Method for evaluating surface defects of semi-conductive shielding layer of high-voltage cable

Abstract

The invention relates to the technical field of electrical digital data processing, and provides a high-voltage cable semi-conductive shielding layer surface defect evaluation method, which comprises the following steps: S1, obtaining shielding layer surface defect morphology characteristics; s2, constructing a finite element model based on defect morphology features, and executing analogue simulation calculation; s3, constructing a standardized response mapping data set based on an analogue simulation calculation result; s4, on the basis of the standardized response mapping data set, constructing a neural network model and executing iterative training; s5, performing finite element simulation based on the geometric shape of the defect contour so as to verify the accuracy of the neural network model; according to the invention, high-flux prediction of electric field distortion caused by surface defects of the semi-conductive shielding layer of the target cable can be realized so as to evaluate the influence on an insulation system. Moreover, the prediction and evaluation speed of the method is far higher than that of traditional simulation, and the method can also show higher precision for untrained defect prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical digital data processing, and in particular to a method for evaluating surface defects of a semi-conductive shielding layer of a high-voltage cable. Background Art

[0002] High-voltage cross-linked polyethylene (XLPE) cables are widely used in modern power transmission systems. However, the surface area of ​​the cable's semi-conductive shielding layer often contains localized geometric defects at the micron level. These defects can arise from various factors, including uneven extrusion processes, residual impurities, carbon black agglomerates, mechanical stress, or uneven cooling. These defects, such as protrusions and pits, cause severe distortion of the local field strength in areas of concentrated electric field concentration, inducing space charge injection and accumulation, accelerating the degradation of the insulation layer's insulation performance, and even causing breakdown accidents. Therefore, high-voltage cables have extremely high requirements for the surface smoothness of the semi-conductive shielding layer.

[0003] In the existing technology, the evaluation methods for semi-conductive shielding layers mainly include white light interferometers, scanning electron microscopes, etc. to evaluate the size, but the scanning area is limited and the speed is slow, making it difficult to achieve high-throughput defect identification of large-area samples; high throughput refers to the ability to quickly process, analyze or evaluate a large number of samples, data or tasks; on the other hand, the current evaluation system and indicators for the surface smoothness of semi-conductive shielding layers are still imperfect. Existing methods mostly ignore the deep coupling relationship between the geometric characteristics of the defect morphology and the local electric field response, only focus on the defect size, lack real physical meaning, and have low prediction and evaluation accuracy.

[0004] Therefore, a defect assessment method that can achieve large-area high-throughput extraction of defects and has higher accuracy is needed. Summary of the Invention

[0005] The present invention provides a method for evaluating surface defects of a semi-conductive shielding layer of a high-voltage cable, which aims to solve the problem in the prior art of being unable to achieve large-area high-throughput extraction of defects.

[0006] The technical solutions of the present invention are as follows:

[0007] A method for evaluating surface defects of a semi-conductive shielding layer of a high-voltage cable comprises the following steps:

[0008] S1. Obtaining the morphological characteristics of the surface defects of the shielding layer;

[0009] S2. Construct a finite element model based on the defect morphology and perform simulation calculations;

[0010] S3. Construct a standardized response mapping dataset based on simulation calculation results;

[0011] S4. Build a neural network model based on the standardized response mapping dataset and perform iterative training;

[0012] S5. Verify the accuracy of the neural network model by performing simulation comparison based on the defect contour; perform finite element simulation based on the geometric shape of the defect contour to verify the accuracy of the neural network model.

[0013] Furthermore, the step S1 includes the following steps:

[0014] S1.1 simulate the cable extrusion process to prepare shielding tape;

[0015] S1.2. Detect defective particles in the shielding layer strip and produce a surface defect image;

[0016] S1.3. Use image processing algorithms to extract defect contours and define three dimensionless geometric descriptors.

[0017] Furthermore, the three dimensionless geometric descriptors are normalized width W, aspect ratio AR and sharpness S; wherein the normalized width W is the defect base width, the aspect ratio AR is the ratio of the defect height to the width, and the sharpness S is the ratio of the full width at half maximum FWHM of the defect top to the base width.

[0018] Furthermore, step S2 includes the following steps:

[0019] S2.1. Measure the resistance-temperature characteristics of the shielding layer and the resistance-temperature-field strength characteristics of the insulating layer used in the shielding layer;

[0020] S2.2. Construct a finite element simulation model of the target cable and obtain the maximum electric field intensity and distortion rate in the defect area by solving the electric field distribution.

