Turnout machine fault diagnosis method, device, equipment and readable storage medium

Abstract

The present invention provides a method, device, equipment and readable storage medium for fault diagnosis of a turnout switch machine, the method comprising: obtaining action curves of different types of turnout switch machines, the action curves comprising A-phase current curve, B-phase current curve, C-phase current curve and power curve; marking each of the action curves for faults to obtain the marked action curves; constructing a sample set based on each of the marked action curves, and constructing a comparison curve set; training a deep neural network model using the sample set and the comparison curve set to obtain a trained deep neural network model; and performing fault diagnosis on the turnout switch machine to be diagnosed using the trained deep neural network model to obtain a fault diagnosis result. In the present invention, the turnout switch machine fault is helped to be diagnosed by integrating the historical sample information of the turnout switch machine, thereby effectively improving the model diagnosis accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of railway intelligent operation and maintenance, and in particular to a turnout machine fault diagnosis method, device, equipment and readable storage medium. Background Art

[0002] When the turnout switch needs to perform a conversion from positioning to reverse position or from reverse position to positioning, the action circuit inside the switch is connected to provide power support. At this time, the system can obtain these electrical signals through sensors. These electrical characteristic curves, as a reflection of the action process of the turnout switch, can indirectly feedback the health status of the turnout switch. Operation and maintenance personnel can use these electrical curves to determine whether the turnout switch is working properly. However, due to the lag in data collection and manual analysis, it is difficult to ensure real-time diagnosis of the turnout switch. At present, the maintenance and diagnosis of the switch equipment still depends on periodic maintenance. This method did not pose a great risk in the past when the number of equipment was small and the railway traffic was not dense. However, with the increase in train scale and the increase in driving speed, the periodic maintenance method can no longer meet the needs of railway development. Therefore, an intelligent fault diagnosis method for turnout switch is urgently needed. Summary of the invention

[0003] The object of the present invention is to provide a turnout machine fault diagnosis method, device, equipment and readable storage medium to improve the above-mentioned problems.

[0004] In order to achieve the above objectives, the present application provides the following technical solutions:

[0005] On the one hand, an embodiment of the present application provides a method for diagnosing a fault of a turnout machine, the method comprising:

[0006] Obtaining action curves of different types of turnout machines, wherein the action curves include an A-phase current curve, a B-phase current curve, a C-phase current curve, and a power curve;

[0007] Performing fault marking on each of the action curves to obtain a marked action curve;

[0008] Building a sample set based on each of the annotated action curves, and building a comparison curve set;

[0009] Using the sample set and the comparison curve set to train the deep neural network model to obtain a trained deep neural network model;

[0010] The trained deep neural network model is used to perform fault diagnosis on the turnout machine to be diagnosed, and a fault diagnosis result is obtained.

[0011] In a second aspect, an embodiment of the present application provides a turnout machine fault diagnosis device, which includes an acquisition module, a labeling module, a construction module, a training module and a diagnosis module.

[0012] An acquisition module is used to acquire action curves of different types of turnout machines, wherein the action curves include an A-phase current curve, a B-phase current curve, a C-phase current curve and a power curve;

[0013] A marking module, used for marking each of the action curves for faults to obtain a marked action curve;

[0014] A construction module, used to construct a sample set based on each of the annotated action curves, and to construct a comparison curve set;

[0015] A training module, used to train the deep neural network model using the sample set and the comparison curve set to obtain a trained deep neural network model;

[0016] The diagnosis module is used to use the trained deep neural network model to perform fault diagnosis on the turnout machine to be diagnosed and obtain a fault diagnosis result.

[0017] In a third aspect, the present application provides a turnout machine fault diagnosis device, the device comprising a memory and a processor. The memory is used to store a computer program; the processor is used to implement the steps of the turnout machine fault diagnosis method when executing the computer program.

[0018] In a fourth aspect, an embodiment of the present application provides a readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the above-mentioned turnout machine fault diagnosis method are implemented.

[0019] The beneficial effects of the present invention are:

[0020] 1. The present invention greatly reduces the fault diagnosis overhead by exploring the commonality of fault modes between different types of turnout switch machines, and can be used for fault diagnosis of different types of turnout switch machine equipment, providing a decision-making basis for the intelligent operation and maintenance of turnout switch machine equipment.

