Electrical equipment defect detection method and system

Abstract

The invention discloses a power equipment defect detection method and system, and belongs to the technical field of image detection, and the method comprises the steps: carrying out the multi-angle image collection of the surface of a transformer through a camera device, and carrying out the image preprocessing through a self-adaptive illumination compensation algorithm; a Faster R-CNN model is utilized to carry out defect area detection on the preprocessed image, and positioning of a defect area is optimized through a non-maximum suppression algorithm; based on the optimized defect area, defect features are extracted by using a multi-modal feature extraction network, feature vectors are generated, and defect classification is carried out; performing correlation analysis on the historical data and the defect classification result through a graph neural network, judging a defect reason and generating a defect detection report, collecting a field maintenance result and updating a defect sample library; and automatic detection and classification of the defects of the power equipment are realized.

Description

Technical Field

[0001] The invention belongs to the technical field of image detection, and in particular to a method and system for detecting defects in electric power equipment. Background Art

[0002] During long-term operation, power equipment may experience performance degradation or failure due to natural aging or external factors such as overload and environmental corrosion. Traditional detection methods mainly rely on manual inspections or simple sensor monitoring, which have problems of low efficiency and high misjudgment rate. In recent years, detection technology based on image recognition has gradually been applied to power equipment status monitoring, but existing technologies are difficult to effectively distinguish between normal aging and fault characteristics, resulting in a high false alarm rate, affecting the reliability of detection results.

[0003] For example, a Chinese patent with authorization announcement number CN115359054B discloses a method for detecting defects in power equipment based on pseudo-defect space generation, including: artificially forging local irregularities with normal images as materials to generate a pseudo-defect image category, and using contrastive learning to decouple feature space regions on the basis of feature extraction and classification networks for normal images, defect images, and pseudo-defect images, embedding constraints on pseudo-defect image features, normal image features, and defect image features, thereby enhancing the anti-interference ability of complex backgrounds in difficult-to-distinguish samples and improving the model's recognition accuracy for hidden defects in power equipment; using a power equipment defect model to predict classification losses and obtain edge pseudo-defect samples based on the classification losses, and after several iterative trainings, using edge pseudo-defect samples to introduce a weakly supervised defect covering strategy for training, thereby alleviating overfitting, helping the model break through the training bottleneck, obtaining optimal model parameters and fixing them, and using the power equipment defect detection model to detect the image to be detected.

[0004] The above existing technologies have the following problems: the quality of the pseudo-defect image directly affects the performance of the model. If the generated pseudo-defect is very different from the real defect, the model may learn wrong features and reduce the detection accuracy. Although the recognition ability of hidden defects is enhanced through contrast learning and pseudo-defect generation, its effect is highly dependent on the design of pseudo-defects and the optimization of contrast learning. There is a lack of consideration of the operating status of the equipment. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention proposes a method and system for detecting defects in power equipment. The method collects multi-angle images of the transformer surface through a camera device, and performs image preprocessing in combination with an adaptive illumination compensation algorithm. The Faster R-CNN model is used to detect defect areas in the preprocessed images, and the positioning of the defect areas is optimized through a non-maximum suppression algorithm. Based on the optimized defect areas, a multimodal feature extraction network is used to extract defect features, generate feature vectors, and classify defects. A graph neural network is used to perform correlation analysis on historical data and defect classification results, determine the causes of defects and generate defect detection reports, collect on-site maintenance results and update the defect sample library. The method and system realize automated detection and classification of defects in power equipment.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for detecting defects in electric power equipment, comprising:

[0008] Step S1: Capture images of the transformer surface using a camera, obtain multi-angle images using an adaptive illumination compensation algorithm, and perform preprocessing;

[0009] Step S2: Use the Faster R-CNN model to detect defect areas on the preprocessed multi-angle images, output the bounding box coordinates and confidence scores of the defect areas, and introduce the non-maximum suppression optimization algorithm to optimize the positioning of the defect areas;

[0010] Step S3: according to the optimized defect area, a multimodal feature extraction network is used to extract features of the defect area from the preprocessed multi-angle image, a feature vector is generated, and based on the feature vector, an improved classification algorithm is used to classify the defects into defects caused by normal aging and defects caused by failures, and a defect classification result is obtained;

[0011] Step S4: Combine the equipment operation history data to build a multi-source data fusion model, and use the graph neural network model to perform correlation analysis on the equipment operation history data and defect classification results to determine whether the defect is caused by normal aging or equipment failure, and output a defect detection report;

[0012] Step S5: Feedback the defect detection report to the equipment operation and maintenance team, collect on-site maintenance results, and build a defect sample library.

