Deep learning-based fault diagnosis algorithm, device, and storage medium for switch machines.

Abstract

This specification provides a deep learning-based switch machine fault diagnosis algorithm, device, and storage medium. The method includes: acquiring data uploaded by various preset sensors to obtain multi-source data; obtaining switch machine behavior features based on the multi-source data using a preset feature extraction network; and determining whether the switch machine is faulty and the type of fault using the switch machine behavior features as input and based on a preset deep learning network model. The technical solution provided in this application solves the problem in the prior art that fault diagnosis of switch machines is impossible when sensors malfunction.

Description

Technical Field

[0001] This document relates to the field of intelligent operation and maintenance technology for switch machines, and in particular to a switch machine fault diagnosis algorithm, device, and storage medium based on deep learning. Background Technology

[0002] As the core unit of the railway turnout switching system, the intelligent operation and maintenance of switch machines is currently the main research direction.

[0003] Existing technologies primarily utilize data-driven methods for fault diagnosis of switch machines. Specifically, current / voltage sensors, sound / vibration sensors, or visual sensors are pre-set, and the status data collected by these sensors is processed to achieve fault diagnosis of the switch machine.

[0004] However, when sensors malfunction, the status information of the switch machine will be partially or even completely lost, making it impossible to complete effective fault diagnosis, reducing the operation and maintenance efficiency of the switch machine, and thus creating potential risks to the safe and reliable operation of the railway. Summary of the Invention

[0005] Based on the above analysis, this application aims to propose a deep learning-based switch machine fault diagnosis algorithm, device, and storage medium to realize switch machine fault diagnosis by utilizing the remaining effective state monitoring information when a sensor fails, thereby improving the operation and maintenance efficiency of the switch machine and the ability of the railway system to operate safely and reliably.

[0006] Firstly, one or more embodiments of this specification provide a deep learning-based fault diagnosis algorithm for switch machines, including:

[0007] Data uploaded by each preset sensor is acquired to obtain multi-source data, which comes from data when the sensor malfunctions or is operating normally.

[0008] Based on a preset feature extraction network, the behavior features of the switch machine are obtained from the multi-source data;

[0009] Using the switch machine's behavioral characteristics as input, and based on a preset deep learning network model, it is determined whether the switch machine is faulty and the type of fault.

[0010] Furthermore, acquiring the data uploaded by each preset sensor includes: for each sensor, collecting all data within a preset time period, i.e.:

[0011] When the sensor is functioning correctly, collect normal data within a preset time period;

[0012] or,

[0013] When a sensor malfunctions during the data acquisition process, it collects both disturbed and normal data within a preset time period.

[0014] or,

[0015] When the sensor malfunctions before data acquisition, the disturbed data is collected.

[0016] or,

[0017] When the sensor is unable to transmit data, assign a value to the data corresponding to the current sensor.

[0018] The preset time period refers to a fixed time period for collecting sensor data.

[0019] Furthermore, the preset feature extraction network includes a first convolutional layer, a fully connected layer, and a second convolutional layer;

[0020] The method of obtaining switch machine behavior features based on the multi-source data using a preset feature extraction network includes:

[0021] The first convolutional layer extracts features from the multi-source data to obtain initial features;

[0022] The fully connected layer nonlinearizes the initial features;

[0023] The second convolutional layer outputs the switch machine behavior features based on the multi-source data and the nonlinearized initial features.

[0024] Furthermore, let the data size of the multi-source data be N×T, where N represents the number of sensors and T is the length of the monitoring information acquired by the sensors;

[0025] The method of obtaining switch machine behavior features based on the multi-source data using a preset feature extraction network specifically includes:

[0026] The first convolutional layer takes N×T of multi-source data as input and N×1 of initial features as output;

[0027] The fully connected layer takes an initial feature of size N×1 as input and the nonlinearized initial feature of size N×1 as output;

[0028] The second convolutional layer takes N×T of the multi-source data and N×1 of the nonlinearized initial features as input, and outputs the switch machine behavior features of size 1×T.

[0029] Furthermore, the deep learning fault classification network model includes a first sub-model and a second sub-model;

[0030] The method uses the switch machine's behavioral characteristics as input and a preset deep learning fault classification network model to determine whether the switch machine is faulty and the type of fault.