[0021] S2.3. Simulate and calculate the defect shape-maximum electric field distortion under different defect shapes and summarize the data samples.

[0022] Furthermore, step S3 includes the following steps:

[0023] S3.1. Construct a structure-electric field response mapping dataset using the defined dimensionless geometric parameters as input.

[0024] S3.2. Standardize the dataset using the mean-standard deviation normalization method.

[0025] S3.3. Use a two-stage data partitioning strategy to partition the dataset.

[0026] Furthermore, step S4 includes the following steps:

[0027] S4.1. Construct a neural network model;

[0028] S4.2. Use algorithms to iteratively update the neural network model;

[0029] S4.3. Selecting the optimal neural network architecture.

[0030] Furthermore, the neural network model is a single hidden layer feedforward neural network model, and the single hidden layer feedforward neural network model includes an input layer, an output layer and a hidden layer, the output layer uses a linear function, and the hidden layer uses a hyperbolic tangent function.

[0031] Furthermore, the verification method of step S5 is to perform finite element simulation based on the real geometric shape and parameterized geometric shape of the defect contour respectively, and compare the simulation results obtained with the prediction results of the single hidden layer feedforward neural network model to verify the accuracy of the single hidden layer feedforward neural network model.

[0032] The beneficial effects of the present invention are:

[0033] This method enables high-throughput prediction of the electric field distortion caused by surface defects in the target cable's semi-conductive shielding layer, thereby assessing its impact on the insulation system. Furthermore, the neural network's prediction speed far exceeds that of traditional simulation. For actual defects, the electric field distortion intensity can be rapidly predicted simply by extracting their geometric descriptors and inputting them into the neural network model. The use of dimensionless geometric descriptors to uniformly describe defects of varying scales improves the neural network's model generalization capability, enabling high accuracy in predicting untrained defects. The prediction method is scalable and applicable to diverse cable structures and operating conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0035] Figure 1 is a flow chart of the defect assessment method of the present invention;

[0036] Figure 2 Schematic diagram of typical interface defect image and contour extraction results of the present invention;

[0037] Figure 3 It is a schematic diagram of the training effect of the neural network model of the present invention;

[0038] Figure 4 It is a schematic diagram of the defect simulation results of the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0040] Example

[0041] This embodiment provides a method for evaluating surface defects of a semi-conductive shielding layer of a high-voltage cable, comprising the following steps:

[0042] S1. Obtaining the surface defect morphology characteristics of the shielding layer;

[0043] S2. Construct a finite element model based on the defect morphology and perform simulation calculations;

[0044] S3. Construct a standardized response mapping dataset based on simulation calculation results;

[0045] S4. Build a neural network model based on the standardized response mapping dataset and perform iterative training;

[0046] S5. Perform finite element simulation based on the geometric shape of the defect contour to verify the accuracy of the neural network model.

[0047] Furthermore, step S1 includes the following steps:

[0048] S1.1 simulate the cable extrusion process to prepare shielding tape;

[0049] S1.2. Detect defective particles in the shielding layer strip and produce a surface defect image;

[0050] S1.3. Use image processing algorithms to extract defect contours and define three dimensionless geometric descriptors.

[0051] Specifically:

[0052] S1.1. Simulate the cable extrusion process to prepare shielding tape.

[0053] A semi-conductive shielding layer tape is prepared by simulating the cable extrusion process through a single-screw extruder. The width and thickness of the semi-conductive shielding layer tape are related to the extrusion temperature and speed of the single-screw extruder. The specific working principle is based on existing technology and will not be further described here.

[0054] It should be noted that, in this process, the extrusion temperature range of the single-screw extruder can be 100-120°C, and the speed range can be 10-25 rpm, which are not limited here;

[0055] Preferably, in this embodiment, the single-screw extrusion temperature range is 120° C., the rotation speed is 15 rpm, and the obtained semi-conductive shielding layer tape has a width of 2.5 cm and a thickness of 1 mm.