[0013] Specifically, the specific steps of acquiring multi-angle images in step S1 include:

[0014] S1.1: Use a camera to shoot the transformer surface from multiple angles to collect transformer surface images;

[0015] S1.2: Perform illumination analysis on the collected transformer surface image to obtain the average brightness and contrast results of the transformer surface image. If the average brightness is less than the preset minimum brightness threshold or greater than the preset maximum brightness threshold, or the contrast is less than the preset minimum contrast threshold, then the transformer surface image is calculated at the coordinate The product of the pixel value at the position and the local illumination enhancement coefficient is calculated, and the transformer surface image at the coordinate The pixel value at point The Gaussian kernel is added to the sum of the illumination smoothing term, and the illumination compensation result is obtained by calculating the ratio of the product and the sum, and then adding the global brightness offset term, and the illumination conditions are adjusted according to the illumination compensation result;

[0016] S1.3: Generate a multi-angle image with uniform illumination according to the illumination compensation result.

[0017] Specifically, the specific steps of step S2 include:

[0018] S2.1: Input the preprocessed multi-angle image into the Faster R-CNN model, introduce a dynamic anchor point generation mechanism, use the feature map to dynamically generate N anchor points on the preprocessed multi-angle image, and output a classification score and bounding box offset information for each anchor point; each anchor point corresponds to a potential candidate region;

[0019] S2.2: Arrange the classification scores in descending order, select the anchor points corresponding to the first M classification scores, generate candidate regions based on the bounding box offset information, and extract features from each candidate region through a convolutional neural network to generate a high-dimensional feature vector;

[0020] S2.3: Take the high-dimensional feature vector of each candidate region as input, use a fully connected layer to map the high-dimensional feature vector to the category space, and apply the Softmax function to the output vector of the fully connected layer to calculate the probability of each category;

[0021] S2.4: According to the output of the Softmax function, the probability of each candidate region belonging to each category is obtained;

[0022] S2.5: Map the high-dimensional feature vector of each candidate region to the bounding box offset space through the regression layer to obtain the offset, and use the offset to adjust the bounding box of the candidate region and output the bounding box coordinates.

[0023] Specifically, the specific steps of step S2 also include:

[0024] S2.6: Use the temperature scaling method to calibrate the category probability of each candidate area to obtain the calibrated category probability, and use the calibrated category probability as the confidence score, combined with the bounding box coordinates, to output the bounding box coordinates and confidence score of the calibrated defect area;

[0025] S2.7: Based on the bounding box coordinates and confidence scores of all detected defect areas, the intersection area and union area of ​​any two bounding boxes are calculated respectively, and the intersection-union ratio of the two bounding boxes is obtained by calculating the ratio of the two. At the same time, all defect areas are sorted from high to low according to the confidence scores to obtain a sorted defect area list;

[0026] S2.8: Select the defect area with the highest confidence, and delete the defect areas with a non-highest confidence whose intersection-over-union ratio with the defect area with the highest confidence is greater than a preset threshold;

[0027] S2.9: Iterate the remaining defective areas according to S2.7-S2.8, and use the screened defective areas as optimized defective areas. At the same time, output their bounding box coordinates and confidence scores.

[0028] Specifically, the specific steps of step S3 include:

[0029] S3.1: Obtain the bounding box coordinates of each optimized defect area, and extract the image block of the defect area from the preprocessed multi-angle image according to the bounding box coordinates;

[0030] S3.2: Use a convolutional neural network to extract image features of the defect area from the image block of the defect area;

[0031] S3.3: According to the optimized bounding box coordinates of the defect area, extract its neighborhood image blocks, and input the neighborhood image blocks into the graph neural network to extract context features;

[0032] S3.4: Concatenate image features and context features to generate a multimodal feature vector.

[0033] Specifically, the specific steps of step S3 also include:

[0034] S3.5: inputting the multimodal feature vector into a multi-scale gradient boosting tree, optimizing the classification result step by step using n decision trees, and outputting the defect category according to the classification result of the decision tree; the multi-scale gradient boosting tree includes n decision trees;

[0035] The process of outputting defect categories is as follows:

[0036] Based on the multimodal feature vector, the prediction result of the t-th tree at the l-th scale on the multimodal feature vector is calculated, and the first dependent variable is obtained by combining the weights of different scales;

[0037] The additional features of different scales are calculated by the scale feature function, and the second dependent variable is obtained by combining the regularization parameters of different scales; the additional features are context features;

[0038] The first dependent variable and the second dependent variable are summed to obtain the local prediction value of the current scale, which is used as the third dependent variable;

[0039] Based on the third dependent variable, the predicted values ​​of all decision trees at the current scale are accumulated and summed to obtain the comprehensive predicted value at the current scale. The weights of different scales are combined and accumulated to obtain the final predicted value after multi-scale fusion, which is used as the fourth dependent variable.

