Ethernet-based train fault monitoring and predicting method and system and storage medium

Abstract

The invention discloses a train fault monitoring and predicting method and system based on the Ethernet and a storage medium, and the method comprises the steps that a vehicle-mounted controller requests to establish communication connection with an edge server corresponding to a target position where a train is located after receiving train operation data, and transmits the train operation data to the edge server; the edge server divides the train operation data into hot operation data and cold operation data, requests to obtain historical operation data corresponding to the hot operation data, analyzes the historical operation data based on a preset analysis model, and generates a first analysis result to perform fault monitoring on the train; and the cloud server searches train parameters corresponding to the cold operation data, analyzes the cold operation data and the train parameters, and generates a second analysis result to perform fault prediction on the train. According to the invention, the train operation data is analyzed and processed through the Ethernet, the cloud edge cooperation mechanism and the preset analysis model formed by training the deep neural network, so that the train fault can be monitored and predicted timely and accurately.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to an Ethernet-based train fault monitoring and prediction method, system and storage medium. Background Art

[0002] With the advancement of science and technology, trains, especially high-speed trains, have developed into modern high-tech transportation tools. In order to ensure the safety, reliability, stability and comfort of train operation, it is necessary to monitor and predict possible failures of trains by collecting various data of various equipment and components during train operation.

[0003] At present, the operating status data of various equipment and devices in the train is often transmitted and processed through the Train Communication Network (TCN). However, with the increase of intelligent equipment and functions in the train, the data types and data levels that need to be transmitted and analyzed are increasing. The traditional train communication network built based on fieldbus can no longer meet the current transmission needs, and it is also difficult to quickly and accurately process and analyze various types of data. As a result, it will not be possible to accurately and timely monitor and predict possible faults in the train, especially for data types with high real-time requirements, which will cause immeasurable losses.

[0004] Therefore, how to timely and accurately monitor and predict possible faults on trains is a technical problem that needs to be solved urgently. Summary of the invention

[0005] The main purpose of the present invention is to provide a train fault monitoring and prediction method, system and storage medium based on Ethernet, aiming to solve the technical problem of how to timely and accurately monitor and predict possible faults on the train in the prior art.

[0006] To achieve the above object, the present invention provides a train fault monitoring and prediction method based on Ethernet, wherein the train comprises an on-board controller and an on-board PHM module connected to the on-board controller via an Ethernet train backbone network, and the train fault monitoring and prediction method comprises:

[0007] After receiving the train operation data transmitted by the on-board PHM module based on the Ethernet train backbone network, the on-board controller determines the target location of the train and requests to establish a communication connection with the edge server corresponding to the target location to transmit the train operation data to the edge server;

[0008] The edge server divides the train operation data into hot operation data and cold operation data, and requests a cloud server that is communicatively connected to the edge server to obtain historical operation data corresponding to the hot operation data, and transmits the cold operation data to the cloud server;

[0009] The edge server analyzes the thermal operation data and the historical operation data based on a preset analysis model to generate a first analysis result, and performs fault monitoring on the train according to the first analysis result;

[0010] The cloud server searches for train parameters corresponding to the cold operation data of the train, analyzes the cold operation data and the train parameters, and generates a second analysis result to predict faults of the train.

[0011] Preferably, the thermal operation data includes a vibration signal corresponding to a motor shaft in a running gear of the train, and the edge server analyzes the thermal operation data and the train historical data based on a preset analysis model, and the step of generating a first analysis result includes:

[0012] Performing time domain analysis and frequency domain analysis on the vibration signal respectively to obtain an original time domain feature group and an original frequency domain feature group of the vibration signal;

[0013] Filtering the vibration signal, and extracting a filtering time domain feature group and a filtering frequency domain feature group from the filtered vibration signal;

[0014] The original time domain feature group, the original frequency domain feature group, the filtered time domain feature group and the filtered frequency domain feature group are formed into a feature matrix, and the historical operation data are formed into a historical feature matrix;

[0015] Based on the input layer, the first hidden layer, the second hidden layer and the output layer in the preset analysis model, the feature matrix and the historical feature matrix are subjected to fault analysis to obtain the first analysis result, wherein the output function of the output layer is:

[0016]

[0017] Among them, a[x] represents the output value of the xth node in the output layer of the preset analysis model, g k , g1, g2 represent the activation functions of the output layer, the first hidden layer, and the second hidden layer respectively, T2 represents the number of nodes in the second hidden layer, and w xj2 represents the third weight between the second hidden layer and the output layer, T1 represents the number of nodes in the first hidden layer, and w j2j1 represents the second weight between the first hidden layer and the second hidden layer, T0 represents the number of nodes in the input layer, and w j1irepresents the first weight between the input layer and the first hidden layer, p 1i represents the i-th row of the feature matrix, p 2i represents the i-th row of the historical feature matrix, B j1 , B j2 , B x Represent the bias values ​​of the first hidden layer, the second hidden layer and the output layer respectively.

[0018] Preferably, the first weight, the second weight, and the third weight are generated by a preset update formula group during the training process of the preset analysis model, and the preset update formula group is:

[0019]

[0020] Among them, w' j1i represents the first weight after update, w' j2j1 represents the updated second weight, w' xj2 represents the updated third weight, μ represents the learning rate during the training of the preset analysis model, and p 3i represents the i-th row of the training matrix during the training of the preset analysis model, δ yj1 is the training error rate of the preset analysis model in the first hidden layer during the yth training, δ yj2 is the training error rate of the preset analysis model for the yth training on the second hidden layer, δ yx The training error rate of the output layer of the preset analysis model for the yth training.

[0021] Preferably, the on-board PHM module includes submodules corresponding to each carriage of the train, and each submodule transmits the collected operation data of each carriage to a node of the Ethernet train backbone network based on the Ethernet marshaling network, and the Ethernet train backbone network transmits the data to the on-board controller;

[0022] The step of performing fault monitoring on the train according to the first analysis result comprises:

[0023] According to the node identifier carried by each of the vibration signals, searching for a target carriage corresponding to the first analysis result;

[0024] Based on the target car, the fault type in the first analysis result, and the type parameter corresponding to the fault type, monitoring information for fault monitoring of the motor shaft of the running part in the target car is generated and output.

[0025] Preferably, the cold running data includes image detection data and sensor detection data of wheel treads in the train, and the step of analyzing the cold running data and the train parameters to generate a second analysis result to predict faults of the train includes:

[0026] Performing smoothing filtering and edge extraction processing on each depth image in the image detection data to obtain a plurality of tread edge images, and converting each of the tread edge images into a binary edge image;

[0027] Generate a pixel matrix corresponding to each of the binary edge images according to the coordinate values ​​of the pixels in the preset coordinate axis in each of the binary edge images, and perform a difference operation on each of the pixel matrices and a reference pixel matrix corresponding to the wheel tread to obtain a first operation matrix;

[0028] Performing a sum operation on each of the first operation matrices to obtain a second operation matrix, and generating a graph analysis result of the wheel tread according to the second operation matrix;

[0029] Generating a sensor analysis result of the wheel tread according to the vibration voltage value, the instantaneous speed value and the train parameter in the sensor detection data;

[0030] The second analysis result is generated based on the graph analysis result and the sensor analysis result to predict the fault of the wheel tread.

[0031] Preferably, the step of generating the sensor analysis result of the wheel tread according to the vibration voltage value, the instantaneous speed value and the train parameter in the sensor detection data comprises:

[0032] Determine whether the vibration voltage value is greater than a preset threshold value, and if so, determine a running section corresponding to the train tread, and search for a section coefficient corresponding to the running section, as well as a first instrument correction coefficient and a second instrument correction coefficient corresponding to the vibration voltage value and the instantaneous speed value, respectively;

[0033] The vibration voltage value, the instantaneous speed value, the section coefficient, and the wheel diameter and the average wheel weight in the train parameters are calculated based on a preset calculation formula to obtain the calculation result, and the preset calculation formula is:

[0034]

[0035] Wherein, H represents the calculation result, D represents the wheel diameter, U represents the vibration voltage value, L represents the road section coefficient, M represents the average wheel weight, V represents the instantaneous speed value, U0 represents the first instrument correction coefficient, and V0 represents the second instrument correction coefficient;

[0036] The calculation result is compared with a preset correspondence between the scratch depth interval and the scratch level to determine the scratch level corresponding to the calculation result, and the calculation result and the scratch level are combined to generate the sensing analysis result.

[0037] Preferably, the step of generating the second analysis result to predict the fault of the wheel tread according to the graph analysis result and the sensor analysis result comprises:

[0038] Determine whether the image analysis result and the sensor analysis result both carry a scratch mark, and if both carry a scratch mark, obtain a first scratch depth and a first scratch level in the image analysis result, and a second scratch depth and a second scratch level in the sensor analysis result;

[0039] generating a similarity value between the first scratch depth and the second scratch depth, and determining whether the similarity value is less than a preset similarity threshold, and whether the first scratch level and the second scratch level are the same;

[0040] If the similarity value is less than a preset similarity threshold, and the first abrasion level and the second abrasion level are the same, an average depth value between the first abrasion depth and the second abrasion depth is generated, and the second analysis result is generated based on the average depth value and the first abrasion level or the second abrasion level, and a fault prediction is performed on the wheel tread according to the second analysis result.

