System and method for detecting railway track faults

Abstract

The invention relates primarily to a railway track (2) defect detection system (1) configured to: • Analyze each recorded data point and determine the difference between said data point and a specified reference data point. • Classify the analyzed data point as a discontinuity data point if the determined difference is greater than a specified threshold; • Select at least one contextual piece of information associated with said discontinuity data point; • Establish a correlation index between the contextual information and the discontinuity data point; and • Classify the discontinuity data point as a railway track defect (2) if the correlation index is less than a second specified threshold.

Description

TECHNICAL FIELD

[0001] The invention falls within the railway field and more specifically relates to the monitoring of defects in railway tracks.

[0002] The invention relates more particularly to a system and method for qualifying data relating to railway track rails in order to detect defects. The invention also relates to the qualification of data relating to the running gear of railway rolling stock. PRIOR ART AND DISADVANTAGES OF PRIOR ART

[0003] The integrity of railway tracks is an essential parameter for rail traffic. Indeed, the number of track defects, and especially their magnitude, must be as low as possible to allow trains to run while ensuring passenger comfort and preventing premature damage to equipment.

[0004] Consequently, the integrity of train running gear and other railway rolling stock is also an essential parameter for safety and comfort of operation.

[0005] It is known to directly measure the profile of a railway track using a measuring device mounted on a railway vehicle, the measuring device being generally an optical system such as a laser rangefinder. The profile of each rail of the track is thus obtained, and consequently any geometric defects and their position on the track are detected.

[0006] However, a dedicated railway vehicle is required to perform these measurements, as the installation of the optical measurement system necessitates prior calibration tests that are dependent on the railway vehicle. Furthermore, the railway vehicle dedicated to the measurements must regularly traverse the entire track, which disrupts commercial train traffic. Finally, this measurement device does not always allow for the accurate characterization of the acquired data, ultimately making it impossible to differentiate between a genuine track defect and a simple discontinuity in the track. OBJECTIVE OF THE INVENTION

[0007] The invention thus aims to provide a system for detecting defects in a railway track which overcomes the disadvantages of the prior art. STATEMENT OF THE INVENTION

[0008] To this end, the invention relates to a system for detecting defects in a railway track comprising two opposing rails, suitable for installation on railway rolling stock and comprising at least data acquisition means configured to be positioned opposite the two rails of the railway track, and computer management means for said acquisition means configured to control the acquisition and recording of data representative of at least one characteristic of the railway track and of contextual information representative of at least one element of railway infrastructure other than the rails, which computer management means are further configured to: Analyze each recorded data point and determine a difference between said data point and a determined reference data point; Classify the analyzed data point as discontinuity data if the determined difference is greater than a determined threshold; Select at least one contextual information associated with said discontinuity data point; Establish a correlation index between the contextual information and the discontinuity data point, and Qualify the discontinuity data point as a railway track fault if the correlation index is less than a second determined threshold.

[0009] The system of the invention may also include the following optional features considered individually or in all possible technical combinations: The data acquisition means include image acquisition means. The image acquisition means include at least two linear or matrix cameras for each rail of the railway track, configured to be positioned opposite the railway rails. The image acquisition means include at least two cameras sensitive to the ultraviolet spectrum of light, respectively configured to be positioned opposite the railway rails. The data acquisition means include means for stabilizing the image acquisition means.The acquisition means include at least one microphone intended for installation on a wheel of the railway rolling stock, with image acquisition means intended for acquiring infrastructure data and the microphone intended for acquiring contextual information. The management means include means for synchronizing data with contextual information. The acquisition means include a plurality of accelerometers intended for installation on the wheels of the railway rolling stock and configured to detect defects in the wheels of said rolling stock. The system includes at least one temperature probe intended for installation on at least one axle of the railway rolling stock.