[0056] S1.2. Detect defective particles in the shielding layer strip and produce a surface defect image;

[0057] The surface quality analyzer is used to detect large-sized particles in a certain area of ​​the semi-conductive shielding layer strip and obtain an image of each surface defect;

[0058] It should be noted that the surface quality analyzer used, the area of ​​the semi-conductive shielding layer strip tested, the reference standard for large-sized particles, and the number of images of each surface defect obtained are not limited here, and can only meet the detection and testing requirements;

[0059] Preferably, in this embodiment, the surface quality analyzer used is OCS surface quality analyzer (SQA100), and the area of ​​the semi-conductive shielding layer strip to be detected is 1m 2 , the criterion for judging large-sized particles is (height>30μm), and the number of images of various surface defects obtained is 165; Figure 2 (a)-(e) show five typical defects, representing smaller size, medium size, larger size, wide shape, and sharp shape, respectively; Width is the defect width, Height is the defect height, and FWHM is the full width at half maximum.

[0060] S1.3. Use image processing algorithms to extract defect contours and define three dimensionless geometric descriptors.

[0061] like Figure 2 As shown in (f)-(j), the defect contour is extracted using image processing algorithms, including clipping, enhancement, and edge detection algorithms, and three dimensionless geometric descriptors are defined;

[0062] It should be noted that the image processing algorithms mentioned above include but are not limited to cropping, enhancement, and edge detection algorithms. Similar effects can also be achieved through other image processing algorithms, such as filtering algorithms and image segmentation algorithms, which are not limited here.

[0063] Preferably, in this embodiment, the edge detection algorithm used is the Canny edge detection algorithm.

[0064] Furthermore, the three dimensionless geometric descriptors are normalized width (W), aspect ratio (AR), and sharpness (S); wherein normalized width (W) is the defect base width, aspect ratio (AR) is the ratio of defect height to width, and sharpness (S) is the ratio of the full width at half maximum (FWHM) of the defect top to the base width;

[0065] Sharpness (S) indirectly reflects the curvature of the defect profile at the top region, providing information about its geometry. Full-width-at-half-maximum (FWHM) is a physical quantity used to describe peak width, defined as the width of the curve at half the peak height. These three dimensionless geometric descriptors enable a unified description of the geometric shapes of different defects, facilitating subsequent analysis.

[0066] Furthermore, step S2 includes the following steps:

[0067] S2.1. Measure the resistance-temperature characteristics of the shielding layer and the resistance-temperature-field strength characteristics of the insulating layer used in the shielding layer;

[0068] S2.2. Construct a finite element simulation model of the target cable and obtain the maximum electric field intensity and distortion rate in the defect area by solving the electric field distribution.

[0069] S2.3. Simulate and calculate the defect shape-maximum electric field distortion under different defect shapes and summarize the data samples.

[0070] Specifically:

[0071] S2.1. Measure the resistance-temperature characteristics of the shielding layer and the resistance-temperature-field strength characteristics of the insulating layer used in the shielding layer;

[0072] The resistance-temperature characteristics of the detected shielding layer are measured, and the measurement temperature range is 30-90°C. The measurement method can adopt the "bridge method" or the like. The measurement method is a conventional technology and is not further described here. The resistance-temperature-field strength characteristics of the insulating layer used to match the shielding layer are measured, and the measurement temperature range is 30-90°C, and the electric field strength range is 10-40 kV / mm. The measurement method can adopt the "three-electrode measurement method" or the like. The measurement method is a conventional technology and is not further described here.

[0073] S2.2. Construct a finite element simulation model of the target cable and obtain the maximum electric field intensity and distortion rate in the defect area by solving the electric field distribution.

[0074] Finite element simulation software was used to construct a two-dimensional axisymmetric finite element model. The multi-layer cable structure (conductor layer, inner shielding layer, insulation layer, outer shielding layer, sheath layer, etc.) was simulated according to the actual size of the target cable. The electric field distribution was then solved under different thermal field and conductivity models to obtain the maximum electric field intensity and distortion rate in the defect area. The construction method of the finite element model and the solution of the electric field distribution are both existing technologies and will not be further explained here.

[0075] Preferably, in this embodiment, the finite element simulation software used is Comsol Multiphysics.

[0076] S2.3. Simulate and calculate the “defect shape-maximum electric field distortion” under different defect shapes and summarize the data samples.

[0077] Add defects of different shapes to the surface of the shielding layer of the finite element model constructed in step S2.2, perform simulation calculations, and collect at least 600 sets of "defect shape-maximum electric field distortion" data samples;

[0078] Preferably, in this embodiment, 800 sets of data samples of “defect shape-maximum electric field distortion” are collected.

[0079] Furthermore, step S3 includes the following steps:

[0080] S3.1. Construct a structure-electric field response mapping dataset using the defined dimensionless geometric parameters as input.