[0041] Preferably, the step of transmitting the cold operation data to the cloud server comprises:

[0042] Searching for the redundant data to be transmitted in the cold running data and the target redundant data corresponding to the redundant data to be transmitted, and determining whether the redundant data to be transmitted and the target redundant data are completely identical, and if they are completely identical, removing any one of the redundant data to be transmitted and the target redundant data from the cold running data to obtain the data to be transmitted;

[0043] If the to-be-redundant data and the target redundant data are not completely the same, the to-be-redundant data and the target redundant data are jointly formed into the to-be-transmitted data;

[0044] Generate an encryption key based on a preset encryption algorithm, and encrypt the data to be transmitted based on the encryption key to obtain encrypted data and transmit it to the cloud server, wherein the preset encryption algorithm is:

[0045] Select any preset prime number from the preset prime number array, and generate a first random number and a first parameter number pair;

[0046] The preset prime number, the first random number and the first parameter pair are calculated based on a first preset formula to obtain a first calculation result, and it is determined whether the first calculation result is a prime number, wherein the first preset formula is:

[0047] y=a*[e h ]*x 2 +b;

[0048] Wherein, y represents the first calculation result, a and b represent the first parameter pair, h represents the first random number, x represents a preset prime number, and [.] represents rounding down;

[0049] If the first calculation result is a prime number, generate a second random number and a second random number pair, and calculate the second random number, the second parameter pair and the first calculation result based on a first preset formula to obtain a second calculation result;

[0050] Determine whether the second calculation result is a prime number, and if it is a prime number, calculate the second calculation result based on a second preset formula to obtain a third calculation result;

[0051] Determining whether the third calculation result is a prime number, and if it is a prime number, determining the third result as a first prime number;

[0052] The steps of selecting any preset prime number from the preset prime number array and generating a first random number are performed, a second prime number is generated, and an encryption key is generated according to the first prime number and the second prime number.

[0053] Furthermore, to achieve the above-mentioned object, the present invention also provides an Ethernet-based train fault monitoring and prediction system, the Ethernet-based train fault monitoring and prediction system comprising a storage, a processor, a communication bus, and a control program stored on the storage:

[0054] The communication bus is used to realize the connection and communication between the processor and the storage;

[0055] The processor is used to execute the control program to implement the steps of the Ethernet-based train fault monitoring and prediction method as described above.

[0056] Furthermore, to achieve the above-mentioned purpose, the present invention also provides a storage medium, on which a control program is stored, and when the control program is executed by a processor, the steps of the Ethernet-based train fault monitoring and prediction method as described above are implemented.

[0057] The present invention discloses a train fault monitoring and prediction method, system and storage medium based on Ethernet. The train is provided with an on-board controller and an on-board PHM module connected to the on-board controller through an Ethernet train backbone network. After receiving the train operation data transmitted by the on-board PHM module based on Ethernet, the on-board controller determines the target position of the train, and after requesting the edge server corresponding to the target position to establish a communication connection, transmits the train operation data to the edge server; the edge server divides the received train operation data into hot operation data and cold operation data, and initiates a request to obtain the historical operation data corresponding to the hot operation data to the cloud server connected to it, and transmits the cold operation data to the cloud server at the same time; after obtaining the historical operation data, the edge server analyzes the hot operation data and the historical operation data through a preset analysis model pre-set therein, generates a first analysis result, and then performs fault monitoring on the train according to the first analysis result; after receiving the cold operation data, the cloud server searches for the train parameters corresponding to the train, analyzes the cold operation data and the train parameters, and generates a second analysis result to predict the fault of the train. In this way, by setting up a cloud-edge coordination mechanism between edge servers and cloud servers, hot running data with high real-time requirements are transmitted to edge servers near the train for processing, while cold running data with relatively low real-time requirements are transmitted to cloud servers for processing, so that data with high real-time requirements can be processed quickly and timely, avoiding the problem of inaccuracy caused by centralized data processing. In addition, the high transmission rate, short communication delay and high reliability of Ethernet further ensure the timeliness and accuracy of data processing. Timely and accurate monitoring and prediction of possible faults on the train are achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a flow chart of the first embodiment of the Ethernet-based train fault monitoring and prediction method of the present invention;

[0059] Figure 2 It is a flow chart of a second embodiment of the Ethernet-based train fault monitoring and prediction method of the present invention;

[0060] Figure 3 It is a flow chart of the third embodiment of the Ethernet-based train fault monitoring and prediction method of the present invention;

[0061] Figure 4 It is a structural schematic diagram of the hardware operating environment involved in an embodiment of the Ethernet-based train fault monitoring and prediction system of the present invention.

[0062] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0063] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0064] The present invention provides a train fault monitoring and prediction method based on Ethernet, wherein the train comprises an onboard controller and an onboard PHM (Prognostics and Health Management) module connected to the onboard controller via an Ethernet train backbone network. Figure 1 , Figure 1 It is a flow chart of the first embodiment of the Ethernet-based train fault monitoring and prediction method of the present invention.

[0065] The embodiment of the present invention provides an embodiment of a train fault monitoring and prediction method based on Ethernet. It should be noted that although the logical order is shown in the flow chart, in some cases, the steps shown or described may be performed in a different order than that here. Specifically, the train fault monitoring and prediction method based on Ethernet in this embodiment includes:

[0066] Step S10, after receiving the train operation data transmitted by the on-board PHM module based on the Ethernet train backbone network, the on-board controller determines the target location of the train and requests to establish a communication connection with the edge server corresponding to the target location to transmit the train operation data to the edge server.

[0067] In this embodiment, the train can be an ordinary train, a motor train, a high-speed train, or a subway, an intercity train, etc. Regardless of the type of train, a train communication network is provided in the train to provide public communication services for train operation control, multi-car coupling control, remote data transmission, vehicle equipment working status monitoring and fault diagnosis, ground inspection and maintenance operations, driver and crew business, and passenger services, so as to realize train information transmission and sharing. In this embodiment, the train communication network is preferably formed by Ethernet, including Ethernet Train Backbone (ETB) and Ethernet Consist Network (ECN). ETB can adopt a linear topology structure, connect each constituting network through the backbone network node, and realize communication between each carriage. ECN connects the terminal equipment in the carriage and is responsible for data communication and network management of each node in the carriage or constituting.

[0068] Furthermore, the train is provided with an on-board controller and an on-board PHM module connected to the on-board controller via the Ethernet train backbone network. The on-board controller is used to control the overall status of various aspects of the train, and the on-board PHM module is used to collect data related to train fault prediction and health management, and cooperate with relevant servers to achieve fault prediction and health management. The on-board PHM module is connected to various types of sensors through the Ethernet marshaling network, using pressure sensors, temperature sensors, speed sensors, vibration sensors, etc. During the operation of the train, various types of data collected by various sensors are transmitted to the on-board PHM module through the Ethernet marshaling network, and the on-board PHM module converts various types of data into train operation data and transmits them to the on-board controller through the Ethernet train backbone network.

[0069] Furthermore, because trains have the characteristics of long-distance operation, in order to perform fault prediction and health management for long-distance trains more timely and quickly, this embodiment constructs a server for fault prediction and health management in a cloud-edge collaborative manner. Edge servers are set up in different regional locations to process data with high real-time requirements during train operation, and cloud servers that communicate with each edge server are set up at a certain location to process data with low real-time requirements during train operation and store train parameters and various historical data. Among them, the cloud server location can be set at the center of each edge server, or the center of an area with a large number of trains.

[0070] Furthermore, the train operation data includes train positioning data, and the on-board controller determines the target position of the train through the train positioning data, and searches for the edge server located at the target position or the closest to the target position. Then the on-board controller initiates a request to establish a communication connection to the edge server found. After receiving the request, the edge server determines its own load and determines whether the amount of data it needs to process exceeds the amount of data it can bear. If it exceeds the amount of data it can bear, it returns to the on-board controller the amount of data it currently needs to process and the prompt information of the waiting time. The on-board controller reselects the adjacent edge server based on the prompt information, and the reselected edge server also performs load judgment. If the load is less than the amount of data it can bear, it establishes a communication connection with the on-board controller and returns to the on-board controller a prompt information that the communication connection is successfully established. If the loads of multiple adjacent edge servers that are re-searched are too high, the on-board controller compares the waiting time returned by each adjacent edge server, and determines the adjacent edge server with the shortest waiting time to re-initiate the request to establish a communication connection. Then, after the communication connection is established, the train operation data is transmitted to the edge server for processing.