[0010] The invention also relates to a method for detecting defects in a railway track comprising two opposing rails, implemented by a detection system installed on railway rolling stock traveling on said track, which system comprises at least means for acquiring data and contextual information configured to be positioned opposite the two rails of the railway track, and computer means for managing said data acquisition means, which method comprises at least the successive steps of: Acquisition and recording of representative data of at least one characteristic of the railway track rails and contextual information; Analysis of each recorded data, and determination of a difference between said data and a determined reference data; If the determined difference is greater than a determined threshold, classification of the associated data as discontinuity data; Selection of at least one contextual information associated with said analyzed data; Establishment of a correlation index between the contextual information and the associated discontinuity data, and Qualification of the discontinuity data as a railway track defect if the correlation index is less than a second determined threshold.

[0011] The detection method may also include the following optional features, considered individually or in all possible technical combinations: The means for managing the qualification system include synchronization means, and the method includes, prior to the analysis of each data point, a synchronization step between said data and contextual information. The synchronization means include geolocation means; the detection system includes communication means connected to the management means and to a wide area network; the method includes a position coordinate acquisition step concurrent with the data acquisition step and a step of associating each data point with each position coordinate acquired at the same time; and the discontinuity data qualification step includes the following substeps: Retrieval of correlation indices established by at least two other detection systems installed on two other railway rolling stock.The aforementioned correlation indices are established from discontinuity data sharing the same position coordinates; a final correlation index is calculated from the three correlation indices from the three detection systems; the discontinuity is definitively classified as a railway track fault if the final correlation index exceeds a predetermined third threshold.

[0012] The invention also relates to railway rolling stock comprising a fault detection system as described above. PRESENTATION OF FIGURES

[0013] Other features and advantages of the invention will become clear from the description given below, which is by way of example and not limitation, with reference to the attached figures, among which: [ Fig. 1 ] There figure 1 represents a schematic cross-sectional view of part of the qualification system of the invention; [ Fig. 2 ] There figure 2 represents a schematic side view of the qualification system of the invention installed on railway rolling stock. DETAILED DESCRIPTION OF THE INVENTION

[0014] It is first clarified that in the figures, the same references designate the same elements regardless of the figure in which they appear and regardless of the form in which those elements are represented. Similarly, if elements are not specifically referenced in one of the figures, their references can easily be found by referring to another figure.

[0015] It is also specified that the figures essentially represent one embodiment of the object of the invention, but that there may be other embodiments which meet the definition of the invention.

[0016] The present invention relates to a method for detecting defects in the rails 3 of railway tracks 2. This method is implemented by a simple and inexpensive detection system 1, which is installed on a commercial train 4. Advantageously, this detection system 1 also makes it possible to detect defects on the wheels 12 of said railway rolling stock 4 on which it is installed.

[0017] The detection system 1 of the invention is designed to be installed on railway rolling stock 4 of the type used for passenger or freight trains. In other words, it is not necessary to use railway rolling stock specifically dedicated to the installation of the detection system 1; any commercial train can be used. This not only avoids the occupancy of railway track 2 by trains solely dedicated to measurements for track maintenance, but also allows measurements to be carried out on a large number of tracks 2, and the frequency of measurements on each track 2 is simply limited by the number of commercial trains 4 running on that track.

[0018] In the following description, the term train will be used to define railway rolling stock 4, but it is understood that the detection system 1 of the invention can be installed on any type of railway rolling stock 4.

[0019] The train 4, on which the detection system 1 of the invention is installed, is put into operation on a railway track 2 comprising two opposing rails 3. The topography of the track 2 is to be determined, and structural defects in the track 2 are to be identified—that is, breaks in the rails 3, surface defects in the rails 3, broken fishplates (not shown) that join two consecutive rails 3, and missing nuts (not shown). The train 4, which extends along a longitudinal axis, comprises at least one car 14 and at least two axles, each axle comprising two axle boxes located at its respective ends. Typically, the train 4 comprises several cars 14 and therefore several axles, and each axle comprises two wheels 12.

[0020] According to the invention, the detection system 1 comprises data acquisition means 5 configured to acquire: data representing at least one characteristic of the rails 3 of the railway track 2 - which will be referred to in the following description as qualification data, and contextual information representing a static infrastructure element of the railway track 2 other than the rails 3, or representative of an element of the train 4 interacting with the rail 3.

