Railway analysis device

Abstract

Railway analysis device A railway analysis device includes a memory (4) arranged to receive railway track measurement data, each comprising at least one railway track measurement value and a dated measurement path identifier, an extractor (6) arranged to determine a set of features from the railway track measurement data, which set of features includes at least two features selected from a group comprising deviation features, leveling features, left features, cant features, rolling stock features, and historical features, and a gradient reinforcement-based machine learning unit (8) trained on railway track measurement data associated with railway track fault values,which railway track defect is chosen from a group comprising at least one defect among a cant defect, a leveling defect, and a left-side defect, and arranged to return one or more railway track defect values. Fig. 1,

Description

Title of the invention: Railway analysis device

[0001] The invention relates to the field of railway analysis and in particular the predictive maintenance of railway tracks.

[0002] The field of predictive maintenance of railway infrastructure has been the subject of study for several decades. Indeed, it is critical to maintain railway infrastructure in good working order in order to guarantee the safety of users and goods. However, even the simplest infrastructures quickly comprise several hundred kilometers of rail, which makes manual maintenance prohibitive in terms of cost.

[0003] In order to automate data collection, machines, such as the EM80H locomotive manufactured by the company Plasser & Theurer (see for example https: / / www.plassertheurer.com / fr / machines-systemes / voitures-automotrices-de-mesur e / em80h) have been developed.

[0004] The American company RailPod (see https: / / rail-pod.com / ?page_id=2263) offers an autonomous measuring device and algorithms for analyzing track geometry and condition (erosion, surface condition, internal structural problems).

[0005] The FleetONE solution from specialist Wabtec Controls Systems (http: / / trackiq.com.au / FleetONE.html) offers a system for acquiring and managing railway maintenance data, primarily for rolling stock. It offers the ability to search and visualize simple data. No predictive rail maintenance functionality appears to be offered.

[0006] None of these solutions offer the possibility of predictive maintenance.

[0007] Predictive maintenance involves trying to determine which sections of railway track are most critical to maintain. This determination is a source of many challenges.

[0008] It appears obvious that predictive maintenance requires comparing measurement signals acquired successively in order to analyze the evolution of track sections. This requires synchronizing the measurement signals between different acquisitions because these vary greatly due to the way in which they are determined (variance on the acquisition devices, determination of the sections by odometry, by GPS or by singular point inducing ambiguities, etc.). Many patent applications have been filed on this subject, such as document EP 2 305 532 which describes the determination of a cross-correlation on overlapping sections, document US 2009 / 0070064 which describes a method for determining a trend from deviations in the vertical position of the track, or document US 2004 / 0204882 which maps track measurement data (MTD for "Measured Track Data" in English) onto data Track Geographical Data (TGD). Most of these methods are inaccurate and / or tedious.

[0009] Then, once the signals are synchronized, it is a question of choosing how to identify the evolutions of the problematic sections. Here again, no commercial offer really exists to date, and the state of the art is mainly based on academic work. The article by Iman Soleimanmeigouni “Track geometry degradation and maintenance modelling: A review”, https: / / journals.sagepub.com / doi / 10.1177 / 0954409716657849 reviews around a hundred articles, none of which provide real satisfaction. Finally, some prospective works have sought to propose using artificial intelligence to carry out predictive maintenance. In general, like the article “Systematic Literature Review on Data-Driven Models for Predictive Maintenance of Railway Track: Implications in Geotechnical Engineering” by Xie et al (https: / / www.mdpi.com / 2076-3263 / 10 / ll / 425 / pdf) suggests that deep and unsupervised learning are the most promising areas. But again, no method seems to have managed to move beyond the academic stage and be sufficiently reliable from an industrial point of view.

[0010] The invention improves the situation. To this end, it proposes a railway analysis device comprising a memory arranged to receive railway track measurement data each comprising at least one railway track measurement value and a dated measurement route identifier, an extractor arranged to determine a set of characteristics from the railway track measurement data, which set of characteristics comprises at least two characteristics chosen from a group comprising deviation characteristics, leveling characteristics, left characteristics, cant characteristics, rolling stock characteristics, and historical characteristics, and a gradient boosting-based machine learning unit trained on railway track measurement data associated with railway track defect values,which railway track defect is selected from a group comprising at least one of a cant defect, a leveling defect and a left defect, and arranged to return one or more railway track defect values.