[0040] Mapping the fourth dependent variable to a defect classification result through a symbolic function;

[0041] S3.6: Based on the defect classification results, the defects are classified into defects caused by normal aging and defects caused by failures.

[0042] Specifically, the specific steps of step S4 include:

[0043] S4.1: Obtain equipment operation history data and defect classification results; the defect classification results include defect categories and their confidence scores;

[0044] S4.2: Use statistical methods to extract features from the equipment operation history data, and fuse the features of the equipment operation history data with the defect classification results to generate a multi-source feature vector;

[0045] S4.3: construct a graph structure based on the multi-source feature vectors, where the nodes represent the equipment status or defects, and the edges represent the association between the equipment status and the defects;

[0046] S4.4: Use graph neural networks to model graph structures and learn feature representations of nodes and edges;

[0047] S4.5: Based on the learned feature representations of nodes and edges, determine whether the defect is caused by normal aging or equipment failure;

[0048] S4.6: Generate a defect detection report based on the determination results, including defect location, category, and cause analysis.

[0049] An electric power equipment defect detection system comprises: an image acquisition module, a defect detection and positioning module, a feature extraction and classification module, a defect analysis module, and a feedback and management module;

[0050] The image acquisition module is used to acquire multi-angle images of the transformer surface and perform preprocessing;

[0051] The defect detection and positioning module uses the Faster R-CNN model to detect defect areas in the image and locates the defect positions through an optimization algorithm;

[0052] The feature extraction and classification module is used to extract features from the detected defect area and classify the defects based on the extracted features;

[0053] The defect analysis module is used to analyze the cause of the defect and generate a test report by combining the equipment operation history data and the defect classification results;

[0054] The feedback and management module is used to feed back the test report to the operation and maintenance team, collect on-site maintenance results, and update the defect sample library.

[0055] Specifically, the defect detection and positioning module includes: a defect detection unit and a positioning optimization unit;

[0056] The defect detection unit uses the Faster R-CNN model to detect defect areas on the preprocessed image and outputs bounding box coordinates and confidence scores of the defect areas;

[0057] The positioning optimization unit is used to introduce a non-maximum suppression algorithm to optimize the positioning of the defect area.

[0058] Specifically, the defect analysis module includes: a data fusion unit and a defect analysis unit;

[0059] The data fusion unit is used to construct a multi-source data fusion model to integrate the equipment operation history data and defect classification results; the equipment operation history data includes temperature, load, and operation time;

[0060] The defect analysis unit uses a graph neural network to perform correlation analysis on the fused data, determines whether the defect is caused by normal aging or equipment failure, and generates a defect detection report.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] 1. The present invention proposes a power equipment defect detection system, and optimizes and improves the architecture, operation steps and processes. The system has the advantages of simple process, low investment and operation costs, and low production work costs.

[0063] 2. The present invention proposes a method for defect detection of power equipment. By combining image recognition, multimodal feature extraction, multi-source data fusion and graph neural network technologies, the method realizes the automatic detection and classification of transformer surface defects. First, the image quality is ensured through adaptive illumination compensation and multi-angle image acquisition; the Faster R-CNN model is used to accurately locate the defect area, and the defects are divided into defects caused by normal aging or failure through multimodal feature extraction and improved classification algorithm; secondly, a multi-source data fusion model is constructed in combination with the equipment operation history data, and the graph neural network is used for correlation analysis to further determine the cause of the defect and generate a defect detection report; finally, a closed-loop optimization mechanism is formed by feedback of on-site maintenance results and updating of the defect sample library to continuously improve the detection accuracy and system performance.

[0064] 3. The present invention proposes a method for detecting defects in power equipment, which has the characteristics of high precision and high efficiency, can effectively distinguish defects caused by normal aging and faults, and reduce false detections and missed detections; at the same time, through a closed-loop feedback mechanism, the system can continuously optimize the model and adapt to complex and changeable actual scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 A schematic diagram of a method for detecting defects in electric power equipment according to the present invention;

[0066] Figure 2 This is a principle flow chart of a method for detecting defects in electric power equipment according to the present invention;

[0067] Figure 3 This is an architecture diagram of a power equipment defect detection system according to the present invention. DETAILED DESCRIPTION

[0068] Example 1

[0069] See also Figure 1 and Figure 2 , an embodiment of the present invention provides: a method for detecting defects in electric power equipment, comprising the following steps:

[0070] Step S1: Capture images of the transformer surface using a camera, obtain multi-angle images using an adaptive illumination compensation algorithm, and perform preprocessing;

[0071] Step S2: Use the Faster R-CNN model to detect defect areas on the preprocessed multi-angle images, output the bounding box coordinates and confidence scores of the defect areas, and introduce the non-maximum suppression optimization algorithm to optimize the positioning of the defect areas;