[0021] The qualification data consists of information containing structural data on the rails 3 and the connecting elements of the consecutive rails 3 (braids, fishplates). In other words, if the analyzed rail 3 or connecting element has structural defects, the qualification data includes indications of the presence of these structural defects. The qualification data can advantageously relate to the train 4 on which the detection system 1 is installed—in particular, the wheels 12 of the train 4—via acceleration data experienced by the wheels 12.

[0022] Contextual information includes elements associated with railway track 2 - other than rails 3 - and train 4 which are nevertheless interacting with rails 2. This acquired contextual information may come from the structural elements of railway track 3 - the fishplates connecting two rails 3 together, the nuts connecting the fishplates to the rails 3, the sleepers 15, the ballast - or from the interaction between the train running 4 and the rails 3 of railway track 2 - for example the sound of the wheels 12 on the rails 3, or the temperature of the axles.

[0023] The detection system 1 also includes computer means for managing the data acquisition means 5. These management means typically include a central processing unit 6 with at least one processor, a graphics card or FPGA (Field-Programmable Gate Array) processor, and data storage means in which one or more data processing algorithms are installed. The management means also include means for time-stamping data and contextual information, a GPS (Global Positioning System) type geographic positioning device, and wireless communication means.

[0024] With reference to the figure 1 , the data acquisition means 5 include image acquisition means 7 connected to the management means and mounted on an armature 16 connected to the train 4.

[0025] This frame 16 comprises a central transverse rod 17 (i.e., extending perpendicularly to the longitudinal axis of the train 4 or the railway track 2) pivotally connected to the car 14 of the train 4 along a transverse pivot axis, and two V-shaped frame portions 18 attached to the ends of the central rod 17, the frame portions 18 being arranged opposite the respective rails 3 of the railway track 2.

[0026] The image acquisition means 7 include four linear or matrix cameras 8, distributed two by two on each portion of V-shaped armature 18. More precisely, for each portion of V-shaped armature 18, two linear or matrix cameras 8 are mounted at the end parts of the arms 19 of the considered V-shaped portion 18. Advantageously, the data acquisition means 5 include lighting devices (not shown) associated with the cameras 8 to improve the quality of the images acquired by the latter.

[0027] Furthermore, each linear or matrix camera 8 is mounted to move freely in a direction parallel to the arm 19 on which it is mounted. Advantageously, each linear or matrix camera 8 is mounted on a motorized rail attached to the arm 19 of the V-shaped frame section 18 and controlled by the central control unit 6. Similarly, the central rod 17 is mounted on the car body 14 of the train 4 via a motorized rotary device controlled by the central control unit 6. In this way, the central control unit 6 can precisely and individually orient each linear or matrix camera 8 so that each one scans the desired section of rail. The linear or matrix cameras 8 are designed to detect surface defects on the rails 3 and breaks in the rails 3.

[0028] The image acquisition means 7 also include two cameras sensitive in the ultraviolet spectrum of light 9 – which will be called UV cameras 9 in the following description – which are mounted in transverse translation (i.e., perpendicular to the longitudinal axis of the train 4 or the railway track 2) at the base 20 of each section of V-shaped reinforcement 18 and are positioned opposite the respective rails 3. Like the linear or matrix cameras 8, each UV camera 9 is mounted on a motorized translation rail controlled by the central unit 6. The central unit 6 can thus precisely and individually orient each UV camera 9 so that they are perfectly aligned with the respective rails 3. The UV cameras 9 make it possible to detect hydrocarbon stains on the surface of the rails 3.

[0029] With reference to the figure 2, the detection system 1 of the invention is connected to the car body considered 14 of the train 4 by means of fastening means 21, which fastening means 21 comprise for example two fasteners 22 cooperating with each other and respectively fastened to the detection system 1 and to the car body considered 14 of the train 4.

[0030] The detection system 1 also includes stabilization means 10 arranged between the frame 16 and the attachment considered 22 of the securing means 21. These stabilization means 10 include at least one damper 23 which allows the constant distance between the image acquisition means 7 and the rails 3 to be maintained, whether the train 4 is moving on the railway track 2 or not.

