Method for predicting expected deceleration of at least one vehicle and corresponding system

Abstract

In a first aspect, the invention provides a computer-implemented method for predicting an expected deceleration of at least one vehicle, in particular at least one rail vehicle, the method comprises the steps of: a) providing, as inputs, at least one braking data (102) relating to the performance of a braking system of the at least one vehicle, at least one environmental data (104) relating to environmental conditions of a route along which the vehicle moves, at least one vehicle data (106) relating to a structure of the at least one vehicle, to the neural network (100); b) predicting an expected deceleration value (108) of the at least one vehicle on the basis of the at least one braking data (102), the at least one environmental data and the at least one vehicle data (106) by means of the neural network (100). In another aspect, the invention provides a corresponding system for predicting an expected deceleration of at least one vehicle.

Description

Technical Field

[0001] The present invention is generally in the field of vehicles; in particular, the present invention relates to a computer-implemented method for predicting an expected deceleration of at least one vehicle, and a corresponding system. Background Art

[0002] The prior art will be described below with particular reference to the rail vehicle industry; however, where applicable, the content of the following description may also be similarly applicable to vehicles in other industries.

[0003] The rail industry is eager to take advantage of new concepts in the management of rolling stock on the network. These new management concepts are aimed at increasing line capacity, rolling stock reliability and resistance to changing environmental conditions.

[0004] The braking performance of a rail vehicle may be impaired due to adverse environmental conditions. Disadvantageously, existing brake systems may only partially limit the impact of environmental conditions. Summary of the invention

[0005] Therefore, one object of the present invention is to provide a solution that enables prediction of the performance of a vehicle, in particular regarding the deceleration performance of the vehicle. This prediction can be used to improve the performance of the vehicle's braking system even under the worst environmental conditions. For example, the vehicle deceleration performance prediction can be used to:

[0006] - reduce the distance of the moving block function ("moving block train"), thereby increasing the capacity of the line;

[0007] - Facilitates the integration of automated vehicle operations by providing information that in prior art would be provided manually by the driver; and

[0008] - Capability to provide infrastructure managers with information on achievable speed reduction conditions on routes / tracks.

[0009] The foregoing and other objects and advantages are achieved according to one aspect of the invention by a computer-implemented method for predicting an expected deceleration of at least one vehicle having the features defined in claim 1, and according to another aspect by a system for predicting an expected deceleration of at least one vehicle having the features defined in claim 10. Preferred embodiments of the invention are defined in the dependent claims, the content of which is to be understood as an integral part of the present description. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The functional and structural features of some preferred embodiments of the system for predicting expected deceleration of at least one vehicle and the computer-implemented method for predicting expected deceleration of at least one vehicle according to the present invention will now be described. With reference to the accompanying drawings, in which:

[0011] - Figure 1 illustrates an exemplary neural network that can be used in a computer-implemented method for predicting an expected deceleration of at least one vehicle in accordance with the present invention;

[0012] - Figure 2 Two exemplary activation functions for a neural network are illustrated;

[0013] - Figure 3 illustrates the states of neurons in a "feed-forward" configuration; and

[0014] - Figure 4 An exemplary vehicle including a computer configured to execute a method for predicting an expected deceleration according to an embodiment of the present invention is illustrated. DETAILED DESCRIPTION

[0015] Before explaining multiple embodiments of the present invention in detail, it should be clarified that the present invention is not limited to the design details and configurations of the components presented in the following description or illustrated in the accompanying drawings in its application. The present invention can adopt other embodiments and be implemented or constructed in different ways in practice. It should also be understood that the wording and terminology have descriptive purposes and should not be interpreted as limiting. The use of "including" and "comprising" and its variants is intended to cover the elements and their equivalents described in detail below, as well as additional elements and their equivalents.

[0016] Reference by example Figure 4 , a first embodiment of a computer-implemented method for predicting an expected deceleration of at least one vehicle V, in particular at least one rail vehicle, is described below.

[0017] The method comprises the following steps:

[0018] Step a): providing the neural network 100 with at least one braking data 102 relating to the performance of a braking system of at least one vehicle, at least one environmental data 104 relating to environmental conditions of a route along which the vehicle moves, at least one vehicle data 106 relating to the structure of at least one vehicle V as input; and

[0019] Step b): predicting / determining, by means of the neural network 100 , an expected deceleration value 108 of at least one vehicle V based on the at least one braking data 102 , the at least one environmental data 104 and the at least one vehicle data 106 .