[0011] This device is particularly advantageous because it makes it possible to provide a systematic, reliable and automated analysis of railway track defects, which was not previously possible.

[0012] According to various embodiments, the invention may have one or more of the following characteristics:

[0013] - the device further comprises a corrector arranged to receive data from current railway track measurements associated with a single measurement route identifier whose measurement values ​​indicate a cant measurement, to analyze this railway track measurement data in order to identify curve starts and ends, and to modify the current railway track measurement data by comparing the curve starts and ends to the curve starts and ends of the reference path measurement data,

[0014] - the extractor determines a set of characteristics which includes at least one characteristic left-hand characteristic, at least one deviation characteristic, at least one leveling characteristic, at least one cant characteristic, at least one rolling stock characteristic and at least one historical characteristic,

[0015] - the measurement data of a railway track further comprises an identifier of section, and the extractor is arranged to process together the measurement data of a railway track associated with the same dated measurement route identifier and the same section,

[0016] - the extractor is arranged to determine a set of characteristics comprising a left characteristic, a deviation characteristic, a leveling characteristic and a cant characteristic of the type maximum value, minimum value, 90th percentile value, 10th percentile value in the railway track measurement data associated with the same previous dated measurement route identifier and the same section,

[0017] - the extractor is arranged to determine a set of characteristics comprising at least at least one historical feature that is a leveling feature, a cant feature, or a rolling stock feature determined for the measurement data of a railway track associated with the same section and a previous dated measurement route identifier,

[0018] - the extractor is arranged to determine a set of characteristics comprising a rolling stock characteristic such as number of trains passed in the last 7, 15 or 30 days, or type of tonnage of rolling stock passed in the last 7, 15 or 30 days, and

[0019] - the gradient boosting based machine learning unit uses a binary classification with logistic loss function.

[0020] The invention also relates to a method for predicting a railway track defect value, comprising the following operations: (a) receive measurement data from a railway track each comprising at least one railway track measurement value and a dated measurement route identifier, (b) determining a feature set from the railway track measurement data, which feature set comprises at least two features selected from a group including deviation characteristics, leveling characteristics, left characteristics, cant characteristics, rolling stock characteristics, and historical characteristics, c) transmitting the feature set determined in operation b) to a gradient boosting-based machine learning unit trained on railway track measurement data associated with railway track defect values, which railway track defect is selected from a group comprising at least one of a cant defect, a leveling defect and a left defect, and arranged to return one or more railway track defect values, d) return the railway track defect value(s) determined for the set of characteristics determined in operation b).

[0021] The invention finally relates to a computer program product comprising instructions for implementing the method when executed on a computer, and a storage medium on which the computer program product is recorded.

[0022] Other characteristics and advantages of the invention will appear more clearly on reading the following description, taken from examples given for illustrative and non-limiting purposes, taken from the drawings in which:

[0023] [Fig-1] [Fig.l] represents a generic diagram of a device according to the invention,

[0024] [Fig.2] [Fig.2] represents an example of a left defect for a railway track rovary,

[0025] [Fig.3] [Fig.3] represents an example of a cant defect for a railway track rovary,

[0026] [Fig.4] [Fig.4] represents an example of a leveling defect for a track railway,

[0027] [Fig.5] [Fig.5] represents an example of a gap defect for a railway track rovary,

[0028] [Fig.6] [Fig.6] represents an example of a cant signal received at the input, and

[0029] [Fig.7] [Fig.7] represents an example of a signal synchronization function of [Fig.6] with a theoretical signal.

[0030] The drawings and the description below contain, for the most part, elements of a certain character. They may therefore not only serve to better understand the present invention, but also contribute to its definition, if necessary.

[0031] [Fig.l] represents a generic diagram of a device according to the invention. The railway analysis device 2 comprises a memory 4, an extractor 6, a gradient reinforcement-based machine learning unit 8 and an optional corrector 10.

[0032] The memory 4 can be any type of data storage suitable for receiving digital data: hard disk, hard disk with flash memory, flash memory in any form, RAM, magnetic disk, locally or cloud distributed storage, etc. The data calculated by the device can be stored on any type of memory similar to the memory 4, or on it. This data can be erased after the device has performed its tasks or retained.

[0033] In the example described here, the memory 4 receives measurement data from a railway track. In the example described here, this measurement data is mostly received from an EM80H locomotive, called a “Plasser”. This type of locomotive is specifically dedicated to measuring railway tracks in order to ensure their monitoring and maintenance. The outputs of this type of locomotive are typically signals as shown in [Fig.6].

