A fault diagnosis method and system for vehicle-mounted cable terminals driven by image data

Abstract

The present application relates to the technical field of electrical equipment detection, and provides an image data-driven on-board cable terminal fault diagnosis method and system, including obtaining the original partial discharge time series signal of the cable terminal; normalizing the signal using the Zscore normalization method; dividing the processed signal into multiple segments, performing a fast Fourier transform on each segment to obtain the frequency amplitude; calculating the relative position of the frequency difference of the frequency amplitude between the segments, and constructing a frequency difference matrix based on the relative position; applying minimum and maximum normalization to convert the frequency difference matrix into a gray value matrix; mapping the element values ​​in the gray value matrix to the RGB channel to form an FDM image; and inputting the FDM image into the EMA‑BCNN defect recognition model to identify the degree of defect. The present invention can simultaneously learn the local and global features of the PD image, thereby realizing accurate identification of the degree of defect of the cable terminal.

Description

Technical Field

[0001] The present application relates to the technical field of electrical equipment detection, and in particular to an image data-driven fault diagnosis method and system for a vehicle-mounted cable terminal. Background Art

[0002] The contents of this section merely provide background information related to the present application and may not constitute prior art.

[0003] In the power supply system of high-speed trains, the insulation performance of cable terminals is crucial to ensure the safe and stable operation of trains. During long-term operation, cable terminals will be affected by thermal, electrical and mechanical stresses, causing the insulation materials to gradually age and deteriorate, which may in turn cause partial discharge (PD) and even lead to catastrophic explosion accidents. Therefore, timely detection and evaluation of the insulation status of cable terminals, especially identifying the severity of insulation defects, is of great significance for preventing faults and ensuring operational safety.

[0004] In the prior art, with the rapid development of image processing technology and machine learning algorithms, researchers have begun to explore cable terminal defect detection methods based on image processing. For example, a Chinese patent with announcement number CN115423740A discloses a method for identifying the size of the internal defect area of ​​an EPR vehicle-mounted cable terminal. This method uses terahertz image detection technology, and through the steps of filtering and denoising the collected terahertz images, marking the defect area, constructing an image data training set, and designing a Deeplabv3+ convolutional neural network model, it realizes the visual recognition of the size of the internal defect area of ​​the EPR vehicle-mounted cable terminal. However, although this method has made some progress in the identification of image defect areas, there are still some problems. For example, it is difficult to characterize the deep pixel information contained in the extracted features, resulting in the target detection results being far from the actual task requirements. In addition, this method needs to be further improved in terms of detection efficiency and accuracy. Summary of the invention

[0005] In order to solve the above technical problems, the purpose of this application is to provide an image data-driven vehicle-mounted cable terminal fault diagnosis method and system, which converts the one-dimensional PD signal into a two-dimensional PD image through the FDM image transformation method, and then uses the EMA-BCNN algorithm to learn the global and local information from the FDM-transformed PD signal, and finally realizes the diagnosis of the severity of the vehicle cable terminal defects.

[0006] The purpose of this application is achieved through the following technical solutions:

[0007] In a first aspect, the present invention provides an image data driven method for diagnosing a fault of a vehicle-mounted cable terminal, comprising:

[0008] Obtain the original partial discharge time series signal at the cable terminal;

[0009] The original partial discharge time series signal is normalized using the Zscore normalization method;

[0010] The normalized local discharge time series signal is divided into a plurality of segments, each segment containing the same number of data points;

[0011] Performing a fast Fourier transform on each segment to obtain the frequency magnitude of each segment;

[0012] Calculate the relative positions of the frequency differences of the frequency amplitudes between the segments, and construct a frequency difference matrix based on the relative positions;

[0013] Apply min-max normalization to convert the frequency difference matrix into a gray value matrix;

[0014] Map the element values ​​in the gray value matrix to the three RGB channels respectively to form an FDM image;

[0015] The FDM image is input into the EMA-BCNN defect recognition model to identify the degree of defects.

[0016] Furthermore, the formula for performing fast Fourier transform on each segment specifically includes:

[0017]

[0018] in, is the discharge time series signal, is the average value of the original partial discharge time series data, is the standard deviation of the original partial discharge time series data, is the number of data points in each segment, is the index, is the original partial discharge time series signal generated by the insulation defect in the cable terminal at time t; , are sequences of different frequencies, e is the base of natural logarithms, For the The value of a frequency point.

