Device for detecting a railway vehicle on a railway track and associated method

Abstract

A detection device (10) for at least one rail vehicle (V) on a railway track (W), the detection device (10) comprising at least one fastening element (11), at least one measuring element (12) configured to acquire a measurement signal (S1) of the vibrations of the running rail (Q) along at least one axis, at least one communication element (14) configured to emit at least one detection alert (AL, Ali), at least one processing element (13) configured to transform the measurement signal (S1) into a plurality of wavelets so as to deduce a plurality of coefficients (α), classify the coefficients (α) so as to determine in the measurement signal (S1) the presence of an approaching rail vehicle (V*), and activate the communication element (14) upon detection of an approaching rail vehicle (V*). Figure 2

Description

Title of the invention: Device for detecting a railway vehicle on a railway track and associated method. Technical field

[0001] The present invention relates to the field of safety of persons on a railway track and more particularly to a detection device enabling an operator working on a railway track to be alerted to the imminent arrival of a railway vehicle.

[0002] As is known, a railway line must regularly undergo maintenance or repair work to ensure that a railway vehicle can travel on it safely. Such operations are carried out by operators who must have direct access to the railway line. In some cases, the movement of railway vehicles on this railway line cannot be interrupted. Therefore, it is necessary to warn the operators working on the railway line of the approach of a railway vehicle so that they can move away from the railway line to avoid any accident.

[0003] Moreover, due to the speed of the railway vehicle, its mass and the low coefficient of friction existing between the rails of the railway track and the wheels of the railway vehicle, the latter requires a significant distance to brake, which can reach several kilometers.

[0004] In the prior art, it is known to position a railway vehicle detection device on a railway track in order, on the one hand, to detect the arrival of a railway vehicle in advance and, on the other hand, to warn the operators.

[0005] A detection device comprising a mechanical component that is moved when a railway vehicle passes is known. Given the speed of a railway vehicle, the detection device must therefore be positioned well in advance of the operators' positions to allow them sufficient time to move away from the railway track. Communication between the mechanical detection device and the operators is not optimal due to the distance, which presents a drawback.

[0006] A detection device is also known which receives the GPS coordinates of railway vehicles but it cannot be used in tunnels for which GPS data is not transmitted.

[0007] An optical detection device is also known, but this device also has limitations in its use. In particular, when a foreign body moves or falls onto the railway track, this can lead to detection errors.

[0008] The invention thus aims to eliminate at least some of these drawbacks by proposing a new system for detecting a railway vehicle on a railway track. The invention relates in particular to a detection system with increased reliability. PRESENTATION OF THE INVENTION

[0009] The invention relates to a device for detecting at least one railway vehicle on a railway track, the detection device comprising: • at least one fastening device configured to be removably attached to a railway track rail, • at least one measuring device configured to acquire a signal measuring the vibrations of the running rail along at least one axis, • at least one communication device configured to send at least one detection alert to an operator positioned on the railway track.

[0010] According to one aspect, the fixing member and the measuring member could be integrated together.

[0011] The invention is remarkable in that the detection device comprises at least one computing element configured to: • transform the measurement signal into a plurality of wavelets in order to deduce a plurality of coefficients, each coefficient representing the measurement signal according to a frequency band and a resolution, • classify the coefficients using a pre-trained classification algorithm to determine the presence of an approaching rail vehicle in the measurement signal, and • activate the communication device when an approaching rail vehicle is detected.

[0012] A wavelet transformation is advantageous compared to other transformations, for example, the MFCC type (Mel-Frequency Cepstral Coefficients), which is based on the spectral analysis of fixed time windows. The wavelet transformation captures both high-frequency transient events and low-frequency components. This multi-scale analysis allows for more precise vibration detection, making it more suitable for non-stationary signals, such as railway vibrations.

[0013] Wavelet transformation is also more robust to additive noise such as environmental or operational noise. The localized nature of wavelet transformation allows for more efficient noise differentiation and attenuation, resulting in clearer and more reliable feature extraction from noisy data. By comparison, a MFCC-type transformation is more sensitive to noise distortion.