[0081] S3.2. Standardize the dataset using the mean-standard deviation normalization method.

[0082] S3.3. Use a two-stage data partitioning strategy to partition the dataset.

[0083] Specifically:

[0084] S3.1. Using the three dimensionless geometric descriptors of the defect as input, construct a structure-electric field response mapping dataset.

[0085] Using the three dimensionless geometric descriptors defined in step S1.3, namely normalized width (W), aspect ratio (AR), and sharpness (S), as input, and performing finite element simulation on the target cable finite element simulation model constructed in step S2.2, the maximum distortion rate of the local electric field is obtained. The maximum distortion rate of the local electric field is then used as output to construct a data set for the "structure-electric field" response mapping;

[0086] S3.2. Standardize the dataset using the mean-standard deviation normalization method.

[0087] Specifically, the "mean-standard deviation normalization" method is used to standardize the "structure-electric field" response mapping data set constructed in step S3.1; "mean-standard deviation normalization" standardization is a common data standardization method, whose purpose is to convert the data into a distribution with zero mean and unit standard deviation. Its main function is to eliminate dimensional effects, accelerate model convergence, improve training efficiency and improve model performance; "mean-standard deviation normalization" standardization method is an existing technology and will not be further explained here.

[0088] S3.3. Use a two-stage data partitioning strategy to partition the dataset;

[0089] The dataset processed in step S3.2 is divided into a "first training set and a test set" in proportion. The first training set is further divided into a "second training set and a validation set" in proportion. This is mainly used to avoid overfitting when subsequently establishing a neural network model and to improve the generalization ability of the model. This method is prior art and will not be further described here.

[0090] It should be noted that the division ratios of the "first training set and test set" and the "second training set and validation set" can be flexibly selected based on factors such as the size and characteristics of the dataset itself and the specific research purpose, in order to better adapt to different project scenarios, while meeting the requirements of training, verification, and testing of the subsequently established neural network model. No limitation is imposed here.

[0091] Preferably, in this embodiment, the division ratio of "the first training set and the test set" is 80%:20%, and the division ratio of "the second training set and the validation set" is 85%:15%.

[0092] Furthermore, step S4 includes the following steps:

[0093] S4.1. Construct a neural network model;

[0094] S4.2. Use algorithms to iteratively update the neural network model;

[0095] S4.3. Selecting the optimal neural network architecture.

[0096] Specifically:

[0097] S4.1. Construct a neural network model;

[0098] A neural network model is constructed based on the "structure-electric field" response mapping data set constructed in step S3, and the neural network model is a single hidden layer feedforward neural network model, consisting of an input layer, a hidden layer, and an output layer; wherein the hidden layer uses a hyperbolic tangent function (tansig) and the output layer uses a linear function (purelin);

[0099] S4.2. Use algorithms to iteratively update the neural network model;

[0100] Specifically, the Levenberg-Marquardt (LM) algorithm is used to iteratively update the weights and biases of the single hidden layer feedforward neural network model constructed in step S4.1;

[0101] Among them, weights are coefficients connecting neurons in the input layer, hidden layer, and output layer. These coefficients determine the degree of influence of each input feature on the output. Bias is another trainable parameter of each neuron in the neural network. It is a constant value added to the weighted sum of neurons. Weights and biases together determine the predictive ability of the neural network. By continuously adjusting weights and biases, the network can learn the complex mapping relationship between input features and outputs. The Levenberg-Marquardt (LM) algorithm is an existing technology and will not be further explained here.

[0102] The specific training results of the network are as follows Figure 3 As shown, in Figure 3 middle:

[0103] (a) is a graph showing the mean square error (MSE) over the training cycle. This graph shows how the MSE of the training set, validation set, and test set change with the training cycle (number of iterations).

[0104] In the figure, the vertical axis Mean Square Error (MSE) is the mean square error, the horizontal axis Training Epochs is the training cycle, Training is the training set, Validation is the validation set, and Test is the test set;

[0105] (b) is a scatter plot comparing the predicted values ​​of the training set and the actual values, which shows the relationship between the predicted values ​​and the actual values ​​of the training set;

[0106] In the figure, the vertical axis (PredictedValues) is the predicted value, which represents the predicted output of the neural network model for the training set data, and the horizontal axis (ActualValues) is the actual value, which represents the actual output value of the training set data; R is the correlation coefficient, and MSE is the mean square error;

[0107] (c) is a scatter plot comparing the predicted values ​​and actual values ​​of the test set, which shows the relationship between the predicted values ​​and actual values ​​of the test set;

[0108] In the figure, the vertical axis (Predicted Values) is the predicted value, which represents the predicted output of the neural network model for the test set data, and the horizontal axis (Actual Values) is the actual value, which represents the actual output value of the test set data; (TrainPredictions) is the predicted value of the training set;

[0109] (d) is a histogram of the prediction error distribution, which shows the distribution of the prediction error of the model of this technical solution.