[0072] Step S3: according to the optimized defect area, a multimodal feature extraction network is used to extract features of the defect area from the preprocessed multi-angle image, a feature vector is generated, and based on the feature vector, an improved classification algorithm is used to classify the defects into defects caused by normal aging and defects caused by failures, and a defect classification result is obtained;

[0073] Step S4: Combine the equipment operation history data to build a multi-source data fusion model, and use the graph neural network model to perform correlation analysis on the equipment operation history data and defect classification results to determine whether the defect is caused by normal aging or equipment failure, and output a defect detection report;

[0074] Step S5: Feedback the defect detection report to the equipment operation and maintenance team, collect on-site maintenance results, and build a defect sample library.

[0075] The specific steps of acquiring multi-angle images in step S1 include:

[0076] S1.1: Use a camera to shoot the transformer surface from multiple angles to collect transformer surface images;

[0077] S1.2: Perform illumination analysis on the collected transformer surface image to obtain the average brightness of the transformer surface image and contrast As a result, if the average brightness is less than the preset minimum brightness threshold or greater than the preset maximum brightness threshold, or the contrast is less than the preset minimum contrast threshold, then the transformer surface image is calculated at the coordinate The pixel value at and the local illumination enhancement coefficient The product of and calculates the transformer surface image in coordinates The pixel value at that point is multiplied by the Gaussian kernel at that point Plus the lighting smoothing term The sum of the product and the sum is calculated by adding the global brightness offset term , thus obtaining the illumination compensation result , and adjust the lighting conditions according to the lighting compensation results, where: Represents the transformer surface image in coordinates The pixel value at , W and H are the width and height of the transformer surface image respectively, x≤W, y≤H, E represents the total number of pixels of the transformer surface image, and Respectively represent the maximum and minimum pixel values ​​of the transformer surface image, represent the local illumination enhancement coefficient, and , Represents the transformer surface image in coordinates The gradient value at represents the adjustment coefficient, Represents the average light intensity in a local area. represents the smoothing coefficient, represents the global average brightness, represents the target brightness, represents the offset coefficient;

[0078] S1.3: Generate a multi-angle image with uniform illumination according to the illumination compensation result.

[0079] The specific steps of step S2 include:

[0080] S2.1: The preprocessed multi-angle image is input into the Faster R-CNN model, and a dynamic anchor point generation mechanism is introduced. The feature map is used to dynamically generate N anchor points on the preprocessed multi-angle image. The dynamic anchor point generation formula is: ,and , and output the classification score for each anchor point and bounding box offset Information, the formula is: ; Each anchor point corresponds to a potential candidate region, and the classification score indicates whether the anchor point contains the target; the bounding box offset information indicates the offset between the anchor point and the true bounding box, wherein, represents the i-th anchor point, Represents the center coordinates of the anchor point, and Respectively represent the width and height of the anchor point, Indicates the scale of dynamic adjustment, represents the scale factor that is dynamically adjusted to accommodate defects of different sizes and shapes, represents the activation function, and Represent the weight and bias of the classification layer respectively, Indicates that the feature map is The eigenvector at and Represent the weight and bias of the regression layer respectively;

[0081] S2.2: Arrange the classification scores in descending order and select the anchor points corresponding to the first M classification scores. The screening formula is: , combine the bounding box offset information to generate candidate regions ,and , And extract features from each candidate region through a convolutional neural network to generate a high-dimensional feature vector, where Anchor j represents the jth anchor point, represents the center coordinate of the jth anchor point, and Respectively represent the width and height of the j-th anchor point, represents the candidate region corresponding to the jth anchor point generated, represents the bounding box offset of the jth anchor point, , , and They represent the horizontal offset of the center coordinate of the anchor point, the vertical offset of the center coordinate, the width offset, and the height offset. Threshold represents the preset score threshold. represents the exponential function;

[0082] Furthermore, the specific steps of extracting features from each candidate region through a convolutional neural network and generating a high-dimensional feature vector include:

[0083] (1) For each candidate region, RoI Pooling is used to extract fixed-size features from the feature map, where RoI Pooling means region of interest pooling. RoI Pooling is a prior art in the art and is not an inventive solution of the present application, and is not described in detail herein;

[0084] (2) Divide the candidate region into 7*7 grids and perform maximum pooling on the features in each grid;

[0085] (3) Output a feature map of size 7*7*A, where A represents the number of feature channels;

[0086] (4) Flatten the feature map of size 7*7*A into a one-dimensional vector as the feature representation of the candidate region.