[0014] Thanks to the invention, a detection alert can be issued reliably and in advance to warn operators on the railway track to ensure their safety.

[0015] In one aspect, the classification algorithm is of the partition clustering type.

[0016] Preferably, the classification algorithm is implemented through unsupervised learning, in particular of the partition clustering type. The use of a clustering algorithm, compared to a neural network, allows for more robust detection of new classes of railway vehicles entering the railway network without requiring retraining. Furthermore, it requires less training data than a neural network. A partition clustering classification algorithm, for example, of the K-Means type, is faster and requires less memory. Also, due to the reduced number of computational operations and memory accesses, a partition clustering classification algorithm places less strain on the detection device's battery, thus improving its autonomy.

[0017] According to one aspect, the communication device includes at least one audible alarm, one visual alarm, or one haptic (vibratory) alarm. Thus, the detection device can directly alert the operators' senses to warn them.

[0018] According to one aspect, the communication device is configured to send at least one computer detection alert to at least one remote electronic device, in particular, electronic equipment worn by an operator. Thus, the alert can be processed remotely from the detection device to activate electronic equipment worn by an operator, in particular, personal protective equipment.

[0019] According to one aspect, the fastening member is configured to be magnetically attached to the running rail. It can thus be conveniently moved when operators are working in different railway areas.

[0020] According to one aspect, the measuring device is an accelerometer or a piezoelectric sensor.

[0021] The invention also relates to a security system comprising at least one detection device as previously described and at least one electronic device, the communication unit of the detection device being configured to send at least one computer detection alert to said electronic device.

[0022] Preferably, the detection device and the electronic equipment are connected by a communication network.

[0023] According to one aspect, the electronic equipment includes at least one audible alarm, one visual alarm, or one vibrating alarm. Thus, the warning is personal to each operator so that they are effectively alerted.

[0024] The invention also relates to a method for detecting a railway vehicle on a railway track by means of a detection device as described above, the fastening member being removably fixed to a running rail of the railway track, the method comprising steps consisting of: • Acquire a vibration measurement signal from the running rail along at least one axis, • Transform the measurement signal into a plurality of wavelets in order to deduce a plurality of coefficients, each coefficient representing the measurement signal according to a frequency band and a resolution, • Classify the coefficients using a pre-trained classification algorithm to determine the presence of an approaching rail vehicle in the measurement signal, and • Issue a detection alert when an approaching rail vehicle is detected.

[0025] According to one aspect, the process comprises steps consisting of: • Issue a detection alert upon detection of an approaching rail vehicle to at least one electronic device, and • Activate an electronic equipment alarm upon receiving the detection alert.

[0026] According to one aspect, the electronic equipment is carried by an operator located near the railway track. PRESENTATION OF THE FIGURES

[0027] The invention will be better understood upon reading the following description, given by way of example, and referring to the following figures, given by way of non-limiting examples, in which identical references are given to similar objects.

[0028] Fig. 1 is a schematic representation of a detection device according to one embodiment of the invention for detecting an approaching railway vehicle.

[0029] Fig. 2 is a schematic representation of the detection device of Fig. 1.

[0030] Figure 3 is a schematic representation of the functioning of the organ of calculation of the detection device.

[0031] Fig. 4 is a schematic representation of the emission of a computer detection alert to an electronic device worn by an operator.

[0032] It should be noted that the figures set out the invention in detail to implement the invention, said figures being of course able to serve to better define the invention where appropriate. DETAILED DESCRIPTION OF THE INVENTION

[0033] With reference to [Fig. 1], a detection device 10 for at least one rail vehicle V on a railway track W is shown. Such a detection device 10 makes it possible to ensure safety on a railway track W, in particular, to protect operators OP traveling on the railway track W.

[0034] In a known manner, the railway track W comprises several running rails Q to support the axles of a railway vehicle V. In this example, with reference to [Fig.1], each running rail Q extends longitudinally along an axis X, laterally along an axis Y and vertically along an axis Z.