[0071] In step S20, the edge server divides the train operation data into hot operation data and cold operation data, and requests a cloud server that is communicatively connected to the edge server to obtain historical operation data corresponding to the hot operation data, and transmits the cold operation data to the cloud server.

[0072] Furthermore, an identifier is set in advance for data with high real-time requirements in the train, and the edge server screens the received train operation data based on the identifier and divides it into hot operation data and cold operation data. Hot operation data is data with high real-time requirements, mainly fault data that affects the real-time operation of the train, such as data related to the train running gear; cold operation data is data with low real-time requirements, such as air conditioning data, component wear data, etc. For hot operation data, the edge cloud server processes it in combination with the previous operation data of the corresponding components. The previous operation data reflects the changes in the operating status of the components, which can make the processing of hot operation data more accurate. This type of operation data is stored in the cloud server as historical operation data. The edge server generates the identifier of the hot operation data as a request to send to the cloud server to request the cloud server to obtain the historical operation data corresponding to the hot operation data.

[0073] For the cold operation data, the edge server transmits it to the cloud server for processing to avoid the edge server occupying server resources due to processing the cold operation data, which affects the speed of processing the hot operation data. In addition, in order to ensure the accuracy of data collection and transmission, the sensors for data collection are usually set up with redundancy, that is, two sensors of the same type are set to collect the same data, so that when the data collection and transmission are normal, the divided cold operation data contains two groups of the same data. When the cold operation data is transmitted to the cloud server, in order to increase the transmission rate, the redundant data in the cold operation data needs to be screened. Specifically, the step of transmitting the cold operation data to the cloud server includes:

[0074] Step S21, searching for redundant data to be stored in the cold running data and target redundant data corresponding to the redundant data to be stored, and determining whether the redundant data to be stored and the target redundant data are completely identical; if they are completely identical, removing any one of the redundant data to be stored and the target redundant data from the cold running data to obtain data to be transmitted;

[0075] Step S22, if the data to be redundant and the target redundant data are not completely the same, the data to be redundant and the target redundant data are jointly formed into data to be transmitted;

[0076] Step S23, generating an encryption key based on a preset encryption algorithm, and encrypting the data to be transmitted based on the encryption key, obtaining encrypted data and transmitting it to the cloud server, wherein the preset encryption algorithm is:

[0077] Step S24, selecting any preset prime number from the preset prime number array, and generating a first random number and a first parameter number pair;

[0078] Step S25, calculating the preset prime number, the first random number and the first parameter pair based on a first preset formula to obtain a first calculation result, and determining whether the first calculation result is a prime number;

[0079] Step S26, if the first calculation result is a prime number, generating a second random number and a second random number pair, and calculating the second random number, the second parameter pair and the first calculation result based on a first preset formula to obtain a second calculation result;

[0080] Step S27, determining whether the second calculation result is a prime number, and if it is a prime number, calculating the second calculation result based on a second preset formula to obtain a third calculation result;

[0081] Step S28, determining whether the third calculation result is a prime number, and if it is a prime number, determining the third result as a first prime number;

[0082] Step S29, executing the step of selecting any preset prime number from the preset prime number array and generating a first random number, generating a second prime number, and generating an encryption key according to the first prime number and the second prime number.

[0083] Furthermore, the components that need to be set up redundantly are taken as the components to be redundant, and the data collected by them are correspondingly the data to be redundant, while the components that are redundant for the components that need to be redundant are the target redundant components, and the data collected by them are correspondingly the target redundant data. In addition, the device identifiers of the components to be redundant and the target redundant devices are carried between the data to be redundant and the target redundant data, indicating the redundant relationship between the data to be redundant and the target redundant data and the devices from which they come. Before transmitting the cold operation data to the cloud server, the edge server first finds the data to be redundant in the cold operation data and the target redundant data corresponding to the data to be redundant based on the device identifiers carried by each data in the cold operation data. Then, the data to be redundant and the target redundant data are compared to determine whether the data to be redundant and the target redundant data are exactly the same. If they are exactly the same, it means that the data collected by the device to be redundant and the target redundant device are the same, and any one of them can be selected and transmitted to the cloud server. At this time, the data to be redundant or the target redundant data can be removed from the cold operation data, and the remaining data can be transmitted to the cloud server as the data to be transmitted for processing.

[0084] On the contrary, if the comparison determines that the redundant data to be processed and the target redundant data are not completely the same, it means that the data detected by one of the redundant devices and the target redundant devices may be wrong. At this time, the redundant data and the target redundant data are jointly formed into the data to be transmitted to the cloud server, and the cloud server makes a judgment based on the previous detection data. The data that is not much different from the previous detection data is selected from the redundant data and the target redundant data for processing as valid data.

[0085] In addition, in order to ensure the security of the transmitted data, the edge server also encrypts the data to be transmitted. An encryption algorithm is pre-set, and an encryption key is generated by the preset encryption algorithm, and then the encryption key is used to encrypt the data to be transmitted, and the encrypted data is transmitted to the cloud server. The preset encryption algorithm is preferably an asymmetric encryption algorithm. Specifically, a preset prime number array is pre-set, and in order to ensure security, the value of the data in the preset prime number array can be set to be larger, for example, the prime numbers between 100 and 200 are formed into a preset prime number array. Any preset prime number is selected from the preset prime number array, and three random numbers are generated by a random number generator, one of which is selected as the first random number, and the remaining two are used as the first parameter pair. Then, the preset prime number, the first random number and the first parameter pair are calculated according to the first preset formula to obtain a first calculation result. It is judged whether the first calculation result is a prime number. If it is not a prime number, a random number is generated according to the random number generator to update the first random number, and it is calculated with the first parameter pair and the preset prime number, and the first calculation result is obtained again for judgment until the first calculation result obtained is a prime number. At this time, three more random numbers are generated according to the random number generator, one of which is selected as the second random number, and the remaining two are used as the second parameter pair. The first calculation result, the second random number and the second parameter pair are calculated according to the first preset formula to obtain the second calculation result, and it is judged whether the second calculation result is a prime number. If it is not a prime number, a random number is generated according to the random number generator to update the second random number, and it is calculated with the first calculation result and the second parameter pair, and the second calculation result is obtained again for judgment until the second calculation result obtained is a prime number. Among them, the first preset formula is specifically referred to as the following formula (1).

[0086] y=a*[e h ]*x 2 +b (1);

[0087] Wherein, corresponding to step S25, y represents the first calculation result, a and b represent the first parameter pair, h represents the first random number, and x represents the preset prime number; corresponding to step S26, y represents the second calculation result, a and b represent the second parameter pair, h represents the second random number, x represents the first calculation result, and [.] represents rounding down. The difference between the first random number and the second random number, and the difference between the first parameter pair and the second parameter pair, make the parameters of the first preset formula change, so that the first calculation result and the second calculation result are generated with different parameters, which is conducive to improving the security of the encryption algorithm.

[0088] Further, the second calculation result is calculated by the second preset formula to determine whether the third calculation result is a prime number. If it is not a prime number, the second parameter pair is updated again based on two random numbers generated by the random number generator, and the second calculation result and the third calculation result are generated in sequence based on the updated second parameter pair until the third calculation result obtained is a prime number, and the third calculation result is used as the first prime number. Thereafter, any preset prime number is selected from the preset prime number array again to generate a second prime number, and an encryption and decryption key is generated based on the first prime number and the second prime number. Among them, the encryption and decryption key includes a public key as an encryption key and a private key as a decryption key. The edge server encrypts the data to be transmitted by the encryption key and transmits it to the cloud server, and the cloud server decrypts it by the decryption key. Because the first prime number and the second prime number are generated based on the prime numbers calculated by the preset prime number array and the first preset formula and the second preset formula, the prime number values ​​in the preset prime number array are relatively large, and the parameter transformation calculation of the first preset formula can make the values ​​of the generated first prime number and the second prime number also larger, and the generation process is also more complicated, thereby increasing the complexity of encryption and decryption, making data transmission safer.

[0089] Step S30: the edge server analyzes the thermal operation data and the historical operation data based on a preset analysis model to generate a first analysis result, and performs fault monitoring on the train according to the first analysis result.

[0090] Furthermore, for various types of thermal operation data that need to be processed, a preset analysis model is pre-built and trained in the edge server through a neural network. The preset analysis model is used to analyze the thermal operation data and historical operation data to generate a first analysis result. The first analysis result includes information such as whether the component corresponding to the thermal operation data has a fault, and if so, the car where the fault point is located, the severity of the fault, and possible causes. The corresponding monitoring information is generated through this type of information in the first analysis result to issue an alarm, thereby realizing fault monitoring of the train.

[0091] Step S40, the cloud server searches for train parameters corresponding to the cold operation data of the train, analyzes the cold operation data and the train parameters, and generates a second analysis result to predict faults of the train.