[0087] S2.3: Take the high-dimensional feature vector of each candidate region, such as 7*7*A, as input and use the fully connected layer to map the high-dimensional feature vector to the category space. The formula of the fully connected layer is: , and apply the Softmax function to the output vector of the fully connected layer to calculate the probability of each category , where Z represents the output vector, is the score of each category, represents the weight matrix of the fully connected layer, X represents the high-dimensional feature vector of the input, and B represents the bias vector of the fully connected layer. represents the probability that the candidate region belongs to category d, D represents the total number of categories, represents the score of category d, represents the score of category k, and ;

[0088] S2.4: According to the output of the Softmax function, the probability of each candidate region belonging to each category is obtained;

[0089] S2.5: Map the high-dimensional feature vector of each candidate region to the bounding box offset space through the regression layer to obtain the offset , and use the offset to adjust the bounding box coordinates of the candidate region Adjust to output precise bounding box coordinates , the adjustment formula is: ;

[0090] S2.6: Use Calibrate the category probability of each candidate area to obtain the calibrated category probability , and use the calibrated category probability as the confidence score, combined with the precise bounding box coordinates, to output the calibrated bounding box coordinates and confidence score of the defect area, where represents the adaptive weight of category d, represents the characteristic function of category d, represents the adaptive temperature parameter;

[0091] S2.7: Based on the bounding box coordinates of all detected defect areas and confidence score , calculate the intersection area of ​​any two bounding boxes respectively and the area of ​​the union , and by calculating the ratio of the two, we get the intersection and union ratio of the two bounding boxes. At the same time, all defect areas are sorted from high to low according to the confidence scores to obtain a sorted defect area list, where represents the coordinates of the upper left corner of the bounding box U, represents the coordinates of the lower right corner of the bounding box U, represents the coordinates of the upper left corner of the bounding box V, represents the coordinates of the lower right corner of the bounding box V, represents the area of ​​the bounding box U, represents the area of ​​the bounding box V, and denote the maximum and minimum functions respectively;

[0092] S2.8: Select the defect area with the highest confidence, and delete the defect areas with a non-highest confidence whose intersection-over-union ratio with the defect area with the highest confidence is greater than a preset threshold;

[0093] S2.9: Iterate the remaining defective areas according to S2.7-S2.8, and use the screened defective areas as optimized defective areas. At the same time, output their bounding box coordinates and confidence scores.

[0094] The specific steps of step S3 include:

[0095] S3.1: Obtain the bounding box coordinates of each optimized defect area , according to the bounding box coordinates, extract the image block of the defect area from the preprocessed multi-angle image ;

[0096] S3.2: Use a convolutional neural network to extract image features of the defect area from the image patch of the defect area ,in, represents a convolutional neural network function, and the convolutional neural network is the prior art content in this field, and is not an inventive solution of the present application, and will not be described in detail here;

[0097] S3.3: Extract the neighborhood image blocks based on the optimized bounding box coordinates of the defect area , and input the neighborhood image blocks into the graph neural network to extract context features ,in, Represents a graph neural network function, and the graph neural network is the prior art content in this field, and is not an inventive solution of the present application, and will not be described in detail here;

[0098] S3.4: concatenate image features and context features to generate a multimodal feature vector;

[0099] S3.5: inputting the multimodal feature vector into a multi-scale gradient boosting tree, optimizing the classification result step by step using n decision trees, and outputting the defect category according to the classification result of the decision tree; the multi-scale gradient boosting tree includes n decision trees;

[0100] The process of outputting defect categories is as follows:

[0101] Based on the multimodal feature vector, the prediction result of the t-th tree at the l-th scale on the multimodal feature vector is calculated, and the first dependent variable is obtained by combining the weights of different scales;

[0102] The additional features of different scales are calculated by the scale feature function, and the second dependent variable is obtained by combining the regularization parameters of different scales; the additional features are context features;

[0103] The first dependent variable and the second dependent variable are summed to obtain the local prediction value of the current scale, which is used as the third dependent variable;

[0104] Based on the third dependent variable, the predicted values ​​of all decision trees at the current scale are accumulated and summed to obtain the comprehensive predicted value at the current scale. The weights of different scales are combined and accumulated to obtain the final predicted value after multi-scale fusion, which is used as the fourth dependent variable.

[0105] Mapping the fourth dependent variable to a defect classification result through a symbolic function;

[0106] The formula is: , where Y represents the defect classification result, represents the multimodal feature vector, represents the weight of the lth scale, represents the weight of the t-th decision tree at the l-th scale, represents the prediction result of the t-th decision tree at the l-th scale, represents the regularization coefficient of the lth scale, represents the additional feature function of the lth scale, L represents the number of scales, n represents the number of decision trees, represents a symbolic function;

[0107] S3.6: According to the defect classification result, the defects are divided into defects caused by normal aging and defects caused by faults, wherein the defect classification result is -1 or 1, wherein 1 indicates defects caused by normal aging and -1 indicates defects caused by faults.