[0019] Furthermore, it also includes:

[0020] index According to the symmetry of the fast Fourier transform result, it is set to traverse only half of the FFT fast Fourier transform result.

[0021] Furthermore, the EMA-BCNN defect recognition model includes a first convolutional layer that sequentially transmits data, a plurality of consecutive bottleneck modules, a second convolutional layer, a soft pooling layer, and a third convolutional layer;

[0022] The first convolutional layer is used for initial feature extraction to obtain the initial feature map;

[0023] Multiple consecutive bottleneck modules, each bottleneck module includes the fourth convolution layer, the depth-wise separable convolution layer, the EMA attention layer and the fifth convolution layer for sequential data transmission; the fourth convolution layer is used to reduce the number of channels and the amount of calculation, and reduce the dimension of the feature map; the depth-wise separable convolution layer is used to extract the spatial features of the initial feature map to enhance the expressiveness of the feature map; the EMA attention layer introduces a multi-scale attention mechanism to perform weighted processing on the feature map to improve the distinguishability of the feature map; the fifth convolution layer is used to restore the number of channels and adjust the feature map to the target dimension;

[0024] The second convolutional layer is used to extract and integrate the feature maps processed by the bottleneck module to enhance the expressiveness of the feature maps;

[0025] The soft pooling layer is used to perform feature aggregation and integrate the target information in the feature map through soft pooling operations;

[0026] The third convolutional layer is used to output the final feature representation.

[0027] Furthermore, the number of bottleneck modules is eleven.

[0028] Furthermore, the EMA attention layer introduces a multi-scale attention mechanism to perform weighted processing on the feature map to improve the discrimination of the feature map. Specifically, the following are included:

[0029] The initial feature map is divided into multiple feature groups along the channel dimension, and different weights are assigned to different regions in the feature groups;

[0030] The EMA attention layer uses three parallel paths to extract features from multiple groups. The first and second paths use 1×1 convolution to perform one-dimensional horizontal and vertical pooling on each feature map, and summarize the outputs of the two channels by multiplication to obtain the first tensor. The third path uses 3×3 convolution to directly perform 3×3 convolution operations on the feature map to obtain the second tensor.

[0031] The first tensor is encoded with two-dimensional global average pooling to encode the global spatial information; the second tensor is reshaped into a dimension corresponding to the output of the 1×1 convolution path, and the global spatial information is encoded with two-dimensional global average pooling; the matrix dot multiplication operation is performed on the encoded output result to obtain the first spatial attention map;

[0032] The second tensor is encoded with two-dimensional global average pooling to encode the global spatial information; the first tensor is reshaped into the corresponding dimension before the combined activation mechanism of the channel features to obtain the second spatial attention map that retains the complete spatial position information;

[0033] The weight values ​​of the first spatial attention map and the second spatial attention map are aggregated, and the output feature map within each group is calculated using the Sigmoid function.

[0034] In a second aspect, the present invention provides an image data driven vehicle cable terminal fault diagnosis system, which is used to implement the image data driven vehicle cable terminal fault diagnosis method described in the first aspect, comprising:

[0035] A data acquisition module, used for acquiring the original partial discharge time series signal of the cable terminal;

[0036] A normalization processing module is used to perform normalization processing on the original partial discharge time series signal using a Zscore normalization method;

[0037] A segmentation module, used for dividing the normalized partial discharge time series signal into a plurality of segments, each segment containing the same number of data points;

[0038] A Fourier transform module for performing a fast Fourier transform on each segment to obtain a frequency amplitude of each segment;

[0039] A difference matrix building module, used for calculating the relative positions of the frequency differences of the frequency amplitudes between the segments, and building a frequency difference matrix based on the relative positions;

[0040] A gray value matrix conversion module, for converting the frequency difference matrix into a gray value matrix by applying minimum and maximum normalization;

[0041] An FDM image acquisition module is used to map the element values ​​in the gray value matrix to three RGB channels respectively to form an FDM image;

[0042] The defect degree recognition module is used to input the FDM image into the EMA-BCNN defect degree recognition model to identify the defect degree.