[0092] Furthermore, for the cold operation data, the cloud server searches for the train parameters corresponding to the cold operation data. For example, if the cold operation data is air conditioning data, the train parameters to be searched are the model and power of the air conditioner, and if the cold operation data is wheel tread data, the train parameters to be searched are the diameter and weight of the wheel, etc. The fixed parameters of the components corresponding to the cold operation data in the train are used to reflect their influence on the cold operation data. After finding the corresponding train parameters, the cloud server analyzes the train parameters and the cold operation data together to obtain a second analysis result. The second analysis result includes information such as whether the components corresponding to the cold operation data may fail and the probability of failure. The corresponding warning information is generated through this type of information in the second analysis result to predict train failures.

[0093] The Ethernet-based train fault monitoring and prediction method of this embodiment is provided with an on-board controller on the train, and an on-board PHM module connected to the on-board controller through the Ethernet train backbone network. After receiving the train operation data transmitted by the on-board PHM module based on Ethernet, the on-board controller determines the target position of the train, and after requesting the edge server corresponding to the target position to establish a communication connection, transmits the train operation data to the edge server; the edge server divides the received train operation data into hot operation data and cold operation data, and initiates a request to obtain the historical operation data corresponding to the hot operation data to the cloud server connected to it, and transmits the cold operation data to the cloud server at the same time; after obtaining the historical operation data, the edge server analyzes the hot operation data and the historical operation data through the preset analysis model pre-set therein, generates a first analysis result, and then performs fault monitoring on the train according to the first analysis result; after receiving the cold operation data, the cloud server searches for the train parameters corresponding to the train, analyzes the cold operation data and the train parameters, and generates a second analysis result to predict the train fault. In this way, by setting up a cloud-edge coordination mechanism between edge servers and cloud servers, hot running data with high real-time requirements are transmitted to edge servers near the train for processing, while cold running data with relatively low real-time requirements are transmitted to cloud servers for processing, so that data with high real-time requirements can be processed quickly and timely, avoiding the problem of inaccuracy caused by centralized data processing. In addition, the high transmission rate, short communication delay and high reliability of Ethernet further ensure the timeliness and accuracy of data processing. Timely and accurate monitoring and prediction of possible faults on the train are achieved.

[0094] For further information, please refer to Figure 2 Based on the first embodiment of the Ethernet-based train fault monitoring and prediction method of the present invention, a second embodiment of the Ethernet-based train fault monitoring and prediction method of the present invention is proposed.

[0095] The difference between the second embodiment of the Ethernet-based train fault monitoring and prediction method and the first embodiment of the Ethernet-based train fault monitoring and prediction method is that the thermal operation data includes a vibration signal corresponding to a motor shaft in the running gear of the train, and the edge server analyzes the thermal operation data and the train historical data based on a preset analysis model, and the step of generating a first analysis result includes:

[0096] Step S31, performing time domain analysis and frequency domain analysis on the vibration signal respectively to obtain an original time domain feature group and an original frequency domain feature group of the vibration signal;

[0097] Step S32, filtering the vibration signal, and extracting a filtering time domain feature group and a filtering frequency domain feature group from the filtered vibration signal;

[0098] Step S33, forming the original time domain feature group, the original frequency domain feature group, the filtered time domain feature group and the filtered frequency domain feature group into a feature matrix, and forming the historical operation data into a historical feature matrix;

[0099] Step S34, performing fault analysis on the feature matrix and the historical feature matrix based on the input layer, the first hidden layer, the second hidden layer and the output layer in the preset analysis model to obtain the first analysis result.

[0100] Furthermore, each carriage of the train includes a running gear for walking, and the running gear includes a motor shaft. The thermal operation data of this embodiment includes a vibration signal corresponding to such a motor shaft. The edge server performs time domain analysis and frequency domain analysis on such vibration signals, obtains multiple time domain features through time domain analysis, and forms an original time domain feature group of the vibration signal, which at least includes skewness, kurtosis, peak factor, margin factor, waveform factor and pulse factor, etc.; at the same time, obtains multiple frequency domain features through frequency domain analysis, and forms an original frequency domain feature group of the vibration signal, which at least includes: spectrum, energy spectrum, power spectrum, bandwidth, power spectrum, etc. At the same time, the vibration signal is filtered, and the filtered vibration signal is subjected to time series analysis and frequency domain analysis. The corresponding time domain features and frequency domain features are extracted through analysis to form a filtered time domain feature group and a filtered frequency domain feature group.

[0101] Furthermore, the original time domain feature group, the original frequency domain feature group, the filtered time domain feature group and the filtered frequency domain feature group are collectively formed into a feature matrix. A feature group is a row in the feature matrix, and in the case where the number of elements in the feature groups is different, the feature group with the missing elements is supplemented with a value of 1, so that each matrix row in the feature matrix has the same number of elements. In addition, the historical operation data is the average value of each of the above-mentioned time domain features obtained from the fault monitoring of the motor shaft of each carriage running part in recent times, and the average value of each of the above-mentioned frequency domain features obtained. Similarly, the average value of each time domain feature is formed into a historical time domain feature group, and the average value of each frequency domain feature is formed into a historical frequency domain feature group, and then each feature group is formed into a historical feature matrix.

[0102] Furthermore, the preset analysis model in the edge server includes an input layer, a hidden layer, and an output layer, and in order to improve the model analysis effect while avoiding model overfitting, the hidden layer is set to two layers, namely, a first hidden layer and a second hidden layer. The input layer, the first hidden layer, the second hidden layer, and the output layer in the preset analysis model are used to perform fault analysis on the feature matrix and the historical feature matrix to obtain a first analysis result, wherein the output function expression of the output layer is shown in the following formula (2).

[0103]

[0104] Among them, a[x] represents the output value of the xth node in the output layer of the preset analysis model, g k , g1, g2 represent the activation functions of the output layer, the first hidden layer, and the second hidden layer respectively, T2 represents the number of nodes in the second hidden layer, and w xj2 represents the third weight between the second hidden layer and the output layer, T1 represents the number of nodes in the first hidden layer, and w j2j1 represents the second weight between the first hidden layer and the second hidden layer, T0 represents the number of nodes in the input layer, and w j1i represents the first weight between the input layer and the first hidden layer, p 1i represents the i-th row of the feature matrix, p 2i represents the i-th row of the historical feature matrix, B j1 , B j2 , Bx represent the bias values ​​of the first hidden layer, the second hidden layer and the output layer respectively. The data obtained by the recent fault monitoring of the motor shaft of each carriage running part is formed into a historical feature matrix, and the historical feature matrix reflects the change trend of the motor shaft vibration signal, so that the first analysis result obtained by the preset analysis model analysis more accurately reflects the current operating state of the motor shaft.

[0105] Furthermore, in order to facilitate the determination of the car where the fault point is located, the on-board PHM module can be divided into sub-modules corresponding to each car of the train, and each of the sub-modules transmits the collected operation data of each car to the node of the Ethernet train backbone network based on the Ethernet marshaling network, and the Ethernet train backbone network transmits it to the on-board controller. The operation data transmitted by each sub-module constitutes the train operation data, which includes hot operation data and cold operation data corresponding to each car. The first analysis result generated based on the vibration data of the motor shaft of the running gear of each car also reflects the fault condition of the motor shaft of the running gear of each car. One vibration data corresponds to one car, and a first analysis result is generated. Therefore, the motor shaft of the running gear of each car can be monitored for faults based on each first analysis result. Specifically, the step of monitoring the train for faults according to the first analysis result includes:

[0106] Step S35, searching for a target carriage corresponding to the first analysis result according to the node identifier carried by each of the vibration signals;

[0107] Step S36, based on the target car, the fault type in the first analysis result, and the type parameter corresponding to the fault type, generate monitoring information for fault monitoring of the motor shaft of the running part in the target car, and output the monitoring information.

[0108] Furthermore, each car is provided with its own corresponding submodule, and each submodule is connected to the node of the Ethernet train backbone network, so one car corresponds to one node, and each car can be distinguished by numbering the node, and the operation data from the same car all carry the node identification corresponding to the car. Similarly, the vibration signal of the motor shaft of each car running part also carries the node identification of each car. After the first analysis result is generated according to the vibration signal, the car from which the vibration signal comes is determined by the node identification carried by the vibration signal, and the car is used as the target car corresponding to the first analysis result.

[0109] Furthermore, the first analysis result includes information reflecting whether the motor shaft has a fault. If the information reflects that the motor shaft has no fault, the target car and such information are generated together as monitoring information indicating that there is no fault, so as to indicate that the motor shaft of the running part of the car has no fault. If the information reflects that the motor shaft has a fault, the fault type in the first analysis result, the type parameter corresponding to the fault type, and the target car are jointly generated as monitoring information and output, so as to monitor the fault of the motor shaft in the running part of the target car through the monitoring information. Among them, the fault type includes at least any one of journal wear, keyway wear, insufficient lubrication, pitting fault, and scratch fault, and the type parameter corresponding to the fault type includes at least a fault severity parameter, such as the degree of wear, the specific location of the fault, such as the specific location of the pitting, etc. By generating such information together as monitoring information, it is convenient to quickly troubleshoot the fault.

