Fault type identification method and device, equipment and storage medium

Abstract

The invention discloses a fault type identification method and device, equipment and a storage medium, and relates to the technical field of fault mining. The method comprises the following steps: constructing a fault type identification model based on a convolutional neural network and a bidirectional long short-term memory network, and automatically extracting local features by using the convolutional neural network to ensure that the model can capture the most essential features of a fault; a nonlinear time sequence is processed by using a bidirectional long-short-term memory network, so that the model can understand how characteristics dynamically evolve and develop along with time and learn the life cycle of a fault; therefore, the learned fault data rule is more stable and reliable, and the mining of the fault rule of the same kind has universality. Moreover, a model decision visualization graph is obtained through an output generation class activation graph of the convolutional neural network, so that fault rule mining has interpretability, and fault type display can be combined for auxiliary judgment.

Description

Technical Field

[0001] This invention relates to the field of fault detection technology, and in particular to a fault type identification method, apparatus, device, and storage medium. Background Technology

[0002] With the rapid development of rail transit, modern condition monitoring technology has enabled multi-point, full-lifecycle data collection for complex conditions, ushering in the era of big data for mechanical health monitoring. However, the operating conditions of vehicles / locomotives are complex, and the types of faults are diverse. Traditional fault diagnosis experience is insufficient to cover all aspects. Therefore, how to fully utilize the accumulated massive amounts of data, uncover the hidden data patterns of various component faults, improve the ability to mine fault data patterns, and thus enhance the ability to identify fault types is a pressing issue that needs to be addressed. Summary of the Invention

[0003] In view of this, the purpose of this invention is to provide a fault type identification method, apparatus, device, and storage medium, which can make the learned fault data patterns more stable and reliable. The specific solution is as follows: Firstly, this application discloses a fault type identification method, including: The system acquires monitoring data from the vehicle system, inputs the monitoring data into a fault type identification model, and obtains the fault type corresponding to the monitoring data based on the output of the fault type identification model; the fault type identification model is trained based on a convolutional neural network and a bidirectional long short-term memory network. Based on the output of the convolutional neural network in the fault type identification model, a class activation map is generated, and the class activation map is mapped to the monitoring data in the form of a heat map to obtain a model decision visualization map. The fault types corresponding to the monitoring data and the model decision visualization diagram are displayed through an interactive interface.

[0004] Optionally, the step of generating a class activation map based on the output of the convolutional neural network in the fault type identification model, and mapping the class activation map to the monitoring data in the form of a heatmap, includes: A class activation map is generated based on the output of the last convolutional layer of the convolutional neural network; wherein the class activation map represents the degree of activation of the convolutional neural network at different locations in the input data; Based on the length of the input data of the convolutional neural network, the class activation map is scaled to obtain a scaled class activation map. The scale-transformed class activation map is superimposed on the input data of the convolutional neural network in the form of a heatmap.

[0005] Optionally, generating the class activation map based on the output of the last convolutional layer of the convolutional neural network includes: Obtain the feature data output from multiple feature channels of the last convolutional layer of the convolutional neural network; The gradient weights for each feature channel are determined using the class activation mapping algorithm. Based on the feature data and gradient weights corresponding to each feature channel, a one-dimensional class activation sequence is generated by weighted summation to obtain the class activation map.

[0006] Optionally, the monitoring data may include any one or more of the following: vehicle information, single-sample impact data, single-sample vibration data, and alarm records.

[0007] Optionally, the convolutional neural network is used to extract feature data, and the bidirectional long short-term memory network is used to perform bidirectional analysis on the feature data; The convolutional neural network is constructed in the order of convolutional layer, max pooling layer, convolutional layer, max pooling layer, and fully connected layer.

[0008] Optionally, the training process of the fault type identification model includes: The initial model is constructed in the following order: convolutional neural network, bidirectional long short-term memory network, dropout layer, fully connected layer, and classification layer. The initial model is trained using the training set, and neurons are randomly dropped through the dropout layer according to a preset dropout rate. The backpropagation algorithm is used to dynamically adjust the network weight parameters based on the prediction error in order to minimize the loss function; The trained model is validated using a validation set until it converges, thus obtaining the fault type identification model.