[0021] 2. In the present invention, the historical sample information of the turnout machine is integrated to help diagnose the turnout machine fault, which effectively improves the model diagnosis accuracy. At the same time, the constructed model can be used for fault diagnosis of different types of turnout machine equipment.

[0022] 3. In the present invention, the data is sampled to the same dimension to construct a sample set through the data alignment method, which effectively solves the problem of inconsistent data format and lays a data foundation for subsequent model training.

[0023] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or be understood by implementing the embodiments of the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 It is a schematic flow chart of a method for diagnosing a fault of a turnout machine described in an embodiment of the present invention;

[0026] Figure 2 1 is a schematic diagram of the structure of a fault diagnosis device for a turnout machine according to an embodiment of the present invention;

[0027] Figure 3 1 is a schematic diagram of the structure of a fault diagnosis device for a turnout machine according to an embodiment of the present invention;

[0028] Markings in the figure: 701, acquisition module; 702, annotation module; 703, construction module; 704, training module; 705, diagnosis module; 7031, alignment unit; 7032, discrete unit; 7033, collection unit; 7034, acquisition unit; 7035, processing unit; 7041, division unit; 7042, training unit; 7043, selection unit; 70421, first training subunit; 70422, second training subunit; 704221, construction subunit; 704222, composition subunit; 800, turnout machine fault diagnosis equipment; 801, processor; 802, memory; 803, multimedia component; 804, I / O interface; 805, communication component. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0030] It should be noted that similar reference numerals or letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0031] Example 1

[0032] like Figure 1 As shown, this embodiment provides a method for diagnosing faults of a turnout machine, which includes step S1, step S2, step S3, step S4 and step S5.

[0033] Step S1, obtaining action curves of different types of turnout machines, wherein the action curves include an A-phase current curve, a B-phase current curve, a C-phase current curve and a power curve;

[0034] In this step, the signal centralized monitoring system can be used to collect various types of turnout machine action curves;

[0035] Step S2, marking each of the action curves for faults to obtain a marked action curve;

[0036] In this step, professionals perform fault diagnosis and annotation on each of the action curves;

[0037] Step S3, constructing a sample set based on each of the annotated action curves, and constructing a comparison curve set;

[0038] In this step, a sample set is constructed using steps S31 to S33;

[0039] Step S31: for each of the marked action curves corresponding to the turnout machine, align the marked A-phase current curve, the marked B-phase current curve, the marked C-phase current curve and the marked power curve according to the timestamp direction, and encapsulate them into a first matrix;

[0040] Step S32: performing data alignment processing on all the first matrices by using a processing method of data clipping and zero-filling, and discretizing each first matrix after the alignment processing to obtain a second matrix;

[0041] Step S31 and step S32 can be understood as: aligning in the direction of the timestamp and encapsulating in the form of a matrix X. For each electrical signal component of the turnout machine action curve (i.e., the marked A-phase current curve, the marked B-phase current curve, the marked C-phase current curve, and the marked power curve) X i ∈X, use zero-filling and data clipping to get 400 equally spaced recording points. That is, the data with less than 400 recording points are padded with 0 values, and the data with more than 400 recording points only retains the first 400 recording points. Use the recording points as the new electrical signal composition to construct a sample set with a unified format.

[0042] At the same time, in order to effectively capture the differences in the action curve data of the turnout machine and better identify the electrical data characteristics of different types of turnout machines during their operation, the data needs to be discretized. Therefore, in step S32, the first matrix after each alignment process is discretized using the z-score method;

[0043] In this step, the data is sampled to the same dimension through data alignment method to construct a sample set, which effectively solves the problem of inconsistent data format and lays the data foundation for subsequent model training.

[0044] Step S33: take the second matrix as a sample, and collect all the second matrices to obtain a sample set.

[0045] In this step, a comparison curve set is constructed using steps S34 and S35;

[0046] Step S34, obtaining a comparison curve corresponding to each sample, wherein the comparison curve is an action curve of the turnout switch machine when it is in the earliest normal operation, and the turnout switch machine corresponding to the comparison curve is the same as the turnout switch machine corresponding to each sample and has the same working direction;

[0047] In this step, each sample has its corresponding turnout switch machine, so the comparison curve corresponding to this sample is also collected by this turnout switch machine, and the working direction of the turnout switch machine when collecting the working curve corresponding to this sample is the same as the working direction of the turnout switch machine when collecting the comparison curve; at the same time, in order to obtain the degradation information characteristics of the turnout switch machine, the action curve of the turnout switch machine when it is in the earliest normal operation is used as the comparison curve;

[0048] Step S35: Process each of the contrast curves in the manner of constructing the sample to obtain a third matrix corresponding to each contrast curve, and use each of the third matrices as an element in the contrast curve set.