[0110] It should be noted that the preset analysis model is formed in advance through training with a large number of training samples, and the training samples include faults such as journal wear, keyway wear, insufficient lubrication, pitting faults, scratch faults, and data indicating the severity of the faults, so that the preset analysis model can analyze the fault type and fault parameters to obtain the fault type and type parameters in the first analysis result. In addition, in order to improve the accuracy of the fault analysis of the preset analysis model, the first weight between the input layer and the first hidden layer, the second weight between the first hidden layer and the second hidden layer, and the third weight between the second hidden layer and the output layer are updated according to the preset update formula group during the training of the preset analysis model. The specific preset update formula group can be found in the following formula (3).

[0111]

[0112] Among them, w′ j1i represents the updated first weight, W′ j2j1 represents the updated second weight, W′ xj2 represents the updated third weight, μ represents the learning rate during the training of the preset analysis model, and p 3i represents the i-th row of the training matrix during the training of the preset analysis model, δ yj1 is the training error rate of the preset analysis model in the first hidden layer during the yth training, δ yj2 is the training error rate of the preset analysis model for the yth training on the second hidden layer, δ yx is the training error rate of the preset analysis model at the output layer during the yth training. Other symbols have the same meanings as those in the above formula (2) and are not described here.

[0113] The number of iterations is preset. When the number of training times for the preset analysis model reaches the number of iterations, the preset analysis model is tested to generate a total error, and it is determined whether the total error is less than a preset threshold. If it is less than the preset threshold, it means that the accuracy of the preset analysis model meets the requirements, and the training of the preset analysis model is completed. If the total error is greater than the preset threshold, it means that the accuracy of the preset analysis model is low, and iterative training needs to continue. At this time, the total error is decomposed into the training error rate of the first hidden layer, the second hidden layer, and each layer on the output, and the first weight, the second weight, and the third weight are updated according to the training error rate of each layer through the preset update formula group until the total error generated by the preset analysis model test is less than the preset threshold.

[0114] This embodiment extracts time domain features and frequency domain features from the thermal operation data, and combines the historical features generated by the previous fault monitoring, and performs analysis using a preset analysis model formed in advance based on neural network training. On the one hand, the time domain features and frequency domain features can accurately reflect the characteristics of the thermal operation data itself, and on the other hand, the historical features reflect the changing trend of the thermal operation data at different fault monitoring times. In addition, the weight of the preset analysis model is updated through the error rate of each layer in each training, ensuring the accuracy of the preset analysis model. In this way, the accuracy of thermal operation data analysis is improved from multiple aspects.

[0115] For further information, please refer to Figure 3 Based on the first and second embodiments of the Ethernet-based train fault monitoring and prediction method of the present invention, a third embodiment of the Ethernet-based train fault monitoring and prediction method of the present invention is proposed.

[0116] The third embodiment of the Ethernet-based train fault monitoring and prediction method is different from the first and second embodiments of the Ethernet-based train fault monitoring and prediction method in that the cold operation data includes image detection data and sensor detection data of the wheel tread of the train, and the step of analyzing the cold operation data and the train parameters to generate a second analysis result to predict the fault of the train includes:

[0117] Step S41, performing smoothing filtering and edge extraction processing on each depth image in the image detection data to obtain a plurality of tread edge images, and converting each of the tread edge images into a binary edge image;

[0118] Step S42, generating a pixel matrix corresponding to each of the binary edge images according to the coordinate values ​​of the pixels in each of the binary edge images in a preset coordinate axis, and performing a difference operation between each of the pixel matrices and a reference pixel matrix corresponding to the wheel tread to obtain a first operation matrix;

[0119] Step S43, performing a sum operation on each of the first operation matrices to obtain a second operation matrix, and generating a graph analysis result of the wheel tread according to the second operation matrix;

[0120] Step S44, generating a sensor analysis result of the wheel tread according to the vibration voltage value, the instantaneous speed value and the train parameter in the sensor detection data;

[0121] It is understandable that the train runs along the track, and the part where the train wheel contacts the rail is called the tread. As the train runs, the wheel tread and the rail may generate friction, which in turn causes the wheel tread to be scratched. Although the wheel tread scratch will not affect the immediate operation of the train, its impact in the long run should not be underestimated. When the wheel tread is scratched, it will cause vibration when the train is running, which will not only damage the train parts, but also the rails. In addition, the greater the depth of the wheel tread scratch, the greater the vibration caused, which reduces the smoothness of the train operation. This embodiment predicts the failure of the wheel tread.

[0122] Specifically, the cold running data collected for the wheel tread includes image detection data and sensor detection data. The image detection data is the image of the wheel tread collected by a binocular vision camera, and the sensor detection data is the data related to the wheel tread detected by detection devices such as a voltmeter and a speed sensor. In addition, the image detection data includes depth images of the wheel tread at multiple angles, and the multiple depth images are smoothed and filtered and edge extracted to obtain multiple tread edge images containing the tread edge contour. Then, each tread edge image is converted into black and white binary to obtain multiple binary edge images.

[0123] Furthermore, a preset coordinate axis is pre-established, and the preset coordinate axis can be an image coordinate axis corresponding to each binary edge map, with the upper left corner of the image as the coordinate origin, the horizontal direction of the image as the X axis, and the vertical direction of the image as the Y axis. Then, for each binary edge map, the coordinate value of each pixel in the preset coordinate axis is obtained, and the coordinate value of each pixel is generated according to the corresponding x-coordinate and y-coordinate position, and each element in the pixel matrix corresponds to the coordinate value of a pixel point.

[0124] Furthermore, before the wheel tread is put into use, its image is collected to generate a reference pixel matrix corresponding to the wheel tread. Each pixel matrix is ​​subjected to difference operation with the reference pixel matrix to obtain a first operation matrix. For a wheel tread that has not been scratched, the value of each element in the first operation matrix obtained by the difference operation should be zero. However, if the value of an element is not zero, it means that the wheel tread has been scratched, causing the pixel points of its imaging to change. Therefore, the means of each first operation can be added to obtain a second operation matrix, and the elements with non-zero values ​​in the second operation matrix can be determined. The non-zero element reflects the scratch, and the number and position of the non-zero element reflect the severity and position depth of the scratch. Therefore, a graphical analysis result of the wheel tread scratch condition can be generated based on the second operation matrix.

[0125] Furthermore, for the sensor detection data, the wheel tread scratch condition is analyzed from the perspective of sensor detection, which includes the vibration voltage value and the instantaneous speed value, and the sensor analysis result of the wheel tread is generated in combination with the train parameters. Specifically, the step of generating the sensor analysis result of the wheel tread according to the vibration voltage value and the instantaneous speed value in the sensor detection data and the train parameters includes:

[0126] Step S441, determining whether the vibration voltage value is greater than a preset threshold value, and if so, determining a running section corresponding to the train tread, and searching for a section coefficient corresponding to the running section, and a first instrument correction coefficient and a second instrument correction coefficient corresponding to the vibration voltage value and the instantaneous speed value, respectively;

[0127] Step S442, calculating the vibration voltage value, the instantaneous speed value, the section coefficient, and the wheel diameter and the average wheel weight in the train parameters based on a preset calculation formula to obtain the calculation result;

[0128] Step S443, comparing the calculation result with the corresponding relationship between the preset scratch depth interval and the scratch level, determining the scratch level corresponding to the calculation result, and generating the sensing analysis result with the calculation result and the scratch level.

[0129] Furthermore, considering the influence of train design and manufacturing process, the train will have a certain degree of scratches after running for a period of time, and slight scratches will not affect the operation of the train. Therefore, the present embodiment is pre-set with a preset threshold value, and the vibration voltage value in the sensor detection data is compared with the preset threshold value to determine whether the vibration voltage value is greater than the preset threshold value. If it is not greater than the preset threshold value, it means that the scratches produced are slight and no treatment is performed. On the contrary, if the vibration voltage value is greater than the preset threshold value, the running section corresponding to the train tread is determined. Different running sections have different road conditions and have different degrees of influence on the scratches. Corresponding section coefficients are pre-set for different sections. After determining the corresponding running section, the section coefficient corresponding to the running section is searched, and then the vibration voltage value, the instantaneous speed value, the searched section coefficient and the wheel diameter and wheel average weight in the train parameters are calculated according to the preset calculation formula to obtain the calculation result, which is the scratch depth. Among them, the preset calculation formula used for calculation can be referred to as the following formula (4).