[0031] Optionally, the data acquisition means 5 may include means for detecting faults in the rail 3, such as eddy current detectors (not shown) attached to the frame 16 and connected to the management means.

[0032] The data acquisition means 5 further include microphones 11 connected to the control means, ideally four microphones 11 arranged in pairs on the wheels 12 of an axle. For each wheel 12, one of the microphones 11 is directed towards the rails 2 to capture the sounds caused by the movement of the wheel 12 on the rail in question 3 – and in particular the noise emitted by a wheel 12 passing over a joint between two rails 3 – and the other microphone 11 is directed to capture the aerodynamic noise caused by the movement of the train 4. Measuring the aerodynamic noise makes it possible to subtract this noise from the recordings made by the microphones directed towards the tracks and thus isolate the noise caused by defects.

[0033] The detection system 1 also includes means for measuring the accelerations experienced by the wheels of the train 4 traveling on the railway track 3. Ideally, the wheels 12 of two axles are each equipped with an accelerometer 13, so the detection system 1 comprises four accelerometers connected to the control means. These accelerometers 13 make it possible to identify defects in the train's running gear, which includes the axles and the wheels 12.

[0034] The detection system 1 includes at least one temperature probe (not shown) connected to the management means and installed on a box of an axle of train 4. This probe measures the temperature of the axle box to identify a possible seizure of said axle box, in particular if the measured temperature is higher than a determined value.

[0035] Finally, the detection system 1 includes means of power supply, for example batteries.

[0036] According to the invention, a method for detecting defects in a railway track 2 implemented by the detection system 1 will now be described.

[0037] With train 4 in operation on track 2, the control system commands the acquisition system 5 to continuously acquire a stream of data and contextual information. More specifically, the linear or matrix cameras 8 and the UV camera 9 acquire characteristic images of the rail surface, the accelerometers 13 acquire acceleration data from the train's wheels, the microphones 11 continuously record the sounds emitted by the wheels 12 traveling on the rails 3, and the temperature of the axle box(es) is continuously recorded by the temperature probe(s).

[0038] In parallel with their acquisition, each piece of railway track data and each contextual information item is associated with a geolocation acquired by positioning means, and where applicable, time-stamped by time-stamping means. Thus, the acquired data and contextual information are synchronized during a second step of the process.

[0039] In a third step, the acquired data is analyzed by the central processing unit 6 of the management information system to determine the difference between each data point and a predetermined reference value. For example, the central processing unit 6 determines whether the amplitude of a discontinuity on the surface of the rail in question 3 exceeds a predetermined threshold amplitude stored in the central processing unit 6's storage system.

[0040] The analysis of this data and the associated discrepancies is performed by algorithms stored in the central processing unit's memory. Furthermore, contextual information is also classified by these algorithms. Advantageously, these analyses and comparisons are carried out using convolutional neural networks.

[0041] According to the invention, several convolutional neural networks are used, including: A first neural network analyzes surface data from image acquisition means to detect defective fasteners, surface defects, breaks in fishplates or rail 3, missing nuts 3 and by extension other defects present on the infrastructure of the railway 2 (for example: defective sleepers, sinkholes, ballast defects); A second neural network to analyze part of the contextual information, in particular the presence of fishplates, sleepers 15 or braids (elements which ensure the connections between metallic structures to ensure their equipotentiality), and A third neural network to analyze another part of the contextual information, in particular the sounds acquired by the microphones 11.

[0042] Other data from accelerometers and contextual information from temperature probes are recorded and associated with position coordinates, and where appropriate, time-stamped.

[0043] In a fourth step, if the central unit 6 detects that the determined deviation for any of the data points is greater than the threshold deviation, then said central unit 6 classifies this data as discontinuity data. This means that this discontinuity data is likely to be associated with a railway track defect 2 – in particular, a rail break 3 or fishplate failure, or a rail surface defect 3.

[0044] Since the acquisition methods 5 are unable to distinguish between a true defect and a normal structural discontinuity—for example, a joint between two rails 3 will be detected as a discontinuity when it is not a railway track defect 2—the process implements a fifth step that qualifies this discontinuity data according to a predetermined confidence level. Typically, this confidence level varies between 70% and 90% depending on the sensor that acquired the data.