[0020] The data 102, 104, 106 provided to the neural network in step a) are selected because of their ability to have an influence on the result, ie the deceleration of the vehicle.

[0021] In other words, at least one braking data 102 may indicate the response of the vehicle's braking system, as well as potential degradation of the braking system and the use of a dedicated brake, which may include, for example, at least one of an electric brake, a magnetic track brake (MTB), a sand spreading device, an eddy current brake, and the like.

[0022] The at least one environmental data 104 can indicate environmental conditions in which the vehicle is moving.

[0023] The at least one vehicle data 106 may indicate vehicle characteristics that may affect deceleration of the vehicle and thus braking distance.

[0024] Preferably, the neural network 100 may include a feed-forward structure.

[0025] A feedforward neural network is a structure in which there are no loops between the layers, but information flows from the input layer to the hidden layer and then to the output layer. This structure is relatively simple and requires limited computing power, making it particularly suitable for implementation in real-time systems such as control units for brake systems.

[0026] refer to Figure 1 An exemplary neural network is shown, and below is an exemplary definition of a neuron-based artificial intelligence structure.

[0027] The input to neuron a is a linear function where b is the bias, w is the weight, p is the data input, and h is the output of the neuron. The output is the same as the activation function:

[0028] a=wp+b

[0029] h=F(a).

[0030] For example, an activation function can be of two types: sigmoid or rectifier.

[0031] Both of them have inherent advantages due to their characteristics:

[0032] -Sigmoid activation function allows more complex data structures and has regular output, thus avoiding value jumps;

[0033] -rectifier type activation function (ReLU) requires less computation.

[0034] Figure 2 Two graphs are shown in Figure 1. The left graph shows the rectifier type activation function. The right graph shows the sigmoid type activation function.

[0035] refer to Figure 3 ,The neurons in the feedforward structure can be represented by input layer, hidden layer and output layer.

[0036] Preferably, the computer-implemented method for predicting an expected deceleration of at least one vehicle, in particular at least one rail vehicle, may comprise, prior to step a) and step b), performing a training of a neural network.

[0037] In this case, neural network training can include:

[0038] - providing as input to the neural network at least one braking data, at least one environmental data, at least one vehicle data and an expected deceleration value, the expected deceleration value being a function of the at least one braking data, the at least one environmental data and the at least one vehicle data;

[0039] - determining a value of at least one parameter of the neural network as a function of the received at least one braking data, at least one environmental data, at least one vehicle data and the expected deceleration value.

[0040] Therefore, the expected deceleration value as a function of the at least one braking data, the at least one environmental data and the at least one vehicle data represents a known expected deceleration value obtained using the at least one braking data, the at least one environmental data and the at least one vehicle data (e.g. a measured expected deceleration value that can be derived from, for example, previous experimental measurement results).

[0041] Preferably, the training is based on a "back propagation" algorithm.

[0042] In other words, a feed-forward neural network can be trained using a back-propagation algorithm to check the consistency of the output (determination of vehicle deceleration). Once the output is accurate, the weights and biases (parameters) are frozen at the end of training.

[0043] Once vehicle deceleration is predicted, this value may also be transmitted to a control device included in the vehicle or to a remote infrastructure manager.

[0044] Preferably, the at least one braking data 102 related to the performance of the braking system of at least one vehicle may include at least one of the following types of data:

[0045] at least one deceleration datum indicating a deceleration value of at least one wheel W or at least one axle of at least one vehicle;

[0046] - at least one data indicating the number of wheels W in a skidding phase of at least one vehicle;

[0047] - at least one data indicative of deceleration of at least one vehicle in an initial phase of braking;

[0048] - at least one datum indicative of an actuation speed of at least one braking device of the braking system;

[0049] - at least one datum indicative of a slip speed of at least one wheel or at least one axle of at least one vehicle;

[0050] - at least one datum indicative of a steady-state value of at least one braking device of the braking system;

[0051] - at least one data indicating activation of a sand spreading device of at least one vehicle;

[0052] - at least one data indicating activation of magnetic brake pads MTB of at least one vehicle;

[0053] - at least one data indicating activation of a system / function arranged to compensate for a missed expected deceleration value;

[0054] - at least one data indicating the presence of a fault in a braking device of the braking system;

[0055] - at least one datum indicative of the number of activations over time of an exhaust valve associated with at least one braking device of a braking system of at least one vehicle.