[0009] Optionally, before training the initial model using the training set, the method further includes: Historical monitoring data of the vehicle system is acquired, and the historical monitoring data is preprocessed to obtain the training set; the data preprocessing includes outlier handling, missing value handling, and duplicate value handling. Data preprocessing of the historical monitoring data includes: Abnormal parameters in the monitoring data are filtered out and deleted according to the threshold value, and the abnormal parameters are filled in according to the data within a preset time period. Identify the missing data in the monitoring data and use cubic spline interpolation to impute the missing data; Duplicate data can be selectively filtered out and deleted from the monitoring data.

[0010] Secondly, this application discloses a fault type identification device, comprising: The fault type determination module is used to acquire monitoring data from the vehicle system, input the monitoring data into the fault type identification model, and obtain the fault type corresponding to the monitoring data based on the output of the fault type identification model; the fault type identification model is trained based on a convolutional neural network and a bidirectional long short-term memory network. The model decision visualization generation module is used to generate a class activation map based on the output of the convolutional neural network in the fault type identification model, and to map the class activation map onto the monitoring data in the form of a heat map to obtain the model decision visualization map. The display module is used to display the fault types corresponding to the monitoring data and the model decision visualization diagram through an interactive interface.

[0011] Thirdly, this application discloses an electronic device, including: Memory, used to store computer programs; A processor is used to execute the computer program to implement the aforementioned fault type identification method.

[0012] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein the computer program, when executed by a processor, implements the aforementioned fault type identification method.

[0013] In this application, monitoring data from an in-vehicle system is acquired and input into a fault type identification model. The fault type corresponding to the monitoring data is obtained based on the output of the fault type identification model. The fault type identification model is trained based on a convolutional neural network and a bidirectional long short-term memory network. A class activation map is generated based on the output of the convolutional neural network in the fault type identification model, and the class activation map is mapped onto the monitoring data in the form of a heatmap to obtain a model decision visualization map. The fault type corresponding to the monitoring data and the model decision visualization map are displayed through an interactive interface. It can be seen that by constructing the fault type identification model based on a convolutional neural network and a bidirectional long short-term memory network, the convolutional neural network automatically extracts local features, ensuring that the model can capture the most essential features of the fault. The bidirectional long short-term memory network processes nonlinear time series, ensuring that the model can understand how features dynamically evolve and develop over time, learning the life cycle of the fault. This makes the learned fault data patterns more stable and reliable, and makes the mining of similar fault patterns more universal. Furthermore, the model decision visualization map obtained by generating a class activation map from the output of the convolutional neural network makes the fault pattern mining interpretable, and combined with the fault type display, it can be used to assist in judgment. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or 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 embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0015] Figure 1 A flowchart of a fault type identification method provided in this application; Figure 2 This application provides a schematic diagram of a specific fault type identification network structure; Figure 3 This application provides a schematic diagram of specific fault type identification network parameters; Figure 4 This application provides a schematic diagram of a specific fault type identification network structure; Figure 5 A flowchart of a specific fault type identification method provided in this application; Figure 6 This application provides a schematic diagram of the structure of a fault type identification device; Figure 7 This application provides a structural diagram of an electronic device. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] In related technologies, fault pattern mining methods are mainly divided into physical model-based methods, expert knowledge model-based methods, and data-driven modeling methods. Physical model-based methods study product failure mechanisms or physical phenomena, effectively combining equipment condition monitoring data with the equipment's dynamic characteristics. They determine model parameters and identify equipment fault types using extensive normal operation and fault data. However, physical model-based methods heavily rely on equipment failure mechanisms and require extensive destructive testing, which is impractical in engineering due to the often high cost of equipment components. Expert knowledge model-based methods transform historically accumulated expert knowledge and logical reasoning into rules / algorithms to build knowledge models, primarily including expert systems and fuzzy logic. When equipment faults cannot be represented by general theories, they can fully utilize expert knowledge and experience to better express the equipment's operating state. However, expert knowledge model-based methods rely on operator experience and are easily limited by expert knowledge during modeling. Furthermore, the process of transforming expert knowledge and experience into algorithms using rules / algorithms is highly complex. Data-driven methods train models based on historical equipment monitoring data and include artificial intelligence-based methods and mathematical statistics-based prediction methods. However, data-driven fault pattern mining methods are highly dependent on data, and their ability to mine fault data patterns still needs improvement. Furthermore, black-box model mining lacks interpretability, meaning the mined patterns lack interpretability. To overcome these technical problems, this application proposes a fault type identification method that makes the learned fault data patterns more stable and reliable, enhances the universality of fault pattern mining for similar faults, and provides interpretability for the fault pattern mining process.