[0043] In a third aspect, the present invention provides an electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the processor implements the steps corresponding to the fault diagnosis method of the vehicle cable terminal driven by image data in the first aspect.

[0044] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps corresponding to the fault diagnosis method of an on-board cable terminal driven by image data in the first aspect.

[0045] In summary, the technical solution of the embodiment of the present application has at least the following advantages and beneficial effects:

[0046] After obtaining the original partial discharge time series signal of the cable terminal, the present invention uses the Zscore normalization method to preprocess the signal to eliminate the deviation and fluctuation in the data, making the signal easier to analyze. Subsequently, the normalized signal is divided into multiple segments of equal length, and a fast Fourier transform is performed on each segment to obtain the frequency amplitude information of each segment. Then, by calculating the difference in frequency amplitude between the segments and their relative positions, a frequency difference matrix is ​​constructed, which can effectively capture the frequency characteristics and structural similarity information in the signal. In order to convert the FDM matrix into a more intuitive image form, the minimum-maximum normalization is applied to convert it into a gray value matrix, and the gray value is mapped to the three RGB channels to generate an FDM image. The distinguishability of the signal is enhanced, and the time series information and frequency characteristics are retained. Finally, the generated FDM image is input into the EMA-BCNN defect recognition model, which combines the EMA attention layer and the bilinear convolutional neural network (BCNN) to simultaneously learn the local and global features of the PD image, thereby realizing accurate recognition of the degree of defects of the cable terminal. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A flow chart of a fault diagnosis method for a vehicle-mounted cable terminal driven by image data provided by the present invention;

[0048] Figure 2 Schematic diagram of image changes of the EMA-BCNN model in the present invention;

[0049] Figure 3 Schematic diagram of defect recognition of the EMA-BCNN model in the present invention;

[0050] Figure 4 It is a schematic diagram of the structure of the MobileNetV3 network in the present invention;

[0051] Figure 5 It is the structural diagram of EMA in the present invention;

[0052] Figure 6 A structural schematic diagram of an image data driven vehicle-mounted cable terminal fault diagnosis system provided by the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0054] Please refer to Figure 1 and Figure 2 The present application embodiment proposes an image data driven vehicle cable terminal fault diagnosis method, comprising:

[0055] S101, obtaining an original partial discharge time series signal at a cable terminal.

[0056] Specifically, a high-voltage AC voltage source is used to provide the voltage required for the test through a 100kVA corona-free transformer, ensuring the safety and stability of signal acquisition. The setting of the partial discharge detection system follows the IEC60885-3:2015 standard to ensure the accuracy and comparability of the experimental results. The test voltage is gradually increased from 0kV, 2kV each time, and maintained for 10 seconds until the selected PD test voltage of 48kV is reached. After reaching the test voltage, the voltage is maintained constant for 1 minute to stabilize the discharge inside the cable terminal. It is ensured that the collected original partial discharge time series signal (PD signal for short) can truly reflect the discharge characteristics of the cable terminal under constant voltage. The PD signals generated by the cable terminal are captured and recorded using the partial discharge detection system. These signals are presented in the form of a one-dimensional time series, containing rich information about the insulation status of the cable terminal.

[0057] S102, using a Zscore normalization method to normalize the original partial discharge time series signal;

[0058] Specifically, assuming that the insulation defect in the cable terminal generates the original partial discharge time series signal at time t, and Represents. The original partial discharge time series signal is normalized using the Zscore normalization method:

[0059]

[0060] in, is the discharge time series signal, is the average value of the original partial discharge time series data, is the standard deviation of the original partial discharge time series data, is the original partial discharge time series signal generated by the insulation defect in the cable terminal at time t.

[0061] S103, dividing the normalized partial discharge time series signal into multiple segments, each segment is m Indicates that each segment contains the same number of data points. The normalized raw partial discharge time series signal Divide Segments, each segment has k data points. N is the length of the original partial discharge time series signal.

[0062] S104, performing a fast Fourier transform on each segment to obtain a frequency amplitude of each segment;

[0063] Specifically, the formula for performing a fast Fourier transform on each segment is as follows:

[0064]

[0065] in, is the number of data points in each segment, is the index, , are sequences of different frequencies, e is the base of natural logarithms, For the The value of a frequency point.