[0034] In the example described here, the memory 4 thus receives four measurement signals in the form of samples: a left measurement signal, a slope measurement signal, a leveling measurement signal, and a deviation measurement signal.

[0035] Figures 2 to 5 show the respective nature of left, cant, leveling and deviation defects. Thus, a left defect is observed when a portion of one rail is raised relative to the other rail, a cant defect is observed when one rail is raised overall relative to the other rail, a leveling defect is observed when both rails have a raised portion when they should remain flat, and a deviation defect is observed when the rails deviate axially relative to the direction they should have.

[0036] In order to be able to use and compare these signals, each of the data composing them is associated on the one hand with a dated route identifier, that is to say an identifier making it possible to know which railway track was measured, and to which date this measurement corresponds, and on the other hand with a section identifier. Indeed, as will be seen below, the measurement data are analyzed in the example described here by sections of 50 m in length.

[0037] Alternatively, the sections could be shorter, for example 25 m, provided that sufficient precision is available on the synchronization of the signals, or longer, for example 100 m. The shorter the length, the more limited the risk of the presence of portions of track that are not homogeneous with each other in the same section. By portions of track that are not homogeneous with each other in the same section, we mean the fact that, by the nature of the route, portions of the same section behave differently with respect to the appearance of defects and that the section is therefore “incoherent” from this point of view. This may be the case if the length of the section is significant, typically 200 m, and part of the section is close to a location subject to a lot of vibration, or a location where the composition of the soil is very different from the rest of the section. As these are railway lines, the characteristic lengths (bends, ground stability, etc.) are quite significant, and the length of 50 m therefore appears naturally advantageous.

[0038] Thus the measurement data is formed by a measurement value of a type of fault, a dated route identifier and a section identifier.

[0039] Alternatively, the device 2 could receive raw track measurement data, and calculate the measurement signals described above. Still as a variant, the dated route identifier could be a simple date, the link between the section identifiers of the same route then being able to be implicit.

[0040] In addition to these signals, the memory 4 receives other information, such as the number of train passages on the railway between two measurements, the tonnage of rolling stock between two measurements, these data forming railway equipment data.

[0041] The measurement data of a given section and a given date can furthermore be associated with the measurement data of this same section of previous measurements. This “past” measurement data forms historical data and can be replicated as historical data associated with the measurement data or simply associated on the basis of its dated measurement path identifier and its section identifier.

[0042] Alternatively, the measurement data could be simplified so that a section identifier is associated with only one measurement data item per measurement type. This amounts to making a section of very small size, or for example to averaging all the measurement data of a given section. The result obtained will be less precise but calculation savings can be made.

[0043] The measurement data received in the memory 4 are used directly or indirectly by the extractor 6 in order to produce a set of features which is provided to the gradient boosting-based machine learning unit 8 to perform the calculation of the predictive maintenance values. The terms “gradient boosting” and “feature” must be interpreted with the meaning given to them in the field of machine learning, that is to say respectively “gradient boosting” and “feature” in English.

[0044] As described above, the measurement data may relate to characteristic measurements of defects that a railway track may have. Thus, the extractor 6 is arranged to determine, for the measurement data associated with a given dated route identifier, a plurality of characteristics that will be provided to the input layer of the gradient-boosting-based machine learning unit 8.

[0045] In the example described here, for each section identifier, each type of fault is associated with four characteristics: the maximum value among the measurement values concerned, the minimum value among the measurement values ​​concerned, the measurement value of the 90th percentile among the measurement values ​​concerned, and the measurement value of the 10th percentile among the measurement values ​​concerned. Alternatively, only some of these values ​​could be retained, such as the measurement value of the 90th percentile among the measurement values ​​concerned.

[0046] The extractor 6, the gradient boosting machine learning unit 8 and the corrector 10 are elements directly or indirectly accessing the memory 4. They can be implemented in the form of appropriate computer code executed on one or more processors. By processors, it is meant any processor suitable for the calculations described below. Such a processor can be implemented in any known manner, in the form of a microprocessor for a personal computer, a dedicated chip of the FPGA or SoC type, a computing resource on a grid or in the cloud, a cluster of graphics processing units (GPUs), a microcontroller, or any other form suitable for providing the computing power necessary for the implementation described below. One or more of these elements can also be implemented in the form of specialized electronic circuits such as an ASIC.A combination of processor and electronic circuits can also be considered. In the case of the gradient boosting machine learning unit 8, dedicated machine learning processors can also be considered.