[0110] In the figure, the vertical axis (Frequency) is the frequency, which indicates the number of times the prediction error occurs, and the horizontal axis (PredictionError) is the prediction error, which indicates the relative error between the model prediction value and the actual value (expressed as a percentage); the closer the prediction error is to zero, the more accurate the model's prediction is.

[0111] Throughout the training process, the single hidden layer feedforward neural network model constructed by this technical solution showed rapid convergence and excellent prediction performance. The mean square error (MSE) dropped sharply in the initial period and converged within only two iterations. The MSE curves of the training set, validation set, and test set overlapped closely, indicating a highly stable and consistent learning process in all data partitions. The predicted values ​​of the training and test data sets were compared with the distortion rates calculated by finite element methods. In both cases, the data points were close to the diagonal, indicating that the model had high prediction fidelity and no overfitting. Error analysis showed that the prediction error followed an approximate normal distribution centered on zero, and most errors were limited to within ±2%, indicating that the single hidden layer feedforward neural network model constructed by this technical solution had excellent prediction performance.

[0112] S4.3. Systematically evaluate the performance indicators of network structures with different numbers of hidden layer neurons and select the optimal network architecture.

[0113] Adjust the number of hidden layer neurons in the single hidden layer feedforward neural network model after iterative update in step S4.2, and compare the comprehensive performance of the adjusted single hidden layer feedforward neural network model using corresponding performance indicators to ultimately select the optimal network architecture;

[0114] It should be noted that the number of neurons in the hidden layer has a significant impact on the performance of the neural network. It is directly related to the model's ability to fit the data, generalization ability, and computational efficiency. The number of neurons in the hidden layer is determined by the following factors:

[0115] 1. Adjust according to data complexity and sample size: If the data is complex and the sample size is large, the number of neurons can be appropriately increased to improve the model fitting ability;

[0116] 2. Comprehensively consider fitting ability and generalization ability: Through cross-validation and validation set evaluation, find the number of neurons that balances training error and generalization error;

[0117] 3. Combine computing resources and efficiency constraints: When computing resources are limited, reduce the number of neurons to increase computing speed.

[0118] Therefore, the adjustment range, optimal number of neurons in the hidden layer, and performance index reflection parameters are not limited here, and can be adjusted according to actual needs;

[0119] Preferably, in this embodiment, the adjustment range of the number of neurons in the hidden layer is 3 to 8, the optimal number of neurons in the hidden layer is 6, and the correlation coefficient (R) and the mean square error (MSE) are selected as parameters reflecting the performance indicators;

[0120] Table 1. Training performance of neural network models with different numbers of hidden layer neurons

[0121] Number of neurons Correlation coefficient (R) Mean Squared Error (MSE) 3 0.9913 0.0026 4 0.9983 0.0005 5 0.9971 0.0008 6 0.9992 0.0002 7 0.9989 0.0003 8 0.9989 0.0003

[0122] The comparison results are shown in Table 1. It can be seen from Table 1 that the number of hidden layer neurons in the range of 3 to 8 has good training performance. When the number of hidden layer neurons is 6, the configuration prediction performance is the best, with a correlation coefficient R of 0.9992 and a mean square error (MSE) of only 0.0002.

[0123] Furthermore, the neural network model is a single hidden layer feedforward neural network model, and the single hidden layer feedforward neural network model includes an input layer, an output layer and a hidden layer, the output layer uses a linear function, and the hidden layer uses a hyperbolic tangent function.

[0124] Furthermore, the verification method of step S5 is to perform finite element simulation based on the real geometric shape and parameterized geometric shape of the defect contour respectively, and compare the simulation results obtained with the prediction results of the single hidden layer feedforward neural network model to verify the accuracy of the single hidden layer feedforward neural network model.

[0125] Specifically, real geometry finite element analysis (FEA) simulation refers to FEA simulation based on the real geometry of the target cable. This simulation restores the actual size and shape of the cable as accurately as possible.