[0018] This application discloses a fault type identification method. See [link to relevant documentation] Figure 1 As shown, the method may include the following steps: Step S11: Obtain monitoring data from the vehicle system, input the monitoring data into the fault type identification model, and obtain the fault type corresponding to the monitoring data based on the output of the fault type identification model; the fault type identification model is trained based on a convolutional neural network and a bidirectional long short-term memory network.

[0019] Acquire monitoring data recorded by the onboard system of the vehicle / locomotive. This monitoring data includes any one or more of the following: vehicle information, single-sample impact data, single-sample vibration data, and alarm records. A fault type identification model is obtained based on a combination of convolutional neural networks (CNNs) and bidirectional long short-term memory (LSTM) networks. The CNNs are used to extract feature data, and the LSTM networks are used for bidirectional analysis of the feature data. In a specific embodiment, the CNNs are constructed in the order of convolutional layers, max-pooling layers, convolutional layers, max-pooling layers, and fully connected layers. Combining CNNs and LSTM networks improves fault detection capabilities; fault detection refers to using data information to uncover data patterns in component faults.

[0020] Convolutional Neural Networks (CNNs) are deep learning models specifically designed to process data with a grid-like topological structure. Monitoring data is input into a CNN for feature extraction and dimensionality reduction. In this embodiment, the CNN includes two one-dimensional convolutional layers, two max-pooling layers, and a fully connected layer. The convolutional layers generate feature maps by performing convolution calculations on each local area of ​​the input data using a set of convolutional kernels. The convolution operation involves sliding the convolutional kernel across the input data, multiplying the weights on the kernel with the corresponding data points in the input data one by one, and then summing the results to obtain the value at the corresponding position in the output feature map. This process processes the local receptive field of the input data and extracts local features. The parameters of the convolutional layer include the number and size of the convolutional kernels, the stride, and the padding. The number of convolutional kernels affects the depth of the output feature map, while the kernel size, stride, and padding affect the size of the output. The main purpose of using pooling layers is to reduce computation by compressing feature dimensions, and this operation can also effectively prevent overfitting. Pooling layers are generally placed after activation layers, and are mainly divided into average pooling and max pooling operations. Average pooling outputs the average value of the feature values ​​in the receptive field of the input feature map, while max pooling outputs the maximum feature value in the receptive field. In this embodiment, max pooling is used to obtain position-independent feature values ​​in the receptive field. Finally, a fully connected layer integrates the features extracted by the convolutional and pooling layers in the convolutional neural network into global features, and performs feature extraction and classification again. Specifically, the output feature map of the last convolutional or pooling layer in the convolutional neural network structure is flattened into a one-dimensional feature vector, and this one-dimensional feature vector is used as the input of the fully connected layer for further feature extraction. Finally, the output of the fully connected layer is connected to a Softmax classifier to complete the fault classification task.