[0047] The extractor 6 is also arranged to establish characteristics in the characteristic set based on the railway equipment data and based on the historical data.

[0048] Thus, for the railway equipment data, the extractor 6 can determine railway equipment characteristics from the number of train passages and the tonnage of rolling stock having circulated in the last 7 and / or 15 and / or 30 days preceding the date associated with the dated route identifier on the associated route. Similarly, historical characteristics can include the deviation characteristics, the leveling characteristics, the left characteristics, the cant characteristics for each section identifier which correspond to the first and / or second and / or third and / or fourth previous dated route identifier, for each section identifier, with each time the number of days elapsed.Historical characteristics may also include field surveys by operators (geometry defects, general condition of tracks and rails), topography (terrain models) or even maintenance work.

[0049] The set of characteristics thus established for each segment identifier is then provided to the input of the gradient boosting-based machine learning unit 8. This is in the example described a gradient boosting tree. In the example described here, the LightGBM library version 3.0.0, maintained by Microsoft, is used from Python with the 'LightGBMClassifier' function and the following hyperparameters.

[0050] - Model type ('boosting_type'): 'gbdt' - the usual gradient boosted tree

[0051] - Number of terminal leaves ('num_leaves'): 50,

[0052] - Maximum tree depth ('max_depth'): no limit,

[0053] - Boosting learning rate ('learning_rate'): 0.1,

[0054] - Number of boosted trees in the model ('n_estimators'): 1000,

[0055] - Objective function ('objective'): binary classification ('binary' with function of logistic loss - log-loss),

[0056] - Random sampling rate (without replacement) on the dataset to construct a tree ('subsample'): 0.7 or 70%,

[0057] - Random sampling rate (without replacement) on the input variables for build a tree ('colsample_bytree'): 50%,

[0058] - Weighting of each class ('class_weight'): balanced for classifications binaries.

[0059] The other hyperparameters are less important for the configuration of the algorithm and are the default ones of the 'LightGBMClassifier' function.

[0060] The Applicant has identified that this type of machine learning is by far the most effective, ahead of logistic regression methods, SVM (Support Vector Machine), or Random Forest methods, but also compared to methods based on deep learning, which are more suited to image processing.

[0061] The training of the gradient boosting-based machine learning unit 8 was carried out from the existing maintenance history for the railway lines of the Gabonese railway, with the measurement of the defects that were actually measured as a label. Ideally, the gradient boosting-based machine learning unit 8 could be specifically trained for each railway line to which it is to be applied.

[0062] The most favorable results were obtained with the use of all the features described above. Alternatively, the gradient boosting-based machine learning unit 8 could use only one feature per feature category, or even just one feature at all.

[0063] The corrector 10 is optional. Indeed, in the above, it was considered that the measurement data were already synchronized with each other. Nevertheless, the Applicant has noted that the history of the measurements taken by the EM80H locomotive is difficult to reconcile between two passes. This is due to: the lack of accuracy of the starting point of a measurement campaign, and the wear and slippage of the locomotive wheels (the measurement location being determined by odometry). It is therefore desirable to synchronize the measurement data with the measurement data already present in the memory 4. The Applicant has discovered that it is particularly advantageous to use the slope measurement data for this purpose.

[0064] [Fig.7] represents an example of implementation of a function allowing to synchronize the measurement data including a new dated route identifier with the measurement data already present in memory 4 for this route.

[0065] In a first operation 700, this function receives the cant measurement data from the measurement data to be synchronized via an Inp() function. Optionally, the Inp() function can also retrieve the reference data associated with the dated route identifier from the measurement data received by the Inp() function. Then, in an operation 710, this data is smoothed using a low-pass filter by a Filt() function. This makes it possible in particular to eliminate certain disturbances linked to the precision of the locomotive, as well as to the state of the railway.

[0066] Once the signal is smoothed, the resulting signal is analyzed by a Detect() function in an operation 720 to determine the points for which the cant measurement exceeds 15 mm. These points physically represent the start and end of each curve. Alternatively, this threshold could be 5 mm or less, or the start and end points of the curve could be determined, for example, based on a derivative of the signal.