[0045] To perform this qualification of discontinuity data, the central unit 6 includes additional algorithms enabling the fusion of data from different convolutional neural networks, temperature sensors and accelerometers.

[0046] As an example, two fusion models, one using the Bayesian method of Markov random fields and the other the probabilistic belief theory of Dempster Shafer, are recorded in the central unit 6 storage means.

[0047] Once the data and contextual information have been synchronized, the fusion model will select one or more contextual pieces of information associated with the discontinuity data, and then establish a correlation index between said discontinuity data and the associated contextual information(s). Obtaining this correlation index depends on the type of fusion model used. For example, the Bayesian method of Markov random fields will establish a correlation link between the discontinuity data and the associated contextual information(s), while the method based on Dempster-Shafer's probabilistic belief theory will establish a probability between 0 (no correlation) and 1 (maximum correlation).

[0048] In particular, if a break in continuity is detected at a rail 3, the fusion model will search the acquired image for the presence of a fishplate or braid. If a fishplate or braid is detected, then the discontinuity data will be qualified as normal data, specifically a joint between two rails 3 – that is, not a defect – with a confidence level determined to be representative of the correlation index determined by the fusion model.

[0049] If no fishplate or braid is detected, the fusion model will analyze the sound produced by the movement of wheels 12 on rail 3. If no particular sound is detected, the fusion model will conclude that there is a rail joint—a normal situation—with a second confidence level determined to be representative of the correlation index determined by the fusion model. If a particular sound is detected, the data will be classified as faulty with the same confidence level.

[0050] Furthermore, contextual information from the UV 9 cameras also allows for the classification of discontinuity data exhibiting a surface defect: if the UV 9 camera returns contextual information characteristic of a hydrocarbon stain, the discontinuity data is classified as such by the fusion model with a third confidence index determined to represent the correlation index determined by the fusion model. Conversely, if the UV 9 camera returns no particular information, the discontinuity data is classified as a rail 3 surface defect with the same confidence index.

[0051] Finally, in the absence of discontinuity data associated with rails 3, the detection system 1 also makes it possible to identify wheel defects 12: if the accelerometer(s) detect a periodic acceleration, the central unit 6 will conclude that there is a defect in the rolling elements.

[0052] The fusion model allows this diagnosis to be refined with information from the temperature probe: if the temperature of the axle in question exceeds a determined value, for example 100 degrees Celsius, then the central unit concludes that there is wheel seizure.

[0053] The detection system 1 also includes, stored in the central processing unit's storage means, at least one algorithm implementing a decision fusion model to qualify discontinuity data with an increased confidence index, making it possible to avoid the need for human operator intervention.

[0054] The decision fusion model can implement one or more of the following methods: Principal Component Analysis (PCA); Latent Structure Projections (LSP); Discriminant Analysis (DA); Qualitative Trend Analysis (QTA); k-Nearest Neighbors (kNN); Decision Tree; Belief Theory; Bond Graphs, and Fuzzy Logic.

[0055] The list above is of course not exhaustive.

[0056] To improve the confidence index of a discontinuity data qualification, the process implements the following sub-steps: The central unit 6 will first retrieve discontinuity data having the same position coordinates and obtained by at least two other detection systems 1 installed on two other respective trains 4, and their respective confidence indices.

[0057] An overall confidence index will thus be calculated by the central processing unit 6 via the decision fusion model. This confidence index is higher when there is a greater number of discontinuity data points with the same coordinates and originating from different trains. In this case, using discontinuity data from three detection systems 1 installed on three different trains 4, the overall confidence index obtained for classifying the discontinuity data is comparable to the confidence index associated with a human operator.

[0058] Thus, the system and detection method of the invention makes it possible to determine in a precise, reliable and reproducible manner the structural defects of the railway track 3 and where applicable of the rolling elements of the train 4 without requiring a complex and costly installation, without requiring specific railway rolling stock and without requiring the intervention of human operators in the processing of the acquired data.

