processor for a noncontact speed measurement system>>, IEEE Transactions on Vehicular Technology, vol. 33, no. 1, págs. 14-22, Feb. 1984.

[10] E. Thomas, ((Design of a correlation system for speed measurement of rail vehicles>>, Measurement, vol. 29, no.2, págs. 157-164, Mar. 2001.

[11] T. X. Mei y H. Li, ((Measurement of Absolute Vehicle Speed With a Simplified Inverse Model>>, Vehicular Technology, IEEE Transactions on, vol. 59, no. 3, págs.

1164-1171, 2010.

[12] J. Real, P. Salvador, L. Montalbán, y M. Bueno, <(Determination of rail vertical profile through inertial methods>>, Proceedings of the Institution of Mechanical Engineers, Part F: Journal of Rail and Rapid Transit, vol. 1, no. 1, págs. 1-10, Ene. 2010.

[13] J. Kawasaki y K. Youcef-Toumi, ((Estimation of rail irregularities>>, presented at the American Control Conference, 2002. Proceedings of the 2002, 2002, vol. 5, págs. 3650-3660 vol.5.

[14] C. Knapp y G. Carter, ((The generalized correlation method for estimation of time delay>), IEEE Transactions on Acoustics, Speech and Signal Processing, vol. 24, no. 4, pages. 320-327, Ago. 1976.

It is an aim of the present invention to provide a speed estimation system based on sensed vibrations but which has reduced practical complexities. This aim is achieved by the use of vibrations sensors located within train cars. In an advantageous embodiment, the sensors are located inside the train over two separated bogies.

In accordance with a first aspect of the present invention there is provided a method for determining the speed of a train, comprising the steps of: providing a first vibration sensor within a car of the train; providing a second vibration sensor within a car of the train, the first and second vibration sensors being positioned at spaced-apart locations a known distance apart along the length of the train; using the first and second vibration sensors to measure vibration during movement of the train; correlating vibration patterns measured by the first and second vibration sensors to identify patterns detected by each sensor relating to a common source; determining the time difference between measurement of said identified patterns at each vibration sensor; and relating the said time difference to the distance between the vibration sensors along the length of the train to determine the speed of the train.

In accordance with a second aspect of the present invention there is provided train speed determination apparatus, comprising a first vibration sensor within a car of the train, a second vibration sensor within a car of the train, the first and second vibration sensors being positioned at spaced-apart locations along the length of the train, means for receiving output signals the first and second vibration signals and processing means for processing the received signals to determine the train speed.

In accordance with a third aspect of the present invention there is provided a train fitted with the apparatus of any the second aspect.

While of course if the vibrations are great enough, they may be detected per se within a car, such vibrations produced by the track irregularities reach the car much more attenuated and the presence of masking noises due to mechanical interferences is frequent.

The interaction of the wheels and the rails is the main source of vibration in railway operations. These vibrations come mainly from rail irregularities and propagate to the bogies and to the car interior through the suspension system, which, in accordance with the present invention, may be modelled as a low-pass filter.

In a preferred embodiment, the system uses measurements taken from vibration sensors such as MEM accelerometers located inside the cars with each one over the vertical of different bogies but inside the cars. The speed is deduced from the delay between corresponding patterns in the accelerometers' output signals. Other sources of vibrations, which act as masking noises, are also present and unavoidable. These are due to, for example, the engines present in the train, movements of passengers and other mechanical vibrations. In spite of these noises and the great attenuation of the desired vibrations, applicant has found that it is possible to estimate the speed in many real situations.

Therefore, the irregularities of the rail tracks can be used to determine speed, using relatively inexpensive sensors (accelerometers) separated by a fixed and known distance (space), and by measuring the delay (time) in the pass-through of these irregularities the speed (space/time) may be determined.

The invention will now be described with reference to the accompanying drawings, in which: Fig. 1 schematically shows a train car including apparatus in accordance with an embodiment of the present invention; and Fig. 2 schematically shows a vibration model.