[0130]

[0131] Among them, H represents the calculation result, D represents the wheel diameter, U represents the vibration voltage value, L represents the road section coefficient, M represents the average wheel weight, V represents the instantaneous speed value, U0 represents the first instrument correction coefficient, and V0 represents the second instrument correction coefficient. In addition, formula (4) is an expression of the relationship between the scratch depth and the vibration voltage, instantaneous speed, wheel mean, and wheel diameter obtained through a large amount of test data. The instrument error of the instrument for detecting the vibration voltage value and the instantaneous speed value represented by the instrument correction coefficient is used to correct formula (4), thereby fully considering various factors affecting the scratch depth, making the calculation of the scratch depth more accurate.

[0132] Furthermore, in order to indicate the severity of the wheel tread scratches, a correspondence between the scratch depth interval and the scratch level is preset. The calculated result is compared with the scratch depth interval in the correspondence to determine the scratch interval where the calculation result is located, and then the scratch level corresponding to the scratch depth interval in the correspondence is found as the scratch level corresponding to the calculation result. The determined scratch level and the calculation result indicating the scratch depth are then formed together as the sensing analysis result. Among them, the larger the value of the calculation result, the deeper the scratch, and the higher the corresponding scratch level, indicating that the wheel tread scratches are more serious.

[0133] Step S45, generating the second analysis result to predict the fault of the wheel tread according to the graph analysis result and the sensor analysis result.

[0134] Furthermore, after the graph detection data and the sensor detection data generate their own graph analysis results and sensor analysis results respectively, a second analysis result can be generated by combining the difference between the two, and the train fault prediction is performed based on the second analysis result. Specifically, the step of generating the second analysis result to predict the fault of the train tread based on the graph analysis result and the sensor analysis result includes:

[0135] Step S451, determining whether the image analysis result and the sensor analysis result both carry a scratch mark, and if both carry a scratch mark, obtaining a first scratch depth and a first scratch level in the image analysis result, and a second scratch depth and a second scratch level in the sensor analysis result;

[0136] Step S452, generating a similarity value between the first scratch depth and the second scratch depth, and determining whether the similarity value is less than a preset similarity threshold, and whether the first scratch level and the second scratch level are the same;

[0137] Step S453: If the similarity value is less than a preset similarity threshold, and the first abrasion level and the second abrasion level are the same, an average depth value between the first abrasion depth and the second abrasion depth is generated, and the second analysis result is generated based on the average depth value and the first abrasion level or the second abrasion level, and a fault prediction is performed on the wheel tread according to the second analysis result.

[0138] Understandably, the graph analysis results generated based on the graph detection data and the sensor analysis results generated based on the sensor detection data are actually to judge whether there is a scratch on the wheel tread and the severity of the scratch from the graph dimension and the sensor dimension, so as to avoid the error of a single detection method and the failure of the detection equipment. When both the graph detection and the sensor detection are correct, the detection results of the two should be relatively close. Therefore, the wheel tread scratch fault prediction can be carried out based on the difference between the two.

[0139] Specifically, determine whether both the image analysis result and the sensor analysis result carry a scratch mark. If both carry a scratch mark, it means that the image detection and the sensor detection have both determined that the wheel tread has a scratch, and the severity of the scratch needs to be further determined. If neither the image analysis result nor the sensor analysis result has a scratch mark, it means that both the image detection and the sensor detection have determined that the wheel tread has no scratch, and no further processing is required. If one of the image analysis result and the sensor analysis result has a scratch mark and the other does not have a scratch mark, it means that the detection error of the image detection and the sensor detection is large. At this time, analysis can be performed based on the previous detection situation, or a prompt message for re-detection can be output.

[0140] Furthermore, for the case where both carry scratch marks, the cloud server obtains the corresponding scratch depth and scratch level in the graph analysis result as the first scratch depth and the first scratch level. At the same time, the scratch depth and scratch level in the sensor analysis result are also obtained as the second scratch depth and the second scratch level. Then, a similarity calculation is performed between the first scratch depth and the second scratch depth to generate a similarity value between the two. In order to standardize the degree of proximity between the two, a preset similarity threshold is pre-set, and the calculated similarity value is compared with the preset similarity threshold to determine whether the similarity value is less than the preset similarity threshold. If it is less than the preset similarity threshold, it means that the first scratch depth and the second scratch depth are close to each other. Otherwise, it means that the difference between the first scratch depth and the second scratch depth is large. At the same time, it is also necessary to determine whether the first scratch level is the same as the second scratch level.

[0141] Furthermore, if it is determined that the similarity value is less than the preset similarity threshold, and the first scratch level is the same as the second scratch level, it means that the detection results of the image detection and the sensor detection are basically consistent, and both are accurate results. At this time, the first scratch depth and the second scratch depth are generated as an average depth value, and any one of the first scratch level and the second scratch level is selected together with the average depth value to generate a second analysis result for train fault prediction.

[0142] Furthermore, if it is determined that the similarity value is greater than or equal to the preset similarity threshold, and the first abrasion level and the second abrasion level are the same, it means that the judgment of the image detection and the sensor detection on the abrasion level is consistent, but the determination on the abrasion depth is quite different. At this time, the first abrasion depth and the second abrasion depth are compared to determine the larger value of the two. Then, based on the larger value, the second analysis result for predicting the fault of the train is generated in combination with the first abrasion level or the second abrasion level. By selecting the larger value of the first abrasion depth and the second abrasion depth to generate the second analysis result for predicting the abrasion fault, it is helpful to ensure the smooth operation of the train.

[0143] Furthermore, if it is determined that the similarity value is greater than or equal to the preset similarity threshold, and the first scratch level and the second scratch level are different, it means that the judgments of the scratch level by the image detection and the sensor detection are inconsistent, and the difference in the determination of the scratch depth is also large. At this time, the first scratch depth and the second scratch depth are compared to determine the larger value of the two, and the higher level of the first scratch level and the second scratch level is screened out. Then, based on the larger value and the higher level, prediction information for predicting train faults is generated. By selecting the larger value and the higher level of the first scratch depth and the second scratch depth to generate the second analysis result for scratch fault prediction, it is beneficial to ensure the smooth operation of the train.

[0144] This embodiment aims at a detection method in which both wheel tread setting diagram detection and sensor detection are carried out in parallel. The detection data of the two detection methods are analyzed to generate respective corresponding detection analysis results, and a second analysis result for predicting wheel tread scratch faults is generated from the difference between the two respective detection analysis results, thereby ensuring the accuracy of the wheel tread scratch fault prediction.

[0145] In addition, the embodiment of the present invention also provides a train fault monitoring and prediction system based on Ethernet. Figure 4 , Figure 4 It is a structural schematic diagram of the equipment hardware operating environment involved in the Ethernet-based train fault monitoring and prediction system embodiment of the present invention.

[0146] like Figure 4 As shown, the Ethernet-based train fault monitoring and prediction system may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a storage 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The storage 1005 may be a high-speed RAM storage, or a stable storage (non-volatile memory), such as a disk storage. The storage 1005 may also be a storage device independent of the aforementioned processor 1001.

[0147] Those skilled in the art will understand that Figure 4 The hardware structure of the Ethernet-based train fault monitoring and prediction system shown in the figure does not constitute a limitation on the Ethernet-based train fault monitoring and prediction system, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0148] like Figure 4 As shown, the storage 1005 as a storage medium may include an operating system, a network communication module, a user interface module and a control program. The operating system is a program for managing and controlling the Ethernet-based train fault monitoring and prediction system and software resources, and supports the operation of the network communication module, the user interface module, the control program and other programs or software; the network communication module is used to manage and control the network interface 1004; the user interface module is used to manage and control the user interface 1003.

[0149] exist Figure 4In the hardware structure of the Ethernet-based train fault monitoring and prediction system shown, the network interface 1004 is mainly used to connect to other system servers and communicate data with other system servers; the user interface 1003 is mainly used to connect to the client (user end) and communicate data with the client; the processor 1001 can call the control program stored in the storage 1005 and perform the following operations:

[0150] After receiving the train operation data transmitted by the on-board PHM module based on the Ethernet train backbone network, the on-board controller determines the target location of the train and requests to establish a communication connection with the edge server corresponding to the target location to transmit the train operation data to the edge server;

[0151] The edge server divides the train operation data into hot operation data and cold operation data, and requests a cloud server that is communicatively connected to the edge server to obtain historical operation data corresponding to the hot operation data, and transmits the cold operation data to the cloud server;

[0152] The edge server analyzes the thermal operation data and the historical operation data based on a preset analysis model to generate a first analysis result, and performs fault monitoring on the train according to the first analysis result;

[0153] The cloud server searches for train parameters corresponding to the cold operation data of the train, analyzes the cold operation data and the train parameters, and generates a second analysis result to predict faults of the train.