[0049] In this step, it can be understood that each sample corresponds to an A-phase current curve, a B-phase current curve, a C-phase current curve and a power curve, so the comparison curve corresponding to each sample also includes an A-phase current curve comparison curve, a B-phase current curve comparison curve, a C-phase current curve comparison curve and a power curve comparison curve. When processing, the A-phase current curve comparison curve, the B-phase current curve comparison curve, the C-phase current curve comparison curve and the power curve comparison curve are also manually annotated. After annotating, the annotated A-phase current curve comparison curve, the annotated B-phase current curve comparison curve, the annotated C-phase current curve comparison curve and the annotated power curve comparison curve are aligned in the direction of the timestamp and encapsulated into a fourth matrix; after encapsulation, data alignment processing is performed on all the fourth matrices using a processing method of data clipping and zero value filling, and each fourth matrix after alignment processing is discretized to obtain the third matrix mentioned above; then all the third matrices are set to form a comparison curve set;

[0050] Step S4, training the deep neural network model using the sample set and the comparison curve set to obtain a trained deep neural network model; in this step, the specific training method is step S41, step S42 and step S43;

[0051] Step S41, dividing the sample set into a training set and a validation set according to a preset ratio;

[0052] In this step, in addition to dividing it into training set and validation set, if you want to test the performance of the model, you can also divide it into training set, validation set and test set;

[0053] Step S42, training the deep neural network model using the training set and the comparison curve set, wherein, in the first N rounds of training, the first sub-model is trained, and starting from the N+1th round of training, the training of the first sub-model is stopped, and the second sub-model is trained; the first sub-model includes a twin multi-dimensional time convolutional network, a fully connected layer and a softmax function, and the second sub-model includes a graph attention neural network layer and a softmax function, and N is a positive integer;

[0054] In order to make the model training more stable, speed up the model fitting, and prevent the model training from taking too long, this embodiment adopts a staged training method to train the model, and achieves rapid convergence by training some sub-models each time. In this step, the value of N can be customized according to user needs. In this embodiment, N is 200, that is, the first sub-model is trained in the first 200 rounds, and the second sub-model is trained from the 201st round; in each round, the cross entropy function is used to define the model loss, and the Adam optimizer is used to update the parameters; the specific training method of this step includes step S421 and step S422;

[0055] Step S421, in the first N rounds of training, the first sub-model is trained, wherein the features of each sample in the training set data and the third matrix corresponding to each sample are respectively extracted through the twin multi-dimensional time convolutional network, all the extracted features are input into the fully connected layer, and the feature sequence is output, and the feature sequence is flattened and input into the softmax function after flattening to obtain the confidence scores of different diagnostic categories, where N is a positive integer;

[0056] This step can be understood as: in order to obtain the degradation information characteristics of the turnout switch machine, the turnout switch machine action curve when the turnout switch machine is operating normally for the earliest time is used as the comparison curve. The turnout switch machine comparison curve information is the electrical expression of the normal operation process of this type of turnout switch machine. Comparing the difference between it and the turnout switch machine action curve can effectively identify the degradation of the equipment. The Temporal Convolutional Network (TCN) has shown strong performance and parallelism in multiple sequence processing tasks. The features of the turnout switch machine action curve and the comparison curve are extracted respectively through a twin 3-layer multi-dimensional TCN network. The data format and specification of the turnout switch machine action curve and the comparison curve are the same. The same feature extractor is used to extract the features of the two respectively, and the fully connected layer is used to learn their difference information, that is, the turnout switch machine degradation information characteristics;

[0057] Step S422: Starting from the N+1th round of training, stop training the first sub-model and train the second sub-model, wherein a sample association network graph is constructed, the sample association network graph is represented by an adjacency matrix, the feature sequence and the adjacency matrix corresponding to each sample are input into a graph attention neural network layer, the adjacency matrix is ​​processed using the graph attention neural network layer to extract graph features; the graph features are flattened and input into a softmax function after flattening to obtain confidence scores for different diagnostic categories;