[0031] The first sub-model extracts global features of the switch machine's behavior characteristics;

[0032] The second sub-model extracts local features of the switch machine's behavior characteristics;

[0033] Connect the global features and the local features;

[0034] Based on the global features and the local features, determine whether the switch machine is faulty and the type of fault.

[0035] Furthermore, the feature extraction network and the deep learning network model together constitute an end-to-end deep learning architecture, which completes the fault diagnosis of the switch machine through a single process. It is regarded as a continuous process, rather than being trained and used independently.

[0036] Secondly, embodiments of this application provide a switch machine fault diagnosis device based on deep learning, including: a data acquisition module, a feature extraction module, and a fault diagnosis module;

[0037] The data acquisition module is used to acquire data uploaded by each preset sensor to obtain multi-source data, which comes from data when the sensor malfunctions or is operating normally.

[0038] The feature extraction module is used to obtain switch machine behavior features based on the multi-source data using a preset feature extraction network.

[0039] The fault diagnosis module is used to determine whether the switch machine is faulty and the type of fault based on a preset deep learning network model, using the switch machine's behavioral characteristics as input.

[0040] Furthermore, the data acquisition module is used to collect all data within a preset time period for each of the sensors, namely: when the sensor is fault-free, normal data within the preset time period is collected; or, when the sensor malfunctions during the collection process, disturbed data and normal data within the preset time period are collected; or, when the sensor malfunctions before collection, disturbed data is collected; or, when the sensor cannot transmit data, a value is assigned to the data corresponding to the current sensor. The preset time period represents a fixed time period for collecting sensor data.

[0041] Furthermore, the preset feature extraction network includes a first convolutional layer, a fully connected layer, and a second convolutional layer;

[0042] The feature extraction module is used to extract features from the multi-source data based on the first convolutional layer to obtain initial features; to nonlinearize the initial features based on the fully connected layer; and to obtain the switch machine behavior features as output based on the second convolutional layer according to the multi-source data and the nonlinearized initial features.

[0043] Thirdly, embodiments of this application provide a storage medium, including:

[0044] Used to store computer-executable instructions, which, when executed, implement the method described in any one of the first aspects.

[0045] Compared with the prior art, this application can achieve at least the following technical effects:

[0046] This application obtains multi-source data through sensors, which comes from data collected during both potential sensor malfunctions and normal operation. Even when some sensors fail or all sensors fail during data acquisition, the data collected by all sensors can still be used as input, allowing for analysis of the switch machine's behavior without removing disturbed information, thus ensuring the integrity of the monitoring data. Simultaneously, feature extraction networks and deep learning models are used to eliminate the impact of data from malfunctioning sensors on the switch machine's behavior analysis and to diagnose faults in the switch machine, thereby improving the switch machine's operational efficiency. Furthermore, this application adopts an end-to-end approach, treating data collection, preprocessing, feature extraction, model training, inference, and result analysis as a continuous process rather than independent steps, making the detection process simpler and faster. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart illustrating a deep learning-based fault diagnosis algorithm for switch machines, provided for one or more embodiments of this specification.

[0049] Figure 2 This is a schematic diagram of a feature extraction network for a deep learning-based switch machine fault diagnosis algorithm provided in one or more embodiments of this specification.

[0050] Figure 3This is a schematic diagram of a deep learning network model for a switch machine fault diagnosis algorithm based on deep learning, provided in one or more embodiments of this specification.

[0051] Figure 4 This is a schematic diagram of the TFPN network structure of a deep learning-based switch machine fault diagnosis algorithm provided in one or more embodiments of this specification. Detailed Implementation

[0052] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.

[0053] In the existing technology, when a switch machine is equipped with only a single sensor, the switch machine's status cannot be monitored when the sensor malfunctions, thus making it impossible to diagnose the switch machine's faults in a timely manner. When multiple sensors are installed, there are no relevant literature reports on how to solve the switch machine fault diagnosis problem in this situation when multiple sensors malfunction.

[0054] To overcome the above challenges, this application proposes a deep learning-based fault diagnosis algorithm for switch machines, such as... Figure 1 As shown, it includes the following steps:

[0055] Step S1: Obtain data uploaded by each preset sensor to obtain multi-source data.

[0056] In this embodiment, a multi-source sensor is configured on a switch machine. For each sensor, all data within a preset time period is collected. When the sensor is fault-free, normal data within the preset time period is collected. When a sensor malfunctions during the collection process, normal data before the malfunction, disturbed data at the time of the malfunction, and disturbed data after the malfunction are collected. When a sensor malfunctions before the preset time period, the data collected by the current sensor is assigned a value. The preset time period represents a fixed time period for collecting sensor data.