[0066] index According to the symmetry of the fast Fourier transform result, it is set to traverse only half of the FFT fast Fourier transform result.

[0067] S105, calculating the relative positions of the frequency differences of the frequency amplitudes between the segments, and constructing a frequency difference matrix based on the relative positions.

[0068] Specifically, for each segment i and j Calculate the absolute value of the difference between the FFT results and average the differences :

[0069]

[0070] By calculating the relative position of the FFT result of each segment and the frequency difference of all other segments, an m×m matrix M is constructed. The frequency difference matrix when k=2 as follows:

[0071]

[0072] in, For the The value of a frequency point.

[0073] S106, applying minimum-maximum normalization to convert the frequency difference matrix into a gray value matrix;

[0074] Specifically, the frequency difference matrix Convert to frequency difference matrix for:

[0075]

[0076] S107, mapping the element values ​​in the gray value matrix to the three RGB channels respectively to form an FDM image; wherein the time series information and frequency characteristics in the original time series data are included in the converted image. At the same time, each element in the matrix N Represents the average difference in frequency components between the i-th segment and the j-th segment. This processing method enables the matrix N to capture the frequency characteristic changes and structural similarity information between different paragraphs in the time series data, which helps to distinguish the degree of defects more accurately in subsequent work.

[0077] S108, such as Figure 3 As shown, the FDM image is input into the EMA-BCNN defect recognition model to identify the degree of defects.

[0078] Specifically, in order to accurately identify different defect levels of cables with similar PD characteristics, the present invention establishes an EMA-BCNN model and combines it with a multi-scale attention (EMA) module. The structure of the EMA-BCNN model is as follows: Figure 4 As shown in Figure 1, features are extracted through the improved MobileNetv3 network, and then the extracted features are input into the bilinear pooling module. Finally, the softmax layer is used for prediction.

[0079] Among them, the EMA-BCNN defect recognition model is as follows Figure 4 As shown, it includes a first convolutional layer for sequentially transmitting data, a plurality of continuous bottleneck modules, a second convolutional layer, a soft pooling layer, and a third convolutional layer;

[0080] The first convolutional layer is a 3×3 convolutional layer, which is used for initial feature extraction to obtain the initial feature map;

[0081] Multiple continuous bottleneck modules. This embodiment uses 11 bottleneck modules. Each bottleneck module includes a fourth convolution layer, a depth-separable convolution layer, an EMA attention layer, and a fifth convolution layer for sequentially transmitting data; the fourth convolution layer is a 1×1 convolution layer, which is used to reduce the number of channels and the amount of calculation and reduce the dimension of the feature map; the depth-separable convolution layer is a 3×3 convolution layer, which is used to extract the spatial features of the initial feature map to enhance the expression ability of the feature map; the EMA attention layer introduces a multi-scale attention mechanism to perform weighted processing on the feature map to improve the discrimination of the feature map; the 1×1 fifth convolution layer is used to restore the number of channels and adjust the feature map to the target dimension;

[0082] The second convolutional layer is used to extract and integrate the feature maps processed by the bottleneck module to enhance the expressiveness of the feature maps;

[0083] The soft pooling layer is used to perform feature aggregation and integrate the target information in the feature map through soft pooling operations;

[0084] The third convolutional layer is used to output the final feature representation.

[0085] Among them, the EMA attention layer introduces a multi-scale attention mechanism to perform weighted processing on the feature map to improve the discrimination of the feature map. Specifically, it includes:

[0086] like Figure 5 As shown, EMA divides the input feature map into g groups along the channel dimension. In this embodiment, g is set to 8 to explore different semantics associated with these groups, expressed as , denoted as . And assign different weights to different regions in these feature groups.

[0087] The EMA attention layer uses three parallel paths to extract features from multiple groups. The first and second paths use 1×1 convolutions, and the third path uses 3×3 convolutions. In the two 1×1 convolution paths, one-dimensional horizontal and vertical pooling is first performed, and then the outputs of the two channels are summed up by simple multiplication to obtain the first tensor. The third path uses 3×3 convolutions to directly perform 3×3 convolution operations on the feature map to obtain the second tensor.