[0058] In this step, the sample association network graph is represented by an adjacency matrix, the vertices of the adjacency matrix are different sample records, and the values ​​in the adjacency matrix are the existence of the association relationship between the two samples, which is 1 if there is an association relationship, and 0 if not; the graph features are extracted using a graph attention neural network layer, which uses self-attention to perform feature aggregation and ensures training stability through a multi-head attention mechanism; the method for constructing the sample association network graph in this step specifically includes steps S4221 and S4222;

[0059] Step S4221, when constructing a sample association subnetwork diagram corresponding to each of the samples, the turnout machine corresponding to each of the samples is recorded as the first turnout machine, and the working curve corresponding to the first turnout machine is used as a node in the sample association subnetwork diagram, and the association between each of the nodes and its first two neighboring nodes is used as the edge of the node; the working curves corresponding to the node and the two neighboring nodes are collected when the first turnout machine is in the same working direction and in the same use cycle; and the time length between the collection time of the working curve corresponding to the node and the collection time of the working curve corresponding to the front neighboring node is less than the preset time length;

[0060] This step can be understood as follows: each sample has a corresponding sample association subnetwork, wherein when constructing this sample association subnetwork, the turnout machine corresponding to this sample is recorded as the first turnout machine, and then the historical working curve of this first turnout machine is obtained, and each historical working curve is taken as a node (recorded as node No. 3), and the historical working curve is screened according to the time distance. The closer the time, the stronger the reference. Therefore, the first two nodes of this node are selected, and they are recorded as node No. 2 and node No. 1 respectively. The collection time of node No. 1 is the earliest, the collection time of node No. 2 is the second, and the collection time of node No. 3 is the latest. These three nodes must meet certain conditions, that is, when collecting the working curves corresponding to these three nodes, the first turnout machine is in the same working direction and in the same use cycle, and the time interval between the collection time of the working curve corresponding to node No. 3 and the collection time of the working curve corresponding to node No. 1 cannot exceed 1 day; at the same time, in this step, the use cycle is: the time period between two maintenance time points of the turnout machine is regarded as a use cycle;

[0061] Step S4222: All the sample-related sub-network graphs are combined into the sample-related network graph.

[0062] Through the above steps, it can be concluded that this embodiment constructs a sample association network diagram on the historical action curve library of the turnout machine, and uses historical sample information to assist in fault diagnosis. This method can achieve extremely high diagnostic accuracy and can be used for fault diagnosis of different types of turnout machine equipment, which can effectively help operation and maintenance personnel identify equipment failures;

[0063] Step S43: After each round of training, a trained model is obtained, and the accuracy of the model is tested using the validation set. When all rounds are completed, the trained model with the highest accuracy is selected as the trained deep neural network model.

[0064] Step S5: Use the trained deep neural network model to perform fault diagnosis on the turnout machine to obtain a fault diagnosis result.

[0065] In actual use, when the turnout machine to be diagnosed is subjected to fault diagnosis, after the model outputs confidence scores of different diagnosis categories, the diagnosis category with the highest confidence score is used as the fault category of the turnout machine to be diagnosed.

[0066] Example 2

[0067] like Figure 2 As shown, this embodiment provides a turnout machine fault diagnosis device, which includes an acquisition module 701, a labeling module 702, a construction module 703, a training module 704 and a diagnosis module 705.

[0068] An acquisition module 701 is used to acquire action curves of different types of turnout machines, wherein the action curves include an A-phase current curve, a B-phase current curve, a C-phase current curve and a power curve;

[0069] A marking module 702 is used to mark each of the action curves for faults to obtain a marked action curve;

[0070] A construction module 703, configured to construct a sample set based on each of the annotated action curves, and to construct a comparison curve set;

[0071] A training module 704 is used to train the deep neural network model using the sample set and the comparison curve set to obtain a trained deep neural network model;

[0072] The diagnosis module 705 is used to use the trained deep neural network model to perform fault diagnosis on the turnout machine to be diagnosed and obtain a fault diagnosis result.

[0073] In a specific implementation of the present disclosure, the construction module 703 further includes an alignment unit 7031 , a discrete unit 7032 and a collection unit 7033 .

[0074] An alignment unit 7031 is used to align the marked A-phase current curve, the marked B-phase current curve, the marked C-phase current curve and the marked power curve according to the timestamp direction for each marked action curve corresponding to the turnout machine, and encapsulate them into a first matrix;

[0075] The discrete unit 7032 is used to perform data alignment processing on all the first matrices by using a processing method of data clipping and zero-filling, and discretize each first matrix after the alignment processing to obtain a second matrix;

[0076] The collection unit 7033 is used to take the second matrix as a sample and collect all the second matrices to obtain a sample set.