[0021] Bidirectional Long Short-Term Memory (BiLSTM) networks address the problem of traditional Recurrent Neural Networks (RNNs) failing to effectively learn long-range dependencies by selectively retaining and transmitting historical information. BiLSTM improves the internal structure of RNNs, enabling them to train and remember information over longer time steps while discarding less important information. Through gate structures (forget gate ft, input gate it, and output gate Ot), BiLSTM discards unnecessary information while retaining more important information, thus achieving the preservation of long-term information and attention to short-term important information, while forgetting unimportant information. The forget gate determines information retention and discard, the input gate determines parameter updates, and the forward propagation information, after being updated and filtered by the forget gate and input gate, reaches the output gate. When processing sequential data, BiLSTM can access not only past time steps but also future time steps; it consists of a forward LSTM network and a backward LSTM network; the output of the LSTM hidden layer is a concatenation of the forward and backward LSTM outputs.

[0022] In some embodiments, the training process of the fault type identification model may include: constructing an initial model in the order of convolutional neural network, bidirectional long short-term memory network, dropout layer, fully connected layer, and classification layer; training the initial model using a training set, and randomly dropping neurons through the dropout layer according to a preset dropout rate; using a backpropagation algorithm to dynamically adjust the network weight parameters according to the prediction error to minimize the loss function; and validating the trained model using a validation set until the model converges to obtain the fault type identification model. For example Figure 2 The diagram illustrates a specific network structure, consisting of two one-dimensional convolutional layers, a max-pooling layer, and a fully connected layer. A bidirectional long short-term memory (LSTM) network is used after the second pooling layer. To prevent overfitting (i.e., the network fits well on the training set but poorly on the test set), a dropout layer is used after the LSM network. For example, a dropout rate of 0.25 is set, causing the network to randomly drop some neurons during forward and backward propagation, reducing interactions between hidden nodes. The dropout layer is disabled in subsequent use. Simultaneously, the network weight parameters are dynamically adjusted using the backpropagation algorithm. The model parameters selected based on ablation experiments are as follows... Figure 3 As shown.

[0023] The final fault type recognition model based on convolutional neural networks and bidirectional long short-term memory networks consists of multiple layers. It first receives one-dimensional input data of length 4096. In the first convolutional layer (Conv1D_1), a convolution operation with a kernel size of 8 and a stride of 2 is used, resulting in an output size of 2045 and 16 channels. Next, a pooling layer (Maxpool_1) is used for downsampling, using a kernel size of 4 and a stride of 4, reducing the output size to 511 while maintaining the same number of channels. The second convolutional layer (Conv1D_2) uses a kernel size of 4 and a stride of 2, resulting in an output size of 254 and increasing the number of channels to 32. Then, a second pooling layer (Maxpool_2) reduces the size to 127. A dropout layer is used to prevent overfitting, maintaining the size at 127. Finally, a bidirectional LSTM (BiLSTM) layer processes the time series data, outputting a sequence of length 127 with a hidden state dimension of 64. The model then passes through two fully connected layers, outputting 64 and 16 units respectively. Finally, a softmax function is used in the output layer to generate classification results for multiple categories. By extracting and compressing features layer by layer, the transformation from a single-sample input to a fault classification output is effectively achieved.

[0024] In some embodiments, before training the initial model using the training set, the method further includes: acquiring historical monitoring data of the vehicle system, performing data preprocessing on the historical monitoring data to obtain the training set; the data preprocessing includes outlier handling, missing value handling, and duplicate value handling; the data preprocessing on the historical monitoring data includes: filtering out and deleting abnormal parameters in the monitoring data according to a threshold value, and filling the abnormal parameters according to data within a preset time period; determining missing data in the monitoring data, and imputing the missing data using cubic spline interpolation technology; filtering out and deleting duplicate data from the monitoring data. Preferably, historical monitoring data of the vehicle system is acquired, including vehicle information, impact / vibration single sample data, and various trend data; vehicle information includes, but is not limited to: vehicle model, vehicle number, axle number, tag number, data start time, and data end time; impact single sample data includes, but is not limited to: sample acquisition time, corresponding rotational speed of the sample, impact single sample data, and fault type; vibration single sample data includes, but is not limited to: sample acquisition time, corresponding rotational speed of the sample, vibration single sample data, and fault type; alarm record data includes, but is not limited to: alarm time, alarm type, alarm level, and alarm rotational speed.