[0154] Further, the thermal operation data includes a vibration signal corresponding to a motor shaft in a running part of the train, and the edge server analyzes the thermal operation data and the train historical data based on a preset analysis model, and the step of generating a first analysis result includes:

[0155] Performing time domain analysis and frequency domain analysis on the vibration signal respectively to obtain an original time domain feature group and an original frequency domain feature group of the vibration signal;

[0156] Filtering the vibration signal, and extracting a filtering time domain feature group and a filtering frequency domain feature group from the filtered vibration signal;

[0157] The original time domain feature group, the original frequency domain feature group, the filtered time domain feature group and the filtered frequency domain feature group are formed into a feature matrix, and the historical operation data are formed into a historical feature matrix;

[0158] Based on the input layer, the first hidden layer, the second hidden layer and the output layer in the preset analysis model, the feature matrix and the historical feature matrix are subjected to fault analysis to obtain the first analysis result, wherein the output function of the output layer is:

[0159]

[0160] Among them, a[x] represents the output value of the xth node in the output layer of the preset analysis model, g k , g1, g2 represent the activation functions of the output layer, the first hidden layer, and the second hidden layer respectively, T2 represents the number of nodes in the second hidden layer, and w xj2 represents the third weight between the second hidden layer and the output layer, T1 represents the number of nodes in the first hidden layer, and w j2j1 represents the second weight between the first hidden layer and the second hidden layer, T0 represents the number of nodes in the input layer, and w j1i represents the first weight between the input layer and the first hidden layer, p 1i represents the i-th row of the feature matrix, p 2i represents the i-th row of the historical feature matrix, B j1 , B j2 , B x Represent the bias values ​​of the first hidden layer, the second hidden layer and the output layer respectively.

[0161] Furthermore, the first weight, the second weight, and the third weight are generated by a preset update formula group during the training process of the preset analysis model, and the preset update formula group is:

[0162]

[0163] Among them, w' j1i represents the first weight after update, w' j2j1 represents the updated second weight, w' xj2 represents the updated third weight, μ represents the learning rate during the training of the preset analysis model, and p 3i represents the i-th row of the training matrix during the training of the preset analysis model, δ yj1 is the training error rate of the preset analysis model in the first hidden layer during the yth training, δ yj2 is the training error rate of the preset analysis model for the yth training on the second hidden layer, δ yx The training error rate of the output layer of the preset analysis model for the yth training.

[0164] Furthermore, the on-board PHM module includes submodules corresponding to each carriage of the train, and each submodule transmits the collected operation data of each carriage to a node of the Ethernet train backbone network based on the Ethernet marshaling network, and the Ethernet train backbone network transmits the data to the on-board controller;

[0165] The step of performing fault monitoring on the train according to the first analysis result comprises:

[0166] According to the node identifier carried by each of the vibration signals, searching for a target carriage corresponding to the first analysis result;

[0167] Based on the target car, the fault type in the first analysis result, and the type parameter corresponding to the fault type, monitoring information for fault monitoring of the motor shaft of the running part in the target car is generated and output.

[0168] Further, the cold running data includes image detection data and sensor detection data of wheel treads in the train, and the step of analyzing the cold running data and the train parameters to generate a second analysis result to predict faults of the train includes:

[0169] Performing smoothing filtering and edge extraction processing on each depth image in the image detection data to obtain a plurality of tread edge images, and converting each of the tread edge images into a binary edge image;

[0170] Generate a pixel matrix corresponding to each of the binary edge images according to the coordinate values ​​of the pixels in the preset coordinate axis in each of the binary edge images, and perform a difference operation on each of the pixel matrices and a reference pixel matrix corresponding to the wheel tread to obtain a first operation matrix;

[0171] Performing a sum operation on each of the first operation matrices to obtain a second operation matrix, and generating a graph analysis result of the wheel tread according to the second operation matrix;

[0172] Generating a sensor analysis result of the wheel tread according to the vibration voltage value, the instantaneous speed value and the train parameter in the sensor detection data;

[0173] The second analysis result is generated based on the graph analysis result and the sensor analysis result to predict the fault of the wheel tread.

[0174] Further, the step of generating the sensor analysis result of the wheel tread according to the vibration voltage value, the instantaneous speed value and the train parameter in the sensor detection data includes:

[0175] Determine whether the vibration voltage value is greater than a preset threshold value, and if so, determine a running section corresponding to the train tread, and search for a section coefficient corresponding to the running section, as well as a first instrument correction coefficient and a second instrument correction coefficient corresponding to the vibration voltage value and the instantaneous speed value, respectively;

[0176] The vibration voltage value, the instantaneous speed value, the section coefficient, and the wheel diameter and the average wheel weight in the train parameters are calculated based on a preset calculation formula to obtain the calculation result, and the preset calculation formula is:

[0177]

[0178] Wherein, H represents the calculation result, D represents the wheel diameter, U represents the vibration voltage value, L represents the road section coefficient, M represents the average wheel weight, V represents the instantaneous speed value, U0 represents the first instrument correction coefficient, and V0 represents the second instrument correction coefficient;

[0179] The calculation result is compared with a preset correspondence between the scratch depth interval and the scratch level to determine the scratch level corresponding to the calculation result, and the calculation result and the scratch level are combined to generate the sensing analysis result.

[0180] Furthermore, the step of generating the second analysis result to predict the fault of the wheel tread according to the graph analysis result and the sensor analysis result includes:

[0181] Determine whether the image analysis result and the sensor analysis result both carry a scratch mark, and if both carry a scratch mark, obtain a first scratch depth and a first scratch level in the image analysis result, and a second scratch depth and a second scratch level in the sensor analysis result;

[0182] generating a similarity value between the first scratch depth and the second scratch depth, and determining whether the similarity value is less than a preset similarity threshold, and whether the first scratch level and the second scratch level are the same;

[0183] If the similarity value is less than a preset similarity threshold, and the first abrasion level and the second abrasion level are the same, an average depth value between the first abrasion depth and the second abrasion depth is generated, and the second analysis result is generated based on the average depth value and the first abrasion level or the second abrasion level, and a fault prediction is performed on the wheel tread according to the second analysis result.

[0184] Further, the step of transmitting the cold operation data to the cloud server includes:

[0185] Searching for the redundant data to be transmitted in the cold running data and the target redundant data corresponding to the redundant data to be transmitted, and determining whether the redundant data to be transmitted and the target redundant data are completely identical, and if they are completely identical, removing any one of the redundant data to be transmitted and the target redundant data from the cold running data to obtain the data to be transmitted;

[0186] If the to-be-redundant data and the target redundant data are not completely the same, the to-be-redundant data and the target redundant data are jointly formed into the to-be-transmitted data;

[0187] Generate an encryption key based on a preset encryption algorithm, and encrypt the data to be transmitted based on the encryption key to obtain encrypted data and transmit it to the cloud server, wherein the preset encryption algorithm is:

[0188] Select any preset prime number from the preset prime number array, and generate a first random number and a first parameter number pair;

[0189] The preset prime number, the first random number and the first parameter pair are calculated based on a first preset formula to obtain a first calculation result, and it is determined whether the first calculation result is a prime number, wherein the first preset formula is:

[0190] y=a*[e h ]*x 2 +b;

[0191] Wherein, y represents the first calculation result, a and b represent the first parameter pair, h represents the first random number, x represents a preset prime number, and [.] represents rounding down;

[0192] If the first calculation result is a prime number, generate a second random number and a second random number pair, and calculate the second random number, the second parameter pair and the first calculation result based on a first preset formula to obtain a second calculation result;

[0193] Determine whether the second calculation result is a prime number, and if it is a prime number, calculate the second calculation result based on a second preset formula to obtain a third calculation result;

[0194] Determining whether the third calculation result is a prime number, and if it is a prime number, determining the third result as a first prime number;

[0195] The steps of selecting any preset prime number from the preset prime number array and generating a first random number are performed, a second prime number is generated, and an encryption key is generated according to the first prime number and the second prime number.

[0196] The specific implementation of the Ethernet-based train fault monitoring and prediction system of the present invention is basically the same as the above-mentioned Ethernet-based train fault monitoring and prediction method embodiments, and will not be repeated here.

[0197] The embodiment of the present invention further provides a storage medium on which a control program is stored, and when the control program is executed by a processor, the steps of the Ethernet-based train fault monitoring and prediction method are implemented.

[0198] The storage medium of the present invention may be a computer-readable storage medium, and its implementation is substantially the same as the above-mentioned embodiments of the Ethernet-based train fault monitoring and prediction method, and will not be described in detail herein.

[0199] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims. All equivalent structures or equivalent process changes made using the contents of the specification and drawings of the present invention, or directly or indirectly used in other related technical fields, are protected by the present invention.

Claims

1. A train fault monitoring and prediction method based on Ethernet, characterized in that: The train includes an onboard controller and an onboard PHM module connected to the onboard controller via an Ethernet train backbone network. The train fault monitoring and prediction method includes: After receiving the train operation data transmitted by the on-board PHM module based on the Ethernet train backbone network, the on-board controller determines the target location of the train and requests to establish a communication connection with the edge server corresponding to the target location to transmit the train operation data to the edge server; The edge server divides the train operation data into hot operation data and cold operation data, and requests a cloud server that is communicatively connected to the edge server to obtain historical operation data corresponding to the hot operation data, and transmits the cold operation data to the cloud server; The edge server analyzes the thermal operation data and the historical operation data based on a preset analysis model to generate a first analysis result, and performs fault monitoring on the train according to the first analysis result; The cloud server searches for train parameters corresponding to the cold operation data of the train, analyzes the cold operation data and the train parameters, and generates a second analysis result to predict faults of the train.