[0057] For example: Assume a preset time period, i.e., the sensor's detection time is 7 seconds after each startup. If the sensor malfunctions before the acquisition period, it generates post-malfunction disturbance monitoring information, and all disturbance information within those 7 seconds is collected. When the sensor is functioning correctly, normal data from the sensor within the preset 7-second time period is collected. If the sensor malfunctions at the 3rd second, normal data from the 1st to the 3rd second and disturbance data from the 3rd to the 7th second are collected. When the sensor cannot transmit data, a value is assigned to the data corresponding to the current sensor. The data obtained under these circumstances are then aggregated to obtain multi-source data. When the sensor subsequently obtains multi-source data at different time points, the next step is performed according to the different time points.

[0058] It should be noted that sensor malfunctions fall into two main categories. The first category involves a sensor that can transmit signals, but some components are short-circuited, open-circuited, or affected by environmental pulse interference. The second category involves a sensor malfunction that prevents signal transmission, in which case a fixed value is assigned to the sensor output.

[0059] Step S2: Based on a preset feature extraction network, obtain the switch machine behavior features according to the multi-source data, such as... Figure 2 As shown.

[0060] In this embodiment, multi-source data is uploaded to the input of the feature extraction network. The size of the multi-source data is N×T, where N represents the number of sensors and T is the length of the monitoring information acquired by the sensors. Feature extraction is performed through a first convolutional layer, which consists of three one-dimensional convolutions. Global average pooling is then performed on the feature data to reduce the number of features at each depth of the input layer to a specified size, thereby achieving feature pooling. This reduces the number of parameters and increases the accuracy and stability of the network. The fully connected layer performs nonlinear transformation on the pooled features to output initial features. The size of the initial features is N×1. The fully connected layer consists of three linear layers, which can achieve feature dimensionality reduction, thereby reducing redundant features, reducing the space required for dataset storage, and reducing the computational training time required for feature bit depth.

[0061] The initial features are normalized using the Sigmoid function, and then the generation of the feature weight vector is supervised by the Binary Cross Entropy (BCE) loss function. The BCE loss function is denoted as loss. sensor The calculation formula is as follows:

[0062]

[0063] Among them, S n This indicates the working status of the nth sensor, with a value of 0 or 1, representing sensor malfunction and normal operation, respectively.

[0064] This indicates the operating status of the nth sensor output by the module;

[0065] N is the total number of sensors.

[0066] The second convolutional layer takes N×T multi-source data and N×1 nonlinearized initial features as input and 1×T switch machine behavior features as output to obtain the switch machine behavior features. The second convolutional layer consists of a one-dimensional convolution and a nonlinear activation function ReLU.

[0067] Specifically, such as Figure 2 As shown, multi-source data is uploaded to the input end of the feature extraction network, and the initial feature extraction is completed based on the first convolutional layer. The first convolutional layer consists of three one-dimensional convolutions. The parameters a, b, and c in the Conv1D(a,b,c) convolutional layer represent the input channel, output channel, and convolutional kernel size, respectively. The stride is 1. The first one-dimensional convolution is Conv1D(N,32,3), where N represents the information collected by N sensors, 32 represents the output of 32 channels, and 3 represents the convolutional kernel size. Then, it is processed by batch normalization (BN), BN(32), where the parameter 32 is consistent with the number of output channels of the upper one-dimensional convolution. Then, the non-linear activation function ReLU is used for output to reduce the time and space complexity.

[0068] The second one-dimensional convolution is Conv1D(32,32,3), where the first 32 represents the information collected by 32 sensors, the second 32 represents the output of 32 channels, and 3 represents the size of the convolution kernel. Then it is processed by Batch Normalization (BN), BN(32), where the parameter 32 is consistent with the number of output channels of the previous one-dimensional convolution. Finally, the non-linear activation function ReLU is used for output.

[0069] The third one-dimensional convolution is Conv1D(32,N,1), where 32 represents the input information collected by 32 sensors, N represents the output N channels, and 1 represents the convolution kernel size. It is then processed by Batch Normalization (BN), where the parameter N is consistent with the number of output channels of the previous one-dimensional convolution. Finally, the non-linear activation function ReLU is used for output.