[0088] The first tensor is encoded with 2D global average pooling to encode global spatial information; the second tensor is reshaped into the dimension corresponding to the output of the 1×1 convolution path, expressed as: The expression of the two-dimensional global pooling operation is as follows:

[0089]

[0090] Among them, Figure 5 As shown, is the height of the feature map, is the width of the feature map, is the coordinate index in the height direction of the feature map, It is the coordinate index in the width direction of the feature map; is the number of channels of the feature map, is the number of groups into which the input and output channels are divided, Indicates the number of channels allocated to each group.

[0091] Perform matrix dot multiplication on the encoded output to obtain the first spatial attention map;

[0092] The second tensor is used to encode the global spatial information using two-dimensional global average pooling; the first tensor is reshaped into the corresponding dimension before the combined activation mechanism of the channel features to obtain a second spatial attention map that retains the complete spatial position information; the second spatial attention map retains complete and accurate spatial position information.

[0093] Subsequently, the weight values ​​of the first and second spatial attention maps are summed up and the output feature map within each group is calculated using the Sigmoid function. This process captures pairwise interactions at the pixel level and highlights the global context of all pixels. The final output of EMA retains the same dimensionality as the input x, so it can be efficiently integrated into modern architectures.

[0094] Based on the same inventive concept, Figure 6 As shown, the present invention provides an image data driven vehicle cable terminal fault diagnosis system, comprising:

[0095] The data acquisition module 201 is used to acquire the original partial discharge time series signal of the cable terminal;

[0096] A normalization processing module 202 is used to perform normalization processing on the original partial discharge time series signal using a Zscore normalization method;

[0097] A segmentation module 203, used for dividing the normalized partial discharge time series signal into a plurality of segments, each segment containing the same number of data points;

[0098] A Fourier transform module 204 for performing a fast Fourier transform on each segment to obtain a frequency amplitude of each segment;

[0099] A difference matrix construction module 205 is used to calculate the relative positions of the frequency differences of the frequency amplitudes between the segments and to construct a frequency difference matrix based on the relative positions;

[0100] A gray value matrix conversion module 206, for converting the frequency difference matrix into a gray value matrix by applying minimum maximum normalization;

[0101] The FDM image acquisition module 207 is used to map the element values ​​in the gray value matrix to the three RGB channels respectively to form an FDM image;

[0102] The defect degree recognition module 208 is used to input the FDM image into the EMA-BCNN defect degree recognition model to identify the defect degree.

[0103] Based on the same inventive concept, the present invention provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, an image data-driven method for diagnosing a vehicle-mounted cable terminal fault is implemented.

[0104] Based on the same inventive concept, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a fault diagnosis method for a vehicle-mounted cable terminal driven by image data.

[0105] The above are only preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A fault diagnosis method for a vehicle-mounted cable terminal driven by image data, characterized in that: include: Obtain the original partial discharge time series signal at the cable terminal; The original partial discharge time series signal is normalized using the Zscore normalization method; The normalized local discharge time series signal is divided into a plurality of segments, each segment containing the same number of data points; Performing a fast Fourier transform on each segment to obtain the frequency magnitude of each segment; Calculating the relative positions of the frequency differences of the frequency amplitudes between the segments, and constructing a frequency difference matrix based on the relative positions; Apply min-max normalization to convert the frequency difference matrix into a gray value matrix; Map the element values ​​in the gray value matrix to the three RGB channels respectively to form an FDM image; Input the FDM image into the EMA-BCNN defect recognition model to identify the degree of defects; The EMA-BCNN defect recognition model includes a first convolutional layer, a plurality of continuous bottleneck modules, a second convolutional layer, a soft pooling layer and a third convolutional layer for sequentially transmitting data; The first convolutional layer is used for initial feature extraction to obtain an initial feature map; A plurality of continuous bottleneck modules, each bottleneck module includes a fourth convolutional layer, a depth-separable convolutional layer, an EMA attention layer and a fifth convolutional layer for sequentially transmitting data; the fourth convolutional layer is used to reduce the number of channels and the amount of calculation, and reduce the dimension of the feature map; The depth-separable convolution layer is used to extract spatial features of the initial feature map to enhance the expressiveness of the feature map; the EMA attention layer introduces a multi-scale attention mechanism to perform weighted processing on the feature map to improve the discrimination of the feature map; the fifth convolution layer is used to restore the number of channels and adjust the feature map to the target dimension; The second convolutional layer is used to extract and integrate features of the feature map processed by the bottleneck module to enhance the expression ability of the feature map; The soft pooling layer is used to perform feature aggregation and integrate the target information in the feature map through the soft pooling operation; The third convolutional layer is used to output the final feature representation.