2. The train fault monitoring and prediction method according to claim 1, characterized in that: The thermal operation data includes a vibration signal corresponding to a motor shaft in a running gear of the train, and the edge server analyzes the thermal operation data and the train history data based on a preset analysis model, and the step of generating a first analysis result includes: Performing time domain analysis and frequency domain analysis on the vibration signal respectively to obtain an original time domain feature group and an original frequency domain feature group of the vibration signal; Filtering the vibration signal, and extracting a filtering time domain feature group and a filtering frequency domain feature group from the filtered vibration signal; The original time domain feature group, the original frequency domain feature group, the filtered time domain feature group and the filtered frequency domain feature group are formed into a feature matrix, and the historical operation data are formed into a historical feature matrix; Based on the input layer, the first hidden layer, the second hidden layer and the output layer in the preset analysis model, the feature matrix and the historical feature matrix are subjected to fault analysis to obtain the first analysis result, wherein the output function of the output layer is: Among them, a[x] represents the output value of the xth node in the output layer of the preset analysis model, g k , g1, g2 represent the activation functions of the output layer, the first hidden layer, and the second hidden layer respectively, T2 represents the number of nodes in the second hidden layer, and w xj2 represents the third weight between the second hidden layer and the output layer, T1 represents the number of nodes in the first hidden layer, and w j2j1 represents the second weight between the first hidden layer and the second hidden layer, T0 represents the number of nodes in the input layer, and w j1i represents the first weight between the input layer and the first hidden layer, p 1i represents the i-th row of the feature matrix, p2i represents the i-th row of the historical feature matrix, B j1 , B j2 , B x Represent the bias values ​​of the first hidden layer, the second hidden layer and the output layer respectively.

3. The train fault monitoring and prediction method according to claim 2, characterized in that: The first weight, the second weight, and the third weight are generated by a preset update formula group during the training process of the preset analysis model, and the preset update formula group is: Among them, w' j1i represents the first weight after update, w' j2j1 represents the updated second weight, w' xj2 represents the updated third weight, μ represents the learning rate during the training of the preset analysis model, p3i represents the i-th row of the training matrix during the training of the preset analysis model, δ yj1 is the training error rate of the preset analysis model in the first hidden layer during the yth training, δ yj2 is the training error rate of the preset analysis model for the yth training on the second hidden layer, δ yx The training error rate of the output layer of the preset analysis model for the yth training.

4. The train fault monitoring and prediction method according to claim 2, characterized in that: The on-board PHM module includes submodules corresponding to each carriage of the train, each of which transmits the collected operation data of each carriage to a node of the Ethernet train backbone network based on the Ethernet marshaling network, and then transmits the collected operation data to the on-board controller by the Ethernet train backbone network; The step of performing fault monitoring on the train according to the first analysis result comprises: According to the node identifier carried by each of the vibration signals, searching for a target carriage corresponding to the first analysis result; Based on the target car, the fault type in the first analysis result, and the type parameter corresponding to the fault type, monitoring information for fault monitoring of the motor shaft of the running part in the target car is generated and output.

5. The train fault monitoring and prediction method according to claim 1, characterized in that: The cold running data includes image detection data and sensor detection data of wheel treads in the train, and the step of analyzing the cold running data and the train parameters to generate a second analysis result to predict faults of the train includes: Performing smoothing filtering and edge extraction processing on each depth image in the image detection data to obtain a plurality of tread edge images, and converting each of the tread edge images into a binary edge image; Generate a pixel matrix corresponding to each of the binary edge images according to the coordinate values ​​of the pixels in the preset coordinate axis in each of the binary edge images, and perform a difference operation on each of the pixel matrices and a reference pixel matrix corresponding to the wheel tread to obtain a first operation matrix; Performing a sum operation on each of the first operation matrices to obtain a second operation matrix, and generating a graph analysis result of the wheel tread according to the second operation matrix; Generating a sensor analysis result of the wheel tread according to the vibration voltage value, the instantaneous speed value and the train parameter in the sensor detection data; The second analysis result is generated based on the graph analysis result and the sensor analysis result to predict the fault of the wheel tread.

6. The train fault monitoring and prediction method according to claim 5, characterized in that: The step of generating the sensor analysis result of the wheel tread according to the vibration voltage value, the instantaneous speed value and the train parameter in the sensor detection data comprises: Determine whether the vibration voltage value is greater than a preset threshold value, and if so, determine a running section corresponding to the train tread, and search for a section coefficient corresponding to the running section, as well as a first instrument correction coefficient and a second instrument correction coefficient corresponding to the vibration voltage value and the instantaneous speed value, respectively; The vibration voltage value, the instantaneous speed value, the section coefficient, and the wheel diameter and the average wheel weight in the train parameters are calculated based on a preset calculation formula to obtain the calculation result, and the preset calculation formula is: Wherein, H represents the calculation result, D represents the wheel diameter, U represents the vibration voltage value, L represents the road section coefficient, M represents the average wheel weight, V represents the instantaneous speed value, U0 represents the first instrument correction coefficient, and V0 represents the second instrument correction coefficient; The calculation result is compared with a preset correspondence between the scratch depth interval and the scratch level to determine the scratch level corresponding to the calculation result, and the calculation result and the scratch level are combined to generate the sensing analysis result.

7. The train fault monitoring and prediction method according to claim 5, characterized in that: The step of generating the second analysis result to predict the wheel tread fault according to the graph analysis result and the sensor analysis result comprises: Determine whether the image analysis result and the sensor analysis result both carry a scratch mark, and if both carry a scratch mark, obtain a first scratch depth and a first scratch level in the image analysis result, and a second scratch depth and a second scratch level in the sensor analysis result; generating a similarity value between the first scratch depth and the second scratch depth, and determining whether the similarity value is less than a preset similarity threshold, and whether the first scratch level and the second scratch level are the same; If the similarity value is less than a preset similarity threshold, and the first abrasion level and the second abrasion level are the same, an average depth value between the first abrasion depth and the second abrasion depth is generated, and the second analysis result is generated based on the average depth value and the first abrasion level or the second abrasion level, and a fault prediction is performed on the wheel tread according to the second analysis result.

8. The train fault monitoring and prediction method according to any one of claims 1 to 7, characterized in that: The step of transmitting the cold operation data to the cloud server comprises: Searching for the redundant data to be transmitted in the cold running data and the target redundant data corresponding to the redundant data to be transmitted, and determining whether the redundant data to be transmitted and the target redundant data are completely identical, and if they are completely identical, removing any one of the redundant data to be transmitted and the target redundant data from the cold running data to obtain the data to be transmitted; If the to-be-redundant data and the target redundant data are not completely the same, the to-be-redundant data and the target redundant data are jointly formed into the to-be-transmitted data; Generate an encryption key based on a preset encryption algorithm, and encrypt the data to be transmitted based on the encryption key to obtain encrypted data and transmit it to the cloud server, wherein the preset encryption algorithm is: Select any preset prime number from the preset prime number array, and generate a first random number and a first parameter number pair; The preset prime number, the first random number and the first parameter pair are calculated based on a first preset formula to obtain a first calculation result, and it is determined whether the first calculation result is a prime number, wherein the first preset formula is: y=a*[e h ]*x 2 +b; Wherein, y represents the first calculation result, a and b represent the first parameter pair, h represents the first random number, x represents a preset prime number, and [·] represents rounding down; If the first calculation result is a prime number, generate a second random number and a second random number pair, and calculate the second random number, the second parameter pair and the first calculation result based on a first preset formula to obtain a second calculation result; Determine whether the second calculation result is a prime number, and if it is a prime number, calculate the second calculation result based on a second preset formula to obtain a third calculation result; Determining whether the third calculation result is a prime number, and if it is a prime number, determining the third result as a first prime number; The steps of selecting any preset prime number from the preset prime number array and generating a first random number are performed, a second prime number is generated, and an encryption key is generated according to the first prime number and the second prime number.

9. A train fault monitoring and prediction system based on Ethernet, characterized in that: The Ethernet-based train fault monitoring and prediction system includes a memory, a processor, a communication bus, and a control program stored in the memory: The communication bus is used to realize the connection and communication between the processor and the storage; The processor is used to execute the control program to implement the steps of the Ethernet-based train fault monitoring and prediction method as described in any one of claims 1-8.

10. A storage medium, characterized in that: The storage medium stores a control program, and when the control program is executed by the processor, the steps of the Ethernet-based train fault monitoring and prediction method as described in any one of claims 1-8 are implemented.

Images