[0025] Data processing for monitoring data includes outlier handling, missing value handling, and duplicate value handling. Considering that outliers generally arise from two causes: first, unreliable data storage due to program malfunctions during vehicle system storage; and second, data storage errors caused by malfunctions in the vehicle's computing program. Therefore, outlier handling includes two aspects: filtering and deleting outlier parameters from the monitoring data based on threshold values, and then filling the outlier parameters with data within a preset time period; for example, data exceeding threshold values ​​are directly removed, such as decibel (dB) data outside the [0, 100dB] range, temperature data outside the [-125°, 125°] range, and alarm data outside the [0, 1, 2] range; and filling the data with mean / mode calculations using short-term data, primarily for dB and temperature data.

[0026] For missing values, cubic spline interpolation is used to impute the missing data. The basic idea of ​​cubic spline interpolation is to represent the interval between every two data points using a cubic polynomial. For each interval... Find a cubic polynomial. Make: ; in, These are the coefficients to be determined. To determine these coefficients, the following conditions must be met: 1. Interpolation condition: The value of the polynomial at each data point is consistent with the original data, that is, for all i, the following condition must be met: ; 2. Continuity condition: The first and second derivatives of the polynomial must be continuous at every data point, that is, for all i, the following condition must be met: ; 3. Boundary conditions: To ensure the uniqueness of the solution, there are usually two common boundary conditions. One is the natural boundary condition, which sets the second derivative to zero at both ends. The other is the closed boundary condition, which requires continuity between the last and last data points, and the first derivative is consistent at both ends.

[0027] Based on the above conditions, a system of linear equations can be constructed to solve it. For simplicity, assume we are given n+1 data points. Then each pair of adjacent data points and The cubic spline interpolation can be calculated using the following steps: 1. Calculate the second-order difference quotient matrix: Construct a tridiagonal matrix containing the second derivative of the interpolation ( ): ; 2. Solving for the coefficients: The coefficients can be solved using the matrix equations described above. Then calculate the other coefficients using the following formula: ; ; .

[0028] 3. Construct the final interpolation function: Using the calculated coefficients, obtain the cubic polynomial form for each interval.

[0029] Furthermore, since the monitoring data from the vehicle-mounted system is stored in the form of data packets, the typical process for storing this data on the ground system involves importing the data packets, parsing their contents, and storing them. Most duplicate data retrieved from the ground system is due to duplicate imports of data packets. Therefore, the preferred approach for handling duplicate data is to directly delete the duplicate values.

[0030] Step S12: Generate a class activation map based on the output of the convolutional neural network in the fault type identification model, and map the class activation map onto the monitoring data in the form of a heatmap to obtain a model decision visualization map.

[0031] Convolutional neural networks lack transparency. This application uses activation maps to visualize the highlighted areas of feature maps, intuitively showing the focus of convolutional neural networks, improving the interpretability of the model, and helping to understand the feature regions that the model focuses on.

[0032] In a specific embodiment, the step of generating a class activation map based on the output of the convolutional neural network in the fault type identification model and mapping the class activation map to the monitoring data in the form of a heatmap includes: generating a class activation map based on the output of the last convolutional layer of the convolutional neural network; wherein, the class activation map represents the activation degree of the convolutional neural network at different positions in the input data; scaling the class activation map according to the length of the input data of the convolutional neural network to obtain a scaled class activation map; and superimposing the scaled class activation map on the input data of the convolutional neural network in the form of a heatmap. That is, because the scale of the class activation map differs from the scale of the input data, a scaling transformation is required first, and then the scaled class activation map is superimposed on the input data of the convolutional neural network in the form of a heatmap. In this embodiment, the input data is a one-dimensional sequence, and the class activation map is just an undulating curve. Without superposition, efficient analysis is impossible. Therefore, superposition allows for a faster understanding of the location of peaks in the heatmap corresponding to the original data, and thus understanding which part of the data the model mainly uses to make its judgments.