[0070] Global pooling is used to average the features of each sensor channel in the depth feature, reducing the time dimension to 1 and generating a depth feature of size N×1.

[0071] The deep features are nonlinearized by a fully connected layer. The feature dimensionality reduction is achieved by a fully connected layer consisting of three linear layers (d, e) to output an initial feature A of size N×1 (where the d and e parameters of the linear layer are the number of input neurons and the number of output neurons, respectively). The first linear layer is a linear layer (N, 32), where N represents N input neurons and 32 represents 32 output neurons. The nonlinear activation function ReLU is then used for the output.

[0072] The second linear layer is a linear layer (32,16), where 32 represents 32 input neurons and 16 represents 16 output neurons, and the non-linear activation function ReLU is used for the output.

[0073] The third linear layer is a linear layer (16, N), where 16 represents 16 input neurons, N represents N output neurons, and the output size is N×1 initial feature A.

[0074] The initial feature A is normalized using the Sigmoid function to generate a feature weight vector B of size N×1. The generation of the feature weight vector B is supervised using the Binary Cross Entropy (BCE) loss function, which enables the multi-source sensor information feature extraction module to generate the feature weight vector B more effectively. The BCE loss function is denoted as loss. sensor The N×1 data is transformed into 1×T data representing the behavior characteristics of the switch machine through the second convolutional layer. The second convolutional layer is Conv1D(N,1,1), where N represents the information collected by N sensors, the first 1 represents one output channel, and the second 1 represents the size of the convolutional kernel. The nonlinear activation function ReLU is then used for output. The output is a 1×T switch machine behavior feature C, which is used for fault diagnosis.

[0075] Step S3: Using the switch machine's behavioral characteristics as input, and based on a pre-defined deep learning network model, determine whether the switch machine is faulty and the type of fault, such as... Figure 3 As shown.

[0076] In this embodiment, the deep learning fault classification network model includes a first sub-model and a second sub-model. The first sub-model consists of a Bidirectional Memory-augmented Long Short Term Memory (BiMLSTM) neural network model, which is responsible for extracting global features of the switch machine's behavior characteristics. The second sub-model consists of a Time-series Feature-Pyramid Network (TFPN) module, which is responsible for extracting local features of the switch machine's behavior characteristic data. The switch machine's behavior characteristics C of size 1×T are used as input. The global features and local features are concatenated to obtain a deep multi-scale feature vector of the switch machine's state information. Based on the global features and local features, it is determined whether the switch machine is faulty and the type of fault.

[0077] Global features represent feature extraction of global information for each switch machine; local features represent feature extraction of a certain area in the switch machine, and finally, multiple local features are fused together as the final feature.

[0078] In this embodiment, BiMLSTM is a recurrent neural network (RNN) architecture that combines the advantages of bidirectional processing and the Memory-augmented Long Short-Term Memory (MLSTM) model. Since traditional LSTM models have poor long-term memory for time series, a filtering module is added to the gating unit of the traditional LSTM model to form a memory gate, thereby enhancing the model's memory for long sequences. Furthermore, in traditional RNNs, the hidden state at each time step is calculated based on the previous hidden state and the current input. However, this only allows information to flow from the past to the future in one direction. Bidirectional processing overcomes this limitation by introducing two MLSTM layers: one processes the input sequence forward (from beginning to end), and the other processes the input sequence backward (from end to beginning). The outputs from both directions are then combined to provide a more comprehensive representation of the input sequence. The forward and backward MLSTM layers operate independently and maintain their respective hidden states. During training, the backward layer receives the input sequence in reverse order, enabling it to capture future contextual information. This bidirectional processing allows BiMLSTM to consider both past and future information when making predictions.

[0079] The TFPN module employs multiple receptive fields to analyze multi-scale features in time-series data from different perspectives, enhancing the richness of multi-scale features and improving the feature discriminability between different data sets. In TFPN, time-series data is processed by convolutional kernels of different sizes (1×1, 3×3, 5×5, and 7×7) to form features within the multi-scale receptive field. Finally, TFPN combines these four spatial features to form the final multi-dimensional features, such as... Figure 4 As shown.