[0126] Finite element analysis (FEA) simulation with parametric geometry refers to finite element analysis (FEA) simulation based on a simplified, parametric geometric model. This simulation constructs the geometric model of the cable by defining geometric parameters such as normalized width (W), aspect ratio (AR) and sharpness (S), rather than relying entirely on the actual geometric shape.

[0127] The maximum electric field distortion values ​​obtained from the above two simulations are compared with the maximum electric field distortion values ​​predicted by the single hidden layer feedforward neural network model constructed by the present technical solution to verify the accuracy of the single hidden layer feedforward neural network model constructed by the present technical solution.

[0128] Table 2 Comparison of maximum electric field distortion values ​​obtained by different methods

[0129]

[0130] The comparison results are shown in Table 2 and Figure 4 As shown, in Figure 4 In the figure, a, b, c, d, and e are FEA simulation results based on real geometric shapes, corresponding to samples 1, 2, 3, 4, and 5 respectively; f, g, h, i, and j are FEA simulation results based on parameterized geometric shapes, corresponding to samples 1, 2, 3, 4, and 5 respectively. The maximum electric field distortion values ​​predicted by the three methods are very consistent in all samples, and the deviation remains within an acceptable range, verifying the accuracy of the single hidden layer feedforward neural network model constructed by this technical solution in predicting the maximum electric field distortion value.

[0131] The above description does not limit the present invention in any form. Although the present invention has been disclosed as above through the embodiments, it is not intended to limit the present invention. Any technician familiar with the profession can make some changes or modifications to equivalent embodiments of equivalent changes using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for evaluating surface defects of a high-voltage cable semi-conductive shielding layer, characterized in that: The following steps are involved: S1. Obtaining the surface defect morphology characteristics of the shielding layer; S2. Construct a finite element model based on the defect morphology and perform simulation calculations; S3. Construct a standardized response mapping dataset based on simulation calculation results; S4. Build a neural network model based on the standardized response mapping dataset and perform iterative training; S5. Perform finite element simulation based on the geometric shape of the defect contour to verify the accuracy of the neural network model.

2. The method according to claim 1, characterized in that The step S1 comprises the following steps: S1.1 simulate the cable extrusion process to prepare shielding tape; S1.

2. Detect defective particles in the shielding layer strip and produce a surface defect image; S1.

3. Use image processing algorithms to extract defect contours and define three dimensionless geometric descriptors.

3. The method according to claim 2, characterized in that The three dimensionless geometric descriptors are normalized width W, aspect ratio AR, and sharpness S; wherein the normalized width W is the defect base width, the aspect ratio AR is the ratio of the defect height to the width, and the sharpness S is the ratio of the full width at half maximum (FWHM) of the defect top to the base width.

4. The method according to claim 1, wherein The step S2 comprises the following steps: S2.

1. Measure the resistance-temperature characteristics of the shielding layer and the resistance-temperature-field strength characteristics of the insulating layer used in the shielding layer; S2.

2. Construct a finite element simulation model of the target cable and obtain the maximum electric field intensity and distortion rate in the defect area by solving the electric field distribution. S2.

3. Simulate and calculate the defect shape-maximum electric field distortion under different defect shapes and summarize the data samples.

5. The method according to claim 1, wherein The step S3 comprises the following steps: S3.

1. Using the defined dimensionless geometric descriptor as input, construct a structure-electric field response mapping dataset. S3.

2. Standardize the dataset using the mean-standard deviation normalization method. S3.

3. Use a two-stage data partitioning strategy to partition the dataset.

6. The method according to claim 1, characterized in that The step S4 comprises the following steps: S4.

1. Construct a neural network model; S4.

2. Use algorithms to iteratively update the neural network model; S4.

3. Selecting the optimal neural network architecture.

7. The method according to claim 6, characterized in that The neural network model is a single hidden layer feedforward neural network model, and the single hidden layer feedforward neural network model includes an input layer, an output layer and a hidden layer, the output layer uses a linear function, and the hidden layer uses a hyperbolic tangent function.

8. The method according to claim 7, characterized in that The verification method of step S5 is to perform finite element simulation based on the real geometric shape and parameterized geometric shape of the defect contour, respectively, and compare the simulation results obtained with the prediction results of the single hidden layer feedforward neural network model to verify the accuracy of the single hidden layer feedforward neural network model.

Images