[0033] In a specific embodiment, generating a class activation map based on the output of the last convolutional layer of the convolutional neural network includes: acquiring feature data from multiple feature channels of the last convolutional layer of the convolutional neural network; determining the gradient weight of each feature channel according to a class activation mapping algorithm; and generating a one-dimensional class activation sequence by weighted summation based on the feature data and gradient weights corresponding to each feature channel to obtain the class activation map. For example... Figure 4 As shown, features are extracted through a sliding window of convolutional neural kernels. A bidirectional long short-term memory network performs bidirectional analysis on the feature data to learn contextual information. Simultaneously, a class activation mapping algorithm (such as Grad-CAM++) is introduced in the last convolutional layer to perform gradient weight class activation mapping, serving as the model's interpretation layer. Based on the feature extraction results of the neural network, a class activation map is drawn using the Grad-CAM++ algorithm, its size transformed to the original input image size, and then a heatmap is overlaid on the original image. The heatmap values ​​correspond to the activation level of the convolutional neural network at that location.

[0034] Step S13: Display the fault type corresponding to the monitoring data and the model decision visualization diagram through the interactive interface.

[0035] For example Figure 5 The flowchart illustrates a specific fault type identification method. Finally, the fault type identification model outputs the fault type corresponding to the monitoring data, along with a model decision visualization of the current detection data, all displayed through an interactive interface. This allows users to analyze the fault type output by the model based on the model decision visualization. Furthermore, the parameters of the fault type identification model are updated iteratively with each fault case update, giving the model strong noise resistance and robustness.

[0036] As can be seen from the above, this embodiment acquires monitoring data from the vehicle system, inputs the monitoring data into a fault type identification model, and obtains the fault type corresponding to the monitoring data based on the output of the fault type identification model. The fault type identification model is trained based on a convolutional neural network and a bidirectional long short-term memory network. A class activation map is generated based on the output of the convolutional neural network in the fault type identification model, and the class activation map is mapped onto the monitoring data in the form of a heatmap to obtain a model decision visualization map. The fault type corresponding to the monitoring data and the model decision visualization map are displayed through an interactive interface. It is evident that by constructing the fault type identification model based on a convolutional neural network and a bidirectional long short-term memory network, the convolutional neural network automatically extracts local features, ensuring that the model can capture the most essential features of the fault; the bidirectional long short-term memory network processes nonlinear time series, ensuring that the model can understand how features dynamically evolve and develop over time, learning the lifecycle of the fault; making the learned fault data patterns more stable and reliable, and making the mining of similar fault patterns more universal. Furthermore, the model decision visualization map obtained by generating a class activation map through the output of the convolutional neural network makes the fault pattern mining interpretable, and combined with the fault type display, it can be used to assist in judgment.

[0037] Accordingly, this application also discloses a fault type identification device, see [link to relevant documentation]. Figure 6 As shown, the device includes: The fault type determination module 11 is used to acquire monitoring data of the vehicle system, input the monitoring data into the fault type identification model, and obtain the fault type corresponding to the monitoring data based on the output of the fault type identification model; the fault type identification model is trained based on a convolutional neural network and a bidirectional long short-term memory network. The model decision visualization generation module 12 is used to generate a class activation map based on the output of the convolutional neural network in the fault type identification model, and to map the class activation map to the monitoring data in the form of a heat map to obtain the model decision visualization map. The display module 13 is used to display the fault type corresponding to the monitoring data and the model decision visualization diagram through an interactive interface.

[0038] As can be seen from the above, this embodiment acquires monitoring data from the vehicle system, inputs the monitoring data into a fault type identification model, and obtains the fault type corresponding to the monitoring data based on the output of the fault type identification model. The fault type identification model is trained based on a convolutional neural network and a bidirectional long short-term memory network. A class activation map is generated based on the output of the convolutional neural network in the fault type identification model, and the class activation map is mapped onto the monitoring data in the form of a heatmap to obtain a model decision visualization map. The fault type corresponding to the monitoring data and the model decision visualization map are displayed through an interactive interface. It is evident that by constructing the fault type identification model based on a convolutional neural network and a bidirectional long short-term memory network, the convolutional neural network automatically extracts local features, ensuring that the model can capture the most essential features of the fault; the bidirectional long short-term memory network processes nonlinear time series, ensuring that the model can understand how features dynamically evolve and develop over time, learning the lifecycle of the fault; making the learned fault data patterns more stable and reliable, and making the mining of similar fault patterns more universal. Furthermore, the model decision visualization map obtained by generating a class activation map through the output of the convolutional neural network makes the fault pattern mining interpretable, and combined with the fault type display, it can be used to assist in judgment.