[0080] For example, BiMLSTM can be used to collect behavioral features of a switch machine. Assuming the memory time is within 5 seconds, the behavioral features within this 5-second period are used as input. Two MLSTM processes are performed: one from left to right and one from right to left. The results are then concatenated as the final output. Data closer to 1 second is more accurate when processing from left to right, while data closer to 5 seconds is more accurate when processing from right to left. Therefore, BiMLSTM can improve the accuracy of feature extraction. Figure 4 As shown, the forward processing of the MLSTM model is from left to right, and the reverse processing is from right to left. The results of the forward and reverse processing are combined to obtain the final output result y.

[0081] In this embodiment, the feature extraction network and the deep learning network model together constitute an end-to-end deep learning architecture. First, data from sensor information is collected to obtain multi-source data. Then, the multi-source data is input into the feature extraction network model for feature extraction to obtain the behavioral features of the switch machine. The features of the switch machine are then input into the deep learning network model for fault diagnosis of the switch machine. The diagnosis of the switch machine is completed in one process. It should be noted that the feature extraction network and the deep learning network model are a continuous process, not independently trained and used, which makes the detection of the switch machine simpler and faster, saving computation time.

[0082] In this embodiment, to verify the superiority of the proposed method, traditional machine learning algorithms widely used in fault diagnosis are selected, including Support Vector Machine (SVM), a combination of feature engineering and SVM, Naive Bayesian Classifier (NBC), a combination of feature engineering and NBC, Random Forest (RF), and a combination of feature engineering and RF; deep learning algorithms, including Long Short-Term Memory (LSTM), Gate Recurrent Unit (GRU), LeNet, AlexNet, VGG, GoogleNet, and ResNet, are used as comparison algorithms, as shown in Table 1. Based on the accuracy of experimental performance indicators, although other algorithms can also diagnose switch machine faults, they all affect the final accuracy of switch machine fault diagnosis. However, the RPMNet algorithm has an accuracy of up to 98% for switch machine fault diagnosis, so the RPMNet algorithm is preferred for switch machine fault diagnosis.

[0083] Table 1 Comparison of experimental performance indicators

[0084]

[0085] This application provides a switch machine fault diagnosis device based on deep learning, including: a data acquisition module, a feature extraction module, and a fault diagnosis module;

[0086] The data acquisition module is used to acquire data uploaded by each preset sensor to obtain multi-source data, which comes from data when the sensor malfunctions or is operating normally; the feature extraction module is used to obtain switch machine behavior features based on the multi-source data using a preset feature extraction network; the fault diagnosis module is used to determine whether the switch machine is malfunctioning based on the switch machine behavior features as input and a preset deep learning network model.

[0087] In this embodiment, the data acquisition module is used to collect all data within a preset time period for each sensor; that is, when the sensor is fault-free, it collects data within the preset time period; or, when the sensor malfunctions during the collection process, it collects the data of the current sensor before the malfunction, the data at the time of the malfunction, and the disturbed data after the malfunction; or, when the sensor malfunctions, it collects all disturbed data within the preset time period; when the sensor has no output, it assigns a value to the data collected by the current sensor. The preset time period represents a fixed time period for collecting sensor data. In this embodiment, the preset feature extraction network includes a first convolutional layer, a fully connected layer, and a second convolutional layer.

[0088] The feature extraction module is used to extract features from the multi-source data based on the first convolutional layer to obtain initial features; to nonlinearize the initial features based on the fully connected layer; and to obtain the switch machine behavior features as output based on the second convolutional layer according to the multi-source data and the nonlinearized initial features.

[0089] This application provides a storage medium for storing computer-executable instructions, which, when executed, implement the steps of the fault diagnosis algorithm for the switch machine described in any of the above embodiments.

[0090] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0091] In the 1930s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the methodology). However, with technological advancements, many improvements to the methodology today can be considered direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that an improvement to the methodology cannot be implemented using a hardware physical module. For example, a Programmable Logic Device (PLD) (e.g., a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program a digital system themselves to "integrate" it onto a PLD, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should understand that by simply performing some logic programming on the method flow using one of these hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0092] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0093] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0094] For ease of description, the above apparatus is described by dividing it into various functional units. Of course, when implementing the embodiments of this specification, the functions of each unit can be implemented in one or more software and / or hardware.

[0095] Those skilled in the art will understand that one or more embodiments of this specification can be provided as a method, system, or computer program product. Therefore, one or more embodiments of this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this specification may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0096] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0097] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0098] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0099] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0100] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0101] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0102] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0103] One or more embodiments of this specification can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a particular task or implement a particular abstract data type. One or more embodiments of this specification can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0104] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0105] The above description is merely an embodiment of this document and is not intended to limit the scope of this document. Various modifications and variations can be made to this document by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this document should be included within the scope of the claims of this document.