[0077] In a specific implementation of the present disclosure, the construction module 703 further includes an acquisition unit 7034 and a processing unit 7035 .

[0078] An acquisition unit 7034 is used to acquire a comparison curve corresponding to each sample, wherein the comparison curve is an action curve of the turnout switch machine when it is in the earliest normal operation, and the turnout switch machine corresponding to the comparison curve is the same as the turnout switch machine corresponding to each sample and has the same working direction;

[0079] The processing unit 7035 is used to process each of the contrast curves in the manner of constructing the sample to obtain a third matrix corresponding to each contrast curve, and use each of the third matrices as an element in the contrast curve set.

[0080] In a specific implementation of the present disclosure, the training module 704 further includes a dividing unit 7041 , a training unit 7042 and a selecting unit 7043 .

[0081] A division unit 7041 is used to divide the sample set into a training set and a validation set according to a preset ratio;

[0082] The training unit 7042 is used to train the deep neural network model using the training set and the comparison curve set, wherein, in the first N rounds of training, the first sub-model is trained, and starting from the N+1th round of training, the training of the first sub-model is stopped and the second sub-model is trained; the first sub-model includes a twin multi-dimensional temporal convolutional network, a fully connected layer and a softmax function, and the second sub-model includes a graph attention neural network layer and a softmax function, and N is a positive integer;

[0083] The selection unit 7043 is used to obtain a trained model after each round of training, and use the verification set to test the accuracy of the model. When all rounds are completed, the trained model with the highest accuracy is selected as the trained deep neural network model.

[0084] In a specific implementation of the present disclosure, the training unit 7042 further includes a first training subunit 70421 and a second training subunit 70422, which are used to:

[0085] The first training sub-unit 70421 is used to train the first sub-model in the first N rounds of training, wherein the features of each sample in the training set data and the third matrix corresponding to each sample are respectively extracted through a twin multi-dimensional time convolutional network, all the extracted features are input into a fully connected layer, a feature sequence is output, the feature sequence is flattened, and after flattening, it is input into a softmax function to obtain confidence scores of different diagnostic categories, and N is a positive integer;

[0086] The second training sub-unit 70422 is used to stop training the first sub-model from the N+1th round of training, and train the second sub-model, wherein a sample association network graph is constructed, the sample association network graph is represented by an adjacency matrix, the feature sequence and the adjacency matrix corresponding to each sample are input into a graph attention neural network layer, the adjacency matrix is ​​processed using the graph attention neural network layer to extract graph features; the graph features are flattened and input into a softmax function after flattening to obtain confidence scores for different diagnostic categories.

[0087] In a specific embodiment of the present disclosure, the second training subunit 70422 further includes a construction subunit 704221 and a composition subunit 704222.

[0088] The construction subunit 704221 is used for recording the turnout machine corresponding to each sample as the first turnout machine when constructing the sample association subnetwork diagram corresponding to each sample, taking the working curve corresponding to the first turnout machine as the node in the sample association subnetwork diagram, and taking the association between each node and its first two neighboring nodes as the edge of the node; the working curves corresponding to the node and the two neighboring nodes are collected when the first turnout machine is in the same working direction and in the same use cycle; and the time length between the collection time of the working curve corresponding to the node and the collection time of the working curve corresponding to the front neighboring node is less than the preset time length;

[0089] The composition subunit 704222 is used to combine all the sample-related sub-network graphs into the sample-related network graph.

[0090] It should be noted that, regarding the device in the above embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment of the method, and will not be elaborated here.

[0091] Example 3

[0092] Corresponding to the above method embodiment, the embodiment of the present disclosure further provides a turnout machine fault diagnosis device, and the turnout machine fault diagnosis device described below and the turnout machine fault diagnosis method described above can be referenced to each other.