[0067] These start and end points are then used by a Match() function in an operation 730 in order to compare them to reference data. These reference data may correspond to measurement data in which absolute confidence is placed, or a plan of the railway track at its construction. The association by the Match() function between the characteristic points of the smoothed signal and those of the reference data can be made by analyzing the similarity of the resulting curves, for example on the basis of the order of these points and the direction and amplitude of the corresponding cant signal. This matching is particularly advantageous because it is fairly simple to implement and makes it possible to take into account the fact that the characteristic points and their sequence is unique to each railway track.Thus, it is possible to achieve an accuracy of around 50 centimeters for the measurement data processed by the corrector 10, as opposed to the accuracies of around ten meters of the state-of-the-art solutions.

[0068] Finally, a Sync() function modifies in an operation 740 the measurement data on the basis of the resettings of the characteristic points identified in operation 730. For this, the start and end of each part of the corresponding signal is aligned on the basis of this resetting. This induces a translation and / or a compression / expansion of the signal concerned. This same transformation is applied to all the signals associated with the same dated path identifier.

Claims

Claims

1. A railway analysis device comprising a memory (4) arranged to receive railway track measurement data each comprising at least one railway track measurement value and a dated measurement path identifier, an extractor (6) arranged to determine a feature set from the railway track measurement data, which feature set comprises at least two features selected from a group comprising deviation features, leveling features, left features, cant features, rolling stock features, and historical features, and a gradient boosting-based machine learning unit (8) trained on railway track measurement data associated with railway track defect values,which railway track defect is selected from a group comprising at least one of a cant defect, a leveling defect and a left defect, and arranged to receive a set of characteristics and return one or more railway track defect values.,

2. A railway analysis device according to claim 1, further comprising a corrector (10) arranged to receive current railway track measurement data associated with a same measurement path identifier whose measurement values ​​indicate a cant measurement, to analyze this railway track measurement data in order to identify curve starts and ends, and to modify the current railway track measurement data by comparing the curve starts and ends to the curve starts and ends of reference path measurement data.

3. Device according to one of the preceding claims, wherein the extractor (6) determines a set of characteristics which comprises at least one left characteristic, at least one deviation characteristic, at least one leveling characteristic, at least one cant characteristic, at least one rolling stock characteristic and at least one historical characteristic.

4. Device according to one of the preceding claims, wherein the measurement data of a railway track further comprises a section identifier, and wherein the extractor (6) is arranged to process together the measurement data of a railway track associated to the same dated measurement route identifier and to the same section.

5. Device according to claim 4, wherein the extractor (6) is arranged to determine a set of characteristics comprising a left characteristic, a deviation characteristic, a leveling characteristic and a cant characteristic of the type maximum value, minimum value, value at percentile 90, value at percentile 10 in the railway track measurement data associated with the same previous dated measurement route identifier and the same section.

6. Device according to claim 4 or 5, wherein the extractor (6) is arranged to determine a set of characteristics comprising at least one historical characteristic which is a leveling characteristic, a cant characteristic, or a rolling stock characteristic determined for the measurement data of a railway track associated with the same section and with a previous dated measurement route identifier.

7. Device according to one of the preceding claims, in which the extractor (6) is arranged to determine a set of characteristics comprising a characteristic of rolling stock of the type number of trains passed in the last 7, 15 or 30 days, or of the type tonnage of rolling stock passed in the last 7, 15 or 30 days.

8. Device according to one of the preceding claims, wherein the gradient boosting-based machine learning unit (8) uses a binary classification with logistic loss function.

9. A method for predicting a railway track defect value, comprising the following operations: a) receiving measurement data of a railway track each comprising at least one railway track measurement value and a dated measurement run identifier, b) determining a feature set from the railway track measurement data, which feature set comprises at least two features selected from a group comprising deviation features, levelling features, left features, cant features, rolling stock features, and historical features, c) transmitting the feature set determined in operation b) to a gradient boosting-based machine learning unit (8) trained on railway track measurement data associated with railway track defect values, which railway track defect is selected from a group comprising at least one defect from among a cant defect, a leveling defect and a left defect, and arranged to return one or more railway track defect values, d) return the railway track defect value(s) determined for the set of characteristics determined in operation b).

10. A computer program product comprising portions of program code for implementing the device according to one of claims 1 to 8 or the method according to claim 9 when said program is executed on a computer.

11. A data storage medium on which the program according to claim 10 is recorded.

Images