Claims

1. A deep learning-based method for fault diagnosis of switch machines, characterized in that, include: Data uploaded by each preset sensor is acquired to obtain multi-source data, which comes from data when the sensor malfunctions or is operating normally. Based on a preset feature extraction network, the behavior features of the switch machine are obtained from the multi-source data; wherein, the preset feature extraction network includes a first convolutional layer, a fully connected layer, and a second convolutional layer; the data size of the multi-source data is N×T, where N represents the number of sensors and T is the length of the monitoring information acquired by the sensors; specifically including: Initial features are obtained by extracting features from the multi-source data through the first convolutional layer. The initial features are nonlinearized using the fully connected layer; The second convolutional layer outputs the switch machine behavior features based on the multi-source data and the nonlinearized initial features; The first convolutional layer takes N×T of multi-source data as input and N×1 of initial features as output. The fully connected layer takes an initial feature of size N×1 as input and the nonlinearized initial feature of size N×1 as output; The second convolutional layer takes the multi-source data of size N×T and the nonlinearized initial features of size N×1 as input, and the switch machine behavior features of size 1×T as output; Using the switch machine's behavioral characteristics as input, and based on a preset deep learning network model, it is determined whether the switch machine is faulty and the type of fault.

2. The method according to claim 1, characterized in that, The acquisition of data uploaded by each preset sensor includes: For each of the aforementioned sensors, all data within a preset time period are collected, i.e.: When the sensor is functioning correctly, collect normal data within a preset time period; Alternatively, if a sensor malfunctions during the data acquisition process, it can collect both disturbed and normal data within a preset time period. Alternatively, when the sensor malfunctions before data acquisition, the disturbed data can be acquired; Alternatively, when the sensor is unable to transmit data, a value is assigned to the data corresponding to the current sensor; The preset time period refers to a fixed time period for collecting sensor data.

3. The method according to claim 1, characterized in that, The deep learning fault classification network model includes a first sub-model and a second sub-model. The method uses the switch machine's behavioral characteristics as input and a preset deep learning fault classification network model to determine whether the switch machine is faulty and the type of fault. The first sub-model extracts global features of the switch machine's behavior characteristics; The second sub-model extracts local features of the switch machine's behavior characteristics; Connect the global features and the local features; Based on the global features and the local features, determine whether the switch machine is faulty and the type of fault.

4. The method according to claim 1, characterized in that, The method further includes: The feature extraction network and the deep learning network model together constitute an end-to-end deep learning architecture, enabling a single process to complete the fault diagnosis of the switch machine.

5. A switch machine fault diagnosis device based on deep learning, characterized in that, include: Data acquisition module, feature extraction module, and fault diagnosis module; The data acquisition module is used to acquire data uploaded by each preset sensor to obtain multi-source data, which comes from data when the sensor malfunctions or is operating normally. The feature extraction module is used to obtain switch machine behavior features based on the multi-source data and a preset feature extraction network; wherein the preset feature extraction network includes a first convolutional layer, a fully connected layer, and a second convolutional layer. The feature extraction module is used to extract features from the multi-source data through the first convolutional layer to obtain initial features; to nonlinearize the initial features through the fully connected layer; and to output the switch machine behavior features through the second convolutional layer based on the multi-source data and the nonlinearized initial features. The fault diagnosis module is used to determine whether the switch machine is faulty and the type of fault based on a preset deep learning network model, using the switch machine's behavioral characteristics as input.

6. The apparatus according to claim 5, characterized in that, The data acquisition module is used to collect all data within a preset time period for each of the sensors, that is: when the sensor is fault-free, normal data within the preset time period is collected. Alternatively, if the sensor malfunctions during the data acquisition process, both disturbed and normal data can be collected within a preset time period; or, if the sensor malfunctions before data acquisition, disturbed data can be collected. Alternatively, when the sensor is unable to transmit data, a value is assigned to the data corresponding to the current sensor; wherein, the preset time period represents a fixed time period for collecting sensor data.

7. A storage medium, characterized in that, include: Used to store computer-executable instructions, which, when executed, implement the method of any one of claims 1-4.

Images