[0056] A steady-state value of at least one braking device of a braking system may be understood as a value after which the required braking force has been reached, after a transient (eg in a pneumatic braking system the transient corresponds to the time taken for the brake cylinder to reach nominal pressure).

[0057] Preferably, the at least one environmental data 104 related to the environmental conditions of the route along which the vehicle moves may include at least one of the following types of data:

[0058] - at least one image data or at least one video data of the route (eg an image or a video acquired of the route);

[0059] - at least one temperature data indicating the temperature along the route;

[0060] - at least one rainfall data indicating the presence of rainfall along the route;

[0061] - at least one humidity datum indicative of a humidity level along the route;

[0062] - at least one adhesion datum indicative of an adhesion level along a route;

[0063] - at least one route data indicating an outline of the route.

[0064] Preferably, the at least one vehicle data 106 related to the structure of at least one vehicle may include at least one of the following types of data:

[0065] - at least one nominal deceleration data indicating a nominal deceleration value of at least one vehicle;

[0066] - at least one wheel or axle data indicating the number of wheels or axles of at least one vehicle;

[0067] - at least one sand spreading device data indicating the number of sand spreading devices of at least one vehicle;

[0068] - at least one data of magnetic pads indicating the number of MTB magnetic brake pads of at least one vehicle;

[0069] - at least one anti-skid system data indicating the fact that an anti-skid system WSP is active for each vehicle bogie or each vehicle axle;

[0070] - at least one deceleration compensation presence data indicating the fact that the vehicle comprises at least one missed deceleration compensation system / function of the vehicle.

[0071] Obviously, various types of brake data 102 related to the performance of the brake system, various types of environmental data related to the environmental conditions of the route along which the vehicle moves, and various types of vehicle data 106 related to the structure of at least one vehicle can be combined in any mode. Some possible combinations are given below by way of example.

[0072] Preferably, in an example embodiment, at least one braking data 102 related to the performance of the braking system of at least one vehicle may include at least one data indicating the slip speed of at least one wheel or axle of at least one vehicle, at least one environmental data 104 related to the environmental conditions of the route along which the vehicle moves may include at least one adhesion data indicating the adhesion level along the route, and at least one vehicle data 106 related to the structure of at least one vehicle may include at least one wheel or axle data indicating the number of wheels or axles of at least one vehicle. This example is mainly intended to monitor the adhesion level and its possible negative impact on the deceleration of the individual wheels of the vehicle.

[0073] In another example, preferably, at least one braking data 102 related to the performance of the braking system of at least one vehicle may include at least one data indicating the number of wheels in a slip phase of at least one vehicle, at least one environmental data 104 related to the environmental conditions of the route along which the vehicle moves may include at least one adhesion data indicating the adhesion level along the route, and at least one vehicle data 106 related to the structure of at least one vehicle may include at least one anti-skid system data indicating the fact that the anti-skid system WSP acts on each vehicle bogie or each vehicle axle. For example, at least one vehicle data 106 related to the structure of at least one vehicle may also include: at least one data of the presence of deceleration compensation, which indicates the fact that the vehicle includes a missed deceleration compensation system / function of at least one vehicle. This example is mainly intended to monitor the adhesion level between the individual wheels of the vehicle and the positive impact on deceleration that may be generated by the missed deceleration compensation system / function of at least one vehicle.

[0074] In yet another example, preferably, at least one braking data 102 related to the performance of the braking system of at least one vehicle may include at least one data indicating a steady-state value of at least one braking device of the braking system, and at least one data indicating that a braking device of the braking system is faulty, at least one environmental data 104 related to the environmental conditions of the route along which the vehicle moves may include at least one of any data items in the above list, and at least one vehicle data 106 related to the structure of at least one vehicle may include at least one of any data items in the above list. This example is mainly intended to monitor the condition of the braking system and the negative impact that a faulty braking device may have on deceleration.