As schematically shown in the embodiment of Fig. 1, a train car 1 runs on a track 2 at a speed V, the car 1 in this case having two sets of wheel bogies, one proximate the rear of the car and another proximate the front. Respective vibration sensors 3 and 4, e.g. MEM accelerometers, are positioned within the car 1 over respective bogies, so that they are a distance d apart along the length of the train. In use, sensor 3 produces an output signal Al, while sensor 4 produces an output signal A2, which relate to the vibration in the vertical direction as shown. Both of these signals are passed to a signal receiving means such as a processing means 5 such as a computer or the like. The processing means 5 is also operable to process the signals and output information based on the processing result to an external train control system as is known in the art. Also shown is an independent speed determination means 6, which may comprise any known speed-determining apparatus, using for example track circuits, GPS, tachometers etc, and the output of which may be combined with the determined speed from the vibration sensors by processing means 5.

Velocity measurement estimation model While moving, train wheels are under the impact of the rail irregularities which generate vibrations. They are propagated from the wheel-set to the bogies through the primary suspension and from there to the train car through the secondary one.

Vibrations of interest can be observed as accelerations induced in some external parts of the train such as the axles or the bogies, or even inside of the cars.

As described above, in this measurement scheme, two MEM accelerometers 3, 4 are installed inside the train car 1 over two bogies separated by a distance d. The acceleration measurement directions of Al, A2 are approximately vertical.

The model used for this study is presented in Fig. 2: here x(t) is the rail profile signal which arrives directly or with a delay T to each one of the bogies. Suspension systems act mainly as low pass filters, characterized by Fl and F2. n(t) represents a source of common noises that pass through two filters Cl and C2 adding n01(t) and nC2(t) to each vibration. Also independent noises ni(t) and n2(t) are considered in the model.

The main objective is to estimate the delay T between xi(t) and x2(t) in spite of the filter and noise characteristics being unknown. Assumptions on this model and delay measurement methods used to minimize the effect of the filters and the noise are described below.

Delay estimation between signals As a starting point Fl and F2 are supposed linear filters with the same frequency response but different gain, then F2(jQ) = aFi(jO). The same assumption will be made for the filters Cl and C2, then C2(jQ) = 3Ci(jQ). According to this, see Fig. 2, if the output of Fl is yi(t) then the output of the second one will be y2(t) = a.yi(t-T) and the delay T has to be estimated. The common noise satisfies n2(t) = 3ni(t). In this case signals at the accelerometers will follow the expressions: a1 (t) = Yi (t) + ci (t) + n1 (t) a2(t) = ay1(t -T) + 13n1(t) + n2(t) Considering the noises are uncorrelated with the signals, the cross-correlation is: Raia2 (-n) = R12 (-n) + R,112 (-n) + (-&) + (-&) + (n) Assuming particular noises are uncorrelated between them and with the common ones: Rala2 (t) = R12 (1) + (t) = aR11 (t -T) + (13) = aR11 eu) * 8(13 -T) + I3RIICIUCI (t) The first term of the addition in the last equation can be interpreted as the autocorrelation of yi(t) shifted T or as the delta has been spread by the autocorrelation function. If yi(t) were a white noise it would be observed as a peak on T over a noise background. If there were no common noises, Ri2(t) = 0, the delay could be estimated as the argument which maximizes the function Raia2(D).

On the other hand, if common noises behave like white Gaussian noises, cross-correlation between them would be a scaled delta: (&) = 13RIClIICI (0). 8(t) Then T could be estimated again as: = Arg<.<. tI12 ()1) In this equation, argument t is scanned between Dmin and Dmax. Dmn avoids selection of the maximum produced by cross-correlation of the common noises but establishes a maximum limit for the velocity estimation. tmax is necessary because of the finite samples and establishes a minimum limit for the velocity estimation.