[0035] As is known, a railway vehicle V comprises several railway trainsets, each consisting of a plurality of cars or wagons (not shown). The railway vehicle V may be a vehicle for the transport of passengers, goods, or other. In this example, the longitudinal axis X represents the direction of travel of the railway vehicle V. In other words, the longitudinal axis X extends from the rear to the front of the railway vehicle V.

[0036] According to the invention, with reference to [Fig.2], the detection device 10 comprises at least one fastening member 11 configured to be removably fixed to a running rail Q of the railway track W, in particular laterally.

[0037] With reference to [Fig. 2], the detection device 10 comprises at least one measuring element 12 configured to acquire a measurement signal SI of vibrations of the running rail Q along at least one axis. The detection device 10 comprises at least one communication element 14 configured to transmit at least one detection alert AL to an operator OP positioned on the railway track W.

[0038] As illustrated in [Fig. 2], the detection device 10 comprises at least one computing unit 13 configured to activate the communication unit 14 upon detection of an approaching rail vehicle V* from the measurement signal SL

[0039] The functions of the computing unit 13 are presented in detail in [Fig. 3]. The computing unit 13 is configured to transform TO the measurement signal SI into a plurality of wavelets so as to deduce a plurality of wavelet coefficients a, each coefficient a representing the measurement signal SI according to a frequency band and a resolution.

[0040] The calculation unit 13 is configured to classify the wavelet coefficients a by a classification algorithm AC, previously trained, so as to determine in the measurement signal SI the presence of a railway vehicle approaching V*.

[0041] The computing unit 13 is finally configured to activate the communication unit 14 upon detection of an approaching railway vehicle V*.

[0042] Advantageously, thanks to the invention, the detection device 10 makes it possible to determine, by means of vibrations in the rail Q, whether a rail vehicle V is approaching, thus providing early warning to the operators OP so they can take safety precautions. Such a detection device 10 can be positioned near the operators OP since detection can be carried out when the rail vehicles V are still some distance away. In other words, detection by vibration analysis allows for early detection, which is advantageous.

[0043] Furthermore, such a detection device 10 can be installed in a tunnel because it does not require a GPS connection and is not affected by foreign objects moving on the railway track. Indeed, the AC classification algorithm advantageously allows for the discrimination of railway vehicles from foreign objects.

[0044] The various elements of the detection device 10 will now be presented.

[0045] With reference to [Fig.2], the fastening member 11 allows the detection device 1 to be connected to the railway rail Q so that the vibrations of the railway rail Q are reliably detected by the measuring member 12. In other words, the fastening member 11 ensures that the measuring member 12 detects the measurement signal S1 very accurately in order to achieve a relevant detection as will be shown later.

[0046] The fastening member 11 can take various forms, for example, a magnet as illustrated in [Fig. 4]. A magnet allows for quick and easy removable mounting and is advantageous for attaching to a metallic railway track Q. It goes without saying that other fastening members 11 could be suitable, in particular a suction cup, a clamp, a mechanical assembly, and the like.

[0047] The fastening member 11 is preferably connected to the core of the railway rail Q so as not to interfere with railway traffic.

[0048] The measuring element 12 is configured to acquire a measurement signal SI of the vibrations of the running rail Q along at least one axis, in particular, the vertical axis Z which is sensitive to variations related to the weight of the axles.

[0049] When a railway vehicle V travels on the railway track W, it causes the railway rails Q to vibrate according to a vibration signal specific to it. This vibration signal is notably a function of its length, mass, configuration, number of axles, type of motorization, speed, and load.

[0050] The measuring element 12 can take various forms, in particular, an accelerometer, a piezoelectric sensor, or the like. The vibration signal from the railway rail Q is converted digitally to obtain a digital measurement signal SI.

[0051] The communication device 14 is configured to transmit an AL detection alert to one or more operators OP. The AL detection alert can take various forms, in particular, an audible alert, a haptic / vibratory alert, a visual alert, or a computer alert ALi. For example, the communication device 14 can take various forms, such as an audible alarm (in particular above 120 dB), an optical alarm (in particular colored), or an electronic communication card configured to transmit a computer detection alert over a communication network, in particular low-speed or high-speed, so as to transmit the computer detection alert to one or more electronic receiving devices 20, in particular at least one central server that collects all the AL detection alerts.