[0108] The specific steps of step S4 include:

[0109] S4.1: Obtain equipment operation history data and defect classification results; the defect classification results include defect categories and their confidence scores;

[0110] S4.2: extract features from the equipment operation history data using a statistical method, and fuse the features of the equipment operation history data with the defect classification results to generate a multi-source feature vector, wherein the statistical method is a prior art content in this field, is not an inventive solution of the present application, and is not described in detail here;

[0111] S4.3: construct a graph structure based on the multi-source feature vectors, where the nodes represent the equipment status or defects, and the edges represent the association between the equipment status and the defects;

[0112] Furthermore, the specific steps of constructing the graph structure include:

[0113] (1) Define nodes, where each node represents a device state or defect;

[0114] For the equipment status node, the characteristics are the characteristics of the equipment operation history data;

[0115] For defective nodes, the features are those of the defect classification results;

[0116] (2) defining edges, where an edge represents the association between the equipment state and the defect, and the weight of the edge is obtained based on the similarity between the equipment state and the defect;

[0117] (3) Taking equipment status and defects as nodes, construct a node set;

[0118] (4) Construct edge sets based on the associations between nodes;

[0119] (5) Obtain the edge weight by calculating the cosine similarity between the feature vectors of different nodes;

[0120] (6) Use the adjacency matrix to represent the graph structure and the node feature matrix to represent the node features;

[0121] The adjacency matrix includes: if the edge connecting any two nodes is in the edge set, the adjacency matrix is ​​equal to the weight of the edge of the two nodes, otherwise, the adjacency matrix is ​​equal to 0;

[0122] The node feature matrix is ​​composed of feature vectors of different nodes.

[0123] S4.4: Use graph neural networks to model graph structures and learn feature representations of nodes and edges;

[0124] Furthermore, the specific steps of S4.4 include:

[0125] (1) Input the adjacency matrix and node feature matrix into the graph neural network model;

[0126] (2) Using a graph neural network model to learn the feature representation of nodes and edges, wherein the graph neural network model is the prior art content in this field and is not an inventive solution of the present application, and is not described in detail here.

[0127] S4.5: Based on the learned feature representations of nodes and edges, determine whether the defect is caused by normal aging or equipment failure;

[0128] S4.6: Generate a defect detection report based on the determination results, including defect location, category, and cause analysis.

[0129] Example 2

[0130] See also Figure 3 Another embodiment provided by the present invention is a power equipment defect detection system, comprising:

[0131] Image acquisition module, defect detection and positioning module, feature extraction and classification module, defect analysis module, feedback and management module;

[0132] Image acquisition module, used to collect multi-angle images of the transformer surface and improve image quality through preprocessing to provide high-quality input data for subsequent defect detection;

[0133] The defect detection and positioning module uses the Faster R-CNN model to detect defect areas in the image and accurately locates the defect position through an optimization algorithm;

[0134] A feature extraction and classification module is used to extract features from the detected defect area and classify the defects based on the features;

[0135] Defect analysis module, which is used to analyze the causes of defects and generate inspection reports by combining equipment operation history data and defect classification results;

[0136] The feedback and management module is used to feed back the inspection report to the operation and maintenance team, collect on-site maintenance results, and update the defect sample library.

[0137] The image acquisition module includes: an image acquisition unit, an illumination compensation unit, and a preprocessing unit;

[0138] An image acquisition unit, which uses a camera to acquire multi-angle images of the transformer surface;

[0139] An illumination compensation unit is used to adjust the image illumination conditions in combination with an adaptive illumination compensation algorithm to ensure image quality consistency;

[0140] The preprocessing unit is used to perform denoising, enhancement and normalization processing on the collected image to generate a preprocessed image.

[0141] The defect detection and positioning module includes: a defect detection unit and a positioning optimization unit;

[0142] The defect detection unit uses the Faster R-CNN model to detect defect areas in the preprocessed images and outputs the bounding box coordinates and confidence scores of the defect areas;

[0143] The positioning optimization unit is used to introduce the non-maximum suppression algorithm, optimize the positioning of the defect area, remove redundant bounding boxes, and improve the detection accuracy.

[0144] The feature extraction and classification module includes: a feature extraction unit and a defect classification unit;

[0145] A feature extraction unit uses a multimodal feature extraction network to extract features of defect areas from preprocessed images and generate a high-dimensional feature vector;

[0146] The defect classification unit, based on the feature vector, uses an improved classification algorithm to classify defects into defects caused by normal aging and defects caused by failures.