[0039] In some specific embodiments, the model decision visualization generation module 12 may specifically include: The class activation map generation unit is used to generate a class activation map based on the output of the last convolutional layer of the convolutional neural network; wherein the class activation map represents the degree of activation of the convolutional neural network at different positions in the input data; The scaling unit is used to scale the class activation map according to the length of the input data of the convolutional neural network to obtain the scaled class activation map. The overlay unit is used to overlay the scale-transformed class activation map onto the input data of the convolutional neural network in the form of a heatmap.

[0040] In some specific embodiments, the class activation graph generation unit may specifically include: The feature data acquisition unit is used to acquire feature data output from multiple feature channels of the last convolutional layer of the convolutional neural network. The gradient weight determination unit is used to determine the gradient weight of each feature channel according to the class activation mapping algorithm; A one-dimensional class activation sequence generation unit is used to generate a one-dimensional class activation sequence by weighted summation based on the feature data and gradient weights corresponding to each feature channel, so as to obtain the class activation map.

[0041] In some specific embodiments, the monitoring data includes any one or more of the following: vehicle information, single-sample impact data, single-sample vibration data, and alarm records.

[0042] In some specific embodiments, the convolutional neural network is used to extract feature data, and the bidirectional long short-term memory network is used to perform bidirectional analysis on the feature data; The convolutional neural network is constructed in the order of convolutional layer, max pooling layer, convolutional layer, max pooling layer, and fully connected layer.

[0043] In some specific embodiments, the training process of the fault type identification model includes: constructing an initial model in the order of convolutional neural network, bidirectional long short-term memory network, dropout layer, fully connected layer and classification layer; training the initial model using a training set, and randomly dropping neurons through the dropout layer according to a preset dropout rate; using a backpropagation algorithm to dynamically adjust the weight parameters of the network according to the prediction error to minimize the loss function; and validating the trained model using a validation set until the model converges to obtain the fault type identification model.

[0044] In some specific embodiments, before training the initial model using the training set, the method further includes: acquiring historical monitoring data of the vehicle system, performing data preprocessing on the historical monitoring data to obtain the training set; the data preprocessing includes outlier handling, missing value handling, and duplicate value handling; the data preprocessing on the historical monitoring data includes: filtering out and deleting abnormal parameters in the monitoring data according to a threshold value, and filling the abnormal parameters with data within a preset time period; identifying missing data in the monitoring data, and imputing the missing data using cubic spline interpolation technology; and filtering out and deleting duplicate data from the monitoring data.

[0045] Furthermore, this application also discloses an electronic device, see [link to relevant documentation]. Figure 7 As shown, the content in the figure should not be considered as any limitation on the scope of use of this application.

[0046] Figure 7 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the fault type identification method disclosed in any of the foregoing embodiments.

[0047] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0048] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon include operating system 221, computer program 222 and data 223 including monitoring data, etc. The storage method can be temporary storage or permanent storage.

[0049] The operating system 221 manages and controls the various hardware devices on the electronic device 20 and the computer program 222 to enable the processor 21 to perform calculations and processing on the massive amounts of data 223 in the memory 22. The operating system 221 can be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the fault type identification method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.

[0050] Furthermore, this application also discloses a computer storage medium storing computer-executable instructions. When the computer-executable instructions are loaded and executed by a processor, they implement the fault type identification method steps disclosed in any of the foregoing embodiments.

[0051] Furthermore, this application also discloses a computer program product, including a computer program that, when executed by a processor, implements the fault type identification method steps disclosed in any of the foregoing embodiments.

[0052] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0053] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0054] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, 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 limitations, 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.