[0052] Preferably, the receiving equipment 20 is positioned near the OP operators working on the railway track W, and preferably, is carried by the OP operators.

[0053] According to a preferred aspect, the receiving equipment 20 configured to receive the computer detection alert ALi is personal protective equipment for an operator. The electronic equipment 20 includes for this purpose at least one warning device, for example, an audible warning device, an optical warning device, a vibrating warning device, etc.

[0054] Advantageously, each operator OP is thus warned in advance and personally of the arrival of a rail vehicle V* on the railway track W on which he / she is working. This allows him / her sufficient time to move away from the railway track to safety.

[0055] According to the invention, as illustrated in [Fig.2], the computing unit 13 is configured to transform the measurement signal SI according to a plurality of wavelets so as to deduce a plurality of wavelet coefficients a, each wavelet coefficient a representing the measurement signal S1 according to a frequency band and a resolution.

[0056] The measurement signal SI is thus mathematically decomposed into wavelet coefficients α, which can be arranged in the form of a wavelet coefficient matrix α. The wavelet transformation is a mathematical transformation that allows for the detailed analysis of vibrational behavior. This transformation is therefore relevant in the present case because it allows for the attenuation of non-vibrational signals, i.e., damping. The wavelet transformation captures both high-frequency transient events and low-frequency components. This multi-scale analysis allows for more precise vibration detection, making it more suitable for non-stationary signals, such as railway vibrations.

[0057] The computing unit 13 is configured to classify the coefficients a using the AC classification algorithm, which has been previously trained to determine the presence of an approaching railway vehicle V* in the measurement signal SI. In other words, the AC classification algorithm makes it possible to detect the vibration signature of a railway vehicle approaching the detection device 10.

[0058] Preferably, the AC classification algorithm has been pre-trained to partition wavelet coefficients α that correspond to an approaching railway vehicle from wavelet coefficients α that do not correspond to an approaching railway vehicle. The detection device 10 can distinguish vibrations caused by a railway vehicle from other sources of vibration such as roads, construction machinery, etc.

[0059] Training is preferably carried out in an unsupervised manner (clustering) but it goes without saying that supervised training could also be carried out.

[0060] For supervised training, measuring devices are positioned at various positions of a railway track W to measure a measurement signal which corresponds to railway vehicles whose characteristics are known, in particular, the travel route, the travel speed, the actual position, the nature of the railway vehicle (railway identifier, mass, number of axles, number of cars, type of motorization, etc.).

[0061] The AC classification algorithm is preferably of the partition "clustering" type, for example, K-MEANS or a DBSCAN.

[0062] The AC classification algorithm of the "clustering" type is capable of grouping the data into at least two clusters (approaching train / no train), ideally three (vibration of the approaching train / vibration of other sources no train / no vibration), but without being limited to three.

[0063] The AC classification algorithm of the "clustering" type can also group vibration data into different classes based, in particular, on the type of railway vehicle, and / or the distance at which it is located, and / or the speed at which it travels, among other things.

[0064] In this example, for training the AC clustering-type classification algorithm, 1000 to 15000 observations are used, but not limited to this number. Ideally, the classes of the training observations are balanced.

[0065] To obtain the observations needed to train the AC clustering algorithm, an inertial sensor network is deployed along the railway tracks to capture the vibrations caused by the passage of railway vehicles. The vibration measurements (observations) are obtained during the passage of the railway vehicles.

[0066] To annotate the observations, information on their speed, the number of wagons, the weight and position of the rail vehicle, and other data can be obtained from a railway database. The railway database may include, in particular, a transport plan, data from a black box of the rail vehicle and / or a rail vehicle driver advisory system containing the GPS position, GPS speed, recommended speed, track gradients, stops and stations, etc.

[0067] Annotations can also be made from video recordings obtained using cameras geolocated by the location of the sensors.

[0068] The AC classification algorithm could also be based on a neural network (NN), a support vector machine (SVM) or a random forest.

[0069] The AC classification algorithm of the SVM type can be trained to detect and classify a railway vehicle as well as its speed, number of wagons, weight and position of the railway vehicle.