[0075] Preferably, when there are multiple types of brake data 102 related to the performance of the brake system of at least one vehicle, each type of brake data may be subjected to a data integration procedure separately before being provided to the neural network.

[0076] For example, if there are 100 braking data (including: 50 data indicating the actuation speed of at least one braking device of the braking system, and 50 data indicating the slip speed of at least one wheel or at least one axle of at least one vehicle), then before being provided to the neural network, the 50 data indicating the actuation speed of at least one braking device of the braking system can undergo their own data integration procedure, and the 50 data indicating the slip speed of at least one wheel or at least one axle of at least one vehicle can undergo their own data integration procedure.

[0077] Preferably, when there are multiple types of environmental data 104 related to environmental conditions along the route along which the vehicle moves, each type of environmental data may be subjected to a data integration procedure separately before being provided to the neural network.

[0078] For example, if there are 100 environmental data (including: 50 temperature data indicating the temperature along the route and 50 rainfall data indicating the presence of rainfall along the route), the 50 temperature data can undergo their own data integration procedure and the 50 rainfall data can undergo their own data integration procedure before being provided to the neural network.

[0079] Preferably, when there are multiple vehicle data types 106 associated with the structure of at least one vehicle, each type of vehicle data may be subjected to a data integration procedure separately before being provided to the neural network.

[0080] For example, if there are 100 vehicle data (including: 50 wheel or axle data and 50 sand spreading device data), the 50 wheel or axle data can undergo their own data integration process and the 50 sand spreading device data can undergo their own data integration process before being provided to the neural network.

[0081] Preferably, the data integration procedure may include at least one of the following:

[0082] - Determination of the sum of the data, determination of the average value of the data, determination of the absolute minimum value among the data, determination of the absolute maximum value among the data.

[0083] For example, the data can be integrated to obtain vectors from the raw data matrix of the time series. This integration stage represents a preliminary analysis of the data. In the case of using raw video streams for pollutant identification (such as defined in "WO2021100003"), the integration of environmental data may be more advanced. The choice of calculation type depends on the properties of the input data.

[0084] Preferably, at least one braking data 102 related to the performance of the braking system of the at least one vehicle, at least one environmental data 104 related to the environmental conditions of the route along which the vehicle moves, and at least one vehicle data 106 related to the structure of the at least one vehicle can be associated via a time variable. Such a time variable can also be provided to or known to the neural network, which in turn can associate the at least one braking data 102 related to the performance of the braking system of the at least one vehicle, at least one environmental data 104 related to the environmental conditions of the route along which the vehicle moves, and at least one vehicle data 106 related to the structure of the at least one vehicle received at the input.

[0085] In another aspect, the invention relates to a system for predicting an expected deceleration of at least one vehicle, in particular at least one rail vehicle, said system comprising at least one computer arranged to perform a method according to any of the preceding claims.

[0086] For example, the computer 101 may include at least one control device, such as a processor, a microprocessor, a controller, a microcontroller, an FPGA, a PLC, a control unit, or a control box.

[0087] Preferably, the computer may be arranged to receive from a communication device of the at least one vehicle at least one braking data relating to the performance of a braking system of the at least one vehicle, at least one environmental data relating to environmental conditions of a route along which the vehicle moves, and at least one vehicle data relating to the structure of the at least one vehicle, for provision to the neural network.

[0088] The communication device may be, for example, a CAN network of the vehicle.

[0089] Preferably, the computer is arranged to receive, from control devices of additional vehicles moving along the route, at least one braking data relating to the performance of a braking system of at least one vehicle, at least one environmental data relating to environmental conditions of the route along which the vehicle moves, and at least one vehicle data relating to the structure of at least one vehicle, for provision to the neural network.

[0090] Preferably, the computer is arranged to receive from a control system of the at least one vehicle at least one braking data relating to the performance of a braking system of the at least one vehicle, at least one environmental data relating to the environmental conditions of a route along which the vehicle moves, and at least one vehicle data relating to the structure of the at least one vehicle for supply to the neural network. The at least one braking data relating to the performance of a braking system of the at least one vehicle, at least one environmental data relating to the environmental conditions of a route along which the vehicle moves, and at least one vehicle data relating to the structure of the at least one vehicle may be generated by the control system according to a predetermined adhesion map.