[0055] The above provides a detailed description of the fault type identification method, apparatus, device, and storage medium provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A fault type identification method, characterized in that, include: The system acquires monitoring data from the vehicle system, inputs the monitoring data into a fault type identification model, and obtains the fault type corresponding to the monitoring data based on the output of the fault type identification model; the fault type identification model is trained based on a convolutional neural network and a bidirectional long short-term memory network. Based on the output of the convolutional neural network in the fault type identification model, a class activation map is generated, and the class activation map is mapped to the monitoring data in the form of a heat map to obtain a model decision visualization map. The fault types corresponding to the monitoring data and the model decision visualization diagram are displayed through an interactive interface.

2. The fault type identification method according to claim 1, characterized in that, Generate a class activation map based on the output of the convolutional neural network in the fault type identification model, and map the class activation map onto the monitoring data in the form of a heatmap, including: A class activation map is generated based on the output of the last convolutional layer of the convolutional neural network; wherein the class activation map represents the degree of activation of the convolutional neural network at different locations in the input data; Based on the length of the input data of the convolutional neural network, the class activation map is scaled to obtain a scaled class activation map. The scale-transformed class activation map is superimposed on the input data of the convolutional neural network in the form of a heatmap.

3. The fault type identification method according to claim 2, characterized in that, Generate a class activation map based on the output of the last convolutional layer of the convolutional neural network, including: Obtain the feature data output from multiple feature channels of the last convolutional layer of the convolutional neural network; The gradient weights for each feature channel are determined using the class activation mapping algorithm. Based on the feature data and gradient weights corresponding to each feature channel, a one-dimensional class activation sequence is generated by weighted summation to obtain the class activation map.

4. The fault type identification method according to claim 1, characterized in that, The monitoring data includes any one or more of the following: vehicle information, single-sample impact data, single-sample vibration data, and alarm records.

5. The fault type identification method according to claim 1, characterized in that, The convolutional neural network is used to extract feature data, and the bidirectional long short-term memory network is used to perform bidirectional analysis on the feature data. The convolutional neural network is constructed in the order of convolutional layer, max pooling layer, convolutional layer, max pooling layer, and fully connected layer.

6. The fault type identification method according to any one of claims 1 to 5, characterized in that, The training process of the fault type identification model includes: The initial model is constructed in the following order: convolutional neural network, bidirectional long short-term memory network, dropout layer, fully connected layer, and classification layer. The initial model is trained using the training set, and neurons are randomly dropped through the dropout layer according to a preset dropout rate. The backpropagation algorithm is used to dynamically adjust the network weight parameters based on the prediction error in order to minimize the loss function; The trained model is validated using a validation set until it converges, thus obtaining the fault type identification model.

7. The fault type identification method according to claim 6, characterized in that, Before training the initial model using the training set, the following steps are also included: Historical monitoring data of the vehicle system is acquired, and the historical monitoring data is preprocessed to obtain the training set; the data preprocessing includes outlier handling, missing value handling, and duplicate value handling. Data preprocessing of the historical monitoring data includes: Abnormal parameters in the monitoring data are filtered out and deleted according to the threshold value, and the abnormal parameters are filled in according to the data within a preset time period. Identify the missing data in the monitoring data and use cubic spline interpolation to impute the missing data; Duplicate data is filtered out and deleted from the monitoring data.

8. A fault type identification device, characterized in that, include: The fault type determination module is used to acquire monitoring data from the vehicle system, input the monitoring data into the fault type identification model, and obtain the fault type corresponding to the monitoring data based on the output of the fault type identification model; the fault type identification model is trained based on a convolutional neural network and a bidirectional long short-term memory network. The model decision visualization generation module is used to generate a class activation map based on the output of the convolutional neural network in the fault type identification model, and to map the class activation map onto the monitoring data in the form of a heat map to obtain the model decision visualization map. The display module is used to display the fault types corresponding to the monitoring data and the model decision visualization diagram through an interactive interface.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the fault type identification method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer programs; wherein the computer programs, when executed by a processor, implement the fault type identification method as described in any one of claims 1 to 7.

Images