[0070] To obtain an annotated observation for training the AC classification algorithm of the SVM type, an inertial sensor network is deployed along the railway track to capture the vibrations caused by the passage of the railway vehicle. The vibration measurements (observations) are obtained during the passage of the railway vehicles.

[0071] To annotate the observations, information on their speed, the number of wagons, the weight and position of the railway vehicle and other data can be obtained from a railway database. The railway database may include, in particular, a transport plan, data from a black box of the railway vehicle and / or a railway vehicle driver advisory system containing the GPS position, the recommended speed, stops and stations, speed, etc.

[0072] Annotations can also be made from video recordings obtained using cameras geolocated by the location of the sensors.

[0073] A series of features (called "features") is extracted from the vibration measurements, such as the duration of the event, the number of peaks, the signal intensity, the maximum amplitude, the average amplitude, the variance of the peaks, the average distance between the peaks, among others, and is used to train the AC classification algorithm of the SVM type.

[0074] Once the AC classification algorithm of the SVM type has been trained and validated, it can be implemented in a detection device 10.

[0075] To train the AC classification algorithm of the SVM type to distinguish the vibrations of the railway vehicle from other sources of vibration such as machines on a construction site or nearby roads, the training procedure is analogous; sensors are placed on construction sites where the railway track W is close to the road and vibration profiles are noted using video recordings, for example.

[0076] In practice, a measurement signal SI of a railway vehicle has a characteristic vibration signature that allows the vibrations of a railway vehicle to be differentiated from vibrations associated with another element. Thus, the AC classification algorithm makes it possible to determine whether the measurement signal SI contains information about the presence of a railway vehicle and also whether the vehicle is approaching.

[0077] According to one aspect, with reference to [Fig.3], the AC classification algorithm is also configured to perform an estimation of several parameters of the determined railway vehicle, in particular, the position, speed, weight and nature of the railway vehicle.

[0078] Thus, thanks to the AC classification algorithm, each nearby rail vehicle can be precisely identified according to several criteria. This makes it possible to strengthen the supervision of traffic on the railway line and improve safety.

[0079] Preferably, the detection device 10 includes a memory for storing the AC classification algorithm and the training database. Preferably, the detection device 10 includes a power supply for powering the electrical components of the detection device 10. The power supply is preferably in the form of a battery.

[0080] A partition clustering type classification algorithm reduces interactions with memory and calculations, thus limiting the use of the electric battery and extending the lifespan.

[0081] Preferably, the computing unit 13 can be accessed by wired or wireless means for maintenance or updates.

[0082] The receiving equipment 20 includes a communication element configured to receive a computer alert ALi from the detection device 10. The receiving equipment 20 includes at least one warning device, in particular, an audible warning device (in particular greater than 120dB), an optical warning device (in particular coloured) or a vibrating warning device.

[0083] Preferably, the receiving equipment 20 comprises a waterproof and shock-resistant housing so that it can be used under the constraints of the railway environment. For example, the receiving equipment 20 can be integrated into a bracelet, a watch, or safety glasses.

[0084] Optionally, with reference to [Fig. 2], the detection device 10 includes a life signal transmitter 15 S2 via the communication element 14. The life signal S2, also called the "heartbeat", is received by the receiving equipment 20 so as to ensure that the detection device 10 is always in a state of operation. The S2 life signal can take various forms, in particular, a periodic signal.

[0085] To this end, the receiving equipment 20 includes a monitoring device configured to receive the S2 life signal via its communication device. The monitoring device is configured to issue a fault alert if the S2 life signal is not received for a specified period, for example, 2 seconds. The fault alert can take various forms, in particular, an audible, visual, or vibrating alert.

[0086] By way of example, the receiving equipment 20 has a green light in the presence of the life signal S2 and a red light in the absence of the life signal S2.

[0087] An example of the implementation of a method for detecting an approaching rail vehicle will be presented with reference to [Fig. 4]. In this example, operators (OPs) are working on a railway track W, and the detection device 10 is positioned upstream of the operators (OPs) on the railway track W. Each operator (OP) is equipped with a receiving device 20 in the form of a wristband, but it is understood that it could be of a different shape.