Claims

1. System for detecting (1) defects in a railway track (2) comprising two opposite rails (3), suitable for being installed on railway rolling stock (4) and comprising at least data acquisition means (5) configured to be arranged opposite the two rails (3) of the railway track (2), and computer management means (6) of said acquisition means (5) configured to control the acquisition and recording of data representative of at least one characteristic of the railway track (2) and contextual information at least representative of an infrastructure element of the railway track (2) other than the rails (3), which computer management means (6) are further configured to: • Analyze each recorded data item and determine a difference between said data item and a determined reference data item; • Classify the analyzed data item as discontinuity data if the determined difference is greater than a determined threshold;• Select at least one piece of contextual information associated with said discontinuity data; • Establish a correlation index between the contextual information and the discontinuity data, and • Qualify the discontinuity data as a railway fault (2) if the correlation index is lower than a second determined threshold.; 2. System (1) according to the preceding claim, characterized in that the data acquisition means (5) comprise image acquisition means (7).

3. System (1) according to the preceding claim, characterized in that the image acquisition means (7) comprise at least two linear or matrix cameras (8) for each rail (3) of the railway track (2) configured to be arranged opposite the rails (3) of the railway track (2).

4. System (1) according to claim 2 or 3, characterized in thatthe image acquisition means (7) comprise at least two cameras sensitive in the ultraviolet spectrum of light (9) respectively configured to be arranged opposite the rails (3) of the railway track (2).

5. System (1) according to any one of claims 2 to 4, characterized in that the data acquisition means (5) comprise stabilization means (10) of the image acquisition means (7).

6. System (1) according to any one of claims 2 to 5, characterized in that the acquisition means (5) comprise at least one microphone (11) intended to be installed on a wheel (12) of the railway rolling stock (4), the image acquisition means (5) being intended to acquire the infrastructure data and the microphone (11) being intended to acquire the contextual information, and in that the management means (6) include means for synchronizing data with contextual information.

7. System (1) according to any one of the preceding claims, characterized in that the acquisition means (5) comprise a plurality of accelerometers (13) intended to be installed on the wheels (12) of the railway rolling stock (4) and configured to detect defects in the wheels (12) of said railway stock (4).

8. System (1) according to the preceding claim, characterized in that it comprises at least one temperature probe intended to be installed on at least one axle of the railway rolling stock (4).

9. Method for detecting defects in a railway track comprising two opposite rails, implemented by a detection system installed on railway rolling stock running on said railway track, which system comprises at least means for acquiring data and contextual information configured to be arranged opposite the two rails of the railway track, and computer means for managing said data acquisition means, characterized in thatit comprises at least the successive steps of: • Acquisition and recording of data representative of at least one characteristic of the rails of the railway track and contextual information; • Analysis of each recorded data, and determination of a difference between said data and a determined reference data; • If the determined difference is greater than a determined threshold, classification of the associated data as discontinuity data; • Selection of at least one contextual information associated with said analyzed data; • Establishment of a correlation index between the contextual information and the associated discontinuity data, and • Qualification of the discontinuity data as a railway track defect (2) if the correlation index is less than a second determined threshold.

10. Method according to the preceding claim, characterized in that the management means (6) of the qualification system (1) comprise synchronization means, and in thatthe method comprises, prior to the analysis of each data item, a synchronization step between said data and the contextual information.

11. Detection method according to the preceding claim, characterized in that synchronization means include geolocation means, in that the detection system (1) comprises communication means connected to the management means (6) and to a wide area network, in that the method comprises a step of acquiring position coordinates concomitant with the step of acquiring the data and a step of associating each data item with each position coordinate acquired at the same time, and in thatthe discontinuity data qualification step includes the following sub-steps: • Recovery of the correlation indices established by at least two other detection systems installed on two other railway rolling stock, said correlation indices being established from discontinuity data sharing the same position coordinates; • Calculation of a final correlation index from the three correlation indices coming from the three detection systems (1), and • Final qualification of the discontinuity as a railway defect (2) if the final correlation index is greater than a third determined threshold.

12. Railway rolling stock (4) comprising a fault detection system (1) according to any one of claims 1 to 8.

Images