[0091] The adhesion map can be understood as a correspondence between geographical locations on a route (eg a railway) and available relative wheel adhesion values ​​and rolling surfaces (eg the track of the route).

[0092] Preferably, the at least one vehicle may include at least one railway vehicle. In addition, there may be more than one vehicle, and they may be associated with each other to form a fleet of vehicles, such as a railway fleet.

[0093] However, the invention is preferably applicable to any type of vehicle. This may include, for example, a railway vehicle / train, a road vehicle, a car, or a truck (e.g., a highway semi-trailer truck, a mining truck, or a truck for transporting wood, etc.), and the route may be, for example, a track, a road, or a path.

[0094] A real-time example embodiment applicable to a railway vehicle or train is described below.

[0095] On a vehicle (e.g., a railroad train comprising a plurality of railroad vehicles), pre-analyzed input data may be fed to a deep (e.g., "feed-forward") neural network to generate predictions of vehicle decelerations. Training may be performed offline by providing input data acquired during vehicle commissioning and comparing the vehicle decelerations generated by the neural network to actual average vehicle decelerations based on measurements.

[0096] The trained neural network can then be incorporated into the brake control system to provide predictions of train deceleration.

[0097] Neural networks can be used in four vehicle architectures, depending on the source of the data used to make predictions.

[0098] Using a local architecture, each vehicle brake system control unit can calculate the expected deceleration of the vehicle based on locally available data.

[0099] With the established architecture, the vehicle brake system control unit may determine the expected deceleration of the vehicle based on data from all other control units of the vehicle's brake system shared over a communication means (eg, a bus of the brake system).

[0100] Using the data from the preceding vehicle, the control unit of the braking system can calculate the expected deceleration of the vehicle based on the data from the preceding vehicle shared by the control device TCMS acting as proxy for the control system of the vehicle.

[0101] Using the vehicle data from the adhesion map, the control unit can calculate the expected deceleration of the train / rail vehicle based on the shared data from the TCMS derived from the adhesion map.

[0102] The advantage achieved is therefore that a solution is provided which allows prediction of the vehicle's performance, in particular the vehicle's deceleration performance, which can be used to improve the performance of the braking system even under the worst environmental conditions.

[0103] Various aspects and embodiments of a computer-implemented method for predicting an expected deceleration of at least one vehicle and a system for predicting an expected deceleration of at least one vehicle according to the present invention have been described. It should be understood that each embodiment can be combined with any other embodiment. In addition, the present invention is not limited to the described embodiments, but can be varied within the scope defined by the attached claims.

Claims

1. A computer-implemented method for predicting an expected deceleration of at least one vehicle, in particular at least one railway vehicle, comprising the following steps: a) providing as input to a neural network (100) at least one braking data (102) relating to the performance of a braking system of the at least one vehicle, at least one environmental data (104) relating to environmental conditions along a route along which the vehicle moves, at least one vehicle data (106) relating to the structure of the at least one vehicle; b) predicting, by means of the neural network (100), an expected deceleration value (108) of the at least one vehicle based on the at least one braking data (102), the at least one environmental data (104) and the at least one vehicle data (106).

2. The method according to claim 1, comprising: Before step a) and step b), performing training of the neural network (100); Wherein, the training of the neural network includes: providing as input to the neural network at least one braking data, at least one environmental data, at least one vehicle data, and an expected deceleration value, the expected deceleration value being a function of the at least one braking data, the at least one environmental data, and the at least one vehicle data; A value of at least one parameter of the neural network is determined based on the received at least one braking data, at least one environmental data, at least one vehicle data and at least one expected deceleration value.

3. The method according to claim 1 or 2, wherein: The training is based on the "back propagation" algorithm.

4. A method according to any one of the preceding claims, wherein: The neural network (100) comprises a feed-forward structure.