[0093] Figure 3 FIG. 8 is a block diagram of a turnout machine fault diagnosis device 800 according to an exemplary embodiment. Figure 3As shown, the switch machine fault diagnosis device 800 may include: a processor 801 and a memory 802. The switch machine fault diagnosis device 800 may also include one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0094] The processor 801 is used to control the overall operation of the turnout machine fault diagnosis device 800 to complete all or part of the steps in the above-mentioned turnout machine fault diagnosis method. The memory 802 is used to store various types of data to support the operation of the turnout machine fault diagnosis device 800, and these data may include, for example, instructions for any application or method operated on the turnout machine fault diagnosis device 800, and application-related data, such as contact data, sent and received messages, pictures, audio, video, etc. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal may be further stored in the memory 802 or sent via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above-mentioned other interface modules can be keyboards, mice, buttons, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is used for wired or wireless communication between the turnout machine fault diagnosis device 800 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 805 can include: Wi-Fi module, Bluetooth module, NFC module.

[0095] In an exemplary embodiment, the turnout machine fault diagnosis device 800 can be implemented by one or more application specific integrated circuits (Application Specific Integrated Circuit, referred to as ASIC), digital signal processors (Digital Signal Processor, referred to as DSP), digital signal processing devices (Digital Signal Processing Device, referred to as DSPD), programmable logic devices (Programmable Logic Device, referred to as PLD), field programmable gate arrays (Field Programmable Gate Array, referred to as FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the above-mentioned turnout machine fault diagnosis method.

[0096] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the above-mentioned turnout machine fault diagnosis method are implemented. For example, the computer-readable storage medium can be the above-mentioned memory 802 including program instructions, and the above-mentioned program instructions can be executed by the processor 801 of the turnout machine fault diagnosis device 800 to complete the above-mentioned turnout machine fault diagnosis method.

[0097] Example 4

[0098] Corresponding to the above method embodiment, the embodiment of the present disclosure further provides a readable storage medium. The readable storage medium described below and the turnout machine fault diagnosis method described above can be referred to in correspondence with each other.

[0099] A readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the turnout machine fault diagnosis method of the above method embodiment are implemented.

[0100] The readable storage medium may specifically be a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or other readable storage medium that can store program codes.

[0101] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for diagnosing faults of a turnout machine, characterized in that: include: Obtaining action curves of different types of turnout machines, wherein the action curves include an A-phase current curve, a B-phase current curve, a C-phase current curve, and a power curve; Performing fault marking on each of the action curves to obtain a marked action curve; Building a sample set based on each of the annotated action curves, and building a comparison curve set; Using the sample set and the comparison curve set to train the deep neural network model to obtain a trained deep neural network model; Using the trained deep neural network model to perform fault diagnosis on the turnout machine to be diagnosed, and obtaining a fault diagnosis result; The deep neural network model is trained using the sample set and the comparison curve set to obtain the trained deep neural network model, including: Dividing the sample set into a training set and a validation set according to a preset ratio; The deep neural network model is trained using the training set and the comparison curve set, wherein the first sub-model is trained in the first N rounds of training, and the training of the first sub-model is stopped from the N+1th round of training, and the second sub-model is trained; the first sub-model includes a twin multi-dimensional temporal convolutional network, a fully connected layer and a softmax function, and the second sub-model includes a graph attention neural network layer and a softmax function, and N is a positive integer; After each round of training, a trained model is obtained, and the accuracy of the model is tested using the validation set. When all rounds are completed, the trained model with the highest accuracy is selected as the trained deep neural network model; The training methods include: In the first N rounds of training, the first sub-model is trained, wherein the features of each sample in the training set data and the third matrix corresponding to each sample are respectively extracted through the twin multi-dimensional time convolutional network, all the extracted features are input into the fully connected layer, and the feature sequence is output. The feature sequence is flattened and input into the softmax function after flattening to obtain the confidence scores of different diagnostic categories, where N is a positive integer; Starting from the N+1th round of training, the training of the first sub-model is stopped and the second sub-model is trained, wherein a sample association network graph is constructed, the sample association network graph is represented by an adjacency matrix, the feature sequence and the adjacency matrix corresponding to each sample are input into a graph attention neural network layer, the adjacency matrix is ​​processed using the graph attention neural network layer to extract graph features; the graph features are flattened and input into a softmax function after flattening to obtain confidence scores for different diagnostic categories.

2. The turnout machine fault diagnosis method according to claim 1, characterized in that: Constructing a sample set based on each of the annotated action curves includes: For each of the marked action curves corresponding to the turnout machine, align the marked A-phase current curve, the marked B-phase current curve, the marked C-phase current curve and the marked power curve according to the timestamp direction, and encapsulate them into a first matrix; Performing data alignment processing on all the first matrices by using data clipping and zero-filling processing, and discretizing each first matrix after the alignment processing to obtain a second matrix; The second matrix is ​​used as a sample, and all the second matrices are grouped together to obtain a sample set.