[0147] The defect analysis module includes: a data fusion unit and a defect analysis unit;

[0148] A data fusion unit is used to construct a multi-source data fusion model to integrate equipment operation history data and defect classification results; the equipment operation history data includes temperature, load, and operation time;

[0149] The defect analysis unit uses graph neural networks to perform correlation analysis on the fused data, determines whether the defect is caused by normal aging or equipment failure, and generates a defect detection report.

[0150] The feedback and management module includes: report feedback unit, sample collection unit, and sample library update unit;

[0151] Report feedback unit, used to send defect detection reports to the equipment operation and maintenance team to guide on-site maintenance work;

[0152] The sample collection unit is used to collect on-site maintenance results, including maintenance records and images after repair, and to build a defect sample library;

[0153] The sample library updating unit is used to add the newly discovered defect samples, including images, feature vectors, classification results and maintenance results, to the defect sample library.

[0154] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in the field may also change, modify, replace and modify the above-mentioned embodiments without departing from the purpose and scope of protection of the present invention, and all of these are within the protection of the present invention.

Claims

1. A method for detecting defects in electric power equipment, characterized in that: include: Step S1: Capture images of the transformer surface using a camera, obtain multi-angle images using an adaptive illumination compensation algorithm, and perform preprocessing; Step S2: Use the Faster R-CNN model to detect defect areas on the preprocessed multi-angle images, output the bounding box coordinates and confidence scores of the defect areas, and introduce the non-maximum suppression optimization algorithm to optimize the positioning of the defect areas; Step S3: according to the optimized defect area, a multimodal feature extraction network is used to extract features of the defect area from the preprocessed multi-angle image, a feature vector is generated, and based on the feature vector, an improved classification algorithm is used to classify the defects into defects caused by normal aging and defects caused by failures, and a defect classification result is obtained; Step S4: Combine the equipment operation history data to build a multi-source data fusion model, and use the graph neural network model to perform correlation analysis on the equipment operation history data and defect classification results to determine whether the defect is caused by normal aging or equipment failure, and output a defect detection report; Step S5: Feedback the defect detection report to the equipment operation and maintenance team, collect on-site maintenance results, and build a defect sample library.

2. A method for detecting defects in electric power equipment according to claim 1, characterized in that: The specific steps of acquiring multi-angle images in step S1 include: S1.1: Use a camera to shoot the transformer surface from multiple angles to collect transformer surface images; S1.2: Perform illumination analysis on the collected transformer surface image to obtain the average brightness and contrast results of the transformer surface image. If the average brightness is less than the preset minimum brightness threshold or greater than the preset maximum brightness threshold, or the contrast is less than the preset minimum contrast threshold, then the transformer surface image is calculated at the coordinate The product of the pixel value at the position and the local illumination enhancement coefficient is calculated, and the transformer surface image at the coordinate The pixel value at point The Gaussian kernel is added to the sum of the illumination smoothing term, and the illumination compensation result is obtained by calculating the ratio of the product and the sum, and then adding the global brightness offset term, and the illumination conditions are adjusted according to the illumination compensation result; S1.3: Generate a multi-angle image with uniform illumination according to the illumination compensation result.

3. A method for detecting defects in electric power equipment according to claim 2, characterized in that: The specific steps of step S2 include: S2.1: Input the preprocessed multi-angle image into the Faster R-CNN model, introduce a dynamic anchor point generation mechanism, use the feature map to dynamically generate N anchor points on the preprocessed multi-angle image, and output a classification score and bounding box offset information for each anchor point; each anchor point corresponds to a potential candidate region; S2.2: Arrange the classification scores in descending order, select the anchor points corresponding to the first M classification scores, generate candidate regions based on the bounding box offset information, and extract features from each candidate region through a convolutional neural network to generate a high-dimensional feature vector; S2.3: Take the high-dimensional feature vector of each candidate region as input, use a fully connected layer to map the high-dimensional feature vector to the category space, and apply the Softmax function to the output vector of the fully connected layer to calculate the probability of each category; S2.4: According to the output of the Softmax function, the probability of each candidate region belonging to each category is obtained; S2.5: Map the high-dimensional feature vector of each candidate region to the bounding box offset space through the regression layer to obtain the offset, and use the offset to adjust the bounding box of the candidate region and output the bounding box coordinates.

4. A method for detecting defects in electric power equipment according to claim 3, characterized in that: The specific steps of step S2 also include: S2.6: Use the temperature scaling method to calibrate the category probability of each candidate area to obtain the calibrated category probability, and use the calibrated category probability as the confidence score, combined with the bounding box coordinates, to output the bounding box coordinates and confidence score of the calibrated defect area; S2.7: Based on the bounding box coordinates and confidence scores of all detected defect areas, the intersection area and union area of ​​any two bounding boxes are calculated respectively, and the intersection-union ratio of the two bounding boxes is obtained by calculating the ratio of the two. At the same time, all defect areas are sorted from high to low according to the confidence scores to obtain a sorted defect area list; S2.8: Select the defect area with the highest confidence, and delete the defect areas with a non-highest confidence whose intersection-over-union ratio with the defect area with the highest confidence is greater than a preset threshold; S2.9: Iterate the remaining defective areas according to S2.7-S2.8, and use the screened defective areas as optimized defective areas. At the same time, output their bounding box coordinates and confidence scores.