5. A method according to any one of the preceding claims, wherein: The at least one braking data (102) related to the performance of the braking system of the at least one vehicle comprises at least one of the following: at least one deceleration data indicating a deceleration value of at least one wheel or at least one axle of the at least one vehicle; at least one data indicative of the number of wheels in a slip phase of said at least one vehicle; at least one data indicative of deceleration of the at least one vehicle during an initial phase of braking; at least one datum indicative of an actuation speed of at least one braking device of the braking system; at least one datum indicative of a slip speed of at least one wheel or at least one axle of the at least one vehicle; at least one datum indicative of a steady-state value of at least one braking device of the braking system; at least one data indicative of activation of a sand spreading device of the at least one vehicle; at least one data indicative of activation of magnetic brake pads of the at least one vehicle; at least one data indicating activation of a system / function arranged to compensate for a missed expected deceleration value; at least one data indicating that a brake device of the brake system has a fault; At least one datum indicative of a number of activations over time of an exhaust valve associated with at least one braking device of a braking system of the at least one vehicle.

6. A method according to any one of the preceding claims, wherein: The at least one environmental data (104) related to environmental conditions of a route along which the vehicle moves comprises at least one of the following: at least one image data or at least one video data of the route; at least one temperature data indicative of a temperature along the route; at least one rainfall amount data indicating the presence of rainfall along the route; at least one humidity data indicative of a humidity level along the route; at least one adhesion data indicative of an adhesion level along the route; At least one route data indicating an outline of the route.

7. A method according to any one of the preceding claims, wherein: The at least one vehicle data (106) related to the structure of the at least one vehicle comprises at least one of the following: at least one nominal deceleration data indicating a nominal deceleration value of the at least one vehicle; at least one wheel / axle data indicating the number of wheels or axles of the at least one vehicle; at least one sand spreading device data indicating a quantity of sand spreading devices of the at least one vehicle; at least one data of magnetic pads MTB indicating the number of magnetic brake pads of said at least one vehicle; at least one anti-skid system data indicating the fact that an anti-skid system WSP is active for each vehicle bogie or each vehicle axle; There is at least one data of deceleration compensation indicating the fact that said vehicle comprises a missed deceleration compensation system / function of said at least one vehicle.

8. A method according to any one of the preceding claims, wherein: When there are a plurality of types of braking data relating to performance of a braking system of said at least one vehicle, each type of braking data is separately subjected to a data integration procedure before being provided to said neural network; and / or When there are a plurality of types of environmental data related to environmental conditions along a route along which the vehicle moves, each type of environmental data is separately subjected to a data integration procedure before being provided to the neural network; and / or When there are a plurality of types of vehicle data related to the structure of the at least one vehicle, each type of vehicle data undergoes a data integration procedure separately before being provided to the neural network.

9. The method according to claim 8, wherein: The data integration procedure includes at least one of the following: Determination of the sum of the data, determination of the average of the data, determination of the absolute minimum value among the data, determination of the absolute maximum value among the data.

10. System for predicting an expected deceleration of at least one vehicle, in particular at least one railway vehicle, the system comprising at least one computer (101) arranged to execute the method according to any of the preceding claims.

11. The system for predicting expected deceleration according to claim 10, wherein: The computer is arranged to receive from a communication device of the at least one vehicle at least one braking data relating to the performance of a braking system of the at least one vehicle, at least one environmental data relating to environmental conditions of a route along which the vehicle moves, and at least one vehicle data relating to a structure of the at least one vehicle, for provision to the neural network.

12. The system for predicting expected deceleration according to claim 10, wherein: The computer is arranged to receive, from control devices of additional vehicles passing through the route, at least one braking data relating to the performance of a braking system of the at least one vehicle, at least one environmental data relating to environmental conditions of the route along which the vehicle moves, and at least one vehicle data relating to the structure of the at least one vehicle, for provision to the neural network.

13. The system for predicting expected deceleration according to claim 10, wherein: The computer is arranged to: receive from a control system of the at least one vehicle at least one braking data relating to performance of a braking system of the at least one vehicle, at least one environmental data relating to environmental conditions of a route along which the vehicle moves, and at least one vehicle data relating to a structure of the at least one vehicle, for provision to the neural network; The at least one braking data related to the performance of the braking system of the at least one vehicle, the at least one environmental data related to the environmental conditions of the route along which the vehicle moves, and the at least one vehicle data related to the structure of the at least one vehicle are generated by the control system according to a predetermined adhesion map.

Images