[0088] The detection device 10 continuously measures the vibrations felt on the railway track Q on which it is mounted. A measurement signal SI is continuously acquired to monitor whether a railway vehicle is approaching.

[0089] At a first instant, the measurement signal SI is processed by the computing unit 13 which detects, thanks to the classification algorithm AC, two vibration signatures corresponding to two railway vehicles on the railway track W. In the present case, thanks to the classification algorithm AC, only one railway vehicle is approaching, that is to say, getting closer to the detection device 10.

[0090] A computer detection alert Ali is then issued by the detection device 10 which is received by the electronic equipment 20 which can then activate one of its warning devices to reactively warn the operator OP so that he moves away from the railway track W.

Claims

Demands

1. A detection device (10) for at least one rail vehicle (V) on a railway track (W), the detection device (10) comprising: • at least one fastening member (11) configured to be removably fixed to a running rail (Q) of the railway track (W), • at least one measuring member (12) configured to acquire a measurement signal (SI) of the vibrations of the running rail (Q) along at least one axis, • at least one communication member (14) configured to transmit at least one detection alert (AL, Ali) to an operator (OP) positioned on the railway track (W), • at least one processing member (13) configured to: • transform the measurement signal (SI) into a plurality of wavelets so as to deduce a plurality of coefficients (a), each coefficient (a) representing the measurement signal (SI) according to a frequency band and a resolution,• classify the coefficients (a) by a previously trained classification algorithm (AC) so as to determine in the measurement signal (SI) the presence of an approaching railway vehicle (V*), and • activate the communication device (14) upon detection of an approaching railway vehicle (V*).

2. A detection device (10) according to claim 1, wherein the classification algorithm (CA) is of the partitioning type "clustering w

3. Z>. Detection device (10) according to any one of claims 1 to 2, wherein the communication member (14) comprises at least one audible or visual warning device.

4. A detection device (10) according to any one of claims 1 to 3, wherein the communication member (14) is configured to issue at least one computer detection alert (Ali) to the attention of less a remote electronic device (20), in particular, an electronic device (20) worn by an operator (OP).

5. Detection device (10) according to any one of claims 1 to 4, wherein the fastening member (4) is configured to be magnetically fixed to the running rail (Q).

6. A detection device (10) according to any one of claims 1 to 5, wherein the measuring member (12) is an accelerometer or a piezoelectric sensor.

7. Security system (1) comprising at least one detection device (10) according to any one of claims 1 to 6 and at least one electronic equipment (20), the communication member (14) of the detection device (10) being configured to emit at least one computer detection alert (Ali) to said electronic equipment (20).

8. Security system (1) according to claim 7 in which the electronic equipment (20) comprises at least one audible alarm, visual alarm or vibrating alarm.

9. A method for detecting a railway vehicle (V) on a railway track (W) by means of a detection device (10) according to any one of claims 1 to 6, the fastening member (11) being removably fixed to a running rail (Q) of the railway track (W), the method comprising steps of: • Acquiring a measurement signal (SI) of the vibrations of the running rail (Q) along at least one axis, • Transforming the measurement signal (SI) into a plurality of wavelets so as to deduce a plurality of coefficients (a), each coefficient (a) representing the measurement signal (SI) according to a frequency band and a resolution, • Classifying the coefficients (a) by a previously trained classification algorithm (AC) so as to determine in the measurement signal (SI) the presence of an approaching railway vehicle (V*), and • Issuing a detection alert (AL,Ali) upon detection of an approaching rail vehicle (V*).

10. A method for detecting a railway vehicle (V) according to claim 9, the method comprising steps consisting of: 14 • Issue the detection alert (AL, Ali) upon detection of an approaching rail vehicle (V*) to at least one electronic equipment (20), and • Activate a warning device on the electronic equipment (20) following receipt of the detection alert (AL, Ali).

11. Method of detecting a railway vehicle (V) according to claim 10 wherein the electronic equipment (20) is carried by an operator (OP) located near the railway track (W).

Images