5. A method for detecting defects in electric power equipment according to claim 4, characterized in that: The specific steps of step S3 include: S3.1: Obtain the bounding box coordinates of each optimized defect area, and extract the image block of the defect area from the preprocessed multi-angle image according to the bounding box coordinates; S3.2: Use a convolutional neural network to extract image features of the defect area from the image block of the defect area; S3.3: According to the optimized bounding box coordinates of the defect area, extract its neighborhood image blocks, and input the neighborhood image blocks into the graph neural network to extract context features; S3.4: Concatenate image features and context features to generate a multimodal feature vector.

6. A method for detecting defects in electric power equipment according to claim 5, characterized in that: The specific steps of step S3 also include: S3.5: inputting the multimodal feature vector into a multiscale gradient boosting tree, optimizing the classification result step by step using n decision trees, and outputting the defect category according to the classification result of the decision tree; the multiscale gradient boosting tree includes n decision trees; The process of outputting defect categories is as follows: Based on the multimodal feature vector, the prediction result of the t-th tree at the l-th scale on the multimodal feature vector is calculated, and the first dependent variable is obtained by combining the weights of different scales; The additional features of different scales are calculated by the scale feature function, and the second dependent variable is obtained by combining the regularization parameters of different scales; the additional features are context features; The first dependent variable and the second dependent variable are summed to obtain the local prediction value of the current scale, which is used as the third dependent variable; Based on the third dependent variable, the predicted values ​​of all decision trees at the current scale are accumulated and summed to obtain the comprehensive predicted value at the current scale. The weights of different scales are combined and accumulated to obtain the final predicted value after multi-scale fusion, which is used as the fourth dependent variable. Mapping the fourth dependent variable to a defect classification result through a symbolic function; S3.6: Based on the defect classification results, the defects are classified into defects caused by normal aging and defects caused by failures.

7. A method for detecting defects in electric power equipment according to claim 6, characterized in that: The specific steps of step S4 include: S4.1: Obtain equipment operation history data and defect classification results; the defect classification results include defect categories and their confidence scores; S4.2: Use statistical methods to extract features from the equipment operation history data, and fuse the features of the equipment operation history data with the defect classification results to generate a multi-source feature vector; S4.3: construct a graph structure based on the multi-source feature vectors, where the nodes represent the equipment status or defects, and the edges represent the association between the equipment status and the defects; S4.4: Use graph neural networks to model graph structures and learn feature representations of nodes and edges; S4.5: Based on the learned feature representations of nodes and edges, determine whether the defect is caused by normal aging or equipment failure; S4.6: Generate a defect detection report based on the determination results, including defect location, category, and cause analysis.

8. A power equipment defect detection system, used to implement a power equipment defect detection method according to any one of claims 1 to 7, characterized in that: include: Image acquisition module, defect detection and positioning module, feature extraction and classification module, defect analysis module, feedback and management module; The image acquisition module is used to acquire multi-angle images of the transformer surface and perform preprocessing; The defect detection and positioning module uses the Faster R-CNN model to detect defect areas in the image and locates the defect positions through an optimization algorithm; The feature extraction and classification module is used to extract features from the detected defect area and classify the defects based on the extracted features; The defect analysis module is used to analyze the cause of the defect and generate a test report by combining the equipment operation history data and the defect classification results; The feedback and management module is used to feed back the test report to the operation and maintenance team, collect on-site maintenance results, and update the defect sample library.

9. A power equipment defect detection system as claimed in claim 8, characterized in that: The defect detection and positioning module includes: a defect detection unit and a positioning optimization unit; The defect detection unit uses the Faster R-CNN model to detect defect areas on the preprocessed image and outputs bounding box coordinates and confidence scores of the defect areas; The positioning optimization unit is used to introduce a non-maximum suppression algorithm to optimize the positioning of the defect area.

10. The power equipment defect detection system according to claim 9, characterized in that: The defect analysis module includes: a data fusion unit and a defect analysis unit; The data fusion unit is used to construct a multi-source data fusion model to integrate the equipment operation history data and defect classification results; the equipment operation history data includes temperature, load, and operation time; The defect analysis unit uses a graph neural network to perform correlation analysis on the fused data, determines whether the defect is caused by normal aging or equipment failure, and generates a defect detection report.

Images