Noises generated by common sources come from vibrations produced by the powering engines, window oscillations, passenger displacement, etc. These unwanted signals could have periodic components that introduce more than one peak in their cross-correlation which can interfere with the estimation of T. Cross-correlation can also be expressed in terms of power spectral density, using a continuous time formalism, as: Raa(n) = Uc1)etdQ To improve the estimation of the desired delay T it is possible to use weighting methods. These aim to equalize the spectral characteristics of the signal in order to obtain a better estimation ofT: R'a2 () = () Ga1a2 (jU)ecffl Weighting can be interpreted in terms of filters for ai and a2. In this case cross-correlation can be expressed as: 12 (-n) = Where is an estimation of the power spectrum density from the data. Using the weighting notation it follows: R ( W -a1a2 with W(jQ) = H1(jQ)H(jfl) This problem per se has been extensively studied. The most commonly-used weightings used for correlation estimation are summarized in Table 1.

Cross-correlation 1 Roth 1 &lal U1) SCOT 1 Gaa (jfl)daa (P) PHAT 1 IGaja2(icO Table 1 -Weighting methods used Regarding the algorithms, as signal delays are varying with velocity, the estimation of cross-correlations has to be from successive windows of signals.

Cross-correlation calculation of two signals x and y of length L (x[n]=O, y[n]=O, for O«=n and ncL) can be calculated using the biased estimator: R[rn] = x[n]y[n + mi = x[n]y[n + m] 71=-co n=O For a given m, only L-ImI samples of each signal are computed. Therefore as the delay value increases, fewer samples are used for estimating the cross-correlation. It also produces a triangular window effect that could make induce an incorrect maximum value. In order to reduce this problem two squared windows of different length are used: 1u1 -N<n<N []4J Nl71f<_n<_N+1571 IM, otherwise CM, otherwise Windowed signals around the centre c and shifted to zero are defined as si[n, ci = ai[n + c]vi[n] and s2[n, c] = a2[n + c]v2[n].

Cross-correlation centred in the sample c follows the equation: (JO Co R[m, ci = s1[n,c]s2[n ± m, ci = a1[n ± c]a2[n ± cf m]v1 [n]v2[n ± nil -w -w According to the last expression, R[m,c] values for «= k «=l are calculated with the same number of samples, 2N + 1, and l set the limits of the delay search and therefore the range for velocity estimation. Their values can be set according to the velocity range expected for a given train or route. In order to do the practical calculation of the delays the following process is used: 1 Signal windowing Signals ai[n] and a2[n] of length M >>2N-'-l are divided into L windows of length 2N+1. The l-th window of the signal called a[n] will be called a,i[n].

2 Weighting functions are estimated from the data according to the following expressions where A1[k] = DFT{a1[n]) a. Cross-correlation: W[k] = 1 b. Roth: W[k]=W9k]= -_1 Gajai[ I c. SCOT: w[k] = w5[kj = 1 d [k]&2a2 [k] d. PHAT: w[k]=w[k]= -_1 Gaa[ II being = [kJAJ1[k] 3 Power spectral density for a window I is estimated as G01a2 F/c, c] = S1 [k, c]S [k, c] with S1[k, c] = DFTfs1[n,c]) = DFT[a1[n + c]v1[n]) 4 Cross-correlations to estimate 1 around a centre c are then calculated as R' [m, c] = DFT1{W[kIOaia2 [k, c]} The maximum value is searched in each window and its time index will be the desired delay, ?, around the time c.

From experiments performed, it has been seen that most relevant information to velocity estimation is under 50Hz.

Results A plot of cross-correlation shows several peaks and some periodic components.

This behaviour is likely due to the nature of the common noises as explained above.

In order to visualize delay evolution, a normalized cross-correlation function is used.

From the cross-correlation estimation, Rr[ln, c] presented above: RW[m,c] RN [m,c]= max(R [m,c]) for each c.