2. The method for fault diagnosis of a vehicle-mounted cable terminal driven by image data according to claim 1, characterized in that: The formula for performing fast Fourier transform on each segment specifically includes: in, is the discharge time series signal, is the average value of the original partial discharge time series data, is the standard deviation of the original partial discharge time series data, is the number of data points in each segment, is the index, is the original partial discharge time series signal generated by the insulation defect in the cable terminal at time t; , are sequences of different frequencies, e is the base of natural logarithms, For the The value of a frequency point.

3. The method for fault diagnosis of a vehicle-mounted cable terminal driven by image data according to claim 2, characterized in that: Also includes: index According to the symmetry of the fast Fourier transform result, it is set to traverse only half of the FFT fast Fourier transform result.

4. The method for fault diagnosis of a vehicle-mounted cable terminal driven by image data according to claim 1, characterized in that: The number of the bottleneck modules is 11.

5. The method for fault diagnosis of a vehicle-mounted cable terminal driven by image data according to claim 1, characterized in that: The EMA attention layer introduces a multi-scale attention mechanism to perform weighted processing on the feature map to improve the discrimination of the feature map. Specifically, it includes: The initial feature map is divided into multiple feature groups along the channel dimension, and different weights are assigned to different regions in the feature groups; The EMA attention layer uses three parallel paths to extract features from multiple groups. The first and second paths use 1×1 convolution to perform one-dimensional horizontal and vertical pooling on each feature map, and summarize the outputs of the two channels by multiplication to obtain the first tensor. The third path uses 3×3 convolution to directly perform 3×3 convolution operations on the feature map to obtain the second tensor. Encoding the global spatial information using two-dimensional global average pooling on the first tensor; reshaping the second tensor into a dimension corresponding to the output of the 1×1 convolution path, and encoding the global spatial information using two-dimensional global average pooling; performing a matrix dot multiplication operation on the encoded output result to obtain a first spatial attention map; The second tensor is encoded with two-dimensional global average pooling to encode global spatial information; the first tensor is reshaped into a corresponding dimension before the combined activation mechanism of the channel features to obtain a second spatial attention map that retains the complete spatial position information; The weight values ​​of the first spatial attention map and the second spatial attention map are aggregated, and the output feature map in each group is calculated using the Sigmoid function.

6. An image data driven vehicle cable terminal fault diagnosis system, used to implement the image data driven vehicle cable terminal fault diagnosis method according to any one of claims 1 to 5, characterized in that: include: A data acquisition module, used for acquiring the original partial discharge time series signal of the cable terminal; A normalization processing module is used to perform normalization processing on the original partial discharge time series signal using a Zscore normalization method; A segmentation module, used for dividing the normalized partial discharge time series signal into a plurality of segments, each segment containing the same number of data points; A Fourier transform module for performing a fast Fourier transform on each segment to obtain a frequency amplitude of each segment; a difference matrix construction module, used to calculate the relative positions of the frequency differences of the frequency amplitudes between the segments, and to construct a frequency difference matrix based on the relative positions; A gray value matrix conversion module, for converting the frequency difference matrix into a gray value matrix by applying minimum and maximum normalization; An FDM image acquisition module is used to map the element values ​​in the gray value matrix to three RGB channels respectively to form an FDM image; The defect degree recognition module is used to input the FDM image into the EMA-BCNN defect degree recognition model to identify the defect degree.

7. An electronic device, characterized in that: The electronic device includes: a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the computer program, it implements the steps corresponding to the fault diagnosis method of the vehicle cable terminal driven by image data as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps corresponding to the fault diagnosis method of the vehicle-mounted cable terminal driven by image data as described in any one of claims 1 to 5 are implemented.

Images