3. The method for diagnosing faults of a turnout machine according to claim 1, characterized in that: The method for constructing the comparison curve set includes: Obtaining a comparison curve corresponding to each sample, wherein the comparison curve is an action curve of the turnout switch machine when it is operating normally at the earliest time, and the turnout switch machine corresponding to the comparison curve is the same as the turnout switch machine corresponding to each sample and has the same working direction; Each of the contrast curves is processed in the manner of constructing the sample to obtain a third matrix corresponding to each contrast curve, and each of the third matrices is used as an element in the contrast curve set.

4. A fault diagnosis device for a turnout machine, characterized in that: include: An acquisition module is used to acquire action curves of different types of turnout machines, wherein the action curves include an A-phase current curve, a B-phase current curve, a C-phase current curve and a power curve; A marking module, used for marking each of the action curves for faults to obtain a marked action curve; A construction module, used to construct a sample set based on each of the annotated action curves, and to construct a comparison curve set; A training module, used to train the deep neural network model using the sample set and the comparison curve set to obtain a trained deep neural network model; A diagnosis module, used to perform fault diagnosis on the turnout machine to be diagnosed by using the trained deep neural network model to obtain a fault diagnosis result; The training modules include: A division unit, used for dividing the sample set into a training set and a validation set according to a preset ratio; A training unit, used to train the deep neural network model using the training set and the comparison curve set, wherein, in the first N rounds of training, the first sub-model is trained, and starting from the N+1th round of training, the training of the first sub-model is stopped and the second sub-model is trained; the first sub-model includes a twin multi-dimensional time convolutional network, a fully connected layer and a softmax function, and the second sub-model includes a graph attention neural network layer and a softmax function, and N is a positive integer; A selection unit is used to obtain a trained model after each round of training, test the accuracy of the model using the validation set, and select the trained model with the highest accuracy as the trained deep neural network model after all rounds are completed; The training unit comprises: A first training sub-unit is used to train the first sub-model in the first N rounds of training, wherein the features of each sample in the training set data and the third matrix corresponding to each sample are respectively extracted through a twin multi-dimensional time convolutional network, all the extracted features are input into a fully connected layer, a feature sequence is output, the feature sequence is flattened, and the flattened sequence is input into a softmax function to obtain confidence scores of different diagnostic categories, where N is a positive integer; The second training sub-unit is used to stop training the first sub-model from the N+1th round of training and train the second sub-model, wherein a sample association network graph is constructed, the sample association network graph is represented by an adjacency matrix, the feature sequence and the adjacency matrix corresponding to each sample are input into a graph attention neural network layer, the adjacency matrix is ​​processed using the graph attention neural network layer to extract graph features; the graph features are flattened and input into a softmax function after flattening to obtain confidence scores for different diagnostic categories.

5. The turnout machine fault diagnosis device according to claim 4, characterized in that: Building blocks, including: An alignment unit, for aligning the marked A-phase current curve, the marked B-phase current curve, the marked C-phase current curve and the marked power curve according to the timestamp direction for each marked action curve corresponding to the turnout machine, and encapsulating them into a first matrix; A discrete unit, used for performing data alignment processing on all the first matrices by using a processing method of data clipping and zero-filling, and discretizing each first matrix after the alignment processing to obtain a second matrix; The collection unit is used to take the second matrix as a sample and collect all the second matrices to obtain a sample set.

6. The turnout machine fault diagnosis device according to claim 4, characterized in that: Building blocks, including: An acquisition unit is used to acquire a comparison curve corresponding to each sample, wherein the comparison curve is an action curve of the turnout switch machine when it is operating normally at the earliest time, and the turnout switch machine corresponding to the comparison curve is the same as the turnout switch machine corresponding to each sample and has the same working direction; A processing unit is used to process each of the contrast curves in a manner of constructing the sample to obtain a third matrix corresponding to each contrast curve, and use each of the third matrices as an element in the contrast curve set.

7. Turnout machine fault diagnosis equipment, characterized in that: include: Memory for storing computer programs; A processor is used to implement the steps of the turnout machine fault diagnosis method as claimed in any one of claims 1 to 3 when executing the computer program.

8. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the turnout machine fault diagnosis method according to any one of claims 1 to 3 are implemented.

Images