RNW [mc] is a two variable function depending on the window centre and the delay value. For a chosen window centre, c, the maximum value of RNW [m,c] will have the value 1 and its corresponding index m will be the desired delay, t. Analysis shows it is clear that velocity information can be detected from cross-correlation in most cases despite other peaks and interferences.

The time when cross-correlation has its maximum value for each window is the delay estimation, ?, for that window. Velocity can be estimated from ? according to 0 = d/?.

Weighting methods of RNW [m, ci make correlation peaks much thinner and in some cases allow more precise estimations. However, they may also emphasize some other peaks making right delay detection more difficult than in the case without weighting. The type of weighting system used may be selected for the particular case.

Experimentation shows that velocity can be estimated in most parts of the route and for several values of velocity.

Conclusions

Preliminary results demonstrate the usefulness of this system for integrating with other positioning system in order to obtain an all-condition positioning system for trains.

It has been shown that velocity information can be detected with the proposed system in spite of the great attenuation of the signal of interest and the interference of unwanted vibrations.

Several weighting methods were tested in order to improve the delay estimation.

Preliminary results suggest that the system behaviour depends on the physical conditions of the rail track and the specific train. When this method is going to be used, it should be tuned for a specific situation train or rail.

The above-described embodiments are exemplary only, and other possibilities and alternatives within the scope of the invention will be apparent to those skilled in the art. For example, in order to improve speed determination, markers may be placed on the tracks to produce easily recognisable vibration patterns when a train passes.

While the above embodiment only uses two vibration sensors, one or more additional vibration sensors may be used along the length of the train, to provide additional vibration information and I or provide redundancy in the case of sensor failure.

Sensors may be placed within the same car, which provides for easier physical set-up, but may equally be placed in different cars, which entails a greater distance d therebetween. This may lead to greater accuracy since the ratio of d to delta d (the error or uncertainty in d) would increase. To facilitate such a set-up, wireless communication techniques such as wi-fi or Bluetooth may be employed between cars.

Claims (13)

1. Claims 1. A method for determining the speed of a train, comprising the steps of: providing a first vibration sensor within a car of the train; providing a second vibration sensor within a car of the train, the first and second vibration sensors being positioned at spaced-apart locations a known distance apart along the length of the train; using the first and second vibration sensors to measure vibration during movement of the train; correlating vibration patterns measured by the first and second vibration sensors to identify patterns detected by each sensor relating to a common source; determining the time difference between measurement of said identified patterns at each vibration sensor; and relating the said time difference to the distance between the vibration sensors along the length of the train to determine the speed of the train.
2. 2. A method according to claim 1, wherein the correlation step comprises applying a weighting to the outputs of each of the first and second vibration sensors.
3. 3. A method according to either of claims 1 and 2, further comprising the step of combining the determined train speed with a value for the train speed derived from an independent source.
4. 4. a method according to any preceding claim, wherein the first and second vibration sensors are located over respective bogies of the train.
5. 5. Train speed determination apparatus, comprising a first vibration sensor within a car of the train, a second vibration sensor within a car of the train, the first and second vibration sensors being positioned at spaced-apart locations along the length of the train, means for receiving output signals the first and second vibration signals and processing means for processing the received signals to determine the train speed.
6. 6. Apparatus according to claim 5, wherein the first and second vibration sensors are provided within the same car.
7. 7. Apparatus according to claim 5, wherein the first and second vibration sensors are provided within the different cars of the train.
8. 8. Apparatus according to any of claims 5 to 7, wherein the first and second vibration sensors are located over respective bogies of the train.
9. 9. Apparatus according to any of claims 5 to 8, further comprising at least one additional vibration sensor positioned within a car of the train.
10. 10. Apparatus according to any of claims 5 to 9, further comprising an independent train speed determination system, and means for combining the speed determinations of the independent system and the processing means.
11. 11. A train fitted with the apparatus of any of claims 5 to 10.
12. 12. A method substantially as herein described with reference to the accompanying figures.
13. 13. Apparatus substantially as herein described with reference to the accompanying figures.